By

Nigussie Dechassa
Gemechu Keneni
Seyoum Bediye
Nigussie Alemayehu

የኢትዮጵያ የግብርና ምርምር ኢንስቲትዩት

[i]
Ethiopian Institute of Agricultural Research

ENARESS (Enabling Next-generation Researchers through Engaging Seasoned Scientists) Project

©EIAR, 2022
Website: [Link]
Tel: +251-11-6462633
[Link]: 2003
Addis Ababa, Ethiopia

ISBN: 978-99944-66-71-9

Correct Citation
Nigussie Dechassa, Gemechu Keneni, Seyoum Bediye, and
Nigussie Alemayehu. 2022. Research Methods A
Handbook for Agricultural Researchers. Ethiopian Institute
of Agricultural Research, Addis Abeba, Ethiopia.

Design and Page layout: Elizabeth Baslyos

[ii]
 Contents
I. FOREWORD i
II. PREFACE iv
SECTION I. GENERAL 7
1. Introduction 7
1.1. What is research 7
1.2. Characteristics of good research 9
1.3. Characteristics of good researchers 11
1.4. Research and scientific method 15
1.5. Research methodology versus research methods 17
SECTION II. BACKGROUND TO RESEAECH METHODS 21
2. The Research Process 21
3. Types of Research 31
3.1. Basic research 32
3.2. Applied research 33
4. Variables in Research 41
4.1. Types of variable 43
4.2. Criteria for selecting variables 55
5. Major Research Methods 56
5.3. Ethnographic method 57
5.4. Case study method 58
5.5. Correlational research method 60
5.6. Modelling method 60
5.7. Action research method 62
5.8. Participatory learning and action research method 63
5.9. Some Special Research Methods 67
6. Principles of Experimentation 87
6.1. Validity of experiments 90
6.2. Applying principles of experimentation 96
7. Classification of Experiments 105
SECTION III. PLANNING RESEARCH 109
8. Research Designs 109
9. Some Practical Considerations in Experimental Research 142
9.1. Field experiments 142
9.2. Quasi experiments 156
9.3. Natural experiments 156
9.4. Ex Post Facto experiments 157
9.5. Causal-comparative studies 158
9.6. Case-control studies 161
10. Requirements of a Good Experiment 164
10.1. Absence of systematic error 164
10.2. Precision 165
10.3. Range of validity 165
[iii]
 10.4. Simplicity 166
10.5. Proper statistical analysis 166
SECTION IV. DATA COLLECTION, MANAGEMENT, AND ANALYSIS 167
11. Collecting Data 167
11.1. Types of data collected in research 179
11.2. Data measurement 194
12. Organizing Data 209
13. Analyzing Data 218
13.1. Univariate Analysis 220
13.2. Bivariate analysis 239
13.3. Multivariate analysis 246
13.4. The concept of parametric and non-parametric statistics 247
14. Illustrations of Analysis of Variance 249
15. Interpreting analytical results 307
16. Possible Errors in Research 307
16.1. Random and systematic errors 309
16.2. Type I Error and Type II Error 312
17. Popular Statistical Packages 314
18. Field Layout for Agricultural Experiments (Case Examples) 318
SECTION V. MANAGING AGRICULTURAL RESEARCH IN ETHIOPIA 326
19. Managing Experiments 326
19.1. Working with Farmers 328
19.2. Major Steps in Undertaking Agricultural Research in Ethiopia 336
20. Proposal Review Process 372
SECTION VI. REPORTING RESEARCH RESULTS 374
21. Writing Research Reports 374
SECTION VII. RESEARCH ETHICS 394
22. Research Ethics 394
22.1. General ethical principles that guide research 395
22.2. Research misconducts 401
23. Bibliography 416

[iv]
 I. FOREWORD

To bridge the widening gap between the dramatically increasing

demand for and the diminishing supply of food, agro-industrial
raw materials and export commodities, Ethiopia needs to increase
agricultural productivity as an urgent priority. To increase the
productivity of the agricultural sector, it is necessary to conduct
agricultural research and avail improved technologies and
information, multiply initial technologies, transfer improved
technologies and knowledge, and build the capacity of key
stakeholders.
Despite the existence of a huge demand for improved
technologies and information, technology transfer, and capacity
development in the sector, the achievements made hitherto in this
regard are not adequate. Moreover, climate change-induced
periodic droughts, moisture stress, floods, land degradation,
outbreaks of pests and diseases, invasion of exotic weeds, and
other vagaries of nature have compounded the already existing
challenges faced by smallholder framers who produce the bulk of
food.
There are thousands of researchers working under the agricultural
research system in Ethiopia. However, most of them have
knowledge and skill gaps. This has slowed down generation and
application of high-quality technologies and information. The
problem of skill gap is exacerbated by the fact that most
researchers may not have been sufficiently exposed to appropriate
tailor-made research skill enhancing training during their time of
formal education. As a result, they often gradually gain new

[i]
knowledge and build their research skills through hands-on
practical activities while working as researchers through trial and
error. Hence, for an inexperienced and novice researcher, it would
take quite a long time before honing their skills and gaining the
deepening their knowledge needed to conduct sound research and
write quality scientific papers, which for many, would be
achievable perhaps only after they are past their prime work life.
Consequently, to a larger extent, at the beginning of their careers,
junior researchers often take a leap in the dark without being well
guided or versed in research knowledge and skills. As a result,
most of the research they conduct is likely to be methodologically
flawed, resulting in compromised qualities of the technologies
developed and information/knowledge generated. In this
connection, junior researchers also face acute challenges in
identifying appropriate research problems, developing relevant
proposals, conceptualizing the process of data collection,
analyzing data, interpreting research results, and writing research
papers articulated for various target audiences.
To tackle these problems, it is, thus, essential to equip junior
researchers coming into the research system with refreshing
necessary tools and means by which they could build their
knowledge and research skills. Therefore, apart from providing
training and mentoring tailored to their needs, availing source
documents on research methods like this Handbook, is
instrumental in building their capacity.
At this juncture, while taking, the opportunity to thank the
Agricultural Transformation Institute (ATI) for financially
supporting the Ethiopian Institute of Agricultural Research
(EIAR) to produce this important document, I extend my sincere
[ii]
appreciations to the ENARESS Team for accomplishing the task. I
strongly believe that not only junior agricultural researchers but
also other researchers in Ethiopia and elsewhere with similar lines
of interest will find this Handbook an invaluable resource for
expanding and deepening their knowledge and skills on planning
and conducting sound agricultural research and for writing quality
research papers.

Feto Esimo (PhD)

Director General, EIAR

[iii]
 II. PREFACE

This Handbook has been prepared to provide a reading material

primarily intended to benefit newly or recently engaged
agricultural researchers as individuals or as members of research
teams. This, however, does not rule out the fact that the handbook
can also serve the interests, at least by way of refreshing some of
the fading memories, of even more experienced researchers in
augmenting their general comprehension and technical dexterity
in the field of agricultural research. The Handbook has been
compiled by gleaning information from relevant books, journals,
manuals, and other literature; but most importantly by articulating
mastery, evoked from own experiences and expertise, of the
authors predicated upon their decades-long engagement as
researchers, trainers, and mentors.
The Handbook is divided into seven sections and 23 chapters.
Section I introduces the concept of research and characteristics of
good research and researchers. Section II walks readers through
theoretical background information, features of research, and the
research process. Section III deals with planning research and
some practical considerations in conducting experimental
research. Section IV deals with the topics of data collection,
management, and analysis. Section V provides information on
how to manage agricultural research. Section VI gives general
information on how to write research reports. Section VII deals
with major research misconducts and how to avoid them. The
Handbook culminates with Section 29, which presents the
Bibliography of the material used to prepare the manuscript.

[iv]
The Handbook is primarily intended for young graduates, who are
laden with the energy and enthusiasm, but lack the necessary
experience and skills to make agricultural research their
professional career. It is also intended to serve others engaged in
the field agriculture such as technical workers, trainers, teachers,
and students who require organized information on how to plan
and conduct research; collect, analyze, and interpret data; and
how to write quality research reports targeting various audiences.
However, this Handbook does not pretend to be an exhaustive
source of knowledge on experimental designs, methods of data
analyses, and writing scientific papers. It is rather framed to be a
repository of research methods whereby junior and novice
researchers would be inspired to further equip themselves with
the necessary knowledge and skills to embark on conducting
quality research by referring to relevant standard textbooks for
deeper understanding of the subject matter.
Nonetheless, it is hoped that the Handbook will also serve
agricultural researchers and students requiring a quick reference
to turn to when faced with the challenges and gaps on the basics
of conducting sound researches, collecting and analyzing data,
and generating information thereof as well as writing research
reports presentable to various target audiences at various levels of
knowledge.
I, on behalf of the authors, would like to put on record the
relentless support provided by the management of the Ethiopian
Institute of Agricultural Research (EIAR) to run the ENARESS
(Enabling Next-generation Agricultural Researchers through
Engaging Seasoned Scientists- Phase II) Project in general and to
produce the Handbook in particular. I also extend, with the same

[v]
breadth and depth, my words of thanks to the Agricultural
Transformation Institute (ATI) for financing this Capacity
Building Project as well as technically fortifying all the way
through the course of its implementation. Dr Getachew Belay
generously lent us a helping hand with reviewing the draft
Handbook critically and his input was instrumental in giving the
document its present shape. ENARESS, thus, extends its sincere
appreciation to him.

Nigussie Alemayehu (PhD)

Project Coordinator, ENARESS Project, EIAR

[vi]
 SECTION I. GENERAL

1. Introduction

1.1. What is research

The term research is often used to refer to activities such as
defining, redefining, and solving problems; observing and
interpreting facts; formulating hypotheses and testing them
through experiments; revising existing theories and laws; and
practically applying information already generated. Specifically,
research is used as a tool to understand and solve problems for the
benefit of people. Discovery of fertilizers to improve soil fertility
and increase crop yields is a good example of output of research.
Methodologies and approaches used to do research in natural
sciences and social sciences are often different. The
methodologies and approaches of research in the two disciplines
differ based on specific characteristics of each subject. In natural
sciences, for instance, measuring objectively is more easily done
(e.g. yield performance) than in social sciences (e.g. measuring
attitudes of farmers towards an agricultural technology). Social
sciences deal with phenomena arising from the activity of
humans. Accordingly, approaches of doing research in this
discipline is mostly different from those used to do research in
natural sciences. The context of social science research also keeps
changing as they deal with complex human behaviors. In fact, this
problem is apparent also in most applied sciences including
agriculture.

[7]
Research may be stimulated by several purposes. The first
stimulating factor to do research could be inquisitiveness for
discovery or interest to discover something new. In fact, most of
the advances in science are attributed to the inquisitiveness to find
something new. The second incentive could be the desire to gain
reputation in science, which stimulates researchers to define and
redefine theories, principles, and other concepts to enhance their
status and gain acceptance and prestige. The third motivating
factor for doing research could be the desire to gain in-depth
knowledge, i.e. the interest to deepen and broaden one’s
knowledge. A person who undertakes research on a subject gets
an opportunity to study that subject in depth. The fourth
motivating factor could be related to the desire to provide services
to society by addressing or solving problems afflicting the
society. In agricultural research, the motives include tackling food
insecurity and climate change. Likewise, need for publications
towards a successful career building in an academic or research
institution and obtaining a higher degree by students may also
require publishing research papers which stimulates them to do
research.

Whatever the motives of the researchers, society enjoys one or

more of the following benefits. The expanded frontiers of
knowledge in various scientific disciplines provides a foundation
for new inventions and discoveries, which may result in
breakthroughs for improving human well-being. Research results
also generate remedy to production problems, increase the
efficiency of work, and reduce the cost of production, thereby
ensuring economic growth and making life easier.

[8]
1.2. Characteristics of good research

Good research is characterized by certain specific attributes. First,

good research is an activity based on the work of others. This
means it builds on previous knowledge and research results of
others. This does not mean that one should copy the work of
others. However, it means that a researcher should meticulously
review the work that others have already done to identify
remaining gaps and focus his/her research on finding solutions to
new problems or bridging remaining gaps.
Second, good research recognizes and avoids prejudice, which
can occur during the planning of experiment, its implementation,
data collection, interpretation, and reporting. Sometimes,
nationality, gender, ethnicity, age, and political views of
researchers may influence them to go for prejudiced evidences or
interpretations. Possible sources of prejudice and how it is likely
to influence evidences and interpretations must be understood and
precautions must be taken against them.

Third, repeatability is an important characteristic of good

research. Repeating good research results in the same finding.
However, repeating bad research does not result in the same
finding. Repeatability/reproducibility is one measure of precision,
and refers to the level of consistency between the results
of similar experiments conducted by different individuals, at
different locations, and with different instruments or it is the
measure of ability to replicate the findings of others. The degree
of precision, i.e. the probability at which the experiment measures
differences between treatments should be high as desired by the
experimenter. The degree of precision is improved through the
[9]
application of proper experimental design with proper
randomization, proper blocking and sufficient replication.
Experiments with sufficient appropriate randomization, blocking,
and replication in time and space, for instance, provide better
validity of the conclusion that could be drawn. If the study is
conducted to obtain information on a given population by
studying only a part of the population, the sample must be
representative to come up with accurate results.

Fourth, research must be generalizable to other settings as it is

universal in nature. The findings and results obtained in one
setting must be true for other settings also.

Fifth, good scientific research is systematic and structured with

specified steps in a sequence. Although it does not rule out
creative thinking, good research rejects mere speculation and
intuition in arriving at conclusions. Good research applies proper
experimental design. The experimental design should be as
simple as possible such that its use is consistent with the
objectives of the experiment. This means researchers should not
rush for the application of more difficult designs unless otherwise
they are compelled to do so. The experimental design to be used
for any experiment should be characterized by absence of
systematic error. This means that the experimental units receiving
one treatment should not differ in any other way from those
receiving another treatment. This is done to ensure an unbiased
estimate of the variance for each treatment. The experimental
design used should also enable calculation of the degree of
uncertainty, i.e. it should be possible to calculate the possibility of
obtaining the observed results by chance alone. The analysis of
data must be adequate and the method of analysis should be
[10]
appropriate with conclusions confined to the realities justified by
the data.

Finally, good research is an enquiry into a new phenomenon and

it generates new questions, which must also be answered.
Research is an apolitical activity: Research is not authoritarian
and should not have any ‘politics’ behind it. It should be
undertaken for the betterment of society and not for selfish or
destructive purposes.

1.3. Characteristics of good researchers

Researchers may have certain intrinsic qualities. However, they

should strive to attain other qualities, which they may not possess.
The following are some of the qualities that all the researchers
must try to possess. Among these qualities, the scientific attitude
and research attitude are the major ones. The first one is
scientific attitude. Scientific attitude is the attitude of a researcher
to promote the use of scientific approaches and rationales to react
to any object. This means researchers should not yield to
superstitions, unfounded information, word of mouth, or
pseudoscience. The second important quality of a researcher is
research aptitude. This quality is related to a researcher’s ability
to have an inquisitive mind to observe things and events occurring
around them. This means a researcher should have a sincere
interest in the subject and an outlook to disentangle and reveal the
unknown. In other words, he or she should have a ‘research mind-
set’. Of course, not all researchers may necessarily have a
research aptitude. The third important quality of a researcher is
perseverance. Perseverance means the traits such as persistence,
patience, tenacity, thoroughness, and determination to achieve
[11]
something. To become a successful researcher, one must be
persevering and do publishing scientific or research papers in
peer-reviewed journals. What is more, a researcher should be
patient enough to wait for results. This is because certain
investigations are time-consuming, involve drudgery, and require
the help of many other persons. The fourth important quality a
researcher should possess is self-motivation. Self-motivation is
the internal state that helps us initiate, continue, or terminate an
action. It is an important quality for researchers. Self-motivated
researchers produce more results than those that are not self-
motivated. The fifth important quality of a researcher is courage
to ask questions. Researchers should ask questions about things,
which they cannot agree or do not understand. They should ask
‘what’, ‘how’, ‘when’, and ‘why’, and find answers by critically
observing, experimenting, consulting, discussing, and reasoning.
The sixth important quality of a researcher is questioning and
receptivity. This means that rresearchers or scientists should not
accept data and interpretations without questioning. A researcher
should assess the data and/or interpretations with an open mind
and keep the required balance between skepticism and receptivity.
On the other hand, a researcher should always handle and tackle
new concepts, ideas, views rather than confronting them with a
negative response or censure. The other quality necessary for a
researcher is objectivity. That means any conclusion reached by a
researcher or a scientist should be objective‚ and be based on
findings that have been revealed through actual data. The
researcher or scientist should remain unbiased to the outcome of
the research, and ensure that any biased or emotional leanings of a
researcher should not in any way influence the conclusion.
Industriousness is the other important quality that is necessary for
[12]
a researcher. This means that a researcher who is indolent and
slovenly, and long for comfort and luxury cannot succeed in
achieving his/her objectives. People who can become good
researchers are the ones that seek or have the propensity and
curiosity to seek new knowledge and make new discoveries and
inventions through sustained and dedicated efforts. Honesty and
truthfulness is the other important quality that a researcher should
possess. Thus, a researcher should be honest in his/her approach.
He or she should follow the established ethics‚ and not commit
any fraud or misconduct. A researcher should record observations
and experimental results honestly, and avoid plagiarism. Open-
mindedness is also an important quality of a researcher. This
means that a researcher should remain neutral. He or she should
not have any pre-conceived notions or bias towards a problem to
be researched or investigated. This implies that a researcher
should be led only by evidence, reason, and logic, but not by
assumptions or speculations. Intelligence is another important
quality necessarily for a researcher. This means that researchers
without the needed qualities of a scientist rarely achieve any
success. However, it is not necessary that scientists or researchers
must be very intelligent to succeed. A scientist or researcher with
just above-average intelligence can succeed by hard work and
perform better than those with higher intelligence perform.
Knowledge is another important quality a researcher should
possess. Thus, a researcher should be expert in his/her fields. He
or she should be the type that always seeks to widen and deepen
his/her knowledge and hone his/her and skills. Thus, when
seeking to do research, a researcher should be eager to glean
information from the relevant literature connected with the
researchable topic, read them critically, analyze them, learn the
[13]
facts, and identify researchable gaps. Imagination is also an
important quality that a researcher should possess. It is the basis
of getting adequate insights into problems and solving the
problems through research. Most scientists are imaginative.
However, imagination coupled with both determination and
vision can yield very good results. Self-confidence is another
quality that a researcher should have. This quality of a researcher
instils inspiration to tackle problems and encourage optimism.
Truthfulness is also a necessary quality for a researcher to
possess. This quality implies that a researcher should repeat
experiments carefully and systematically, if required, and never
manipulate results under any circumstances. Team spirit is
another important quality that a researcher should possess. This is
because individual researchers or scientists can solve not all
problems because most researchable problems are vast with wide
biophysical and socio-economic perspectives. Thus, researchable
problems need multidisciplinary approaches that require experts
from a number of fields in the area. Under such conditions,
researchers should be eager to work in a team with
complementarity.

The researcher who conducts any research should also be reliably

capable, dedicated, disciplined and ethical with good personal
integrity. Therefore, a researcher should practice all relevant
research ethical guidelines while conducting, reporting, and
publishing research results, i.e. he or she must report with
complete frankness and absence of flaws.

[14]
1.4. Research and scientific method

The terms “research” and “scientific method”, are closely related.

Research is an inquiry into the nature of, the reasons for, and the
consequences of any particular set of circumstances. Research
suggests the researcher is interested not only in getting results but
also in the repeatability of the results. On the other hand, the logic
common to all research methods and techniques is usually given
the name of scientific method. Scientific method is the pursuit of
truth as determined by logical considerations. The ideal of science
is to attain a systematic interrelation of facts. Scientific method
attempts to achieve this objective through experimentation,
observation, logical arguments from accepted postulates and a
combination of these three in varying proportions.
In scientific method, logic aids in formulating propositions
plainly and accurately so that the possible alternatives also
become clear. Through logic, the consequences of such
alternatives are developed, and when these are compared with
observable phenomena, the researcher or the scientist is enabled
to state which alternative is most in harmony with the observed
facts. All this is done through experimentation and survey studies,
which are the integral parts of scientific method.

Experiments are conducted to test hypotheses and discover new

relationships, if any, among variables. However, the conclusions
drawn on the basis of experimental data are generally criticized
for either faulty assumptions, poorly designed experiments, badly
executed experiments or faulty interpretations. Thus, researchers
must be vigilant while choosing the experimental design to use
and must state only probable inferences. The purpose of survey
[15]
investigations may also be to provide information or data to work
as a basis for researchers to arrive at a conclusion.

Therefore, scientific method is a general term used to express the

different processes that lead to building sciences. The main
advantage of scientific method is that it is unbiased. One does not
have to believe a particular scientist but he or she can conduct the
experiment again and determine whether the results are true or
false. The conclusions will hold true regardless of the researcher’s
state of mind, religious persuasion, state of consciousness, or the
subject of the investigation. However, this is not the case with
various pseudoscientific theories or beliefs.

Scientific method depends on observation, measurement,

prediction, experimentation, and verification. These features of
the scientific method differentiate science from other fields of
knowledge. The scientific method is based on proofs rather than
beliefs or arguments, a character that distinguishes science from
faith or authority.

The scientific method is, thus, based on certain basic postulates.

These include the following: reliance on empirical evidence,
utilization of relevant concepts, commitment to objective
considerations,, presupposition of ethical neutrality, i.e., aiming at
nothing but making only adequate and correct statements about
population objects, resulting in probabilistic predictions, making
its methodology known to all concerned for critical scrutiny or for
use in testing the conclusions through replication, aiming at
formulating most general axioms or what can be termed as
scientific theories.

[16]
The scientific method often consists of the following main steps:
making observations or gathering information; developing a
hypothesis to predict results; designing an experiment; conducting
the experiment and collecting data; evaluation and conclusion;
and accepting, modifying, or rejecting the hypothesis.

1.5. Research methodology versus research

methods

The term “research methods” mean the methods and procedures a

researcher employs to accomplish a research task. The methods
need to include detailed procedures of how to start, implement,
and complete a research project. Thus, in this “Handbook”,
research methods are presented to help researchers in planning
and executing sound research and drawing relevant conclusions
for creating new knowledge and improved agricultural
technologies.
There is a certain distinction between the terms “research
methodology” and “research methods”. Research methodology
can be defined as the systematic study of the research process
starting from the planning process to reporting the results. It
encompasses the approaches we use during planning,
implementation, and reporting stages of a research project.
Certainly, writing research reports and speaking about research
are also part of the wider field of research methodology.

Specifically, research methodology is the science of studying how

research is done scientifically. Thus, under research methodology,
the various steps are studied that should be followed by a
researcher in investigating his/her research problem along with

[17]
the logic behind them. In this connection, it is important to
understand that researchers should know not only how to develop
certain indices or tests, how to calculate the mean, the mode, the
median or the standard deviation or Chi-square, how to apply
particular research techniques, etc., but they should also know
which of these methods or techniques are relevant, which are not
relevant, and what those techniques mean and why.

Researchers also need to know the assumptions behind the

various techniques and the criteria on which they base the use of
certain techniques and procedures applicable to certain problems
but not to others. This signifies that a researcher needs to select or
design his/her methodology that is pertinent to investigate his/her
particular research problem since the applicable methodology
varies from problem to problem.

Thus, research methodology has a wider dimension and research

methods do constitute a part of the research methodology. This
means the scope of research methodology is wider than that of
research methods. Thus, under research methodology, a
researcher considers not only research methods but also the logic
behind his/her choice of the methods in the context of the
research problem to be investigated, and should explain why
he/she opted to use particular methods or technique but not
others.

Therefore, why a research study has to be done, how the research

problem has been identified and defined, what hypothesis has
been formulated, what and why particular methods will be
adopted to do the research, what data will be collected, why a
particular technique of analyzing the data has been selected, and
[18]
similar other questions are usually addressed through research
methodology. In other words, research methodology is about the
research methods and the logic behind selecting a particular
method and how the researcher proceeds and concludes the
research project.

On the other hand, research methods are understood as all those

methods/techniques that are used for conducting research.
Research methods include all techniques the researchers use in
performing research activities. Thus, the term “Research
methods” refers only to a part of the wider field of “Research
methodology”. They are applied during the execution phase of a
research project or experiment. However, research methodology
is required starting from the planning stage itself. For a
researcher, there are many methods and techniques, which can be
selected for completing the research task. The researcher should
know not only the theory and practice of these methods or
techniques but also needs which of these methods are relevant and
how they are selected.

Since the objective of research, particularly that of applied

research, it to seek a solution to a problem, the available data and
the unknown aspects of the problem have to be related to each
other to find a solution to the problem. Based on this stipulation,
research methods can be categorized into the following three
groups:

(a) Methods which are concerned with data collection

These methods are used when already available data are
not sufficient to reach a conclusion.

[19]
 (b) Statistical techniques used for establishing relationships
between the data and the unknowns
(c) Methods which are used to evaluate the accuracy of the
results obtained.

Research methods falling in the above-stated last two groups are

generally taken as the analytical tools of research.

[20]
 SECTION II. BACKGROUND TO
RESEAECH METHODS

2. The Research Process

Research process consists of a series of steps necessary to conduct

research and the desired sequencing of these steps. It consists of a
number of closely related activities. However, such activities
overlap continuously rather than following a strictly prescribed
sequence. Thus, the various steps involved in a research process
are not mutually exclusive. Furthermore, the activities do not
necessarily follow each other in any specific order. Therefore, at
each step in the research process, the researcher has to anticipate
the requirements of the subsequent steps constantly. However, the
following are the major steps in a research process:
(a) Formulating the research problem
(b) Extensive literature survey
(c) Developing the hypothesis
(d) Preparing the research design
(e) Determining sample design
(f) Collecting the data
(g) Executing the project
(h) Analyzing data
(i) Testing hypothesis
(j) Interpreting the data, and
(k) Preparing the report or presenting the results

[21]
2.1. Formulating the research problem

Formulating the research problem consists of identifying the

problem to be addressed, gauging its severity, and knowing what
research was previously done by others to solve it.
A research problem generally refers to some difficulty faced to
which a solution is sought in the context of either a theoretical or
a practical situation. The research problem can also be a
continuance of research started or completed previously.
There are two types of research problems. These are problems
that relate to states of nature and problems that relate to
relationships between variables. At the beginning, a researcher
must identify the problem that he wants to address. He/she may
first describe the problem in a broad term. The problem must be a
researchable one. It should not be a problem that can be solved
with a quick fix, without needing research. For example, if the
problem is shortage of irrigation water in a particular area, there is
no need to do research to solve this problem. This is because that
problem would not lend itself to scientific investigation. The
problem can be fixed quickly if there are enough financial and
material resources. For example, rain harvesting or harnessing
lakes or rivers or digging wells to supply water for irrigation
needs only initial capital investment and running costs for
maintenance. Therefore, the viability of a specific solution has to
be considered before framing a research problem. The
formulation of a general topic into a precise research problem,
thus, constitutes the first step in a scientific enquiry.
Two steps are involved in framing the research problem, namely,
understanding the problem carefully, and rephrasing it into
[22]
meaningful terms for analysis. To understand the problem well, it
is good to discuss on it with colleagues and peers as well as with
others having special expertise and experiences in the subject
matter. Thus, the research problem may be identified by the
researcher, society, other researchers, etc. At research and higher
learning institutions, the researcher can seek support from
experienced senior staff members to further elucidate and narrow
down the problem and rephrase it in operational terms. In
governmental organizations, the problem may be already
earmarked for a solution. Thus, the researcher can discuss on the
problem with the relevant organization and get further
clarifications on what respects are involved in its possible
solutions.
After formulating the problem, the researcher should write down
a brief description of the problem. Thus, the problem should be
defined clearly since this will help in making distinctions between
relevant data and irrelevant ones. At this juncture, it is also
important to validate the objectivity and rationality of the
background facts about the problem. The objective is solving a
researchable problem, which is key for defining the data to be
collected, the characteristics of the relevant data, relations to be
explored, the choice of techniques to use for the investigations,
and the form of the final report.
2.2. Literature survey

The main output of reviewing literature is information as to what

data and other materials exist to assist in postulating the research
problem in a meaningful setting. In this connection, the researcher
should scrutinize all available literature on the topic so that he/she
[23]
will have information as to what research has been done before on
the problem and what results have been found, thereby identifying
gaps that need to be bridged through his/her intended research. If
the researcher finds from the literature that there already exists a
solution to the problem, there is no need to do additional research
and that solution can be adopted. Thus, if a solution is already
available, there is no need to re-invent the wheel.
There are usually two types of literature that a research should
review. These are the conceptual literature and empirical
literature. The conceptual literature is literature concerned with
concepts and theories on the topic whereas empirical literature is
literature concerned with hitherto conducted studies to solve
similar or related problems (usually not earlier than ten years ago
and preferably not older than five years ago).
Thus, for literature review, abstracting and indexing journals and
published or unpublished bibliographies are required. Academic
journals, conference proceedings, reports, books etc., should be
used depending on the nature of the problem. In particular, key
literature related to earlier studies, if any, which are similar to the
study intended, should be reviewed carefully.
2.3. Formulating hypotheses

Formulating working hypotheses is a basic requirement in the

process of most research problems. Therefore, after reviewing
literature, the researcher should state a working hypothesis or
hypotheses. Developing hypotheses is important since they
provide the focal point for the research. Researchers formulate
hypotheses using inductive logic. The most important aspect of a
hypothesis is that it must be ‘falsifiable’. In a hypothesis, the
[24]
researcher expresses his expectations. Thus, he/she makes a
prediction by putting a hypothesis to a test. Suppose a researcher
wants to investigate the effect of protein rich feed on carcass
weight of a bullock, he or she can formulate a hypothesis and
describe it predictively as follow: Protein rich feed increases
carcass weight of a bullock and its meat yield. If this prediction is
proved true, then the researcher should accept the hypothesis.
After the experiment, if the researcher finds that this prediction
does not hold true, the researcher will have to reject the
hypothesis.
A hypothesis should be very specific and restricted to the piece of
research in hand since it has to be tested. A hypothesis guides the
researcher by delimiting the area of research and keeping him/her
on the appropriate track of the research question. It sharpens the
researchers’ thinking and focuses attention on the more important
aspect of the problem. It also indicates the type of data required
and the type of methods of data analysis to be used.
Approaches for formulating research hypotheses:
(a) Discussing with colleagues and experts about the problem
and the prospect of seeking a solution;
(b) Examining data concerning the problem, if available, for
possible trends, peculiarities, and other clues;
(c) Results of review of previous research on similar
problems; and
(d) Exploratory personal investigations, which involves
original preliminary studies, field interviews done to
secure insights into the practical aspects of the problem.
In certain circumstances, some problems may be encountered
where a working hypothesis is not required. This is the case with

[25]
exploratory or formulative research, which does not aim at testing
the hypothesis.

2.4. Preparing research design

After formulating the research problem clearly, the researcher

will be required to prepare a research design, i.e., he/she will have
to state the conceptual structure within which research can be
conducted. The preparation of such a design facilitates research to
be as efficient as possible, yielding maximal information. The
purpose of planning a research design is to provide for collection
of relevant evidence with minimal effort, time, and money.
However, how much of all these can be achieved depends mainly
on the research purpose. Research purposes may be grouped into
four categories, namely, (i) Exploration, (ii) Description, (iii)
Diagnosis, and (iv) Experimentation.
Exploration is defined as the initial research into a hypothetical or
theoretical idea. It is where a researcher has an idea or has
observed something, and seeks to understand more about it to lay
the initial groundwork for future research. Description refers to
seeking more information about a newly discovered initial
groundwork or phenomenon by collecting as much information as
possible instead of making guesses or elaborate models to predict
the future. Diagnosis refers to studies that aim to quantify a
problem or a situation or a phenomenon to determine its existing
level. Experimentation is conducting experiments by
manipulating variables.

[26]
If the purpose of the research study is exploration, it is
appropriate to formulate a flexible research design, which
provides an opportunity for considering many different aspects of
a problem. However, if the purpose happens to be an accurate
description of a situation or of an association between variables, a
suitable design will be one that minimizes bias and maximizes the
reliability of the collected data.
There are several research designs, which include, experimental
and non-experimental hypothesis testing. Experimental designs
can be either informal designs (such as before-and-after without
control, after-only with control, before-and-after with control) or
formal designs (such as completely randomized design,
randomized block design, Latin square design, simple and
complex factorial designs, etc.), out of which the researcher must
select one for his own project. Planning the research design that is
appropriate for a particular research problem requires
consideration of the following points: (i) the means of obtaining
the information; (ii) availability and skills of the researcher and
his staff (if any); (iii) explanation of the way in which selected
means of obtaining information will be organized and the
reasoning leading to the selection; (iv) the time available for
research; and (v) the cost factor relating to research, i.e., the
finance available for the purpose.

2.5. Determining sampling design

All the items in any field of inquiry constitute a ‘population’. A

complete enumeration of all the items in the ‘population’ is
known as a census inquiry. If it was possible to cover all terms of

[27]
a population in an inquiry, no element of chance would exist and
the highest accuracy would be obtained. However, census inquiry
is not possible in practice under many circumstances. Besides, an
inquiry of all items of a population will involve a lot of time,
money, and energy. Hence, only a few items are selected for a
study. The items so selected constitute what is technically called a
sample. The researcher must select a sampling design. A
sampling design is a definite plan determined before any data are
actually collected for obtaining a sample from a given population.
Samples can be either probability samples or non-probability
samples. With probability samples, each element has a known
probability of being selected in the sample but the non-probability
samples do not allow this possibility. Probability samples are
those based on simple random sampling, systematic sampling,
stratified sampling, cluster/area sampling whereas non-probability
samples are those based on convenience sampling, judgement
sampling and quota sampling techniques.
2.6. Conducting research

The research should be conducted appropriately and timely to

obtain adequate and dependable data. A hypothesis should be
tested or the research questions should be answered by
conducting proper studies and evaluating the results.

2.7. Collecting data

In research, it is necessary to collect appropriate data. Several

ways of collecting data exist. These ways differ widely in terms
of cost, time, and the requirement for other resources. Primary
[28]
data are collected either through experiment or through survey.
When the researcher conducts an experiment, he observes data
that require quantitative measurements through which to test the
hypothesis. However, in the case of a survey, data can be
collected by observation, through personal interviews, telephone
interviews, by mailing of questionnaires, through schedules, etc.

2.8. Analyzing data

Data analysis requires a number of closely related activities such

as establishing categories of data through coding, tabulating them,
and drawing statistical inferences. Data coding is often done at
this stage through which the categories of data are transformed
into symbols that may be tabulated and counted. Editing improves
the quality of the data for coding. After coding, the data are ready
for tabulation. Tabulating data is the method through which the
data are organized in tables. Computers are often used for
organizing data because they not only save time but also enable
researchers to handle bulky data on a number of variables.
Analyzing the data is done through computing means, standard
deviations, percentages, coefficients, etc., by applying various
well-defined statistical formulae often using statistical software.
In the process of analysis, relationships or differences supporting
or conflicting with original or new hypotheses should be
subjected to tests of significance to determine with what validity
data can be said to indicate any conclusions.

[29]
2.9. Testing hypothesis

Accepting, modifying or rejecting a hypothesis come at the end of

a research process. Hypothesis testing follows data analysis. In
this process, the researcher askes the following questions to test
the hypotheses, if any, he/she had formulated earlier. Is the
hypothesis verified or falsified by the results of the research?
Various tests, such as Chi-square test, t-test, F-test, etc., have
been developed by statisticians for the purpose. The hypotheses
may be tested using one or more of such tests, depending upon
the nature and object of research inquiry. A hypothesis is
accepted if the data generated through the research support it; and
rejected if the data generated through the research do not support
it.
On the other hand, sometimes, the researcher may not have
formulated a hypothesis in the research process, perhaps, because
he/she dealt with descriptive or exploratory research that does not
require any hypothesis. In such a case, the conclusion of the
current research drawn based on the data generated now can be
stated as hypotheses to be tested in subsequent research to be
conducted in the future.

2.10. Interpreting data

This is the step in which the researcher interprets the findings of

his/her research and arrives at some form of generalization or
building up of a theory. If the researcher had no formulated a
hypothesis in the research process, he should explain his/her

[30]
findings on the basis of some theory. This is known as
interpretation. Interpretations may lead to new questions and
further research.

2.11. Preparing research reports

If the research problem is successfully solved or if new findings

are obtained, the researcher should communicate the results of
his/her research.

For this purpose, the researcher should write up a report and

communicate his/her findings to the audiences that have interests
or stakes in the work done. These mainly include the scientific
community, the wider public, or any other stakeholders. The
reports are often prepared in the forms of scientific papers,
conference proceedings, posters, technical reports, policy briefs,
etc.
In general, reports should be written carefully with inclusion of
all required contents, including title, abstract, introduction, a clear
statement of the problem investigated, objective of the research,
methodology adopted, results and discussions, and the conclusion.

3. Types of Research
Research is often categorized as basic research, applied research,
descriptive research, analytical research, quantitative or
qualitative research, conceptual or empirical research,
experimental or non-experimental research, retrospective or
prospective research, and cross-sectional or longitudinal research.
[31]
First, based on the application of the results, research is broadly
divided into two, i.e., basic research and applied research.
3.1. Basic research

Basic research, also called fundamental research or pure research,

is the type of research that is conducted to generate new
knowledge to enhance the science in question but not for
immediate practical application of the results. Basic research is
the basis of most of the scientific advances made so far.
Basic research is sub-divided into two categories, namely, pure
basic research and strategic basic research.
Pure basic research
Pure basic research is the type of basic research that is conducted
as experimental or theoretical research mainly to generate new
knowledge without a specific application. Pure basic research is
conducted regardless of long-term scientific, economic or social
benefits but just for the enhancement of knowledge in the field of
study concerned.

Strategic basic research

Strategic basic research is research that is conducted as
experimental or theoretical research mainly meant for generating
new knowledge with no specific application in mind. However, a
main difference of strategic basic research from pure basic
research is that the former is aimed at particular broad areas in the
expectation of valuable findings or resolving certain problems.

[32]
3.2. Applied research

Applied research is the type of research done to generate new

knowledge and innovation for immediate application and
practical benefits. In other words, it is original research conducted
chiefly to obtain knowledge with a specific useful application.
Specifically, it is concerned with the application of a theory to
find a solution to a problem or problems. Most of the experiments
conducted in agriculture are applied in nature, which are aimed at
solving specific problems or improving existing practices. There
are two types of applied basic research:
Evaluation applied research
Evaluation applied research is the type of research in which data
are collected, evaluated, and decisions are made based on the
results found. This type of applied research consists of organized
gathering of data on the variables of interest to make a decision.
For instance, (missing plot) fertilizer trials, herbicide trials,
insecticide trials, and fungicide trials, comparing performances of
varieties, etc. are types of evaluation applied research.
Developmental applied research
Developmental applied research, sometimes referred to as
research and development (R & D), is the type of research done
for developing effective products for use. The main objective of
developmental applied research is extension of the results of
applied research towards finished products or processes. In
agriculture, developmental agriculture research involves the
process of transferring agricultural knowledge and innovations,
new crop varieties, animal breeds, etc. to the farming
communities. In social sciences, developmental research involves

[33]
the process of transferring knowledge gained through research
into operational programmes such as farmer training materials,
teacher-training materials, and student learning materials, etc.
Descriptive versus analytical research

Descriptive or exploratory research

Descriptive or exploratory research is the type of research done to
determine, describe, or identify a situation, a matter, an object, or
a phenomenon, as it exists now. Descriptive or exploratory
research is done in areas that have attracted little attention or it is
for checking the possibility of research in the particular domain or
area. Descriptive or exploratory research is done when little is
known about a given phenomenon, an object, or a situation. The
objective of descriptive or exploratory research is developing
hypotheses rather than testing them.

In descriptive or exploratory research, the researcher observes,

describes, and documents aspects of the phenomenon or situation
or object under study through observation or survey. In such
research, variables are not manipulated or cause and effect
relationships cannot be studied or looked for. The researcher
describes the phenomenon or object to find out the frequency of
its occurrence, and to categorize the information. The results
provide information base for formulating hypotheses that may
lead to detailed research, mainly experimental research. The
results of this type of research are useful for evaluating a situation
or a phenomenon and using it as decision-making tool by
managers and policymakers.

[34]
Descriptive or exploratory research can be done in both natural
sciences and social sciences. In natural sciences including
agriculture, this type of research is used for describing nature
such as anatomy of an organism by observation through the naked
eye, or through microscope. In biology, systematic or taxonomic
studies are also descriptive studies that involve naming and
classifying organisms.

In social sciences, the term Ex post facto research is often used

for descriptive research studies. The major feature of this method
is that the researcher has no control over the variables; he can
only report what has happened or what is happening. Most ex post
facto research activities are used for descriptive studies in which
the researcher seeks to measure such items as, for example,
frequency of shopping, preferences of farmers for crop varieties,
or similar data. Ex post facto studies also encompass research
activities conducted to investigate causes although the researchers
have no control over the variables.

Descriptive research encompasses all types of fact-finding

enquiries. The methods of research used in descriptive research
are survey methods of all kinds, including comparative and
correlational methods.

Analytical research
Analytical research is the type of research done to determine why
the state of affairs is the way it is or how this may have happened.
Analytical research is often used to designate associations
between variables and analyse them for possible cause and effect
relationships. In analytical research, a researcher uses already

[35]
available or generated data or information and analyses it
critically to address the problem or to provide answers.

Thus, in analytical research, unlike in descriptive or exploratory

research, the researcher has to use facts or information already
available or generated and analyse them critically to address the
problem or to provide answers about the phenomenon under
investigation. Experimental studies, quasi-experimental studies,
and correlational studies are types of analytical study.

Quantitative versus qualitative research

The distinction between quantitative research and qualitative
research is primarily based on design of the research used. Thus,
the classification of the two types of research mainly hinges on
the nature of data collected.
Quantitative research
Quantitative research approach is characterized by much greater
control over the research environment and in this case some
variables are manipulated to observe their effect on other
variables.

In quantitative research, the knowledge generated is based on

gathering numerical data and analysing them. Usually,
quantitative research is deductive in nature such that the
researcher begins with an existing theory, where ideas have been
already reduced into variables, and then collects evidences to
verify if the theory is supported by his/her findings. Thus, a
quantitative research approach involves generation of data in
quantitative terms which can be subjected to rigorous quantitative
analysis. The quantitative research approach is further sub-
divided into inferential, experimental, and simulation approaches
[36]
to research. The objective of inferential approach to research is to
form a database from which to infer characteristics or
relationships of the population. This often means survey research
where samples from the population are investigated to determine
the characteristics, and it is then inferred whether the population
has the same characteristics.

Simulation approach involves the construction of an artificial

environment within which relevant information and data can be
generated. This permits an observation of the dynamic behavior
of a system (or its sub-system) under controlled conditions. The
term ‘simulation’ in the context of business and social sciences
applications refers to “the operation of a numerical model that
represents the structure of a dynamic process. Given the values of
initial conditions, parameters and exogenous variables, a
simulation is run to represent the behavior of the process over
time.” Simulation approach can also be useful in building models
for understanding future conditions.
Qualitative research
Qualitative approach to research is concerned with subjective
assessment of attitudes, opinions, and behavior. Research in such
a situation is a function of the researcher’s insights and
impressions.

Under this approach, research consists of collecting data in a

narrative form rather in a numerical form to understand a
particular situation or phenomenon or process. This type of
research depends mainly on qualitative quantifications rather than
numerical quantification. Qualitative research is exploratory and
inductive in nature, as it begins with the postulation that truth as

[37]
subjective rather than being objective. Qualitative research or its
design is customarily used under any condition that little is known
about a particular phenomenon or concept. After the concepts,
ideas, and themes are identified and known about a phenomenon,
through qualitative research, it is usual to investigate the
phenomenon further using a quantitative approach.

Mostly, the techniques of interviews and focus group discussions

are used in collecting data under qualitative research approach.
The results are generated either in non-quantitative form or in the
form which are not subjected to rigorous quantitative analysis. In
this approach, the data are mostly analysed through coding and
describing them verbally as a narration.
Conceptual research versus empirical research

Conceptual research
Conceptual research is often applied to certain abstract ideas,
concepts, or theories. It is a type of research in which the
investigator usually divides a theory or a concept into several
components for understanding it better. Conceptual analysis of
the information is the usual method but other methods of analysis
could be combined with this method of analysis for better results.
Conceptual research is most frequently used in social sciences
(e.g. philosophy). Philosophers and thinkers often use it to
advance new concepts or theories to reinterpret current ones.
Empirical research
Under empirical research, data are collected through observation
and experiments. The researcher formulates a hypothesis based on
the facts he or she observes, which can be falsified or verified
through research by observation (survey) or by conducting

[38]
experiments. Under this type of research, the researcher generates
data to prove or disprove the hypothesis, using a suitable design
and by manipulating the materials in question. Empirical research
is often applied when it is required to know if certain variables
affect some other variables. Data generated through empirical
research are considered the most powerful basis for accepting or
rejecting a hypothesis. Empirical studies are the most common
methods of research in scientific studies because observation and
experimentation provide direct evidence for accepting or rejecting
hypotheses.
Experimental research vs. non-experimental research

Experimental research
Quantitative research designs, which are used for measuring the
relations between independent variable and dependent variables,
could be either experimental or non-experimental. An
experimental research is the type of research in which the
independent variable is directly manipulated by the researcher by
keeping other variables constant as much as it could be. In this
type of research, a cause and effect relationship is established by
excluding all other factors that might be responsible for a
particular outcome.

The cause and effect relationship is investigated in two ways.

First, the applied treatment is supposed to be the cause of change
or difference. Therefore, it is important to give the treatment a
sufficient time to cause the change. Secondly, after observing the
situation or the phenomenon, the researcher seeks to understand
the cause of change. In this situation, the researcher begins from
the effects to outcomes and determine causation. If a quantitative

[39]
study is conducted beginning from the cause and proceeds to the
effects, it is called experimental research. In this case, the
independent variable can be observed, introduced, controlled, or
manipulated by the investigator.
Non-experimental research
Non-experimental research is the type of experiment used to
describe, differentiate, or examine associations between or among
variables, groups, or situations. Thus, if a quantitative study
begins from the effects to trace the cause, it is non-experimental
research. In this case, unlike in experimental research, it is not
possible to observe, introduce, control or manipulate the
independent variable since the postulated cause has already
occurred. In this type of research, manipulating variables,
assigning treatments randomly, and comparing groups are
normally absent. Thus, with a hindsight the researcher relates the
outcomes to the causes.

Non-experimental designs are often conducted because of certain

conditions. These include the following:

In a situation where variables cannot be manipulated, for instance,

experiments involving human subjects need care, and there are
limitations to conducting experiments involving humans. This is
attributed to ethical obligations or to the fact that certain variables
should not be manipulated. For example, it is unethical to conduct
research on the effect of smoking on the health of 12-year-old
boys in a town or city or village.

In some other instances, independent variables have already

occurred and, therefore, manipulation is impossible. In this case,
on the basis of timing of data collection, non-experimental
[40]
designs can be categorized into cross-sectional or longitudinal. In
cross-sectional research, data on the recognized variables are
collected at one point in time, and the relationships between them
are decided. However, in a longitudinal study, data are collected
at different points over time.
Prospective versus retrospective research
A non-experimental study could be prospective or retrospective.
If data are collected at the present time and the results are traced
to the past, the research is called ‘retrospective’. However, if the
researcher starts collecting data now to generate results in the
future, the research is ‘prospective’ research. To find the results
of a prospective study, which is slow, a researcher has to
normally collect enough data to analyse the results and draw
conclusions.

However, there are certain studies that have both the prospective
and retrospective components. In this case, past and future trends
are studied together.
Semi-experiment or quasi-experiment
A semi-experiment or quasi-experiment is the type of experiment
in which part of the research is experimental but the other part of
it is non-experimental.

4. Variables in Research

A variable is an entity which varies over different situations (e.g.,

time, individuals, places, etc.). In nature, there are entities that
vary over different situations and there are also entities that
remain fixed.

[41]
In general, variables that are studied during a research process can
be of two different types. These are measurable and non-
measurable. Height, weight, length, income, number of family
members, number of pods per plant, number of grains per panicle,
yields of crops per unit area of land, number of insects per plant,
area under disease, milk yield per cow per day, lactation milk
yield, age at first calving, calving interval, prices of different
commodities at different markets at a particular point of time,
etc., are a few examples of measurable variables. The above
measurable characteristics/variables are known as quantitative
characters/variables.

There are certain characteristics which cannot be measured

directly but can be grouped, ranked, and categorized; these are
known as categorical or qualitative characters/variables. These
include race (e.g., Aryans, Dravidians, Caucasoid, Black,
Mongoloid), gender (e.g., male or female), complexion (e.g.,
good, fair, better), colour (e.g., red, green), and taste (e.g., sweet,
sour, bitter), creed (religion, ideology), etc. Such qualitative
variables are non-measurable, and also known as attributes.

Each variable or characteristic is generally associated with the

chance factor. In statistical sciences, a researcher is interested in
studying not only the variable but also their probability
distributions. As such, the variable with chance/probability
factor—the variate—is of much interest rather than simply a
variable. Thus, the variate is defined as variable with chance or
probability factor. For instance, rainfall is a variable which varies
over seasons, but when it is associated with probability
distributions, then it becomes a variate.

[42]
Therefore, a variable is a character, condition, or concept that can
be measured and can take different values. Crop yields, plant
height, leaf area index, plant canopy width, crop quality, carcass
weight, etc. are variables because they can take different values.
In social and health science research, gender, age, agreement,
educational level, attitude, programme, treatment, medicines,
diets, blood pressure, body weight, etc. are some of the variables.
Sex could be considered as a variable while its values, say male
or female for particular person, could be considered as a variate.

When conducting experimental research, the researcher monitors

a character under controlled conditions to see if that character
causes any variations in another character. The researcher handles
a character or characters keeping all other characters fixed as far
as possible. The method of determining a cause and effect
relationship is to separate and exclude all the factors that might
result in an outcome and examine only one factor that the
researcher is interested in determining. The specific factor the
researcher is interested in testing is a variable.

Variables are necessary in both experimental and non-

experimental research. In literature, variables are categorized into
different types in accordance with their nature, purpose, use, etc.
However, types of variable are neither exhaustive nor exclusive
but rather sometimes they are overlapping.

4.1. Types of variable

Independent and dependent variables

Independent and dependent variables are necessary for
investigating cause and effect relationships in research. In

[43]
experimental research, the independent variable, also known as
experimental variable or predictor variable, is the one
manipulated by the investigator. However, the dependent variable
or outcome variable is measured for response due to the
independent variable, which may be a treatment, programme, or
cause.

In an experiment, an independent variable is the apparent cause

and the dependent variable is the expected effect. When a
researcher manipulates an independent variable or variables, some
changes may happen in characters such as yield, carcass weight,
attitude, gender, educational level, blood pressure, which are
dependent variables. For example, yield of a crop (a dependent
variable) depends on many factors or causes such as rainfall
regime, temperature, fertilizer rate, water supply, and cultivation
practices (independent variables).

When dealing with dependent variables, most are easy to observe,

but some require special equipment for measuring them. For
instance, it is very easy to determine the height of crop plants, but
measuring photosynthesis is not so easy as measuring plant
height. For measuring photosynthesis, the researcher needs a
special apparatus and has to go through arduous procedures.
Therefore, under some conditions, a researcher may need to
determine a relation existing between characters by calculating
some variables that are difficult to observe directly from
characters that can be observed easily. This is often done through
modelling.
In some studies, researchers may be unable to control and
manipulate independent variables. For instance, the variables such

[44]
as gender, age, education level, body weight, race, ethnicity, etc.
cannot be manipulated. Such variables are called status variables
or attribute variables because they mirror the characteristics or
status of the study population. These attributes are already fixed
and cannot be changed. Researchers cannot manipulate status
variables. However, researchers can treat status variables as
independent variables and investigate their effects employing
appropriate research designs.

Independent variables that can be manipulated, changed, or

controlled are called active variables or treatment variables.

An attribute is a precise value on a variable. For instance, the

variable gender has 2 attributes, male and female; and a common
variable in social research, ‘agreement’ might be defined to have
5 attributes, 1 = strongly disagree; 2 = disagree; 3 = neutral; 4 =
agree; and 5 = strongly agree.

Independent variables may also differ depending upon their

nature of values. They can take numerical values or text values.
Examples of variables with numerical or text values are shown in
Table 1 below.
Table 1. Examples of variables with numerical values and variables with text values

Fertilizer doses (kg ha–1) 0, 50, 100, 150

Seed rate (kg ha–1) 80, 160, 240
Gender Male, female
Crops Wheat, teff, sorghum, and maize
Cultivars of rice Fikru, Jarso, and Morka
Breeds of cattle Boran, Horo, Foggera and Holstein Friesian

[45]
Quantitative and qualitative variables
Quantitative variables are variables that can be measured and
quantified directly whereas qualitative variables are variables that
cannot be measured directly. Examples of quantitative variables
include such characters as height, weight, length, etc. Examples
of qualitative variables or characters include creed, colour,
religion, ideology, gender etc. Such qualitative variables are also
called categorical variables. However, there are cases when some
qualitative variables can be considered as quantitative variables.
For instance, seed dormancy is a qualitative character but if a
researcher takes into account the number of days taken for
germination after breakage of the dormancy, this character may
turn out to be a quantitative character.

Qualitative variables can be categorized as nominal and ordinal

variables, whereas quantitative variables are categorized as
continuous and discrete variables.
Nominal and ordinal variables
A nominal variable is the one having one, two, or more sub-
categories based on the degree of variation; for instance, the
variable ‘educational status’ can have the sub-categories as non-
literate, literate, matriculate, graduate, postgraduate, and
doctorate; and the variable ‘gender’ has male and females as sub-
categories.

On the other hand, ordinal variables do not have absolute values.

They are organized in either ascending or descending order based
on the degree of variations. Since ordinal variables have no
absolute values, the actual variation in the adjacent ranks may not
be equal. Thus, ranking categorizes them equally, although they
may not actually be equally spaced. For instance, the variable
[46]
‘agreement’ can have five sub-categories as 1- strongly disagree,
2-disagree, 3-neutral, 4-agree, and 5-strongly agree.
Continuous and discrete variables
Continuous variables or discreet variables are the type of
variables that are usually quantified or determined in quantitative
research. In such research, numerical data are generated to
explain, predict, or control the phenomena of interest.
Continuous variables possess continuity in measurement. As a
result, they may take any value. For instance, the height of a plant
may range from 8–9 cm as 8.01, 8.1, 8.312, 8.5, 8.88, 8.9, and
8.99.

Discrete variables, on the other hand, take only discrete values.

For instance, the ‘size of a family’ can take only integer values
such as 5, 6, 7, 8, and 9. A discrete variable is one which takes
only an integer value within a given range. For example, the
number of grains per panicle of a particular variety of teff varies
between 40 and 60 grains. This means, if one takes out any
panicle of that particular teff variety, the number of grains in it
will take any value within this range. But one cannot expect that
the panicle taken at random should have a number of grains 50.6
or like that; it will be 50 or 51.

The nature of a character can indicate if that character is a

continuous or a discrete variable. In some cases, some discrete
variables are treated as continuous variable. These include a
population when expressed in millions, billions, etc. The
population of Ethiopia according to 2005 census is 120,185,422.
However, after estimation, it is possible to write 120 million.

[47]
Continuous variables are often further divided into interval and
ratio variables. An interval variable is similar to an ordinal
variable indicated above. However, the intervals are spaced
equally (ordinal variables have no absolute values and the
intervals are not equally spaced). In variables such as this one, the
intervals can be measured numerically along a range. For
instance, the difference between 6°C and 7°C is same as the
difference between 15°C and 16°C. The interval is the same, and
hence, the difference between the characters is also the same.
Dependent variable
A dependent variable is a type of variable whose values are
dependent on the values taken by the other variables and their
relationship. Generally, in relational studies, a variable is
influenced by other related variables. In a production function
analysis, there exists a functional relationship between the output
and the factors of production. Here, the output is considered as
dependent variable which depends on the factors of production
like land, labour, capital, and management. In socioeconomic
studies, the adoption index (dependent variable) with respect to
the adoption of a particular technology may depend on a number
of socioeconomic factors like age (x1), caste (x2), education (x3),
family type (x4), social status (x5), economic conditions (x6), area
under cultivation (x7), and the size of holding (x8). Thus, one can
write y = 𝑓(𝑥1, 𝑥2, 𝑥3, 𝑥4, 𝑥5, 𝑥6, 𝑥7, 𝑥8). In this example, y is
the dependent variable and x1 ... x8 are the independent variables.
One can use a functional relationship to predict the values of a
dependent variable for a given set of values of the variables x 1 ...
x8. As such, y is also known as predicted variable and x 1 ... x8 are
known as predictor variables.

[48]
Independent variable
In any relational analysis, variables which help to predict the
dependent variable using the functional relationship are known as
independent variables. In the above example, x1 ... x8 are said to
have no association among themselves and are termed as
independent variables. These variables independently help in
predicting the dependent variable. Generally, in regression
analysis, predictor variables are synonymous with independent
variables.
Explanatory variables
Independent variables are sometimes known as explanatory
variables, i.e. any variable which explains the response of the
dependent/predicted variable. In a simple regression analysis,
there are only one predictor and one response variable. In a
multiple regression analysis, generally, there is one response or
predicted variable with more than one
predictor/explanatory/independent variables. In the case of system
of simultaneous equations model, there may be more than one
response variable and more than one
predictor/independent/explanatory variable. Moreover, the
response variable(s) in one equation may be the explanatory
variable in the other equation.
Stimulus variable
The idea of stimulus and response variables is familiar in
agriculture, socioeconomic, and clinical studies. A stimulus is a
type of treatment applied to the respondents to record their
response. In clinical studies generally, the doses, concentrations,
different chemicals, etc., form a stimulus, whereas the response
may be in the form of quantitative response. When a stimulus is

[49]
applied to a record response, the response may be either-or type,
or it may be measurable. In an either-or type of response, a
respondent will either respond or not respond after being applied
with the stimulus. The different concentrations of a particular
chemical in controlling a particular pest of a particular crop may
kill the pest or may not; the response is either to kill or not to kill.
On the other hand, applications of insulin at a particular dose can
help in reducing the blood sugar level, a measurable response.

In socioeconomic studies, stimulus variables may be in the form

of action variable like documentary film, field demonstration, and
method demonstration, an effect of which would be measured in
the form of adoption or no adoption of a particular technology for
which documentary/field demonstration/method demonstrations
were used.
Control variable
Control variables are independent variables that can affect not
only the outcome of the study but also the relationship between
the dependent and other independent variables unless otherwise
they are held constant or eliminated from the model. Thus, control
variables are the type of independent variables which could be
controlled by a researcher to effectively study the effects of other
independent variables. For example, a number of laboratory
experiments are conducted at a controlled room temperature as
variable temperature affects the outcome. Thus, the independent
variable can be categorized into two groups, that is, the control
variable group and the moderator variable group. In comparison
to the control variable groups, the effect of moderator variables is
studied, keeping the control variables at constant or eliminating or
minimizing them. Depending upon the objective of the research,
[50]
it is up to the researcher to determine the moderator variables and
the control variables.
Dummy variable
In many research studies, particularly concerned with the
qualitative characters, it is very difficult to guide/put a study
under mathematical treatment. To overcome this problem, one of
the techniques is to assign numbers against the quality
parameters. For example, in a study concerned with gender-
related issues, the male may be assigned number 1, while 0 for the
female or vice versa. In a study of plant type, a bushy type of
plants may be assigned number 1, erect type number 2, and tree
type number 3, and so on. Thus, in each of the above cases,
quality characters are designated by different numbers. As such,
quality characters take the values 0, 1, 2, etc., and are known as
dummy variables. Dummy variables have special implications,
particularly in relational analysis. For example, instead of an
ordinary simple regression analysis in case of numeric variables,
one should go for probit, logit, and nomit regression analysis
when encountered with dummy variables but the mathematical
detail is beyond the scope of this handbook.
Preference variable
Preference variables are generally discrete type of variables
whose values are arranged either in a decreasing or an increasing
order. For example, in a survey of acceptability of crop varieties
by farmers, respondents were asked to grade the varieties by
using five different codes, namely,1 for excellent; 2 for very
good; 3 for good; 4 for poor; and 5 for bad. It may be noted here
that there is no relationship between the difference in grades 1
and 2 with that of the difference between any two consecutive

[51]
grades and vice versa. Similarly, in a study of constraint analysis,
farmers may be asked to indicate the importance of the following
constraints in accordance with their perception by using codes
from 1 to 8: 1 for weather, 2 for finance, 3 for irrigation, 4 for
marketing, 5 for price of input, 6 for price of output, 7 for non-
availability of good quality seed, and 8 for non-availability of
appropriate technology. The farmers, based on the way they
perceive, have to arrange the above eight constraints from 1 to 8,
with 1 being allotted to the most important constraint and 8 being
allotted to the least important constraint.
Multiple response variables
In multiple response variables, a variable can assume more than
one value. In a social, economic, market research, etc., it becomes
very difficult for respondents to select a particular option against
the other alternatives; rather, they opt for combinations of
absence. A typical example is the use of modern-day high-tech
mobile phones. If a respondent is asked to indicate the purposes
of using mobile phones in his/her daily life, the respondent could
score more than one category out of the following options: 1 for
talking to people at distant places, 2 for using GPS for daily life
activities, 3 for using its camera to take pictures, 4 for listening to
music, 5 for using it to add, subtract, and multiply numbers in
their daily life, etc. To some users, talking to people at different
places along with listening to music may be of equal and top
priority. To others, it may be the talking to people at distant
places along with the use for adding, subtracting, and multiplying
numbers that is of equal or top priority. Thus, in both cases, the
user has no preference to select one at top most priority rather
than the other ones. As such, there are multiple responses to a
given query.
[52]
Target variable
A target variable is almost synonymous with that of a dependent
variable in a classical regression analysis. The main objective is
to target a variable (predicted variable) whose value(s) is required
to be predicted taking the values of the other variables and also
the relational form.
Weight variable
A weight variable specifies the weightage to be given to different
data sets/subsets. A weight variable may be continuous or discrete
in nature. In a given data set, if “0” is assigned to any row of data,
then that particular row could be ignored. On the other hand, a
weightage given to rows 1, 2, 3, etc., means the rows concerned
have different importance.
Extraneous variables
Extraneous variables are variables that cannot be controlled but
could have a significant influence on the results of the research.
Extraneous variables fit the category of neither dependent nor
independent variable. For instance, take two variables, irrigation
and yield. Yield is affected not only by irrigation but also by
many other factors. For example, when there is a yield increase in
response to applying a fertilizer, only part the response can be
explained by the effect of the fertilizer applied. Likewise, the
relationship between the performances of milking cows may be
influenced by factors like age, conditions of shelter, weather
condition, etc. Thus, factors like age, conditions of shelter, and
weather conditions are extraneous factors/variables in a relational
study of the milk yield of cows. Overall, the values of these
variables are not directly obtained from the system under study,
but these variables may affect the dependent/response/predicted

[53]
variables. Thus, unexplained variability may occur due to
variability in other factors and not all the other factors or
variables that influence yield can be categorized as dependent or
independent variables.

In this connection, yield can be written as a function of several

inputs.
Yield = f (inputs).

The inputs could be fertilizer, pesticide, growth hormones,

irrigation, or some others. However, the dependent variable
(yield) is influenced not only by the quantity of inputs supplied
(independent variable), but also by some other factors such as
contents of nutrients other than that supplied as fertilizer, soil
moisture content, soil aeration, depth of soil, drainage, soil
acidity, temperature, sunlight, shading, frost, and incidence of
pest and diseases. Such factors that cannot be controlled in an
experiment are called extraneous variables or confounding
variables.

Uncontrolled differences are common occurrences in biological

research. A researcher doing an experiment may come across
uncontrolled variations when no preliminary precautions and
safeguards were taken before hand to address the error. Such
problems can occur when the results of a given piece of research
produce surprising results.

Therefore, in an experiment, the researcher should hold all

extraneous variables constant except for a single or a couple of
other independent variable the effect of which he/she intends to

[54]
study. This is an ideal situation; however, guaranteeing such an
ideal situation is difficult in most cases. This is because the
influence of not all extraneous variables can be excluded.
However, the researcher can limit the influence of external
variables to the minimum by using the principles of
experimentation and including a ‘control’ treatment.

Extraneous variables may again be of two types: (a) participant

variables and (b) situational variables. Participant variables are
extraneous variables, which are related to the individual
characteristics of each participant. Thus, in the above example,
age is the participant variables. On the other hand, conditions of
shelter, and weather conditions are situational variables, which
are mostly related to environmental conditions. Generally, under
experimental research conditions, extraneous variables are
controlled by researchers.

4.2. Criteria for selecting variables

Generally, while selecting a variable, a researcher should keep in

mind the objective of the study, hypothesis to be tested, variables
which should be mutually exclusive (non-overlapping), variables
which should be clearly understood by him/her, techniques that
are available or to be developed to measure the variables, the
number of variables to be measured, and availability of time and
resources. The variables to be selected should be guided by the
objective and specific objective of a research program. What is
more, the variable to be included must be clearly defined since
any ambiguity in the definition of variables may jeopardize the
findings of a research programme. The variables to be selected
must be related to the hypothesis to be tested in a study, and must
[55]
be non-overlapping in nature. A variable which is difficult to
measure with an adequate level of accuracy should be avoided. In
addition, the number of variables to be measured must not be too
many or too few.

5. Major Research Methods

Scientists follow different approaches for doing research in

different disciplines. In social sciences, most research is
conducted using qualitative and non-experimental methods.
However, this does not mean that experimental methods are never
used in social sciences. However, for conducting research in
natural sciences, mostly quantitative and experimental research
methods are employed. It is also possible that both qualitative and
non-experimental methods are employed for conducting research
in natural sciences. Thus, depending on the field of study, natural
or social sciences, researchers opt for research designs that meet
their needs, applying qualitative or quantitative methods
depending on the circumstances, time, resources, and the type of
variable involved as indicated below.

5.1. Natural observation method

If experimentation is not possible, researchers usually use natural

observation as one fundamental method of research. The
observations are made in natural setting of the situation or
phenomenon where the researcher objectively observes and
records the data. In social sciences, health, and ecology,
observation, which is also called natural experiments or
observational studies are common.

[56]
Likewise, observation is a common method of research in
physical sciences such as astronomy, geology, oceanography,
meteorology, etc. Using the data collected by observing a natural
phenomenon, a researcher can test a hypothesis. In social
sciences, it sometimes happens that researchers set on research on
humans or animals by observing them in their natural
environment and collect data on their activities and responses. In
natural sciences also, it is common to observe animal or groups of
animals, plant or plant communities, or some physical phenomena
to collect data. Observation is also the main method of research in
ecology even though experimental methods can also be used in
ecological research.

The major limitation of the observation method is the difficulty

faced in observing behaviour or a process or a phenomenon
without disruption and the data generated from the method may
not be amenable to statistical analysis. The results obtained
through observation is done without the controlled conditions of a
laboratory and, therefore, conclusions about cause-and-effect
relationships cannot be drawn with certainty.

5.2. Historical method

A historical research method is used in relation to past events in

view of the current events. In general, a historical researcher
states a problem, formulates a hypothesis or hypotheses, gathers
and analyses primary data, tests the hypotheses and draws a
conclusion. Historical research utilizes historical sources like
documents, remains, etc. to study events or ideas of the past,

[57]
including the philosophy of persons and groups at any remote
point of time.

A researcher of history attempts to reconstruct the past usually by

formulating a hypothesis. However, it is not easy to use scientific
method in historical research. This is because scientific research
is about prediction, but a researcher of history cannot predict.
This is because past events occur with no plans. They also occur
with the presence of uncontrolled variables. Therefore, in
historical research, experimental methods with ‘randomization’
and ‘control’ are not possible.

Researchers of history rely on reported observations from other

primary or secondary sources of doubtful origin. Primary sources
of information for historical research include eyewitness
accounts, remains or relics, and oral testimonies. Eyewitness
accounts and the records kept and written by concrete participants
of events are good sources of information. Remains or remnants
are objects related with a person, group, or period. These include
fossils, skeletons, tools, weapons, utensils, clothes, paintings, and
coins, which may provide evidence about past events. Oral
testimony is an explanation of a witness or participant. Secondary
sources of information for historical research include history
books, reports, and encyclopaedias, etc., but are also less reliable.

5.3. Ethnographic method

Ethnography is the description of peoples and cultures with their

customs, habits, and mutual differences. Ethnographic research
method is the method used in research that studies the present
events rather than past events involving people. It is a qualitative

[58]
research method used to study any group of people for describing
their socio-cultural activities, patterns, diversity, etc. at home and
abroad. The method involves the collection of data on many
variables, which cannot be statistically but descriptively narrated
and documented.
The most common approach to conduct ethnographic studies is
participant observation, in which case the researcher happens to
live with the people he/she studies, chronicles and extensive field
notes. The researcher participates in the routine activities of the
people being studied, restructures their relations and events in the
field notes taken on the spot or as soon as possible after they
happened.

It is also possible to conduct ethnographic research using the

method of direct observation for collecting data. In this case, the
researcher does not participate in the events but merely observes
the events occurring unresponsively to avoid any bias. Modern
tools such as hidden camera, recorders, and video can be used in
the direct observation.

5.4. Case study method

Case study is a research method in which a phenomenon is

investigated in its real-life context based on an in-depth
investigation of a group, event, or individual to explore the
underlying causes.
Case studies are often used to assess something employing both
qualitative and quantitative methods. This method can be used
also in agricultural and life sciences to study a crop, animal, or a

[59]
single instance of some physical phenomenon. Case studies are
also valuable when researchers are unable to conduct
experimental studies. Case studies often provide in-depth
information about a phenomenon and give way to a broader study
later.
Case studies are often used in both natural and social sciences. A
case can be anything that a researcher wants to investigate, which
includes a person, a group, an event, a decision, a process, a
policy, period, institution, or any other system that can be studied
historically. The case may be a plant, a group, process, a disease,
a community, or any other similar unit. A case study is
appropriate to discover intensively the background, status, and
environmental interactions of a given social unit, individual,
group, institution, or community. Case studies are common to
identify the social, natural, and background factors that might
have caused something to an individual or a group.
As common features, case studies need to focus on a selected
case; provides an in-depth understanding of an issue; and data can
be collected in many different ways such as observations,
interviews, and secondary data review.
Case studies, despite their importance, have also limitations. First,
case studies are mostly descriptive in approach, and not analytical
or explanatory. Secondly, case studies involve a lone case or just
a few cases only because of which they may not fully represent
the target population. Thirdly, since case studies are based on
information obtained from different people or other sources, there
could be a higher chance of excluding important details. Fourthly,
since retrospective data of past occurrences are collected in case

[60]
studies, the problem is connected with memory that may constrain
accuracy of information.

5.5. Correlational research method

Correlational research method is often employed to assess if

change in one or more variables are related to change in other
variables. If two variables are correlated, it is possible to use the
association to forecast the value of one variable based on the
other.
There are certain types of correlations, including predictive
correlation and relationship correlation. A predictive correlation is
done to learn which variables from a set is the best to predict the
dependent variable. The objective of such a correlational study is
to maximize prediction. In a relationship correlation study, the
objective is to understand variables that are theoretically related
to a dependent variable.

The outcome of a correlational study method is valuable for

framing hypotheses, which could later be tested in experimental
and quasi-experimental studies. However, correlations do not
indicate cause and effect relationship. Thus, correlations are used
for predicting outcome but not for predicting cause.

The limitation in a correlation study is that it only shows if two

variables are correlated, but does not show a cause-and-effect
relationship.

[61]
5.6. Modelling method

Modelling is the method of research used to replicate real-world

systems in which an equation or formula can be used to estimate
the effect of a physical process on some other process. Research
methods used to solve a problem without conducting experiments
are called ‘models’.

Both physical and conceptual models are important components

of scientific research. There may be simple models which
emanate from regression and are used most frequently as
empirical models, the main group of models based on observed
facts, which enable researchers to predict what would happen in
certain conditions. The dependability of models to predict events
or phenomena depends on the thoroughness of database created
through experience. An empirical model may be a simple
relationship or a complex multiple equation. Models are often
built mathematically through computer programming. These
computer-based models serve their purpose as physical models,
but are determined entirely by mathematical relationships
between variables that are defined numerically.

For instance, the Universal Soil Loss Equation (USLE) is an

empirical model designed to predict long-term average soil losses
from a landscape under a particular cropping and management
systems. Various factors affecting soil losses through erosion
such as rainfall erosivity, soil erodibility, gradient of slope, length
of slope, vegetative cover, and soil conservation practices are
considered in the equation to predict expected soil losses for a
particular area.

[62]
Similarly, several empirical models have been developed for
estimating water consumptive use of crops from routinely
measured meteorological data. Many of them have been
determined and tested locally in Ethiopia. It is possible to select a
suitable empirical formula such as Penman–Monteith to calculate
the reference crop evapotranspiration (ETo), and using the crop-
coefficient, crop evapotranspiration (ETc) can be estimated.
Fitting models is common in research areas including agriculture,
and others.

5.7. Action research method

Action research is a method of research employed to solve

existing problems faced by a society by involving the
collaboration and cooperation of all stakeholders—researchers,
bureaus of agriculture, extension workers, research institutions,
farmers, and other beneficiaries in teams. Two activities are
included in action research. These are “action” and “research”. It
means “research in action”. Action research is done by involving
different stakeholders, namely, bureaus of agriculture,
development agencies, and community organizations by applying
innovative ideas, techniques, and technologies in solving issues
and problems for promoting socio-economic development. Action
research is ‘learning by doing’. Before engaging in action
research, a group of stakeholders identify a problem, plan and
undertake the action research to solve it, and evaluate the success
of their efforts.

Action research is conducted mainly to develop new skills or new

approaches and to solve problems with direct application. A

[63]
problem is diagnosed first, remedial action is immediately
planned, implemented, and effects monitored.

For example, if a sudden crop pest invasion occurs in an area and

there are no proven control measures, there would be no enough
time to formulate a hypothesis, validate it through experiments,
and then an immediate action is required to implement the control
measure. Action research has four basic characteristics. These
include being situational, collaborative, participatory, and self-
evaluative.
It is situational: A problem is identified in a specific context, and
action research is done to solve it.

It is collaborative: Action research is done in collaboration with

stakeholders, namely, researchers, extension workers,
practitioners, farmers, or non-professionals.

It is participatory: All the team members participate in

executing the action research work.

It is self-evaluative: Monitoring and evaluation is done

constantly.

If necessary, changes may be made to the action research based

on the results obtained from the monitoring and evaluation team.
Action research differs from other types of research basically
because it focuses on solving real problems as well as it is used in
real situations as against experimental research.

[64]
5.8. Participatory learning and action research
method

Participatory research methods are used for assessing

developmental issues such as natural resources management in an
area, watershed management, environmental concerns, and youth
& gender issues in local communities in a relatively short period
of time in a multi-disciplinary approach.

The major thrust of any participatory methods (PMs) is to give

the local people the opportunity to participate and play active
roles in decision making process that affects their lives. This is
often done by empowering the local community to actively
participate at all stages of the research process. In such cases, the
researchers play only catalytic and supportive roles. They will
refrain from domineering and rather act as facilitators by
involving also other stakeholders in all aspect of the research
process. Such an approach makes possible for the official or
extension agents to use their own criteria to take samples from the
local people for data collection as well as to understand the local
environment with definite priorities. Participatory methods allow
the local people to attain their developmental goals using the
skills of outsiders. Participatory learning and action research are
of several types among which the following are widely used.
Rapid rural appraisal
Rapid Rural Appraisal (RRA) is a quick technique to discover
quickly information necessary for formulating or designing a
project. The method involves local people in collecting and
analysing data by employing the systems such as mapping,
transect walks, and ranking.
[65]
RRA gave rise to Participatory Rural Appraisal (PRA) since the
focus shifted from quick data gathering to the empowerment and
involvement of end-users, and the thrust of the researcher
focussed more on facilitation, empowerment, and assisting
towards sustainability of the action.
Participatory rural appraisal (PRA)
Participatory Rural Appraisal (PRA) is a set of tools and
techniques used with households to gather and analyse
information on community resources, problems, potential and
needs. PRA brings together, on the one hand, development needs
defined by the community members and, on the other, skills of
Government, donor agencies and NGOs. It integrates traditional
knowledge systems and external technical knowledge in the
development process. PRA is a way of learning from, and with
community members to investigate, analyze and evaluate
constraints and opportunities and make informed and timely
decisions regarding development projects.
PRA techniques are also applicable to urban settings is not only
limited to a project appraisal but it is also valuable in an entire
project cycle. The main difference between RRA and PRA
depends on who leads the assessment. If the learning process is
done mostly by outsiders such as extension workers, it is called
Rapid Rural Appraisal (RRA), whereas if it is continuous research
and an action done by the local community, it is called PRA.

Participatory Action Research (PAR) is an innovative approach to

research. By enabling people to participate in a program of action
and empowering them, it is possible to change the lives of poor
rural people thereby changing their environment through their
own methods.
[66]
Participatory learning and action (PLA)
Participatory Learning and Action (PLA) is a type of qualitative
research, which is implemented with the active participation of
community members to achieve a better understanding of a
problem encountered by the community for finding a solution.
PLA shares the same techniques with RRA, PRA, or PAR.
Participatory Learning and Action (PLA) is assumed as an
umbrella term that includes a variety of participatory approaches.
PLA is used widely in natural resource management (watershed
management, soil and water conservation, forestry, wild life,
biodiversity, fisheries, etc.), village planning, community health,
food security, and poverty eradication programmes.
Some of the tools used in PLA, like other approaches commonly
used in social studies, include semi-structured interviewing which
can be done as individual interviews, natural group interviews,
focus group discussions, community meetings, and brainstorming.
Another major tool used is transect walk, a kind of exploratory or
observational walk. Social transects and land use transects are
common.

5.9. Some special research methods

In agricultural and life sciences, research should not necessarily

involve experiments based on hypothesis formation and testing. It
may be simply descriptive as in anatomy, morphology, or
ecology. The approaches adopted are dependent upon the nature
of the discipline and particular topic. The following are some
additional types of research employed for specific purposes.

[67]
5.9.1. Prototypes and pilot studies
Prototypes and pilot studies are two special forms of scientific
discovery through research. When designing a new apparatus or
equipment, a prototype is made. Making a prototype is a common
method in technology or applied science as a bridge between
theory and practical application. There are two types of prototype.
These are laboratory prototype and pilot plant. A prototype is
necessary when a company or institution substantially modifies an
existing equipment design. It is also useful when a researcher
assembles equipment based on published work and sketchy
details.
Similarly, when beginning a novel type of experiment, a pilot
study is conducted. A pilot study is the procedural analogue to a
prototype. Pilot studies can give valuable information on whether
a detailed experiment will succeed and how one should cope with
uncontrollable variables.
5.9.2. Directed research/targeted research
When research is done in response to a request from someone or
an organization to explore and address a specific problem, it is
called directed research or targeted research.
5.9.3. Clinical research
Clinical research deals with questions to the normal functioning
of the animal body and investigating diseases using animal
subjects such as new treatments, new medicines, new techniques
by using existing treatments, screening techniques, and new
diagnostic techniques. Clinical trials are conducted based on
specific plans, which are usually called a protocol. The protocol
clarifies what the types of diseased animals are to be enrolled for

[68]
the study, schedules of tests and procedures, drugs, dosages,
length of study, and measurable outcomes.
5.9.4. On-farm research
On-farm research is a type of research conducted to evaluate,
refine, or develop farmer-oriented need based-research under
existing biophysical and socio-economic conditions of
particularly resource-poor farmers. On-farm research is usually
meant to take promising results of a given research from the
experimental field or laboratory to the farmers’ fields.
5.9.5. Farmers’ participatory research
Farmer participatory research (FPR) is a kind of agricultural
research system that ensures participation of farmers by including
indigenous knowledge in the research process. Thus, participatory
research is a type of research done on a population together with
the population (stakeholders). It involves engaging the target
population to choose research questions, design the study, and
address ethical concerns.

In farmers’ participatory research, there are methods used to

measure performances of candidate technologies or practices.
Ranking method is one of the simplest performance evaluation
methods used in this type of research. In this method, study
subjects are ranked from best to worst in a group.
Ranking exercises
In participatory research evaluation, four types of ranking
techniques are commonly used, including problem ranking,
preference ranking, matrix ranking, and wealth ranking.
Experiments where farmers (both males and females) are
involved, like any other studies, also require generating,
[69]
analyzing, and interpreting at least two types of data. The first,
just like any other on the station trials, is collection, analysis, and
interpretation of data on performance assessment of the
technology under consideration which is already thoroughly
discussed elsewhere in this Handbook. The second, which needs
proper discussion and illustration, is collection, analysis, and
interpretation of data on attitudes of participant farmers towards
the technology in a gender segregated manner as a follow up of
on the station research where farmers’ views and circumstances
maybe disregarded. The aim here is to identify the farmer
preferred technologies that can be easily adopted under their own
real circumstances, thereby improve their production systems,
boost productivity and enhance their wellbeing in a way that can
be sustained.
Problem ranking
Problem ranking is a participatory technique that allows analyzing
and identifying problems that stakeholders face. The objective of
problem ranking is to obtain perceptions of local people on their
important problems and implement adequate improvements and
solutions. Problem ranking is usually done by making groups of
men and women involving all the socio-economic groups
participate in the process. During the discussion, the participants
list out their problems. They then rank the problems according to
importance and select the most important ones. These problems
are prioritized by discussing the causes and effects. The results
provide the starting base for discussions on possible solutions to
the priority problems.

[70]
Preference ranking
Preference ranking is similar to problem ranking, but in this
technique, participants assess different options they are given
using their own criteria. Scores can be assigned for identifying the
preferences. For example, the problems of access to fertilizer and
availability of fertilizer can be compared on a 5-point scale, by
giving 5 for the most important and 1 for the least important. You
can use a blackboard or flipchart, and draw a priority-setting
background in a tabular format. Criteria are listed in columns and
possible actions in rows. Then, the group has to evaluate possible
actions and list them in the left column of the table. In this case,
the researcher distributes the scoring cards among the members so
that each member gets one set of cards for each of the scoring
criteria. The number of cards in each criterion set must be equal
to the number of actions being ranked. For instance, if there are
four actions, then each criterion set should contain a number from
1 to 4. To assess the ‘feasibility’ of a specific action, the
participant must display the appropriate card. After all criteria are
evaluated, the scores are jotted down in the table. The individual
scores are added and the sum is indicated in the right column of
the table.

To complete the process of preference ranking, all the participants

have to study the final results and make comments on them. The
researcher needs to explain to them clearly that the scores are
meant to assist in decision-making. The researcher also asks the
participants to make decisions considering both the trends
revealed by the total scores and the suggestions from the
discussion.

[71]
Preference score index
One common method of measuring and analyzing attitudes of
participant farmers towards the technology is to make a farm walk
and calculation of the preference score index by comparing each
test treatment with the standard or local check. Farmers positive
and negative votes at a specific time are collected from all over
the locations and the preference score index are calculated from
the mean preference scores by the total number of farmers as
follows:

𝑁𝑜. 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑣𝑜𝑡𝑒𝑠 − 𝑁𝑜. 𝑜𝑓 𝑛𝑒𝑔𝑒𝑡𝑖𝑣𝑒 𝑣𝑜𝑡𝑒𝑠

𝑃𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑖𝑛𝑑𝑒𝑥 =
𝑇𝑜𝑡𝑎𝑙 𝑁𝑜. 𝑜𝑓 𝑣𝑜𝑡𝑒𝑠

For example, assume that five genotypes of faba bean (variety 1-

5) were subjected to a farmers’ participatory selection along with
a standard check (Numan). Assume also that 16 farmers were
invited to respond through votes in favor or against a particular
variety as compared to the standard check at, for instance, the full
pod stage. The farmers were invited to make a farm walk and
preference ranking for each variety as presented in Table 2 below.
The higher preference index being the better, Variety 4 received
the highest preferential scoring of unity, of course at the full pod
stage, followed by Variety 2. On the other hand, Variety 3
received the least number of votes (least preferential scoring) as
the least preferred by the farmers even lesser than the standard
check, Numan. Note that if the experiment is replicated over
locations and years, preference data pooled over locations and
years are needed as a decision tool to promote selected varieties
to the next stage. It should also be noted that the preference
[72]
pattern could change at another time for some other traits.
Likewise, preference ranking may change by different group of
farmers dwelling in a different locality.
Table 2. Farmers’ preference ranking of different faba bean varieties as compared to the
standard check

Genotype Total number of farmers participated

Positive votes Negative votes Preference score Rank
Variety 1 14 2 0.75 3
Variety 2 15 1 0.87 2
Variety 3 7 9 -0.12 5
Variety 4 16 0 1.00 1
Variety 5 13 3 0.62 4
Numan --- --- --- ---

Examples of preference ranking

The participants are asked to rank the treatments (e.g. genotypes)
in order of preference, i.e. 1 = most preferred and 5 = least
preferred for the example given above. The respondents have the
opportunity not only to identify which treatments are the best but
also which are the least preferred. The average of rank values
given by all farmers for each treatment should be calculated in
order to determine the treatments most preferred (the smaller
value is the better) and the ones which were least preferred. For
the hypothetical example given above, Variety 4 and Variety 2 in
that order were the most preferred, whereas Variety 3 and Variety
5 were the least preferred, Variety 1 being somewhat in the
middle (Table 3).

[73]
Table 3. Preference ranking six faba bean varieties by six participant farmers

Variety Participant farmers

Farmer 1 Farmer 2 Farmer 3 Farmer 4 Farmer 5 Farmer 6 Ranking

average

Variety 1 3 2 3 3 3 3 2.83
Variety 2 2 3 1 2 1 2 1.83
Variety 3 6 5 6 5 6 5 5.50
Variety 4 1 1 2 1 2 1 1.33
Variety 5 5 6 5 6 5 6 5.50
Numan 4 4 4 4 4 4 4.00

Or else, the ranking average values given by all farmers for each
treatment could be calculated based weights that are given in
reverse order where the highly preferred item with a rank of 1 has
the highest weight (of 6 in the above example) and the least
preferred item has the lowest weight (of 1 in the above example).
Weight count depends on the total number of treatments and, in
our example where we have six treatments, for instance, the
weight of each of them based on their preferences will be 6 for
the rank of 1, 5 for the rank of 2, 4 for the rank of 3, 3 for the rank
of 4, 2 for the rank of 5 and 1 for the rank of 6. The ranking
average is calculated as follows:

𝑿𝟏𝑾𝟏 + 𝑿𝟐𝑾𝟐 + 𝑿𝟑𝑾𝟑 + ⋯ + 𝑿𝒏𝑾𝒏

𝑹𝒂𝒏𝒌𝒊𝒏𝒈 𝑨𝒗𝒆𝒓𝒂𝒈𝒆 =
𝑻𝒐𝒕𝒂𝒍 𝒓𝒆𝒔𝒑𝒐𝒏𝒔𝒆 𝒄𝒐𝒖𝒏𝒕
Where W = weight of the ranked position and X = rank.

[74]
Pair-wise ranking
First, the treatments are listed in the matrix table across the first
column and across the top row as given below (Table 4).
Comparisons are made in all possible pairs taking two treatments
at a time, comparing with each other by participant’s group
through discussion and the better of the two treatments is marked
by writing its name in the cell. The names of both treatments are
written, if in case, they have equal performance and choosing
between the two becomes difficult. Compare the first treatment
with the second, the first treatment with the third, and keep on
comparing them until all the treatments are compared in all
possible pairs. Finally, make count of the number of times each
and every treatment was selected to be better. The treatments
which are named more numbers of times are considered to be the
best and, accordingly, Var 4 with a score of 5 points followed by
Var 1 with a score of 4 points could be considered as varieties of
better choice, whereas Var 3 as a variety of not chosen and Var 5
with a score of unity could be considered as varieties of least
preference (Table 5).

Table 4. Preference ranking of six faba bean varieties by participant farmers

Var 1 Var 2 Var 3 Var 4 Var 5 Numan

Var 1 -- Var 1 Var 1 Var 4 Var 1 Var 1
Var 2 -- -- Var 2 Var 4 Var 2 Var 2
Var 3 -- -- -- Var 4 Var 5 Numan
Var 4 -- -- -- -- Var 4 Var 4
Var 5 -- -- -- -- -- Numan
Numan -- -- -- -- -- --

The same result could or else be presented based on weights to be

generated from the number of times a given treatment is selected.
[75]
In the hypothetical example given above, for instance, Var 4 = 5,
Var 1 = 4, Var 2 = 3, Numan = 2 and Var 3 = 0. To assign weight,
the scores of all treatments are considered to make a total of
100%, such that:

100 = 5X + 4X +3X + 2X + 0X

100 = 14X

X = 100/14

X = 7.143
Then the proportional preference of the Vars could be presented
as:

5  7.143 = 35.715 for Var 4, 4  7.143 = 28.572 for Var 1, 3 

7.143 = 21.429 for Var 2, and 2  7.143 = 14.286 for Numan,
indicating that Var 4 received the first rank followed by
Var 1.

Matrix ranking
Matrix ranking is an important PRA tool to investigate
preferences of farmers for a particular technology over others,
with respect to crop or animal based technologies. It is done to
know about the preferences of households for different activities,
resources or items and reasons or criteria for preferences. The
preferences and criteria for the same are also studied in the
process.

[76]
In matrix ranking or matrix scoring, a group of options is
evaluated by applying different criteria and assigning scores to
each criterion. It uses a matrix with options along the horizontal
axis and the elicited criteria along the vertical axis. This technique
enables the facilitators to find out the reasons for the preferences
and priorities in various practices such as crops, cultivars, tree
species, soil and water conservation measures, irrigation methods,
fertilizers, plant protection measures, improved heifers, and other
technologies.
Matrix scoring can reveal interesting differences among group
members. The researcher starts discussing on a particular
technology. For example, if it is improved varieties of maize, list
all the maize varieties that the framers grow. Then, list the
reasons for preferring the given varieties in short phrases. Then,
draw up a table of matrix by placing the reasons in rows and
technologies in columns. Then, the researcher enquires ask the
farmers to rank the varieties of their choice for each season.
Indicate the variety that is best suited for the first season, second
season, and so on, until all the varieties are covered. If five
technologies are considered for ranking, the best one for the
specific reason is to be given a score of 5, the next best 4, the next
3, the next 2, and the last 1. Calculate the total score for each
technology over all the reasons mentioned. Repeat the process for
all participants. Based on the total score, the researcher makes
inferences on adoption of technologies and uses these inferences
for future planning strategies.
Direct matrix ranking
Draw up a matrix table where treatments are put across the top
row, and selection criteria are put down the side of the first

[77]
column as shown in Table 5 below. Then select participant
farmers (an individual or group), let them establish commonly
agreed up on selection criteria and allow them to the field to rank
the treatments based on their own selection criteria. In this
process, details of why a given treatment is rated as good or bad
need to be properly noted and the rankings should be directly
recorded onto the matrix. Participant farmers (both males and
females) should also be given opportunity to select their best
choice for overall performances and, in this process, it is
advisable to keep men and women participants separate in order
to avoid dominance of the former and generate better information.
The use of mixed groups, where male and female farmers should
make arguments and come up with common preferences, may
also need to be considered in certain cases particularly when they
share similar selection criteria.

Table 5. Direct matrix ranking of six faba bean varieties for different selection criteria by a group
of participant farmers

Criteria Var 1 Var 2 Var 3 Var 4 Var 5 Numan

Pod load 3 2 6 1 5 4
Pod size 3 1 6 2 4 5
Seed size 4 1 5 2 6 3
Diseases resistance 3 2 5 1 4 4
Overall performance 3 2 6 1 5 4

Var 4 and Var 2 received better preferences for both separate

traits and for the overall performances, Var 3 and Var 5 being the
least preferred (Table 5).

[78]
Wealth ranking
A participatory exercise where key informants rank households in
the community according to their evaluation of each household’s
resources is called wealth ranking. Wealth ranking can provide
important insights into the social values of community members.
The ranking obtained from such an exercise is a weighted average
of the household’s resources. In this case, higher weights are
implicitly given to resources considered socially more important
by the key informant. Therefore, data from wealth ranking may
not correspond with a ranking of households based on a survey of
assets. Wealth rankings are often carried out actually as wealth
categorizations using pre-defined wealth categories, such as
wealth terciles (3 categories) that might be defined as “poor”,
“intermediate”, and “rich”. The specific criteria for falling into
each category can be developed and made explicit by the key
informants. Wealth is a relative category and, therefore, research
scientists should be aware that members in the communities
possess knowledge of the wealth positions of community
members.
Assessing how wealth interacts with technologies is necessary
since the adoption and adaptation of an intervention may differ by
the endowment in resources or productive assets. Knowledge of
participant’s wealth categories may provide information to the
researcher to guide recommendations of how the intervention may
match the differing resources in the community.
Wealth ranking approach may be an important alternative and
complement to other methods, such as conventional surveys that
estimate poverty rates and asset-based wealth indices, and may be
cheaper to administer.

[79]
In wealth ranking, people are ranked and placed on different steps
of the social ladder based on their wealth. The objective is to have
a first-hand information about the stratification of local society
into rich, middle, and poor categories as viewed by people
themselves. The exercise helps officials concerned to have an idea
about the differences in wealth in every neighbourhood leading to
an overall understanding of the socio-economic conditions of the
population in the territory. This exercise is also helpful for
selecting the beneficiaries for various schemes. The most
common version of this technique involves a focus group
discussion involving members of a locality. The facilitator
introduces the procedure using local terms for wealth and poverty
and encourages them to first discuss how they define these terms
and how they would describe a poor or rich household. Using
these criteria, cards are made and actual ranking is done by
sorting cards. Wealth ranking may be difficult in urban and semi-
urban areas, where people tend to be less familiar with their
neighbours than in rural settings. Moreover, in some
communities, relative wealth ranking is a very sensitive topic and,
therefore, it is better to conduct this technique in a private setting
allowing participants to talk freely.
Trend analysis
Trend analysis is a PRA tool of collecting information and
attempting to spot a pattern. Trend analysis is often used to
predict future events. However, it could also be used to estimate
uncertain events in the past.
Trend analysis shows the changes in the past few years regarding
the daily life pattern of people in a locality. It can be in the form
of a line graph, which indicates the relationship between two

[80]
types of data. In agriculture, trend analysis can be done for the
items such as the change in crop yield (increased or decreased),
cost of cultivation of crops, wasteland area, farm produce prices,
consumption of fertilizers and pesticides, land use changes,
cropping pattern, population growth, coverage of improved
cultivars, tree species grown, fuel used, flow of income (increase
or decrease), changes in credit, and status of transport and
communication facilities. Seasonal calendars and daily activity
charts are two common trend analysis techniques.
Seasonal calendars
Seasonal calendars or seasonal diagrams are tools, which help to
explore changes taking place over a certain period of time. In
such participatory exercise, attempts are made to determine the
seasonal activities as understood and practised by the
stakeholders. Seasonal features such as labour supply shortage),
pest and disease incidence, harvesting of crops, ploughing
activities, sowing of crops, intercultural operations, input
requirements, irrigation requirement, weather elements such as
temperature and rainfall, soil moisture status, and fuel and fodder
availability can be assessed based on seasonal calendars.

Seasonal calendars are often compiled based on semi-structured

interviews and group discussions. Calendars can be drawn on a
large piece of paper, traced in the sand, or on a concrete floor
using stones or seeds. A series of seasonal variables can be
included in one calendar to give an overview of the situation
throughout the year. Important seasonal events such as festivals
can also be shown. The researcher draws a horizontal line all the
way across the top of the paper explaining that the line represents
a year. He or she will ascertain from the group how they divide
[81]
the whole year as months or seasons. Accordingly, the
participants have to mark the seasons along the top of the
horizontal line. It is convenient to start the calendar by discussing
the rainfall pattern. The participants may put stones, grains,
sticks, or leaves under each month of the calendar according to
the relative quantities of rainfall. It is better to scale it from 0 to
10, no stones meaning no rain, and 10 means maximum rain. The
seasonal calendar can be made on paper too, by replacing the
stones with dots in each month with a marker. After completing
the seasonal calendar for rainfall, other aspects of interests can be
studied.
Daily activity charts
Daily activity charts or daily activity clocks show graphically
how the community members spend their day. The diagrams also
make it easy to compare the daily activities of different groups of
people such as women versus men, employed versus unemployed,
rich versus poor, married women versus widows and young
versus old. Daily activity charts show who works for longer
hours, who concentrates on a small number of activities, who
must divide their time for a multitude of activities, and who has
the most leisure time. Just as a seasonal calendar shows the
busiest times of the year, a daily activity chart shows the busiest
times of the day and, therefore, will be useful in monitoring
changes in the time use of project beneficiaries and whether these
vary by gender. It is a useful tool for gender analysis.

For conducting the exercise, it is important to organize separate

focus groups of women and men making sure that each group
includes people from different socio-economic categories.
Explain the purpose of the exercise. Ask women and men groups
[82]
to produce their own charts. They should first focus on the
activities of the previous day. Illustrate all kind of activities such
as agriculture work, livestock care, childcare, cooking, and sleep.
Ignore slight variations in daily activities of persons, but go for
daily activities typical of the group.
Gender analysis
The term gender refers to socially constructed and culturally
determined differences attributed to men and women and their
roles in the household, community, or the society at large. These
differences and roles change over time and vary widely within
and between cultures. At the same time, the word sex is used to
mean the real biological differences between a male and a female.
A systematic study to document and understand the roles, needs,
and priorities of women and man and their relations within a
given context is called gender analysis.
Most of the PLA tools can be used for gender analysis too. Semi-
structured interviews and daily activity charts are typical tools for
the analysis. However, it is important to access and record the
data in disaggregated and specific terms with respect to men and
women. This is important for understanding the gender division
of labour, how the people use natural resources, rely on them, and
have access to alternatives. Gender analysis enables researchers to
explore and highlight the underlying inequalities in the
relationship between men and women by asking fundamental
questions such as who does or uses what, how, and why, and who
benefits, and how much. When conducting discussions or
interviews, both men and women should be allowed to provide
their answers and their views. If necessary, separate meetings
should be conducted for men and women, because in some

[83]
cultures, women are reluctant to attend meetings and to speak
their minds in the presence of men or their husbands. In such
cases, a woman facilitator should assist the discussion and help
women to find out what they wish to discuss and how.

Gender roles and responsibilities are not only structured by the

socio-economic position of a household but also by other factors
such as age. For example, women from poor families shoulder
most of the burden of domestic work. In contrast, rich women
may not bear the same burden of work than their unfortunate
counterparts do. However, they are generally more restricted in
social mobility. Similarly, age gives some more freedom for the
women as older women cutting across class categories enjoy in
social mobility.
Satisfactory levels of advance preparations are necessary for
organizing PLAs. A PLA activity should be conducted by a multi-
disciplinary team constituting a team leader and subject matter
specialists based on the topic of study. In a PLA study, a team of
people usually works for two to three weeks on discussions,
analyses, and fieldwork. It is necessary to prepare the required
logistics such as accommodation, meals, transportation, funds to
purchase refreshments for community meetings, video or audio
recorder, and supplies such as chart paper and marker pens, laptop
comers, etc. It is also necessary to train new team members. The
results of the appraisal results are often affected by the length of
time the exercise is done, scheduling and assignment of report
writing, critical analysis of data, and recommendations made.
Creating a conducive environment is necessary before conducting
a participatory learning. It is important to meet the stakeholders
with an open mind. The researchers have to tell them about the
[84]
motive of the study and the purpose for which the results are
used. It is important to establish a good report with the
stakeholders. It is also necessary to identify stakeholders who are
willing to share their experiences.

PLA usually starts off with a semi-structured approach. However,

it is necessary to make room for including unexpected or
unrelated issues or topics that may come up during the
discussions. A PLA covering relatively a few topics in a small
area normally takes between 10 to 30 days. However, a PLA with
a wider scope over a larger area can take several months.
Immediately after the fieldwork has been done, it is important to
write reports. A preliminary report should be made within a few
days after completing the fieldwork, and the final report should be
made available to all participants and the local institutions that
were involved.

Among different PLA techniques, correct sequencing is

important. The PLA team must decide the most appropriate
techniques to elicit the desired information. Depending on the
topic under investigation, the techniques can be combined in
different ways. Mapping exercises are good to begin with as they
involve several people, which stimulate much discussion and
enthusiasm. These also provide the appraisal team an overview of
the area. After the maps are drawn, it is a good idea to proceed
with transect walk, along with some of the people who have
constructed the map. As these local persons know the area clearly,
it gives much information. Seasonal calendars and timelines can
reveal changes and trends throughout a single year or over several
years. Ranking techniques such as wealth ranking that have some

[85]
sensitivity in divulging information, may be done only after
establishing good rapport with the people.

Brainstorming serves as a good introduction for more structured

and focused discussions at a later stage. Preference ranking is a
good opener at the beginning of a group interview and helps focus
the discussion. Subsequently, individual interviews can be
conducted on the preferences of local people already spelt out and
the reasons. At the end of the whole exercise, some clear cut ideas
should emerge.

On-farm verification trials

On-farm verification trials reflect a high level of confidence on

the part of the researcher that a new technology is technically
feasible and effective. It is often a straight comparison of new
technology and current farmer’s practice, usually done in large
plots.

Although an on-farm verification trial is a farmer-managed

practice, its implementation is done jointly by the farmers,
researchers, and extension workers. These stakeholders have
different roles and responsibilities in the implementation of the
trials. Researchers help farmers and extension workers by
providing on the spot training and technical backstopping.
Extension workers play their role in site and farmer selection,
supporting the farmers in their daily routines of field and crop
management. Farmers also play their roles by allocating
experimental plots and undertaking the activities of crop
management and protection.

[86]
Unlike on-farm verification trials for crops, on-farm verification
trials for livestock, researchers take the lead because of the
knowledge-intensive nature animal research. However, extension
workers and farmers actively participate in conducting g less
knowledge-intensive activities particularly at the initial stage of
the work. The necessary technological packages and animal
health care are provided by the research system but the input of
experimental animals may be provided by farmers as deemed
necessary. The day to day data collection activities are handled by
trained enumerators who are provided with technical
backstopping by researchers.

Meta-analysis
Meta-analysis is a statistical procedure for combining the results
from several previously made independent studies. This helps to
improve the dependability of the results. Meta-analysis gives a
more objective appraisal of the collected evidences by different
researchers than a traditional review of literature on the problem.
It is also helpful to provide a more precise estimate of a treatment
effect and explain heterogeneity between the results of individual
studies. Meta-analyses may help not only to mediate studies that
might otherwise seem to have conflicting results but also to
establish strategic implications of past studies.

6. Principles of Experimentation

The conclusion of any finding from research has to be verified

before it is accepted. That is why experiments are required.

[87]
The term ‘experiment’ originates from the Latin term Experiri,
which means, ‘to try’. Experiments are conducted to decide the
fate of two or more competing hypotheses or research questions
on a particular phenomenon. The knowledge that is generated
through experiments is empirical knowledge, which is based on
observation and experience.

Experiments may range from laboratory experiments to field

experiments. Laboratory experiments are done under highly
controlled conditions, and repeatable results are always found
regardless of how many times the experiments are repeated.
When doing field experiments, it is actually the laboratory
methods that are extended to field or real situations with some
precautions against uncontrollable variations as in most
agricultural experiments.

An experiment is a test designed and conducted by purposely

altering some variables with the intention of falsifying a
hypothesis involving selected variables. It involves the creation of
a situation in which the researcher manipulates one or more
independent variables and measures the outcome variables.
Experiments are the most powerful methods for establishing
causal relationships. A researcher conducting such a study
exercises greater control over the objects of study.

For establishing cause and effect relationships, a researcher has to

isolate and eliminate all the factors that might be responsible for a
particular outcome and test only one that are directly wanted to
measured. The researcher can control certain causes such as
fertilizer rate, herbicide rate, etc., but certain causes may occur
naturally or outside the researcher’s influence (such as flood,

[88]
drought, or pest attack). These also have to be controlled by the
researcher as much as possible.

Experimental methods are widely used in agriculture and life

sciences for determining cause and effect relationships. The
relationship or question is generally expressed as a statement of
hypothesis that must be subjected to falsification. In an
experiment, the experimental area is divided into experimental
plots or experimental units. The term ‘plot’ is synonymous with
‘experimental unit’. It refers to the unit of material to which a
treatment is applied. It can be a plot of land, a pot, an animal, a
person, a group of person, etc. The objects under comparison such
as variety, fertilizer rate, seed rate, feed types, etc. are assigned to
each unit.
An independent variable of a specific experimental interest
(treatment) is known as a factor. Normally, the term is used when
an experiment involves more than one variable. Level or rate
refers to the degree or intensity of a factor. There could be several
levels for a factor including a zero level. Another term often used
with experiments is blocks, which refers to categories of subjects
within a treatment group.

Blocking is common in agricultural field experiments to take care

of heterogeneity in experimental units (plots). Blocks are also
possible in other kind of experiments, for example, in a veterinary
study, test animal subjects may be divided into age groups old-
aged, middle-aged, and youngsters forming three blocks.

Variability is a distinguishing feature of biological materials.

Variations may occur among experimental plots, even if the same
treatment is applied to all plots. As a result, there could be
[89]
significant differences in the outcome such as yield. The
variations that occur among plots may exhibit a pattern. Thus,
most of the time, adjacent experimental plots or units produce
similar yields compared to distant plots. This implies that there is
heterogeneity (non-uniformity) in experimental plots. Therefore,
it is necessary to define whether the differences among
experimental units are due to real treatment effects or because of
other extraneous factors. Applying appropriate statistical tools can
enable a researcher to get over this difficulty.

There could be limitations in conducting experimental research

because of ethical issues. Experiments involving human beings
and animals need strict examination in terms of ethics. For
example, a psychologist might want to know, if the way parents
discipline their children has an effect on how their children
behave. Asking the parents to discipline their children in a certain
method would be neither practical nor ethical. Similarly, to see
the effect of smoking, drinking, alcohol, etc. on the health of
humans, it would be unethical to conduct experiments on these
matters.

6.1. Validity of experiments

A major limitation of experiments is that the conclusions and

inferences made out of the experiment should have a wide range
of validity or authenticity applicable to other settings and regions.
An experiment is valid only if observed changes in the dependent
variable are actually due to the independent variable and the
results of the experiment apply to the real world outside the
experimental setting such as a laboratory or a research field. If the
experiment in question fails to achieve these conditions, it is of no
[90]
value. The first condition that the observed changes in the
dependent variable are actually due to the independent variable is
related to internal validity, and the second condition that the
results of the experiment apply to the real world outside the
experimental setting is external validity.

An experiment is said to have high internal validity when the

factors of the treatment that have been manipulated (independent
variables) actually have a genuine effect on the observed or
dependent variables. In other words, the treatment alone should
make the difference rather than other extraneous factors. A study
with high internal validity permits conclusions on true cause and
effect relationships. After the experiment is completed, a
researcher is likely to make some claims that his/her research
findings have implications for other settings and at other times.
When a researcher makes such a claim, he/should examine the
external validity of these claims, that is, the extent to which the
findings of the study can be generalized. A researcher would
achieve nothing if his/her observed relationships between the
variables were valid only in the experimental setting!

While planning and conducting experiments, a researcher may

confront several threats to internal validity and external validity.
We may not be confident enough to say that a relationship exists
between the independent and dependent variables, if threats to
internal validity are high. Similarly, our confidence in stating
whether the results of the study are applicable to other settings
and groups is also compromised, if the threats to external validity
are not under control.

[91]
6.1.1. Threats to internal validity
The factors that affect internal validity of a research design are:
history, maturation, testing effects, instrumentation, sampling
bias, attrition, and selection bias.
History
While the experiment is under implementation, if some external
events other than the treatments such as drought, flood, or pest
incidence occur, these can influence the results, and internal
validity is at stake. These kinds of events during the period of
experimentation are usually denoted by the term ‘history’. By
including a control group from the same population as the
experimental group sharing the same history, we can reduce the
effects of history largely. The threat from history is relatively low
in experiments of short durations.
Maturation
Maturation refers to changes that can occur in the subjects for
over a period of time influencing the results of a research study.
Unlike history, effects of maturation are not the results of specific
events but occur because of ageing, changes in physical,
intellectual, or emotional characteristics that happen naturally
over time, etc. These effects are more problematic with respect to
human and animal subjects, because during the experimental
period, physical or psychological changes taking place within the
subjects can jeopardise the conclusions. As in the case of dealing
with the effects of history, the effects of maturation can be
overcome by using a control group chosen from the population
from which the experimental group is drawn. The effects of
maturation are likely to be low in short duration studies.

[92]
Testing effects
Testing effects is a threat to internal validity that happens due to
the effects of a previous test upon the performance after a second
test. For example, exposure to a pre-test usually influences
observations from a post-test. This threat is more obvious in
experiments involving human beings. If the intervening time
between pre-test and post-test is short, testing can become a
serious threat to internal validity. In most cases, the use of a
research design that does not include a pre-test can eliminate
testing as a potential threat to internal validity. However, if pre-
treatment data are required, the only option is to use data
collection techniques without revealing the actual intention to
participants.
Instrumentation
Changes in the calibration of instruments, questionnaires, or
interviewing technique can create problems to internal validity.
Standardized instruments, data collection procedures, and the
training of observers are some of the measures, which reduce
problems from instrumentation. Most threats from
instrumentation can be eliminated by specifying the measurement
procedures. Occasionally, the instruments being used for the
experiment have to be recalibrated, if their readings become
doubtful. For example, voltage fluctuations and temperature
variations can do havoc on data. However, by keeping these
factors constant during the course of investigation, these can be
prevented from influencing other variables. This stabilization
approach is the best for variables that have a disruptive influence,
for example, voltage or temperature variations. In many instances,
[93]
a researcher can use a standard sample for managing
simultaneously uncontrollable or unknown variables.
Sampling bias
A sampling bias occurs when the assignment of treatments or
subjects to experimental groups or plots is done in a way likely to
prejudice the outcomes. Avoiding using extreme scorers and
employing a random sampling method or assigning treatments to
experimental units randomly will minimize the threat due to
sampling bias.
Attrition
Attrition or mortality is the differential loss of individuals from
treatment or comparison groups, which may bias the results. Loss
of plants, animals, and human subjects are threats of attrition. In
experiments with human subjects, this is often a problem when
research participants are volunteers. They may drop out of the
study, if they find that it is consuming too much of their time or
the task is too arduous. To reduce mortality threats, a researcher
needs to select large groups and must make certain that they are
true representative of the population.
Selection bias
Selection bias refers to the possibility that differences exist
between individuals in treatment and control groups at the start of
the study and that those differences may affect the results. For
example, one group may differ from another group in age, ability,
gender, ethnic composition, or other ways. If the differences in
features of a group affect the result of a study, these amount to a
serious threat to internal validity. Internal validly threats due to

[94]
selection can be minimized by adopting random selection,
random assignment of subjects and blocking.
6.1.2. Threats to external validity
External validity is under threat whenever conditions inherent in
the research design are such that the generalizability of the results
to other settings or groups is limited. In general, a researcher can
reduce threats due to external validity by taking various
precautions to ensure that the selected sample, the environment,
and the background are representative of the population.

Threats to external validity of results may be related to the

population, the extent to which a sample is representative of the
population from which it was selected; or to the ecology, that is,
the characteristics of the context or setting of the study are
representative of the context and setting to which the results are to
be extended.

There are two types of external validity. These are population

validity and ecological validity.
Population validity
The question of population validity arises when the selected
sample is not a true representative of the population. The more
representative the sample, the more confident a researcher would
be in generalizing from the sample to the population. Another
question is how extensively the results can be applied. When a
particular research result works across many different kinds of
settings—even those not represented in the sample— external
validity is high.

[95]
Ecological validity
The extent to which the results of a study can be simulated from
the background environmental conditions of the study to other
environmental conditions is often called ecological validity. It is
possible that a researcher may have created a set of environmental
conditions for the research work and the results may be valid only
under those conditions. That is why plant breeders strive to prove
the appropriateness of their varieties (before release) for wide
adaptation in terms of performance consistency across a range of
physical environments, which is confirmed by spatial and
temporal replication of the test environments even if it is
laborious, costly, and time consuming. Threats to ecological
validity aggravate when the environmental background of the
study is not representative of the population.

6.2. Applying principles of experimentation

Both internal and external validities require careful consideration

while designing experiments. The influential ideas for
experimental designs applicable to laboratory and field
experiments in agriculture, medicine, other life sciences, and
social sciences can be traced back to Sir Ronald A. Fisher (1890–
1962). Fisher was a geneticist, who joined Rothamsted
Experimental Station in 1919. He was asked to do something on
the accumulated data over the years from agricultural
experiments. He systematically introduced three principles of
experimentation—randomization, replication, and local control or
blocking to take care of validity issues and broadened insights
into the applicability of analysis of variance (ANOVA). The
publication of Fisher’s Statistical Methods for Research Workers”
[96]
in 1925 and “The Design of Experiments” in 1935 gradually led
to the acceptance of the three principles of experimentation
worldwide. Therefore, one can say that the experimental designs
have their origin in agricultural research. Before Fisher’s
pioneering work came to limelight, most researchers used
systematic schemes rather than randomization to assign
experimental units to the levels of a treatment, often leading to
possible bias and inaccurate interpretation of the data.

Designing experiments is used in many other fields of application

in addition to agriculture. These fields include medicine,
psychology, and sociology. Since the majority of present-day
experimental designs originated from agricultural experiments, it
is not surprising to see several terms of agriculture such as
treatment, yield, plot, and block in experimental designs related to
other disciplines.

Fisher developed the concept of randomization based on small

samples by consulting William Sealy Gosset (known popularly as
‘Student’). The three principles that Fisher vigorously
championed—randomization, replication, and local control—
remain the foundation of good experimental design still now, and
they are considered as the major principles of experimentation to
increase both internal and external validity.
6.2.1. Randomization
Randomization prevents the effect of extraneous factors. This
means that the researcher should design or plan the experiment in
such a way that the variations caused by extraneous factors can all
be combined under the general heading of “chance.”

[97]
In experiments, assigning plots or subjects to various treatments
must be done in such a way that treatment and control groups are
comparable in all respects except for the treatment. For example,
if a researcher selectively assigns treatments to participants, plots,
or any other experimental units, personal bias may occur.
Therefore, randomisation of treatments is an important principle
to avoid personal bias, and it ensures that each treatment has an
equal chance of being assigned to any experimental plot.
For example, if a researcher grows one variety of teff, say, in the
first half of the parts of a field and we grow the other teff variety
in the other half, then it is just possible that the soil fertility status
may differ in the first half in comparison to the other half. If this
is so the results would not be truthful. In such a case, the
researcher may assign the variety of teff and grow it different
parts of the field on the basis of a random sampling technique.
Thus, the researcher may apply randomization principle and
protect the experiment against the effects of extraneous factors
(difference in soil fertility status). As a result, the researcher can
have a better estimate of the experimental error.
Randomization is also helpful to deal with uncontrolled variables.
When a researcher resorts to random selection of samples, he or
she can easily translate the biasing effects of uncontrolled
variables to random unbiased errors. For example, ‘time’ can
influence an experiment as equipment may show minor changes
because of the period factor. By randomizing the order of sample
measurements, the researcher can evade temporal influences of
this kind from potentially biasing the outcome.
In social science studies, groups can be created in two ways:
through matching and randomization. The subjects may be
[98]
matched according to characteristics such as family lineage, age,
weight, health condition, soil fertility levels, or some other
characteristics. This involves distributing these subjects so that
each subject in one group exactly matches characteristics of
subjects in other groups. However, it would be difficult to have
equal groupings; therefore, randomization is preferred to
matching. Random assignment makes sure that the experimental
groups are probabilistically alike at the time of assignment.
Randomization can be done with a table of random numbers or by
drawing lots.
In fact, randomization is based on the statistical principle of
normal distribution. If you subject the population data to a normal
distribution, most individuals will fall within the middle range of
values for a given characteristic with progressively fewer ones
towards either extreme (remember the bell curve of normal
distribution). Differences between groups will average out and
become more comparable.
6.2.2. Replication
Experiments should be repeated more than once to increase
reliability. The process of repeating the treatment more than once
is known as replication (a coined term from repetition and
duplication). Replication increases statistical accuracy and
precision. The treatment must be replicated at least twice to
estimate experimental error. However, too many replications are
costly, and beyond a certain limit, additional advantage due to
precision is not significant if you consider the additional cost
involved. The number of replications needed depends on the
magnitude of experimental error likely to be obtained in the
experiment and the degree of precision wanted. The experimental

[99]
error can be kept to the minimum by providing more degrees of
freedom for the experimental error.
Degrees of freedom is an interesting concept in the analysis of
variance. The number of degrees of freedom in a distribution is
the number of values that are independent of each other, which
cannot be deduced from each other. Generally, a rule of thumb is
followed to determine the number of replications. Choose the
number of replications to provide not less than 10 degrees of
freedom for estimating experimental error or, in other sources, 12
degrees of freedom as the lower limit, the latter being the most
commonly followed in most experiments. By following this rule
of thumb, with five treatments in a randomized block design, for
instance, a minimum of four replications are needed.
6.2.3. Blocking/local control
Blocking or local control is the method by which the extraneous
factor, the known source of variability, is made to vary
deliberately over as wide a range as necessary. This is done in
such a way that the variability it causes can be measured and
eliminated from the experimental error. This signifies that the
experiment should be planned in such a way that a two-way
analysis of variance is done. The total variability in the data is
divided into three components, which are attributed to the
treatment (varieties of teff in the above case described in the
Principle of Replication above), the extraneous factor (soil
fertility status in the same case indicated above), and
experimental error.
It was Sir Ronald Fisher who observed that by dividing the
experimental fields into different blocks, and then conducting
experiments, the data gathered and inferences drawn would be
[100]
more reliable. This process of blocking or local control helps the
researcher to separate out differences due to confounding
variables and by excluding them from the estimates of error
effects.
Blocks refer to categories of subjects within a treatment group. In
agricultural field experiments, variations between experimental
plots can be reduced largely through blocking. When the blocks
differ widely, there will be substantial reduction in experimental
error. However, plots within a block should be homogenous.
When the fertility pattern of an experimental field is known,
blocks can be oriented in such a way that soil differences between
blocks are maximized and those within blocks are minimized.
When fertility patterns are not known, avoid using long and
narrow blocks, but rather opt for compact blocks as nearly square,
since closer plots can be more alike than those that are farther
apart.
We can have blocks in experiments related to animals too.
Suppose we want to test two drugs. We might divide the diseased
animals into three blocks—old aged, middle aged, and young
ones; and further divide the groups into a group treated with drug
A and another treated with drug B. This arrangement allows us to
find out how the treatment affects different age groups. In the
example, we might observe that drug B out performs performed
drug medicine A except for older patients where drug A
outperformed drug B. This phenomenon is due to interaction
between the treatment (the drug) and subject characteristics (age).
Interaction between treatment factors can occur in factorial trials
when there are more than two factors.

[101]
6.2.4. Control in experiments
Normally, an experiment demonstrates that an event occurs after a
certain treatment is given, and that the event does not occur in the
absence of the treatment. By dividing the entire experimental
units into two groups, the ‘control group’ and the ‘experimental
group’, you can confirm the effect of a variable, even if many
unknown variables interfere. In such experiments, a ‘control
group’ is useful, especially when one must deal with several
uncontrolled variables. The two groups should be as similar as
possible except for the one aspect whose effect is being tested. In
this context, you should particularly note that a ‘controlled
experiment’ is not an experiment under control; it simply means
that the experiment is designed with the ‘experiment’ and the
‘control’ groups.

In agriculture, you resort to controlled experiments when there are

a number of alternative treatments. These are tested along with a
standard treatment (control). Each treatment is applied to each
experimental unit, and then several observations are made on each
unit. Enough precautions are taken to separate out experimental
and random variations from true variations. Experiments
conducted by observing adequate precautions and stipulations
help the researcher to separate out differences between the
treatments from the uncontrolled variation that is assumed to be
present. When planning for experiments, the treatments to be
included are a point of concern. The researcher should know very
well why these particular quantities for the treatments, and why
these combinations alone and not others have been fixed.
Take sufficient precautions so that the two samples—
experimental and control— are identical in all important respects.
[102]
The dependent variable is measured in both groups for an
identical period before the treatment. The treatment is then
introduced into the experimental group only, and the dependent
variable is measured in both groups for an identical period after
introducing the treatment. The treatment effect is determined by
measuring differences in the dependent variable in the control
group and experimental group.

Many researchers believe that for all kind of experiments, a

‘control’ must be included. ‘Control’ need not be insisted in all
the experiments, but examine each case individually and decide
whether a control is essential.
Unlike field experiments, in which standard checks are used,
concepts of precision and accuracy are applied in laboratory-
based experiments as a reference or controlling mechanism
towards ensuring reliability of measurements. Accuracy, in this
sense, represents how close those sample measurements are to the
true measurement. Precision, on the other hand, represents the
closeness to one another of the measurements of duplicate,
triplicate, etc. samples coming in one particular batch. This
implies that data may be accurate but not precise, or precise but
not accurate.
In laboratory experiments, a standard sample of known value is
used as a reference for comparing with the test treatments,
ensuring the accuracy of measurements, and controlling batch
effects. Any deviation in the measurements of two standard
samples that lies within 5% is usually considered tolerable. When
the deviation exceeds 5%, the whole measurement in that
particular batch should be repeated. Similarly, laboratories can
establish their own local standards by running repeated
[103]
measurements of particular samples and setting tolerance
thresholds for potential use as a local reference instead of global
standard references.
Furthermore, to ensure accuracy of measurement, it is also
important the equipment of the measurement is properly
calibrated. A value of a standard sample is usually set by a
certified lab globally, regionally or nationally. To control
reliability of lab results, proficiency tests or ring tests are usually
carried out at different levels i.e. globally, continentally,
regionally or nationally. Participation in these tests is of great
help for ensuring accuracy of lab results.
Likewise, precision/accuracy of lab-based measurements is also
ensured by applying known international laboratory standards,
commonly known as the International Standard Organization
(ISO) such as ISO 5725.
Although most scientists decide for the need of a control,
understand that there are experiments where a control can be
harmful! For example, if there is already an effective cure for a
disease, it is unethical to conduct an experiment comparing a new
cure with a negative ‘control’, i.e. no treatment particularly when
we are dealing with humans. In this case, the treatments ought to
be the new cure and the one currently in use, instead of a ‘no
treatment’ control. Similarly, in a trial with some new pesticides,
if there is ‘no treatment’ control, then the pest may multiply on
those plots and spread to other fields.

[104]
7. Classification of Experiments

Hypothesis experiments
Hypothesis testing experiments are generally classified as true
experiments, quasi-experiments, and ex post facto experiments
based on the applicability of randomization and manipulation of
treatments.
True experiments
True experiments are characterized by three main features,
namely, randomly formed groups, manipulation of treatments and
measurement of dependent variables.

Random assignment is the most feasible practical strategy to

ensure that groups are statistically equivalent. Note that random
assignment is different from random sampling. Random sampling
is one method for selecting subjects to participate in a study,
whereas random assignment is a method for assigning subjects
such as people, animals, or plants to a treatment group, but not a
method for selecting the study subjects. A true experiment
requires randomly formed groups. However, there are cases when
one could use convenience sampling to select study participants.

The second requirement of a true experiment is manipulation of

treatments, which means that the investigator can determine who
or what receives what treatment. If the researcher decides who or
what gets what, then manipulation has occurred. The third
requirement is that groups are compared and the dependent
variables are measured. In most experiments, at least two groups
are compared by measuring the dependent variable. In agriculture
and life sciences, true experiments are often grouped under three
[105]
categories—laboratory experiments, pot culture experiments, and
field experiments.
Laboratory experiments
In most research and academic institutions, there are a number of
laboratories engaged in soil chemistry, crop quality, animal feeds
and nutrition, animal health, food science and nutrition, crop
protection, biotechnology, etc. Laboratory-based
studies/experiments are either the basis or supplement to field
level experiments and they deserve due attention in planning,
implementation, data analysis, and archiving of the results.
Laboratory experiments are carried out under controlled
conditions, and chances of errors are comparatively less. In such
experiments, one or more independent variables are deliberately
varied and measurements on the dependent variables are taken in
an artificial setting, which could eliminate all the possible
extraneous variables. Laboratory experiments are commonly
employed for basic studies in disciplines such as microbiology,
physics, chemistry, biochemistry, physiology, biotechnology, and
other science disciplines. However, laboratory experiments with
crops are rare in agriculture as growing of crops under artificial
conditions in a laboratory has only limited practical value.
Pot culture experiments
Pot culture experiments are conducted under controlled
conditions but not so strictly as in under pure laboratory
conditions. Thus, a pot culture experiment may face greater
unwanted variations than laboratory experiments. However, they
are extremely useful in solving fundamental problems in
agriculture. In general, pot culture studies alone or in conjunction

[106]
with laboratory studies are common for basic and some applied
studies. Pot culture experiments are most often used in certain
disciplines such as plant pathology, soil science, plant physiology,
horticulture, etc. A negative result obtained from a pot culture
experiment would, in all probability, be negative in field trails
also, while a positive result need not necessarily be so.

Field experiments
Field experiments are routine in agricultural sciences in which
independent variables are manipulated and measurements are
made on the dependent variable from the experimental units or
plots in their natural setting. Using modern techniques and
application of appropriate statistical methods, field experiments
have the potential to show significant results. The heterogeneity
experienced in the fields can be eliminated largely by suitable
experimental designs, layouts, and statistical analysis.

Quasi-experiments
Unlike true experiments, in a quasi-experiment, groups of
subjects are formed not based on random assignment but by non-
random methods. Quasi experiments are just like a true
experiment with the only difference of missing randomly assigned
groups. Quasi-experimental design serves in place of true
experiments in situations where the latter cannot be applied for
ethical or practical reasons. For instance, when using human
subjects, it is often impossible to do random assignment,
especially when they are part of intact groups such as schools,
colleges, communities, neighbourhoods, etc. Although groups
might be reasonably similar in a practical sense, using data from
intact groups limits the conclusions that can be drawn concerning
the treatment effects.
[107]
Ex Post Facto experiments
This type of experiment is used in situations wherein the
researcher devises an experiment to examine the effect of a
treatment after it has already taken place. Ex post facto means
‘after the fact’. Thus, the causal event of interest has already
happened without the manipulation of treatments by the
researcher. These kinds of experiments are useful when human
subjects or animals in true situations are involved, and the
researcher comes to the scene only ‘after the fact’. Therefore,
most ex post facto studies are observational studies as no
manipulation of treatment is possible. In such situations, the
investigator begins the investigation by identifying the outcome
variables and attempts to identify possible causes, which is
independent variables responsible for the outcome. As an
example, take the case of two groups of animals in a vet clinic,
one treated with a particular medicine and the other with an
alternate medicine. If the researcher is deliberately selecting the
patients and giving the medicines randomly, then it would be a
true experiment. On the other hand, if the researcher has no role
in medication and she/he just takes observations after somebody
else gave the medicines, it would be an ex post facto study.
Quasi-experimental research and ex post-facto are commonly
used in health and social science fields of study.

[108]
 SECTION III. PLANNING RESEARCH

8. Research Designs

A research design is the planning of conditions for collecting and

analysing data in a way that combine relevance to the research
purpose and economy of procedure. Research design required for
facilitating smooth conduct the various research operations,
thereby making the research as efficient as possible and
generating maximal information with minimal expenditure of
effort, time, and money. It simply constitutes the plan of
collecting, measuring, and analysing data. The planning of
research design and its execution with utmost sincerity and
accuracy will lead to valid inference and generalization about the
subject of the research undertaken.
A research design involves different steps that include what the
researcher will do from formulating the hypothesis to the final
analysis of data. Specifically, designing research includes the
following aspects for consideration: what the study is about, why
the study is conducted, where the study will be conducted, the
type of data required, where the required data can be found, the
duration of the study, the sampling design, the techniques of data
collection to be used, how to analyze the data, and how the report
will be prepared.
Research design is often divided into the following sections:

(a) The sampling design, which deals with the method of

selecting items to be observed

[109]
 (b) The observational design, which relates to the conditions
under which the observations are to be made
(c) The statistical design, which deals with the question of
how many items are to be observed and how the
information and data gathered are to be analyzed;
(d) The operational design, which deals with the techniques
by which the procedures specified in the sampling,
statistical and observational designs can be carried out.

Based on the above information, the following characteristics

of a research design have been identified:

 It is a plan that indicates the sources and type data

relevant to address the research problem
 It shows the approach will be used for gathering and
analyzing the data
 It shows the time and cost budget required to
implement the research process

Thus, a research design should contain at least the following

points: a clear statement of the research problem, procedures and
techniques to be used for gathering data, the population to be
studied, and methods to be used for processing and analyzing the
data.
A good research design is the one that enables a researcher to
attain the objective of his/her research in the best possible way.
Ensuring objectivity, reliability, validity, and generalization are
the main characteristic features of a good research design. The
major types of research design may differ based on the type of
research to be conducted. A researcher must decide upon the

[110]
proper type of research design, which is suitable for meeting the
objective of the research. In doing so, a researcher must keep in
mind the type of data required and its collection and analysis, so
as to extract the hidden information from the subject area of the
research.
A good research design is often the one that minimizes bias and
maximizes reliability of the data collected and analyzed. The
design which gives the smallest experimental error is considered
the best design in many investigations. In the same way, a
research design which leads to the generation of maximal
information and provides an opportunity for handling many
different aspects of the research problem is regarded as the most
appropriate and efficient design. Consequently, the question of a
good design is related to the objective of the research and type of
problem to be studied. In this connection, a design may be
suitable for researching one problem, but may not be suitable in
one respect or other in the context of some other research
problem. Therefore, one single design cannot serve the purpose of
all types of research problems. As a result, for choosing a
research design appropriate for a particular research problem, the
following factors should be considered:

(a) The means of obtaining information

(b) Availability and skills of the researcher and his/her staff
(c) The objective of studying the problem
(d) The nature of the problem to be studied
(e) Availability of time and money for the research work to be
done.

[111]
Accordingly, for exploratory or formulative research, in which the
major aim is discovering new ideas and insights, the research
design that is appropriate must be flexible enough to consider a
number of different aspects of a phenomenon. However, if the
purpose of the research describing a situation or an association
between variables, which is common with descriptive research,
accuracy is the main consideration. For such research the most
appropriate design is the one that minimizes bias and maximizes
the reliability of the evidence collected. For research dealing with
testing of a hypothesis to reveal a causal relationship between
variables, the most appropriate design is the one that permits
inferences about causality in addition to the minimizing bias and
maximizing reliability.

However, in practice, categorizing a particular study under a

particular type of research is sometimes difficult since a given
research type may have elements of two or more of the functions
of different types of research. Therefore, it is only on the basis of
its primary function that a study can be categorized either as an
exploratory or descriptive or hypothesis-testing type of research.
It is in accordance with this fact that a research design is chosen
for a particular type of study. Besides, the availability of time,
money, skills of the research staff and the means of obtaining the
information must be given due considerations when selecting the
relevant type of research design such as experimental design,
survey design, sample design, etc.

There are different commonly used research designs. These

include research design for exploratory research, research design
for descriptive and diagnostic research, and research design for
hypothesis-testing research.
[112]
8.1. Research design for exploratory research
Exploratory research is also known as formulative research. The
main purpose of such research is formulating a problem for more
precise investigation or developing working hypotheses for
conducting further research on a situation or a phenomenon. The
main thrust of such designs is the discovery of ideas and theories
that are mostly unexplored. Exploratory research can also be
viewed as a study meant to enhance knowledge
Therefore, the research design that fits exploratory research must
be flexible enough to provide an opportunity for considering
different aspects of the problem under investigation and
investigate it with an open mind. This is because the research
problem under such a study, which is broadly defined at the
beginning, is gradually transformed into one with more precise
meaning. This may necessitate modifications in the research
procedure for gathering relevant data. As a whole, there are three
methods available in the context of research design for
exploratory research. These include survey of literature, survey of
experience, and analyzing of ‘insight-stimulating’ examples.
Surveying literature is the process of gathering knowledge from
past works in the concerned field of study. It is the most fruitful
and simple method for identifying a research problem or
formulating a hypothesis or hypotheses. Under this method,
hypotheses stated or formulated by previous workers may be
reviewed and evaluated for their usefulness on the basis of the
future research outlook. In the process, a lot of efforts, time,
money, and energy could be saved if the researcher is well aware
of the facts of what has been done in the past with respect to the
proposed area under study. Thus, the researcher should review
[113]
and build upon the research already done by others. However, if
hypotheses have not yet been formulated, the researcher has to
review the available material for deriving pertinent hypotheses.
In addition to conducting a survey, conducting formal or informal
discussions with experts, researchers, administrators, and
professionals may provide effective clues for formulating
problems.

Experience survey refers to the survey of people who have had

practical experience with the research problem. The purpose of
such a survey is to get insights into the relationships between
variables and new ideas relating to the research problem. For such
a survey people who are knowledgeable and can add new ideas
may be selected prudently as respondents to represent different
types of experience. The respondents may then be interviewed by
the investigator. But the interview must ensure flexibility so that
the respondents should be allowed to raise issues and questions
which the researcher has not previously contemplated. It is often
advisable to send a copy of the questionnaire to the respondents
(experts, administrators, group of affected people, researchers,
professionals) prior to the interview schedule. This advance
preparation will give the respondents an opportunity for thinking
before hand over the various issues to be raised for effective
responses. Thus, an experience survey may give the investigator
the opportunity to define the problem more concisely and
formulate the research hypothesis clearly. The survey may as well
provide evidence regarding concrete possibilities for conducting
different types of research.
Analysis of ‘insight-stimulating’ examples is also a productive
approach for formulating hypotheses for research. It is especially
[114]
useful for topics on which there is little accumulated experience
to serve as a guide. This method is meant for intensive study of
selected instances of the phenomenon in which one is interested.
To this end, the existing records, if any, may be examined, the
unstructured interviewing may be conducted, or some other
methods may be used. Attitude of the researcher, the intensity of
the study, and the ability of the researcher combine diverse
information into an integrated interpretation are the major
characteristics which make this method an appropriate procedure
for evoking insights.
A few examples of ‘insight-stimulating’ cases include reactions of
strangers, marginal individuals, individuals from different social
strata, etc. In general, sharp contrasts or striking features are
considered relatively more useful while adopting this method of
hypotheses formulation. Thus, in an exploratory or formulative
research, whatever method or research design is used adopted, the
important aspect is that the method should remain flexible. This is
important to investigate a number of varying features of a
research problem may be considered by the researcher whenever
they arise.
8.2. Research design for descriptive research

Descriptive research is the type of research concerned with

describing the characteristics of a particular individual, or of a
group or a phenomenon or a situation. Thus, studies concerned
with specific predictions, with narration of facts, and
characteristics concerning individuals, groups, or situations or
phenomena are descriptive research. Most of the social science
research comes under this category.

[115]
In terms of research design, the two types of research share
common features. In both types of research, the researcher should
define clearly the data to be measured and the methods for
measuring the data together with a clear definition of the
‘population’ he/she wants to study. Under this type of research,
the objective is to obtain thorough and accurate information.
However, the research design to be used and the procedure to be
followed have to be carefully planned. The research design must
guarantee protection against bias and maximize reliability, and
economical completion of the study. Thus, it should not be
flexible and must be focused on the following six requirements:
formulating the objective of the study (what the study is about
and why is it being made?), designing the methods of data
collection (what techniques of gathering data will be adopted?),
selecting the sample (how much material will be needed?),
collecting the data (where can the required data be found and with
what time period should the data be related?), processing and
analyzing the data, and reporting the findings.

8.3. Research design for hypothesis-testing

research
A hypothesis-testing type of research is generally concerned with
studies where a researcher wants to verify the hypothesis with
respect to one or more objectives of the study. This type of design
my concern both experimental research and diagnostic type of
research. Hypothesis-testing research (generally known as
experimental study) is the type of study in which hypotheses or
causal relationships between variables are tested.

[116]
Experimental research requires procedures that will not only
minimize bias and increase reliability, but also will permit
drawing inferences. Experimental research meets this
requirement. Thus, research design for such research often mean
the design of experiments.
However, in the diagnostic type of research also, there is a scope
for testing the hypothesis, and diagnosing the causal relationship
among the variables in a research design.
Thus, hypothesis-testing type of design can broadly be
categorized under: research design for diagnostic studies, and
research design for experimental studies.
8.4. Research design for diagnostic studies
Diagnostic research is the type of research that determines the
frequency with which something occurs or it is associated with
something else. Thus, studies conducted to ascertain whether
certain variables are associated are examples of diagnostic
research. In this type of research, one or many hypotheses relating
to the concerned population are tested. In the diagnostic type of
research design, generally the observations are recorded from the
existing population; an association between the variables and the
related hypotheses is tested. A diagnostic type of study deals with
problems as well as solutions. The main objective is to diagnose
the problem accurately to work out for some solutions and, in the
process, to find out the relationship among the variables
associated.

[117]
8.5. Research designs for experimental studies
In an experimental type of studies, a researcher wants to test the
hypothesis with respect to the populations through
experimentation where the independent variable is varied or
manipulated and the response of the dependent variable is
measured. Experimental designs, in this case, may be of two
types: (a) informal and (b) formal.
8.5.1. Informal experimental research designs
Informal research designs are based on or associated with a less
sophisticated form of analysis. Informal experimental design can
be of three categories: before and after without a control design,
after only with a control design, and before and after with a
control design, after only with a control design, and before and
after with a control design.
The after-only design without control
In ‘after-only’ experimental designs without control or post-test
only without control, data are collected after the subjects have
been exposed to the independent variable. In this design, baseline
data are created depending on the respondents’ recall of the
situation before the intervention or from existing records.
Changes in the dependent variable are reported as the difference
between the baseline data and the observations made ‘after’ the
intervention. This type of design is adopted when you do not have
the time to take ‘before’ observations directly. For example, if
you want to study the impact of a campaign by the government,
which must be implemented as a contingency measure, probably
you may employ this method. In this case, baseline data are
created from the records, and any change in trends is assumed to
be because of the new programme. The main problem with after-
[118]
only designs is that they do not have any control over other
confounding factors that could have affected the post-test
observations.
The after-only design with control group
In this type of experiment, the experimental groups and the
control groups are evaluated and compared only after the
implementation of an intervention. The ‘after only design with
control group’ is also sometimes called ‘post-test only control
group’ design, wherein subjects are randomly assigned to either a
control group or an experimental group. The groups are not pre-
tested. One group is exposed to a treatment or a series of different
treatments, and then both groups are post-tested. Thus, two
groups or areas (test area and control area) are selected and the
treatments introduced into the test area only. The dependent
variable is then measured in both groups or areas at the same
time. The effect of the treatment is evaluated by subtracting the
value of the dependent variable in the control group (area) from
its value in the test group (area). The differences of the values are
subjected to statistical analysis to determine treatment effects.
Most of the comparative experiments being taken up in
agriculture fall under this category.
The fundamental basic postulation in this design is that the two
groups or areas are identical with respect to their behaviour
towards the situation in question. If this postulation is not true,
there is the possibility of extraneous variation occurring in the
treatment effect. However, unlike the before-and-after without
control design, data can be collected in this design with no
occurrence of variations due to extraneous variables with the
passage of time.

[119]
The before-and-after design without control
This type of experiment is also known as the pre-test/post-test
without control. In this type, a single experimental group is
chosen and the impact of the dependent variable on it is measured
before introducing the treatments. After the treatments are
introduced, the dependent variable is again measured and the
difference between the pre-test values and post-test values are
estimated. The before-and-after design is an upgraded version
compared to the after-only design, because the researcher is
considering the probable effect of the independent variable by
observing differences between the observations on the dependent
variable before and after the test. The effect of the treatment is
equal to the level of the situation after the treatment minus the
level of the situation before the treatment. Thus, this kind of study
normally includes two sets of cross-sectional data collection
points on the same population to find out the change in the
phenomenon or variable between two points in time, the second
being undertaken after a certain period.
The before-and-after design without control is considered as the
most suitable design for measuring the impact of a programme or
activity. It is also one of the most common designs in evaluation
studies. We get the overall impact of introducing the treatments
from the differences in observation before the treatment and after
the treatment.
The main downside of this design is that as time elapses,
considerable extraneous variations may occur in the treatment
effect.

[120]
The before-and-after design with control group
In this type of experiments (also called pre-test/post-test control
group design), two test groups are established; an experimental
group, which would be exposed to an independent variable and a
control group, which would not be subjected to the variable under
study. Both groups are pre-tested. After exposing the
experimental group to a treatment or different treatments, both
groups are post-tested and the dependent variable is measured for
an identical time-period. The treatment effect is determined by
subtracting the change in the dependent variable in the control
group from the change in the dependent variable in the
experimental group. The before-and-after design with control is
designed based on the assumption that the confounding factors, if
any, would affect equally the experimental and control groups.
Therefore, any differences in the data collected from the two
groups can be credited to the impact of the treatment alone.
This design is superior to the above-mentioned three designs
because it avoids extraneous variations occurring because of
passage of time and from non-comparability of the test and
control areas. However, at times, due to lack of historical data,
time or a comparable control area, a researcher may have to prefer
to select one of the first three informal designs stated above.
8.5.2. Formal experimental designs
Compared to informal experimental design, the researcher has
more control over the experiments in the formal experimental
design. Also more advanced statistical tools can be used for the
analysis of the data from the experiments. Formal experiments are
of two types: laboratory experiments and field experiments.

[121]
Formal experimental designs deal with the planning and conduct
of experiments including the system of assigning the selected
treatments to the experimental plots and analysing the generated
data to obtain valid conclusions. The main objective of an
experimental design is to establish a causal relationship between
the independent and dependent variables. Another major goal is to
extract the maximum information with the minimum utilization of
resources. A good experimental design allows valid comparisons
between treatments. In the case of agricultural field experiments,
soil heterogeneity is the main source of variation. The selected
experimental design must balance the requirements and
limitations of the problem in which one works so that the
experiment can provide the best conclusion about the hypothesis
being tested. An experimental design recognizes the independent,
dependent, and confounding or external variables, and specifies
the manner in which the principles of experimentation and
statistical aspects of an experiment are taken up.

In these types of design, mostly the techniques of analysis of

variance are applied. Depending upon the situation and the
objective of the study, the analysis of data arising out of
experimental designs is taken up as per one-way analysis of
variance, two-way analysis of variance, analysis of covariance,
and so on. The analysis of variance is based on certain
assumptions. So the validity of the result is not independent of
these assumptions. The experimental designs may be one-factor,
two-factor, or more-factor experiments, depending upon the
number of factors put under the experimentation with different
levels of each factor.

[122]
Whatever the number of factors may be, these designs are based
on certain basic principles and follow three basic designs. The
principles of designs of experiments are replication,
randomization, and local control.

Replication is the repeated application of the treatment among the

experimental units. It helps in the estimation of the effects of
different components of the experiments. Along with
randomization, it leads to valid estimation and, along with local
control, it leads to reductions in experimental error.
Randomization is the technique of unbiased allocation of
treatment among the experimental units. Randomization along
with replication provides valid estimations of the component
experiment. It reduces bias on the part of the experimenter and
others.

Local control or blocking is the technique by virtue of which

experimental error is reduced. The above three basic principles
form the backbone of the experimental designs.
Single factor experiments
An experiment in which only a single factor is varied while all
other factors are maintained uniform are called single factor
experiments. The treatments selected include different levels of a
single factor. For example, suppose that you want to compare a
group of crop varieties for yield, where the single variable factor
is the variety and the factor levels (i.e. treatments) are different
varieties. In such experiments, only the variety varies in the
experimental plots. The management practices such as fertilize
application, irrigation, weed management, etc., are applied
uniformly to the experimental plots. Examples of other single
[123]
factor experiments include testing several rates of a single
fertiliser element; insecticide trials, where several insecticides are
tested; and plant population trials, where several plant densities
are tested, etc.

For dealing with single factor experiments, two groups of

experimental designs are used: complete block designs and
incomplete block designs. Complete block designs are used for
experiments with a small number of treatments and is
characterized by blocks, each of which contains at least one set of
treatments. Completely randomised design (CRD), randomised
block design (RBD), and Latin square designs (LSD) are the three
most commonly used complete block designs. Complete block
designs are often used for simple experiments with relatively
small numbers of treatments.

The three basic designs are completely randomized designs,

randomized block designs (randomized complete block design),
and Latin square design.
Completely randomized design (CRD)
The completely randomized design (CRD) is the simplest of all
designs where only two principles of design of the experiments,
that is, replication and randomization, are used. A completely
randomized design (CRD) is one where the treatments are
assigned completely at random so that each experimental unit has
the same chance of receiving any one treatment. The principle of
local control is not used in this design as the experimental units
are assumed homogeneous and, as such, a one-way analysis of
variance is adopted in analyzing the data.

[124]
It is assumed that the differences observed between experimental
units receiving the same treatment, if any, are due to experimental
errors and this error must be kept to the minimum.

One needs as many experimental units as the sum of the number

of replications of all the treatments. If we have five treatments A,
B, C, D, and E replicated 4 times, respectively, then according to
this design, we require the whole experimental area to be divided
into 20 experimental units of equal size or 20 homogeneous
experimental units. Thus, a completely randomized design is
applicable only when the experimental unit is homogeneous in
nature. Under laboratory conditions, where other conditions
including the environmental condition are controlled and all
experimental units are mostly homogeneous in nature, completely
randomized design is the most accepted and widely used design.

Therefore, CRD can be used for experiments with homogenous

experimental units such as laboratory experiments and pot culture
experiments, where near uniformity can be ensured and
extraneous factor effects are relatively easy to manage. For field
experiments in agriculture, however, CRD is not normally used
because of field heterogeneity. In clinical or epidemiological
studies, the experimental units are human subjects, and an
experiment with completely randomized design is referred to as
randomized controlled trials (RCTs).
One advantage of this design is that it is the simplest of all
experimental design and flexible in adopting different numbers of
replications for different treatments. The CRD is a design that
accommodates any number of treatments and any number of
replications within a limit. It has also maximum error degree of

[125]
freedom, t (r-1) and no other experimental design provide higher
error degree of freedom with similar number of plots and
treatments.

In a practical situation, it is very useful because sometimes

researchers come across the problem of differential availability of
experimental materials. Sometimes, response from particular
experimental unit(s) may not be available; even then, the data can
be analysed with the help of the CRD design.

On the other hand, the disadvantage of this design is that the basic
assumption of homogeneity of experimental units, particularly
under field condition, is rare. That is why this design is suitable
mostly in laboratory condition or greenhouse condition rather
than under field conditions.

The second disadvantage of this design is that the principle of

“local control”, which is very efficient in reducing the
experimental error, cannot be used. Thus, the experimental error
in CRD is comparatively higher. As a result, with increase in the
number of treatments especially under field condition, it becomes
very difficult to use this design, because of the need to obtain
more number of homogeneous experimental units.
Randomized complete block design (RCBD)
Randomized complete block design (RCBD) or simply
randomized block design is a widely used design in field
experiments in agriculture because of its simplicity and
flexibility. A unique feature of RCBD is the presence of blocks of
equal size each of which contains all the treatments.

[126]
 It is the most widely used design that takes into account
variability among the experimental units. A randomized block
design uses all the three basic principles of experimental design
explained above, namely replication, randomization, and local
control.
For example, when conducting field trials to compare fertilizers,
plant varieties, or whatever, there is concern that some parts of a
field may be more fertile than others. So, if one were comparing
three fertilizers, say, it would not be a good idea to use one
fertilizer here, another fertilizer over there, and the third fertilizer
way out back because the effects of the fertilizers would be
confounded with the natural fertility of the land.
The design takes account of the heterogeneity of an experimental
unit in one direction. In RCBD, the whole variation is partitioned
mainly into three components: (a) factors that are varied by the
experimenter at his/her own wish, (b) variability in experimental
units, and (c) variability in extraneous factors, i.e. the
experimental error.

In a randomized complete block design, the whole experimental

area is divided into a number of homogeneous block/groups, and
each block/group consists of as many experimental units as the
number of treatments. Blocking/grouping is done in such a way
that the variation among the experimental units within the
block/group is minimum (homogeneous) and the variation among
the experimental units of different blocks/groups is maximum
(heterogeneous).
The advantages of this design includes that it is the simplest of all
block designs; it has a very simple layout; all the three principles

[127]
of design of experiments are given due consideration and it is
more efficient than CRD. In addition, there is no restriction on the
number of treatments or replicates. There will be no problem if
some treatments are replicated more times than others.
Furthermore, missing plots can also be easily estimated.

The disadvantage of this design is that the number of treatments

cannot be very large. If the number of treatments is very large,
then it is very difficult to have a greater homogeneous
block/group to accommodate all the treatments. In practice, with a
large number of treatments in RCBD, the possibility of
experimental error would be high unless the plot size (for crop
research) is reasonably small to maintain intra-block
homogeneity. In addition, unlike CRD, flexibility of using
variable replications for different treatments is not possible and
missing observation, if any, is to be estimated first and then
analysis of data is to be undertaken.
Latin square design (LSD)
A Latin square design, sometimes called row-column design, is a
method of placing treatments so that they appear in a balanced
fashion within a square block or field. This design is useful
particularly in some experimental areas where that soil
heterogeneity varies in two perpendicular directions, that is, north
to south and east to west or south to north and west to east. It can
also be used under conditions where individual units can be
grouped based on the two variable characteristics.
The Latin square design is more widely used in livestock research
than in crop research. This is because there is more need to block
two directional variations in experimental animals; say initial
weight and breed, initial weight and age, and age and breed, as
[128]
illustrated elsewhere in this Handbook with mathematical
examples.

This means that LSD has the capacity to handle two sources of
variation independently from the blocking criterion as in the case
of RCBD. Treatments appear once in each row and once in each
column. Replicates are also included in this design. Because of
this type of allocation of treatments, the total variations among
the experimental units are partitioned into different sources,
namely, row, column, treatments, and errors. In this design, it is
assumed that there is no interaction effect between row, column,
and treatment factors. If the number of treatments is also equal to
the number of replications r for each treatment, then the total
number of experimental units needed for this design is t  t.
One advantage of the Latin square design is the improvement
over the other two basic designs, CRD and RCBD, as it takes care
of the heterogeneity or the two-way variations in the experimental
units. Sometimes, when an experimenter does not have an idea
about the heterogeneity among the experimental units or does not
have the time to check the heterogeneity pattern, he/she can opt
for the LSD design. On the other hand, the disadvantages include
the rigidity of requiring that the number of treatments be equal to
the number of rows and columns as well as the fact that the layout
of the design is not as simple as in the case of CRD or RCBD.
Crossover design
The residual effect of a treatment may continue even after the
application of another succeeding treatment if the same
experimental unit is sequentially used again and again. A
crossover design, where the experimental units cross over from
one treatment to another during the course of the trial periods,
[129]
provides a technical solution to the carryover effects of previous
treatments with better efficiency compared to a parallel design or
non-crossover design where experimental units receive separate
treatments.
In livestock research (e.g. animal nutritional studies), where the
price of experimental animals is high and the number of animals
available for different experiments is limited, the same animals
may be used repeatedly for different experiments. The problem
associated with such an approach is that it makes difficult to
prove if the differences in performances between the animals is
due to a carryover effect of the previous study or if it is due to the
actual treatment effect of the follow up study. Therefore, reusing
the experimental animals is possible only with certain pre-
conditioning treatments to remove the “carryover effects” from
the treatments given in a previous study. The preconditioning
treatment is necessary to avoid the confounding effect the initial
treatment on the effect the new treatment. Thus, one practical way
to minimize carryover effects is providing a sufficient resting or
wash-out period between two treatment periods. The other is the
use of counterbalancing of treatments for each animal when
treatments are assigned to the experimental animals, i.e. each
treatment should follow each of the other treatments an equal
number of times.
Augmented design
These designs may be minimum variance designs for estimating
contrasts of check effects, of new variety effects, of new variety
versus checks, or of all check and new varieties.
In augmented designs, the goal is to compare existing (control)
treatments with new treatments that have an experimental
[130]
constraint of "limited replication". To understand limited
replication, think about experiments that may only allow a single
representation of the new treatment, this limitation may be due to
the cost associated with the experiment, limited resources, or
limited number of new units that can be used in the experiment.
In contrast, the existing treatments, referred as checks, are
generally replicated multiple times. With augmented design one
can estimate the following:
a) Differences between checks and new treatments,
b) Differences among new treatments,
c) Differences among check treatments, and
d) Differences among new and check treatments combined.
Graeco-Latin square design
As mentioned earlier, the Latin square design is useful to block
two directional variations in the experimental unit (double
blocking) into rows and columns, and then each of the treatments
(designated in Latin letters) are randomly allocated to the
experimental units only once in each row and in each column.
The Graeco-Latin square design, as a modification of the Latin
square design, consisting of two Latin square designs (one using
the letters A, B, C, etc. and the other using Greek letters α, β, χ,
etc.). When the two Latin squares are superimposed on each
other, in such a way that each treatment (Latin letter) appears just
once in each row, once in each column and once with each Greek
letter (triple blocking) the Graeco-Latin square design is formed
as illustrated below in Table 6. The two Latin square designs
become orthogonal as every Latin letter coincides exactly once
with a Greek letter.
[131]
Table 6. Schematic format of the Graeco-Latin square design

Row Column
C1 C2 C3 C4 C5 C6 C7
R1 A B C D E F G
R2 B C D E F G A
R3 C D E F G A B
R4 D E F G A B C
R5 E F G A B C D
R6 F G A B C D E
R7 G A B C D E F

Row Column
C1 C2 C3 C4 C5 C6 C7
R1 α  β  χ  
R2 β  χ   α 
R3 Χ   α  β 
R4  α  Β  χ 
R5  β  Χ   α
R6  Χ   α  β
R7   α  β  χ

Row Column
C1 C2 C3 C4 C5 C6 C7
R1 Aα B Cβ D Eχ F G
R2 Bβ C Dχ E F Gα A
R3 Cχ D E Fα G Aβ B
R4 D Eα F Gβ A Bχ C
R5 E Fβ G Aχ B C Dα
R6 F Gχ A B Cα D Eβ
R7 G A Bα C Dβ E Fχ

In livestock, for instance, we can conduct an experiment in which

different diets (designated in Latin letters), breeds of animals
[132]
(columns), weight groups (rows) and feed concentrations (Greek
letters) using the Graeco-Latin square design. Graeco-Latin
squares are result in efficiency in a study of the effect of one
treatment factor (Latin letters) in the presence of three other
factors (columns, rows and Greek letters). Like the Latin square
model, the Graeco-Latin square model, also assumes that there
are no interactions effects between treatment and the other
factors. The model and skeleton of ANOVA for the Graeco-Latin
square design is presented in Section 19 under the illustrations of
analyses of variance.
Incomplete block designs
Incomplete block designs, suited for experiments with a large
number of treatments (or larger plots) are characterized by blocks,
each of which contains only a fraction of the treatments under
test. In an incomplete block design, not all treatments are tested in
each block, i.e. there is an incomplete set of treatments in each
block. Then, "incomplete" in this design simply means all
treatments do not occur within the same block. Compare this to
the Randomized Complete Block design, where all treatments
must be tested in every block. Otherwise, the ICBD and RCBD
are identical, with experimental units divided into groups (blocks)
that are similar, but differing from group to group. This design
should only be used if the experimental situation forces blocks to
be too small. An example is varietal tests, which usually involve
many varieties. Lattice (or α-lattice) and group balanced block
designs are the common incomplete block designs.
The advantage is associated with the fact that small blocks are
more homogeneous than large blocks, so experimental error is
lower. Use incomplete block designs is advised to increase

[133]
precision when there is variability within larger blocks. The
disadvantages of the incomplete block designs, on the other hand,
include the rigid requirement for a fixed number of treatments, a
fixed number of replication, or both and the complexity of the
analysis. There is also unequal precision for certain comparisons
of treatment means.
Multiple factor experiments

Factorial experiment
Factorial experiments are experiments that simultaneously
involve more than one factor, with each factor being at two or
more levels. Several factors may affect concurrently the
characteristic under study and the experimenter is interested in the
main effects and the interaction effects of the factors on the
characteristic under study. A factorial experiment is particularly
useful when little is known about factors and all combinations
have to be analyzed in order to conclude which combination is the
best.
The effect of a factor alone is called a main effect whereas the
effect of different factors acting together is called an interaction.
The term factorial is used to indicate that all possible
combinations of the factors are considered. Note that the
background experimental design could be completely
randomized, randomized complete block, Latin-square design,
and incomplete blocks depending on situations.
This type of experiment is chosen mostly to save time and other
resources and also to know the interaction effects of different
factors. Factorial experiments are suitable for agricultural, social

[134]
science, and other studies in which the effect of more than one
factor is required to be compared.
For example, consider an experiment to test the effect of protein
content and type of feed on milk yield of dairy cows. The first
factor is the protein content and the second is type of feed. Protein
content is defined in three levels, and two types of feed are used.
Each cow in the experiment receives one of the six protein x feed
combinations. This experiment is called a 3 x 2 factorial
experiment, because three levels of the first factor and two levels
of the second factor are defined. An objective could be to
determine if cows’ response to different protein levels is different
with different feeds or what we call interaction. The main
characteristic of a factorial experiment is the possibility to
analyze interactions between factor levels.
Likewise, to investigate the effect of different varieties of a crop
with different rates of fertilizer, different irrigation regime with
different crop genotypes, etc., factorial experiments are
conducted.
Similarly, in socioeconomic studies, one may be interested in
relating the economic status and educational status of the people
in adopting modern agricultural practices. Thus, factorial
experiments are the methods for answering the questions related
to more than one factor at a time for their individual effects as
well as their interactions.

The advantages of factorial experiment include that the effects of

more than one factor at a time can be estimated; the interaction
effects can be estimated which is not possible in single-factor
experiments, and factorial experiments save resources compared

[135]
to single-factor experiments. Factorial experiments could also be
set up with lesser number of replications because the required
error degree of freedom could be easily achieved by virtue of
involving a number factors at a time.

The major disadvantage of factorial experiments is that, when the

number of factors or the levels of factors or both increase, then
the number of treatment combinations will also increase,
requiring an increasingly larger experimental areas/units and
larger blocks/group size. In addition, as the block/group size
increases, it is very difficult to maintain homogeneity among the
plots/units within the block/group. Thus, there is a possibility of
increasing the experimental error vis-a`-vis decreasing in
precision of experiment. Factorial experiments are also more
complicated than single-factor experiments. The other peculiar
limitations of a factorial experiment are that a failure in one
experiment may result in a loss of a huge amount of information
compared to a single-factor experiment.
There are different types of factorial experiments:
(a) Based on the number of factors and level of factors.

Depending upon the number of factors used in the experiment,

a factorial experiment is a two-factor, three-factor ... p-factor
factorial experiment when the number of factors put under
experimentation are 2, 3 ... p, respectively.

(b) Based on the equality or inequality in the levels of factors put

under experimentation, a factorial experiment is either
symmetrical or asymmetrical. If the numbers of levels for all
factors are the same, it is symmetrical; otherwise, it is an

[136]
 asymmetrical factorial experiment. For example, a two-factor
factorial experiment with five varieties and five different rates
of nitrogen is a symmetrical factorial experiment, but a
factorial experiment with five varieties and any other number
of doses of nitrogen (not equal to five) is an asymmetrical
factorial experiment.

Symmetrical factorial experiments are denoted by pq form, that

is, q factors each at p levels. When asymmetrical factorial
experiments are denoted by p q r s, where the 1st, 2nd, 3rd, and 4th
factors have p, q, r, and s levels, respectively. Factorial
experiments can also be divided into complete block designs
(factorial RCBD and factorial LSD), incomplete block designs
(balanced lattice and confounding), and split plot designs (split
plot, split-split plot, and strip-plot).

To illustrate the interpretation of interaction effects, assume that

an agronomist tested a factorial combination of two varieties (V1
and V2) with two rates of nitrogen (N1 and N2). For example, N1
may stand for no nitrogen application and N2 for an application
of 50 kg N ha–1. Assume also that the result hypothetically
showed three scenarios as detailed below.
Scenario 1. The graph below (Figure 1 scenario 1) showed that
the application of nitrogen increased yields of both varieties at the
same rate as depicted from the parallel lines, which implies that
varieties did not show significant differential response to
fertilizer, i.e. there was no interaction between varieties and
nitrogen level. Note that the difference between the two varieties
remained the same irrespective of the rate of fertilizer applied, the
first variety (V1) being the best under both nitrogen levels and the

[137]
second nitrogen level (N2) being the best regardless of the
varieties. In this case, we can obviously explain the main
treatment effects without any reservation as the interaction effect
was non-significant (scenario 1 in Figure 1).
Scenario 2. The application of nitrogen increased yields of both
varieties but not at the same rate as depicted from the non-parallel
lines which implies that varieties showed differential response to
fertilizer, i.e. interaction, which maybe or may not be detected by
the ANOVA table. Still, as could be witnessed from the graph
(scenario 2 in Figure 1), the first variety (V1) maintained its
relative rank despite the deferential response and the two lines
seem to cross each other at some point if they are allowed to
extend. Variety V1 stood the best under both nitrogen levels and
the second nitrogen level (N2) was also the best in biological
terms regardless of varieties.
Some authors argue that whenever the interaction effect is
significant, the main effects cannot be explained because of the
interdependence between the main factors. However, explanation
of results of the interaction effects depend not on the magnitude
but on direction of the interaction response as witnessed here. In
this case (scenario 2), for instance, we can obviously explain the
main treatment effects without any reservation as the interaction
effect maybe significant but did not cause a relative change in
rank order of varietal performance, i.e. “non-cross over type” of
interaction.
Scenario 3. The application of nitrogen not only increased yields
of both varieties but also caused a rank order change in varietal
performance, i.e. resulted in a “cross-over type” of interaction. In
this case, comparison of the main effects (varietal and nitrogen
[138]
levels) would be difficult and we cannot obviously explain the
main treatment effects without any reservation as the interaction
effect was a cross-over type.
We cannot right away answer the question, “which variety is the
best?”, as it automatically raises another pertinent question, “with
which nitrogen level”? Likewise, the question, “which nitrogen
level is the best?” raises another question, “with which variety”?
In this case, therefore, we should explain the result in terms of the
best combination of variety by nitrogen levels.
For poor farmers who could not afford application of nitrogen at
all, variety V1 is the best; whereas for well-to-do farmers who
could afford application of nitrogen, variety V2 is the best.
Therefore, when significant interactions exist, the interpretation
of the main effects must be modified in the light of the interaction
types.

[139]
[140]
Figure 1. Interaction plots of grain yields (q ha–1) of two varieties (V1 and V2) grown under two
nitrogen levels (N1 and N2) showing different scenarios when there is no, a non-cross
over and a cross-over types of interactions between the two.

Split-plot design
A split-plot design is used for experiments involving two or more
factors each with different levels, and differing in the requirement
of plot size and accuracy. A split plot design is characterized by a
“nested” blocking structure: split plots are nested within whole
plots, which may be nested within blocks. This arrangement is
applied when the practical limit for plot size is much larger for
one factor compared with the other. A good example is an
experiment conducted to compare irrigation treatments and plant
population densities.
The split plot arrangement can also be used when greater
precision is desired in one factor relative to the other; e.g., if
several varieties are being compared at different fertilizer levels
and the factor of primary interest is the varieties, then it should be
assigned to the subplots and fertilizer levels assigned to the main

[141]
plots. In this case, an extra factor (second) could be introduced
into a study by dividing the large experimental units (whole unit)
for the first factor into smaller experimental units (sub-units) on
which the different levels of the second factor will be applied.
Since sub-unit variance is generally less than whole unit variance,
the sub-unit treatment factor and the interaction are generally
tested with greater sensitivity/precision.

9. Some Practical Considerations in

Experimental Research

9.1. Field experiments

A field experiment is a true experimental study performed outside

the laboratory; that is, in the real-world situations such as farms,
forests, grasslands, ponds, polluting sources, rivers, watersheds,
factories, vet clinic, and so on. For example, an ecologist studying
an ecosystem will not be able to move the entire system into the
laboratory for conducting an experiment. In agriculture, field
experiments along with control are widely used, as the crop
growing conditions cannot be recreated in the laboratory.
Although field experiments are carried out in a more or less
realistic situation such as a farm, the degree of control over the
independent variables is comparatively less unlike a laboratory
experiment wherein all the conditions for the experiment will be
under the control of the investigator. Therefore, the necessity of a
‘control’ group normally does not arise.
Although once considered ineffective as a scientific instrument of
research, the world began to note and recognize the potential of
field experiments by the classical works of Sir Ronald Fisher (in
[142]
1920s and 1930s), who introduced the principles of
experimentation and analysis of variance. During the last five to
six decades, they have been identified with unlimited possibilities
by the application of appropriate statistical methods. Before
Fisher, although soil heterogeneity was known to exist, it was
either ignored or attempts were made to even out by various
means. At present, the presence of soil heterogeneity is fully
recognized, and its effects are eliminated by suitable experimental
designs, layouts, and statistical analysis.

In field experiments, the researcher manipulates one or more

independent variables in field situations taking all precautions to
have control over the environment. However, as everything is not
under the control of the investigator, it is recommended to have a
‘control group’ in field experiments. In such experiments, you
begin the experiment by creating two or more treatment groups,
which are almost equivalent so that measurements on characters
should be similar among the groups, and that the groups should be
able to react almost uniformly to a particular treatment. This
equivalency between groups can be ensured by applying the
principles of experimentation (randomization, replication, and
local control), which take into account the variation between
treatment groups and between the individuals in each group. Once
equivalent groups have been formed, the researcher tries to treat
them identically except for the one variable that she/he wishes to
isolate.

A major advantage of a field experiment is that it is practical and

amenable to experimentation without artificially introducing
confounding variables. A field experiment follows all the steps of
the scientific method starting from the formulation of a research
[143]
hypothesis. In scientific disciplines such as agriculture, biology,
ecology, and geology, field experiments are long considered as a
sound experimental practice following various steps of the
scientific method. Researchers can also use modern techniques to
record observations. However, a major concern is the cost of field
studies, as they tend to be very expensive compared to laboratory
studies.

In agriculture, especially those involving crops, field experiments

are routine. The following are some practical issues, which are to
be considered while planning and conducting field experiments in
agriculture.
Choosing designs
Researchers should choose appropriate designs for their
experiments. The most suitable design depends largely on the
number and nature of the proposed treatments. For pot culture
experiments, completely randomized design without blocking can
be conveniently used. Randomized block design with local
control may become necessary when soil heterogeneity or
differences in experimental materials are suspected. However, if
the fertility gradient is in two directions, by length and breadth,
then blocks are to be considered in both directions following a
Latin square (row-column) design. If the number of treatments is
large, and a uniform area within the block cannot be attained, an
incomplete block design like lattice or α-lattice design may be
used. For example, if many genotypes are to be tested, it is
desirable to use an incomplete block design.
In experiments involving a combination of treatments, wherein
one of the factors requires a large-sized plot as in irrigation, it
might be advisable to adopt a split-plot design with irrigation
[144]
treatments assigned to main plots. If both the factors need large
plots for effective application of treatments like irrigation and
land configuration treatments, a strip-plot design will be more
efficient. Similarly, if many treatments are to be tested in a
factorial scheme, it is necessary to adopt a confounded design
than a simple randomized block design. Because of its simplicity
and flexibility, simple randomized complete block design is the
most preferred design for research workers in agriculture. With
the usually adopted plot sizes ranging from 20 to 40 m2,
randomized complete block design can be adopted safely for
experiments with up to 20 treatments without appreciable loss of
efficiency.

Experiments should be replicated, i.e. repeated more than once, to

increase reliability. The greater the number of replications, the
lesser would be the errors due to chance. Fixing the number of
replications for an experiment is also a problem for novice
researchers. Replications needed for a field experiment is
dependent on factors such as the natural variability of the
experimental materials, the experimental design adopted, the
number of treatments, and the desired degree of precision.

Generally, we follow a rule of thumb to determine the number of

replications. Choose the number of replications to provide not less
than 10 or 12 degrees of freedom as suggested above for
estimating experimental error. For example, as the error is based
on (t − 1) (r − 1) degrees of freedom in a randomized complete
block design, if the treatments to be tried is seven, then a
minimum number of three replications are needed.

[145]
Size and dimension of plots
Size of plots is another crucial factor, which affects experimental
error. Smaller plots are unreliable, but too large plots are
unmanageable and misuse resources and labour. For the same plot
area, the perimeter dimension of a square plot is lesser compared
to a rectangular plot and, therefore, only lesser number of plants
are exposed to border effects in square plots. While laying out
plots, the length and width of plots must be measured correctly, as
even a small error in plot dimensions can greatly affect the
experimental results. This is especially so when we project the
yield from a small net plot area to a large area; for example, a
minor error in projecting the yield from 12 m2 to one hectare
might be substantial.

Fix the plot dimensions after careful consideration of planting

methods and spacing. For a broadcasted crop, a plot of 20 m2 (5
m × 4 m) might be alright but, when row to row spacing and plant
to plant spacing are involved, the length and width dimensions
should be fixed as multiples of spacing. For example, if the
spacing given is 60 cm × 30 cm, it means that row-to-row spacing
is 60 cm and plant-to-plant spacing is 30 cm. Instead of fixing 5
m × 4 m dimensions for the plot, it may be slightly varied and
take 4.8 and 4.2 m so that we will get 8 rows and 14 plants per
row, leaving exactly half the spacing at the borders (30 and 15
cm). If you take 5 m × 4 m, then you will get 8 rows and 13 plants
only, leaving wider borders of 50 and 20 cm, thus not only
wasting land but also complicating border effects.

For crops like rice, the size of plots generally ranges from 8 m2 to
25 m2, depending on the nature of treatment and uniformity of
plots. For field crops like rice, decide the size of plots in such a
[146]
way that a minimum net area of 5 m2 is available for harvesting,
leaving border rows and rows for destructive sampling. Normally,
fertilizer trials require larger plots than a variety trail. An
irrigation trial requires still larger plots to take care of buffer
strips between plots. Similarly, for those experiments involving
pesticides, the width of the plots should be decided by the
coverage of the sprayer being used.
Selection of treatments
Treatments for the experiment are to be fixed after carefully
considering many aspects. There may be cases when, before
going for experimentation in a detailed way, it could be advisable
to try the treatments in an observation trial. The preliminary tests
often reveal possible difficulties that might be involved in the
application of the treatments or the unsuitability of some
treatments under field conditions. The treatments must be selected
in such a way as to achieve the objectives and to increase the
precision of the experiment. This is especially important when
studying the effects of inputs in increasing doses such as
herbicides, fertilizers, water, fungicides, or insecticides. In such
experiments, it is more beneficial to determine how the
experimental units respond to increasing rates of treatments than
to know whether different rates are significantly different. The
major objective of these kinds of experiments should, therefore,
be to determine a response curve with a range of rates going well
beyond the optimum. For determining the shape of the response
curve, too many treatments are not required; but four or five well-
structured rates could be enough. For example, if the optimum
rate of nitrogen is expected to be around 100, then suitable
treatments for the trial might be 0, 50,100, and 150 or 0,

[147]
40,80,120 and 160 kg/ha. This indicates that rates in equal or
multiple increments within the expected range of the response are
most efficient in establishing equations for a rate: response curve.
Border effect and harvest area
In agricultural field experiments, an ‘experimental plot’ is a unit
of land area on which random assignment of treatments is made.
The whole unit of land receiving a treatment (gross plot size) may
not be the harvest area (net plot size). Harvest area is decided
excluding the ‘border plants’. The phenomenon in which the yield
or other characteristics of crops in the border areas of a plot differ
from those of the central portions is called the “border effect”.
Border effects can occur when plants in a plot facing the outer
margin of unplanted free space are advantaged in terms of
resource like nutrients, moisture and even space as compared to
those in the inner side or centre of the plot higher relative
competition is expected, i.e. the advantage is not available to
inner plants. Thus, if we are not accounting border effects, the
difference between the performances of the varieties in the field is
liable to be overestimation. Apart from the possibility of bias, the
border effects would aggravate error variances by increasing
heterogeneity among plants. In fertilizer trials, chances are that
the nutrients at different doses supplied to various plots might
seep up to the border plants in the unfertilized plots, which
ultimately affect the response of border plants to the fertilizers
applied. In the case of trials involving irrigation, more precautions
to avoid border effects are needed, as lateral movement of
irrigation water is to be expected, and double bunds and buffer
areas between the plots are required in addition to excluding
border plants from observations.

[148]
The size, shape, and orientation of plots should be in such a way
that experimental error is reduced to the minimum. In varietal
yield trials involving crops like rice, where inter-varietal
competition is expected, plots with at least six rows should be
used leaving out one row on each side of the plot as border, so
that four centre rows are available for harvest. Some plants at
both ends of each row should also be excluded as border plants.
Destructive sampling
Some experiments involve the measurement of several plant
characters. In certain cases, it may also involve destruction of
plants as in the case of dry matter production. In weed
management trials, for instance, periodic monitoring of dry matter
production of weeds is common by uprooting weeds. Uprooting
of weeds enable crops in the weed uprooted area to be vigorous
and more competitive. You may follow some techniques to avoid
plant competition near the spot from where uprooting is done.
Leaving an area in the centre of the plot for final harvesting, and
using the surrounding areas for sampling is a common method.
Another option is to set up an area entirely for destructive
sampling lengthwise. For example, if you are taking a plot size of
6 m × 4 m, you could provide 2 m lengthwise entirely to conduct
sampling.
Experimental site
A researcher should ensure that a site selected for an experiment
is as uniform as possible in all aspects. By following efficient
blocking, a researcher can take care of heterogeneity to a certain
extent. It is often observed that quantitative characters measured
from neighbouring plots sown simultaneously with the same
cultivar and treatment are not uniform and differ mildly or widely

[149]
depending on several factors. The most obvious and probably the
most important features that magnify the differences are related to
soil heterogeneity, which is amplified by one or any of the
following features:
Slope: Slope of the land affects soil uniformity. If the
experimental site is not uniform in terms of slope, soluble
nutrients may move to lower slopes along with runoff water.
Differences in slope may create differences in soil texture and soil
depth too. Obviously, an ideal experimental site is one that has no
slope. In the absence of such levelled lands, we can go for lands
with uniform and gentle slope having predictable fertility
gradients using efficient blocking along the contours.

Cropping history: Previous crops and various treatments given

to them must be taken into consideration. For example, various
management practices such as fertilizer schedules, plant
protection, and crop rotation are sources of additional soil
heterogeneity. Therefore, it is necessary to standardize the field to
obtain uniform fertility by some measures. A commonly adopted
practice by research centres in Ethiopia is to have a clear crop
rotation plan, to sow the area with a green manure crop or cover
crop for at least one season, and/or to equalize the nutrient
contents of soil before conducting an experiment. Another
method is to assess the fertility gradient by uniformity trials and
proper block arrangement.

Grading: Levelling or grading the field is a normal operation in

farms for various purposes such as irrigation and drainage. It is
done by removing top soil from elevated areas and spreading and
filling crevices and uneven areas. However, for an experiment,

[150]
such areas are not ideal as this operation causes an uneven depth
of surface soil and at times exposes the subsoil and, as far as
possible, it is advisable to avoid such sites. Nevertheless, if there
is no option except the present site, assess the pattern of soil
heterogeneity through uniformity trials and follow blocking or
adjustments through analysis of covariance techniques.

Presence of large trees, buildings, and other structures: These

structures influence the surrounding areas by shading. Large trees
with their massive root system offer stiff root competition.
Sometimes, shade may not be a problem but top soil is removed
for construction and subsoil is exposed. Avoid such problem sites.
Sometimes, certain trees species fix nitrogen and make the soil
and their canopies more fertile than across the other surface of the
field. Thus, they create heterogeneity.

Unproductive site: Avoid waste and unproductive lands for

experimentation. A field with very poor or problem soils should
not be selected unless it is an experiment designed specifically to
evaluate such problem conditions.
Sowing
Sowing and transplanting are important operations to ensure
uniformity. In general, over sowing seeds by 25–50% may be
required to ensure that enough seedlings emerge to establish the
required plant population. Use seeds from the same seed lots and
the seeds should be sown on the same day completing all the
formalities in all the plots.
If there is difficulty to cover all the plots in one day, cover at least
all the plots of one block at a time at least for blocked designs.

[151]
For uniform germination of seeds, compact the soil around the
seeds immediately after sowing and ensure soil moisture.
Thinning and gap filling
Excess plants in the experimental plots should be removed within
a reasonable time. The removal of excess plants is called thinning.
Thinning should be done within 6–10 days after seedling
emergence to avoid unevenness in the growth of the crop. While
removing excess seedlings, take care to avoid disturbance to the
remaining plants. Seedlings should be left as equally spaced as
possible. A common practice is to lay a two metre pole or stick
along the row and remove the seedlings in excess of the
calculated numbers. The most vigorous seedlings of similar size
are to be retained. Diseased, damaged, and weak seedlings may
be removed while thinning. As far as possible, the same person
should be engaged to thin all plots in a block to reduce bias.
Thinning should be done within the shortest time in the whole
replication in a day. Normally, transplanting seedlings in vacant
places is not recommended because the transplanted plants need
some recovery time from the stress, may be weaker and less
productive than the normal plants. A researcher has two
alternatives in dealing with missing plants. One is to manipulate
the thinning process so that the areas adjacent to the vacant space
have more plants or to ignore the gaps at the seedlings stage and
adjust the number of plants at harvest or adjust plot size at
harvest.
Manures and fertilizers
Manures and fertilizers required for each plot should be spread
uniformly so that each plant in the experimental plot gets its due
share. Plants in a plot may behave differently if these are not

[152]
applied uniformly. Sub-divide the experimental plot into smaller
units to enable uniform application to smaller areas. Applying
fertilizers to each row of plot separately can also be considered.
The usual procedure is to weigh or measure the fertilizer required
per row, and uniformly spreading it on each row by hand.
Cultural operations
All the common cultural operations including ploughing, seedbed
preparation, levelling, weeding, earthing up, and plant protection
must be done uniformly to all plots, covering all the plots in one
block at a time. Differences in cultural operations can aggravate
experimental errors. Sufficient protection from animals, birds, and
intruders must be given. Damaged and disturbed plots should not
be included for observation, but they should be considered within
a limit as missing plots at the time of statistical analysis.
Off-type plants
Sometimes, you may notice a few exceptionally tall or vigorous
plants in the plots. These off-type plants may have come from the
seeds of a previous crop or may be natural mutants. You cannot
simply ignore these plants as they may already have affected
surrounding plants. Their removal creates missing hills or gaps in
the row and thus affects the surrounding plants. Therefore, they
are normally allowed to mature, but just before harvest, such
plants are counted and removed. The yield of the plot can then be
computed as:

[153]
 𝑎+𝑏
Y=( )𝑦
𝑎

where Y is corrected plot yield; y is actual grain

yield from normal plants in the plot; a is normal

plants in the plot; and b is number of off type

plants.

For example, take the case of a plot with a total plant stand of
100, in which 4 plants are off types. Suppose that after removing
the 4 off types, the remaining 96 plants yielded a harvest of 2 kg.
The corrected plot yield is 2.083 kg. This correction is applied
based on an assumption that the competition effects offered by the
off-types to the surrounding plants are similar to those of the
normal plants.
Competition effects
Competition between plants for various resources is common in
crop communities. Competition between crops and weeds for the
same resources such as solar energy, soil nutrients, and moisture
should be kept to the minimum. Competition effects between crop
plants within a plot (intraspecific competition) should also be
considered and kept uniform. Make sure that the plant response
represents the actual conditions being tested in terms of intra-
varietal competition. Intra-specific competition is also most
obvious when the experiment involves different varieties or
genotypes of a crop. Different cultivars or genotypes sown in
adjacent plots may be subjected to different environmental
[154]
conditions depending upon their location relative to adjacent
plots. In such trials, the plants near the borders may experience
the effects of varietal competition or edge effects.
Competition could also occur for nutrients when neighbouring
plots receive different rates of fertilizers. Plants that receive a
higher rate tend to be more vigorous and more competitive.
Likewise, another source of competition is seepage of water from
fertilized plots, which may spread nutrients to the root zone of the
adjacent plots. Ultimately, this may favourably influence the plots
receiving smaller doses of fertilizers.
Pathways or alleys
It is common to see vigorous and luxuriously growing crops along
the pathways and alleys in a farm. This happens mainly because
of less competition along pathways and alleys where there are
free spaces than those in the centre of the plot.
Missing hills or plants
Missing hills or plants are common problems for farm
researchers. It is very difficult to ensure full stand for all plots in
an experiment even for the vigilant researcher who have taken all
precautions. Poor germination, bird damage, insect and disease
attacks, physical mutilation, and similar causes may be
responsible for the death of a few hills or plants in a plot. Plants
near to the spots of missing plants perform better than the normal
plants because of less competition compared to other plants
creating more problems. Discard all the plants immediately
adjacent to a missing hill, and harvest only those that are fully
competitive. While computing total yield, this loss of plants must
be accounted for. If seeds of the test genotypes vary in age, they
should be rejuvenated at the same time and under the same
[155]
condition to minimize initial variations resulting from the
disparity in age.
9.2. Quasi experiments

A quasi-experiment is sometimes called a ‘natural experiment’

when it involves testing a hypothesis based on data gathered by
observing a natural system. The variables are not controlled or
manipulated in such experiments. The term ‘natural experiment’
refers to mean events, interventions, or policies, which are not
under the control of researchers, but amenable to observation and
recording of data. We may not be able to design a true experiment
with natural phenomena; but we can collect data by simply
observing a phenomenon for testing a hypothesis concerned with
it. Natural experiments are observational studies, and these may
not have a ‘control’ in the traditional sense of a randomized
experiment. ‘Natural experiments’ are almost similar to ‘natural
observation’. The only difference is that the natural experiment is
done to test a hypothesis while natural observation may be simply
a descriptive study.

9.3. Natural experiments

Natural experiments are accepted as study designs in many

disciplines wherever controlled experimentation is difficult. For
example, astronomy usually relies on natural experiments of this
type, as it is impractical to do experiments involving celestial
bodies. In social sciences, this usually entails observing humans
or animals in real-life situations.

[156]
In life sciences, natural experiments involve observing an animal
or groups of animals, a plant or plant community, or some
physical phenomena. Natural experiments are also feasible in
health science when a randomized controlled trial is difficult to
conduct because of ethical issues. A typical example of a natural
experiment is to relate the quality of water in a particular locality
and the occurrence of diseases like cholera.

9.4. Ex Post Facto experiments

Ex post facto studies resemble a typical experiment because

groups are compared; however, a key difference is that there is no
manipulation of independent variables. The causal event of
interest has already happened without the manipulation of
treatments by the researcher. In such experiments, the ‘effect’
becomes the dependent variable and the probable cause becomes
the independent variable. Ex post facto experiments are useful
when human subjects in true situations are involved, and the
researcher comes to the scene only ‘after the fact’. Most ex post
facto studies are observational studies as no manipulation of
treatment is possible. In such situations, the investigator begins
the investigation by identifying the outcome variables and then
tries to identify possible causes, i.e. independent variables
responsible for the outcome. Ex post facto studies are often
criticized because of the post hoc fallacy. Causal-comparative
studies and case-control studies are examples of ex post facto
studies.

[157]
9.5. Causal-comparative studies

In a causal-comparative study, the researcher attempts to trace an

effect that has occurred without any manipulation to its possible
causes. This type of study may look like a type of descriptive
research only because it describes conditions that already exist.
Actually, it is a retrospective study, as it starts with effects and
investigates back to its probable causes. Without any
manipulation, the independent variables have already occurred
and, therefore, the same degree of controls as in an experimental
study is not possible. Because of this ex post facto nature,
interpretation of results requires caution. Nevertheless, certain
control procedures such as matching can be adopted to improve
the reliability of results and consequent interpretations. Normally,
a causal-comparative research is planned to approximate the
conditions of a true experiment in a situation that does not allow
control or manipulations of relevant variables.

In a causal-comparative study, the researcher selects two groups

of subjects, the experimental group (who have subjected
themselves to the independent variable assumed to be the cause)
and comparison group to act as control (hence the name ‘causal-
comparative’ experiment). The two groups may differ in two
ways. One group may possess a characteristic that the other does
not possess or each group has the characteristic but to differing
degrees, which make it a point to define the independent variable
differentiating the groups clearly.

Note that in causal-comparative research, random samples are

chosen from two already existing populations, unlike a single
population as in experimental research. However, the aim is to
[158]
have groups that are as similar as possible on all relevant
variables except the independent variable as in experimental
studies.

Example: suppose you want to determine the relation of aerial

spraying of pesticide X (as the independent variable) on the
occurrence of cancer (as the dependent variable) in humans. It is
unethical and impractical to conduct an experiment with humans
dividing them into groups and by requesting them to be subjected
to the aerial spraying treatment with pesticide X. In such a
situation, you can use the ‘causal-comparative method’. You can
enlist several people who were exposed to pesticide X spraying in
the past. In fact, they have already applied the independent
variable in your question, that is, aerial spraying of pesticide X
and exposed to it. To test your hypothesis, you must gather data
on two groups of people, i.e. an ‘experimental’ group of people
who have subjected themselves to your independent variable
assumed to be the cause (exposed to aerial spraying of pesticide
X), and a ‘comparison’ group (to act as control) of similar people
with the same background who were not exposed to aerial
spraying.

After collecting sufficient data, a statistical test is conducted to

decide whether the people exposed to pesticide X spraying scored
significantly higher than those who were not exposed (the
comparison group). By comparing the incidence of cancer in the
two samples, you can check whether there are significantly more
cases of cancer in the experimental group. If you find no
significant difference between the two groups of people on the
incidence of cancer, then you accept the null hypothesis and reject
the research hypothesis. On the contrary, if there are significantly
[159]
more cases of cancer in the experimental group, you may infer
that your study supports the alternate hypothesis and conclude
that exposure to aerial spraying of pesticide X increases the
occurrence of cancer.

You must apply caution in attributing cause and effect

relationships as above and come to a firm conclusion without
doing further studies. In this kind of observational studies, it is
very difficult to rule out all causes other than the one being
investigated. For example, the supposed cause of an observed
effect may be a third variable. Understand that a causal-
comparative study lacks randomization, no manipulation of
variables, and true control is absent. The investigator cannot
randomly allocate subjects to treatment groups because they are
already in those groups. The groups may be different on some
other important variables too, for example, gender, age, or health,
in addition to the identified independent variable.

Causal-comparative research is prone to ‘post hoc fallacy’, the

conclusion that because two factors go together one must be the
cause and the other the effect. Post hoc fallacy occurs when a
conclusion presumes a causal relationship without sufficient
grounds. For example, linking cancer with pollution, pesticide
application, fast foods, or other such causes after doing simply a
study as above requires caution. This fallacy is named after a
Latin phrase post hoc ergo propter hoc, meaning ‘after this,
therefore because of this’, which suggests that one event caused
another event simply because it took place first. It is likely that
other factors may have caused the second event. As there is a
danger of confusing symptoms with causes, these kinds of studies
should test not just one hypothesis but other logical alternatives or
[160]
competing hypotheses too before rushing for a post hoc fallacy.
Therefore, the results of any ex post facto study should not be
taken for granted but to be taken as a guide only for deeper and
more rigorous research. However, these kinds of studies facilitate
decision-making and provide guidance for further studies.

9.6. Case-control studies

A causal-comparative study when used for epidemiological

studies particularly for studying infrequent events with
‘matching’ is usually called case-control study. Sometimes, a
case-control study is also referred to as a case-referent or case
comparison study. In a typical case-control study, the
characteristics of a group of people with a specified medical
condition such as a disease, risk factor, a treatment, or an
intervention (cases) are matched with a control group of people
without the condition but are otherwise similar (controls) (hence
the name, case-control study). Case-control studies are usually
retrospective in nature, and the major objective is to verify
retrospectively the exposure to the risk factor of interest from
each of the two groups of subjects—cases and controls. In fact,
these types of studies are ideal to estimate odds. The magnitude to
which each participant was previously exposed to the independent
variable can be easily identified. For example, in the case of a
disease like cancer, the investigator looks backwards in time for
exposures that might have caused the disease. By examining
medical case records and through interviews, the researcher can
build up the history of the subject’s ‘case’. Case-control studies
are useful for studying conditions with long intervals between
exposure and outcome. Statistical analysis is also possible to find

[161]
out whether there are any significant differences between the
groups, and it allows the researcher to arrive at a firm conclusion.

The first step in conducting a case-control study is to decide on

the research question to be answered. Based on the researchable
question, formulate a hypothesis and then decide what will be
your observations, and how they are measured. The next step is to
specify the features of the study group, and decide how to form a
valid control group. After having finalized these aspects, the
‘case’ group and ‘control’ group are compared often by
conducting interviews. Medical records are also used to build up
history of the subject’s life. Sometimes, blood tests or other tests
are also conducted after drawing samples.

In situations where the outcome is rare, a case-control study may

be the only possible option to get enough information from a few
subjects. Similarly, in situations where there is relatively a long
period between an exposure and the disease, case control studies
are ideal. As some of the subjects have been intentionally chosen
because they have the disease in question, case-control studies are
more cost efficient when compared to cohort or cross-sectional
studies.

The main weakness of case-control study, as in the case of causal

comparative, is that as no intervention is attempted and no effort
is made to alter the course of the condition, it is purely an
observational study and, therefore, does not provide the same
level of evidence as randomized controlled trials. Another
problem is sampling bias. Ideally, you should use random
sampling, and the cases chosen should be a random sample of all
the subjects with the disease or problem. Most often, however,

[162]
you have no other choice but to study a sample of those subjects
whom you are able to recruit (convenience sampling). There are
issues in selecting the controls too. To enable the controls to
represent the same population as the cases, pair wise matching of
participants is often followed. For each subject in a group, the
researcher finds a match in the other group with the same or very
near score on the control variable. If a volunteer does not have a
suitable match, he or she will be eliminated from the study. The
resulting matched groups are very similar with respect to the
recognized extraneous variable.

Case-control studies are very poor in establishing cause and effect

relationships, and it is very difficult to rule out all causes other
than the one being investigated. A case-control study usually
depends on retrospective data, and because such studies rely on
memory, the possibility of recall bias cannot be ruled out. Recall
bias is the inclination of people with a disease or condition to
recall events better than the people who do not have that
condition. In certain cases, recall bias can be eliminated by using
data recorded for other purposes before the outcome had
occurred. The success of this strategy, however, is limited by the
low availability and reliability of the data collected. Because of
the problems of matching, recall bias, and the presence of
confounding variables, utmost care is needed while evaluating
results.

Case-control studies are also prone to the ‘post hoc fallacy’, as

discussed earlier in the case of causal-comparative studies, the
conclusion that because two factors go together one must be the
cause and the other the effect. Because there is a danger of
confusing symptoms with causes, case-control study should test
[163]
not just one hypothesis but other logical alternatives or competing
hypotheses too before rushing for a post hoc fallacy. Therefore,
the results of any case-control study should also not be taken for
granted but to be taken as a guide only for deeper and more
rigorous research. They are, of course, useful for generating
hypotheses. The hypotheses thus formed can be tested more
thoroughly by other methods such as randomised controlled trials.

10. Requirements of a Good Experiment

A well-planned and properly laid out experiment is important for
success. A researcher can ensure success in his/her research
endeavour, if he/shee is able to meet the requirements such as the
absence of systematic error, precision, wide range of validity,
simplicity, and proper statistical analysis.

10.1. Absence of systematic error

Chances of experimental error are higher, if a researcher treats all

experimental units or plots alike. Random errors are out of scope
but one can take enough safeguards to ‘control’ or reduce
systematic error to the minimum. In other words, the units
receiving one treatment should show only random differences
from units receiving other treatments including control, and they
must be allowed to respond independently of one another.

Blocking, proper plot techniques and proper data analysis are the
three techniques followed in agricultural research for the control
of systematic error. In field experiments, where substantial
variations are expected within an experimental field, employing
proper blocking can reduce experimental errors. This is called

[164]
local control. Proper plot techniques include measures adopted to
reduce soil heterogeneity, competition effects, and mechanical
errors. In experiments where blocking is not done effectively,
proper data analysis can help to control errors. Analysis of
covariance can be adopted in such cases. Suppose you want to lay
out an experiment in a coconut garden where the yield levels of
coconut trees differ considerably. Using the initial yield as the
covariate, the final yield after the experiment can be adjusted to
the values that would have been attained had all experimental
trees started with the same yield.

10.2. Precision

If there are no systematic errors and the estimates of treatments

differ from its true value only by random errors, the experiment
has more precision. You measure random errors usually by
standard error. A standard error should be sufficiently small. If it
is large, the experiment itself is useless. The precision of any
experiment depends on the intrinsic variability of the
experimental material, the accuracy of the experimental work, the
number of experimental units, and the design of the experiment.
Therefore, these aspects have to be considered carefully for better
precision.

10.3. Range of validity

The conclusions and inferences made out of the experiment

should have a wide range of external validity applicable to other
settings and regions. Necessary precautions should be taken to
avoid or reduce threats to both internal validity and external
validity. In agricultural research, most experiments show a great
[165]
deal of internal validity indicating that they are valid in highly
controlled conditions; but external validity may be low when the
experimental results are applied to actual field situations.
Therefore, experimental designs in agriculture focus more on the
problems of external validity.

10.4. Simplicity

The experiment should be simple in design and analysis, for

example, simplicity of design, simplicity of analysis, and simple
analytical tools. Select a suitable design so that its lay out,
observations, and statistical analysis can be done easily. In the
case of field experiments, follow simple designs like CRD, RBD,
or simple factorial combinations. Because of the problems in
getting large area for experiments and huge costs involved, the
modern trend is to avoid complicated designs. Similarly, a
researcher should include only those methods and techniques that
you can do with the available resources and facilities.

10.5. Proper statistical analysis

The data collected during the course of investigation must be

subjected to appropriate statistical treatment. Without making
artificial assumptions, a proper statistical analysis of the results
should be possible. For this, the investigator should have a basic
knowledge on statistics; otherwise, seek the help of a statistician.

[166]
 SECTION IV. DATA COLLECTION,
MANAGEMENT, AND ANALYSIS

11. Collecting Data

Data collection is an important feature of the whole research
process. When you collect data, your intention is to make
inferences based on the collected data. Sometimes, the
information you are trying to find out is already available in
records but needs to be extracted. However, this may not be the
case always, and in most situations, data have to be generated by
doing some research.

Data are measured facts used as a basis for reasoning, calculation,

or decision-making. Datum is singular and data is plural.
However, in technical writing, the plural form “data” is always
used.

Before anything else, data collection must start from the

determination of the type of data required based on the objective
of the research, sampling procedures to be followed, and the type
of instruments needed to collect the data from the selected
sample.
The specific types of parameters to be recorded during data
collection are different for different studies and difficult to
exhaustively cover here. Several steps are involved in collection
of research data.

[167]
Sampling
A parameter, a value which represents a certain feature of a
population, is usually unknown (and therefore has to be
estimated). The population mean, for example, is a parameter that
helps in indicating the center or an average value of a population
characteristic. A population parameter for a given population does
not vary as it is a fixed value but it obviously requires a complete
enumeration of all items in the ‘population’ or the procedure
known as a census inquiry in cases such as population census. It
can be presumed that in such an inquiry, when all items are
covered, highest accuracy is obtained. Nevertheless, this type of
inquiry is practically not affordable as it involves a great deal of
time, money and energy.
The common method is possible to study only a part of total
population, i.e. a representative sample, to obtain a fairly accurate
sample statistic. Sampling is the process of taking a
predetermined number of observations from a larger population,
which is meant for statistical analysis. Thus, sampling is a
procedure of selecting a representative sample from a big mass of
population.
A statistic is calculated from a sample which is randomly drawn
from a population and a sample statistic which is obviously used
to estimate the corresponding unknown value of the population
parameter. A good example is a mean value of a sample which is
used to give information about the mean of the corresponding
population from which that sample was initially drawn.
For better representation of a population by a single or multiple
samples, a good sampling design, i.e. the basic framework or road
map for the selection of a sample, is required.
[168]
The characteristics of a good sample design include: good
representation of the corresponding population (with a small
sampling error), enables to monitor and control any systematic
bias that may be encountered in the process, and the results of
such a sample study can be applied for the population with a
reasonable level of confidence.
A major problem of sample evaluation is the occurrence of
sampling errors, as only a part of the population is measured in
sampling. Although there will not be any sampling errors in
complete enumeration, possibilities of non-sampling errors are
greater. If you want to have high degree of precision, take large
samples and use superior instruments for measurements.
Sampling and non-sampling errors
In sampling, instead of studying a whole population, you study a
fraction of the population, the sample, and infer the situation to
the population under investigation based on the sample statistics.
This is actually an inductive process. Inferences are made by
observing a small section of people, animals, or plants and
extrapolate them for the whole population, which they represent.
Note that representative sampling is a prerequisite for
successfully averaging out random errors and avoiding systematic
errors. The quality and usefulness of your inference depends on
how best representative is the sample.

The error occurring because of the likely faults in the sampling

process is called sampling error. Sampling error is the extent to
which a sample drawn from a population differs from the original
population. Errors can also occur due to other reasons like errors
in measurement, investigator bias during data processing, and
interpretation. Such errors are called non-sampling errors. It
[169]
includes personal error also, for example, things hidden or
forgotten. When the population is large and the sample size is
small, chances of errors are more. The degree of precision can be
increased by taking larger samples and by using superior
instruments for measurements.

In census, sampling errors are absent, only non-sampling errors;

while in sampling, both types of errors occur. When the sample
size increases, sampling error decreases, but non-sampling error
increases. When the sample size becomes equal to the population,
there is no sampling error as there is no difference between the
population and sample.

In order to minimize occurrences of errors in data collection, one

should follow good data collection practices. A researcher should
try to review the data collected immediately after completing the
measurement to detect any anomalous data, for example,
unusually high or unusually low readings. The doubtful figures
need to be checked again by taking observations from the specific
plot. Always keep the sample until after the analysis of data is
completed. Another source of error is the differences between
individuals in taking the measurements. Keep the number of
persons involved in data collection to the minimum. Similarly,
allow the same individual to evaluate all the plots in one block. It
is important to double check row dimensions and plot size as
plots are harvested, and avoid mislabelling data by counter-
checking to reduce errors.
Sampling methods
Sampling methods are broadly classified as probability sampling
and non-probability sampling.

[170]
Probability sampling methods
In probability sampling, each constituent of the population has a
known probability of being selected. Because of this character,
sampling error can be estimated—a major advantage of
probability sampling. When extrapolating data from samples to
that of population, values are presented plus or minus the
sampling error. Common probability methods include simple
random sampling, systematic random sampling, and stratified
random sampling.
Simple random sampling
As the name indicates, it is the simplest form of probability
sampling where individual units are drawn from the population at
random. In this sampling method, every item in the population
has an equal chance of being included in the sample. For a small
and almost a uniform population the simple random sampling
method is most preferred. When you attempt sampling, ensure
that every constituent of the population has an equal chance of
being selected. You have to fix up the number of samples based
on the size of population, and every element of population has the
same probability of being selected. In such a design, the selection
process starts by picking some random point in the list and then
every nth element is selected until the desired number is secured.

If a researcher has to select a sample of 150 items from a

population of 7500 items, then he/she can list the names or
numbers of all the 7500 items on slips of paper and conduct a
lottery. The researcher can use the random number tables as
another method of random sampling. To select the sample, each
item is assigned a number from 1 to 7500. Then, 150 four-digit
random numbers are selected from the table. The researcher will
[171]
have to select some random starting point and then a systematic
pattern is used in moving through the table. The researcher might
start in the 4th row, second column, and proceed down the column
to the bottom of the table and then move to the top of the next
column to the right. When a number exceeds the limit of the
numbers in the frame, in our case over 7500, it is simply passed
over and the next number selected that does fall within the
relevant range. Since the numbers were placed in the table in a
completely random fashion, the resulting sample is random.
Thus, data collection will be made from the representative
samples only and not from the entire population. For example,
when you conduct an experiment with crops, you may not
measure the heights of all the crop plants in a plot but only those
of sample plants selected through simple random sampling.
Techniques of simple random sampling
Perfect randomness is an important factor in sampling because
many of the common statistical techniques used to process results
are valid on truly generated random data only. The method of
selecting a sample at random depends to some extent on the size
and nature of the population. The following are some of the
commonly used methods to select samples:

Drawing lots: The lottery method is a commonly used procedure

for random sampling especially when the population is small.
Names of all the items or individuals in a population are written
on pieces of papers, and random draws are made.

Selective-service numbers: In this method, the units are arranged

in a particular order, for example, alphabetically, numerically, or
geographically. Then, you use selective-service numbers for
[172]
selecting the sample. For example, every 10th, 20th, or Nth unit can
be selected.

Random numbers: The most commonly followed approach for

scientific experiments. Usually, random numbers prepared by
Fischer and Yates or Tippett are used to draw samples. Presently,
computer generated random numbers are also in use.

Grid system: Grid system is mainly employed for studying an

area, especially in ecology and geography using quadrats. A
quadrat is a small plot to isolate a standard unit of area for
studying the distribution of an item over a large area, and
particularly suitable for plants, slow-moving animals such as
snails and millipedes, and some aquatic organisms. The actual
size of the quadrat is decided by the habitat being sampled and by
the purpose of the study. Usually, square quadrats with sides 0.5
m are used in plant sampling from experimental plots. In the
simplest form of using quadrats, the quadrat is thrown to fall at
‘random’ within the site. However, this is unsatisfactory in most
cases as personal bias can creep into the throwing act, that is, it
may not be truly random. Moreover, this ‘throwing’ would be
possible only in cases where quadrats of small size are used.
Normally, it may not be possible to throw anything larger than a 1
m × 1 m quadrat. Therefore, a better method to use quadrat is to
map the entire area and then to lay a numbered grid over the map.
A random number table is then used to select the grids for
observations.
Systematic random sampling
Sometimes, a variation of simple random sampling—systematic
random sampling— is adopted if the entire population is finite or

[173]
can be listed. It is also called an Nth name selection technique.
This method of sampling is useful when sampling frame is
available in the form of a list. The selection starts by taking some
random point in the list and then every nth element is selected
until the desired number of samples is obtained. Thus, first, the
required sample size is determined, suppose it is 50 from a
population of 1000. Then, the number of intervals is found out by
dividing the population by the sample size. One would select the
first item by choosing a randomly selected item, and then, every
40th (Nth record) item is selected from a list of the population until
the sample of 50 items is completed. For example, if the first
sample is the 6th item, subsequent samples will be 46th, 86th, 126th,
and so on until we get 50 samples. This kind of systematic
sampling is as good as random sampling
Stratified random sampling
Stratified random sampling is a commonly used probability
method when large samples are involved. In such cases, it is
considered superior to random sampling as it helps to reduce
sampling error. Stratified sampling is adopted usually for
sampling heterogeneous populations. The population is divided
into homogeneous subgroups or strata, and from each subgroup or
stratum, a random sample is drawn. In other words, the
population is stratified into a number of non-overlapping sub-
populations or strata and sample items are selected from each
stratum. Before going for sampling, however, the researcher has
to identify relevant strata with some common characteristics and
their actual representation in the population. Afterwards, for each
stratum, subjects are selected in proportion to its frequency in the
population using random sampling procedure.

[174]
Cluster sampling and area sampling
Cluster sampling refers to the method of grouping the population
and then selecting the groups or the clusters rather than individual
elements for sampling. Suppose a Zonal Agricultural Office
wishes to sample farmers in a Peasant Association or a district
who have bank accounts. The bank has issued 2000 bank account
books to its farmer customers. The sample size is to be kept say
150. For cluster sampling this list of 4500, the bank account book
holding farmers could be formed into 45 clusters of 100 card
holders each. Three clusters might be selected for the sample
randomly. The sample size must often be larger than the simple
random sample to ensure the same level of accuracy. This is
because the potential for order bias and other sources of error is
usually high. The clustering approach can, make the sampling
procedure relatively easy and increase the efficiency of field
work, particularly for conducting interviews. Area sampling is
similar to cluster sampling but is often applied when the
geographical area to be studied is too vast. Therefore, first, the
total area is first divided into a number of smaller non-
overlapping area. These are called geographical clusters. Then,
the smaller areas are randomly selected, with inclusion of all units
in the sample. Area sampling is particularly useful when the list
of the target population is not available. It also enables making of
efficient field interviewing for interviewers can administer a
number of interviews at each location.
Multistage sampling
It is a type of stratified sampling suitable for infinite populations,
where a list of members is absent, or when the individuals are
living in widely scattered groups. This is also called cluster or

[175]
area sampling. It is an extension of cluster sampling. The method
is meant for big investigations extending to a considerably large
geographical area like the whole of a country. The population is
first divided into different stages, and random samples are drawn.
Initially, the population is divided into first stage sampling units,
from which a random sample is selected. This sample is then
divided into second stage units, and again a sample is selected. In
this way, a random sample is selected at each stage. There must
be at least two stages in this type of sampling.

For example, the first stage may be to select large primary

sampling units such as states, then zones, then districts, then
towns and finally certain families within towns or peasant
associations. Then technique of random-sampling is applied at all
stages.
Non-probability sampling methods
This sampling method involves purposive or deliberate selection
of particular units of the population under study for constituting a
sample or samples. In non-probability sampling, samples are
chosen from the population using some non-random procedures.
This method of sampling includes convenience sampling, quota
sampling, judgment sampling, and snowball sampling. A major
disadvantage of non-probability sampling is that the extent to
which the sample differs from the population remains unknown
and, therefore, it is very difficult to estimate sampling error. Non-
probability sampling methods are generally used for qualitative
studies, and include the following.

[176]
Convenience sampling
Convenience sampling is a non-probability sampling technique,
which is also called accidental sampling. It is most often used in
descriptive research where the concern of researchers is to get an
inexpensive estimate of facts. As the name implies, selection of
samples is based on convenience. Convenience sampling occurs
when samples are drawn from the population based on ease of
access. However, such a procedure may give very biased results
particularly when the population is not homogeneous. This non-
probability method is often used during preliminary stages of
research to get a rough estimate of the results.
Judgment sampling
Judgment sampling is often used when the universe (population)
is small and a known characteristic of it is to be studied
intensively. In judgement sampling, the researcher’s judgement is
used for selecting items which he/she regards as representatives
of the population. For example, a judgement sample of college
students might be taken to secure reactions to a new method of
teaching. Judgement sampling is often used in qualitative research
when the researcher intends to formulate a hypothesis rather than
to generalize to larger populations.
The researcher selects the sample based on some judgment,
especially when the entire population is inaccessible. This is
actually an extension of convenience sampling. Suppose that an
investigator has to take samples from several districts. The
researcher may decide to select samples from one representative
district only instead of several districts, having convinced that the
chosen district is truly representative of all the districts to be
sampled. Although bias can occur in judgment sampling, you can

[177]
still have good representation of population, if you can do it
objectively.
Quota sampling
Quota sampling is an important form of non-probability sampling.
In general, in stratified sampling, the cost of taking random
samples from individual strata is often expensive. Therefore,
interviewers are given quotas to be filled from different strata.
The selection of the samples is left to the interviewer’s
judgement. This is what constitutes quota sampling. The size of
the quota for each stratum is generally proportionate to the size of
that stratum in the population. Quota samples generally happen to
be judgement samples rather than random samples.

Thus, in quota sampling too, the researcher first identifies the

strata and their proportions as they are represented in the
population. However, after selecting the stratum, samples are
drawn from each stratum using the procedure of convenience or
judgment sampling unlike stratified sampling, where each stratum
is filled by random sampling.
Snowball sampling
Snowball sampling is a special non-probability sampling method
for situations where the desired sample characteristic is rare. In
certain occasions, locating experimental subjects may be very
difficult or costly. In such cases, the researcher can try referrals
from initial subjects to generate additional subjects. Although this
technique reduces difficulties of researchers in locating samples
and cut the cost on searching, this may increase sampling bias.

[178]
11.1. Types of data collected in research
Data that researchers collect are generally grouped into two types,
namely, primary and secondary data. The grouping is based on
their source.
11.1.1. Primary data
Primary data are original data that are collected by researchers
from the actual research he or she has done. Thus, when the data
are collected by direct observation or survey, they are called
primary data. Primary data, also known as raw data, give first-
hand information to the researcher after processed statistically.
For primary data, specific method of data collection is required.
However, for secondary data, what is required is mere
compilation of the data from its source.

Secondary data are data that have been already collected or

generated by other researchers or agencies, and are used for
research purposes by other researchers. Thus, secondary data are
not original data. If the data are collected from already published
books, census reports, journals, theses, project reports, published
statistics, and similar documents, they are called secondary data.

However, sometimes, some data could be considered both

primary and secondary data. For example, weather data is the
primary data to the Ethiopian Meteorological Agency but it
becomes secondary data when used by other researchers as the
main or supportive information for a research purpose.

Based on the type of research design, primary data may be

obtained from experimental fields or through survey studies. Data
from experiments are collected based on the results of an
investigation to explore the hidden facts in accordance with the
[179]
research objectives set. However, in descriptive research data are
collected through a sample survey technique or a census survey
method from the existing area under study. Methods commonly
used in primary data collection include observation, interviews,
questionnaires, schedules, and survey.
Observation method
Observation of nature is the fundamental basis of science. It is a
purposeful and selective way of watching and recording an
interaction or phenomena as it occurs. When you conduct an
experiment, you observe and record information on several
features. For example, if it is a growth analysis study of plants,
you will observe and record number of leaves, leaf area, total dry
weight, stem weight, height, root weight, and so on. Similarly,
when you are conducting a titration study with chemicals or
designing a device and testing its functions, you are making
observations. In social sciences too, observation is made in
several situations such as the behaviour of a group, personality
traits of an individual, and functions of a worker.

Observations can be divided into three types as direct observation,

natural observation, and participant observation based on the way
in which data are collected.
Direct observation
Direct observation is the norm in natural sciences such as physics,
chemistry, biology, and most applied sciences. It can also be used
in social sciences. The researcher watches the events with a
dispassionate perspective. In all kinds of experiments, researchers
note down observations directly as the variables show signs of
changes. However, the observer should try to be as unobtrusive as

[180]
possible to avoid bias in observations. Researchers can use
modern techniques such as microscopes, hidden cameras,
recorders, and videos to record observations.
Natural observation
In natural observation, researchers passively observe and record
the behaviour, phenomena, or some other events in its natural
background. The researcher will not in any way interfere with the
system. This kind of observation has been a successful method in
such physical sciences as astronomy, geology, oceanography, and
meteorology. In social sciences, humans or animals are observed
as they go about their activities in real life situations. In natural
sciences, this may involve observing plants, animals, or some
physical phenomena in natural settings. In health science-related
areas such as human anatomy, observation is the primary method
used to describe the construction of the body. In ecology,
observation has always been of importance although experimental
methods have also been utilized.
Participant observation
Participant observation is one of the most common methods for
collecting qualitative data. In participant observation, researchers
live with the subjects being observed as an active participant and
record extensive field notes. They take part in the daily activities
of people, reconstructing their interactions and activities in field
notes taken on the spot or as soon as possible after their
occurrence. Compared to direct observation, participant
observation often requires long periods, months, or even years of
intensive work. Time is a major constraint with participant
observation as a data collection method.

[181]
Observation is used to collect data from structured experiments as
well as from descriptive research studies. This method of data
collection is used more commonly in experimental research than
in social science studies. This method of data collection is
planned in advance, and carefully and thoughtfully selected. Data
collecting using the method of observation often yields accurate
findings and results in sound conclusions. If observation is taken
accurately, subjective bias may also get reduced. When using this
method of data collection, a researcher should bear in mind the
following facts: (a) what to observe, (b) how observations are
required to be noted, (c) how to ensure the accuracy of the data,
etc.

There are two kinds of observation. These are structured

observation and unstructured observation. In a structured
observation, the definition of units to be observed, the style of
observation, the method of recording, the standardization of
condition of data collection and selection of the pertinent data
observation, etc., are all settled well in advance. However, when
data collection is done with no prior planning, the method is
known as unstructured observation. Generally, the structured
observation is used in descriptive type of studies, whereas the
unstructured observations are mostly used in an exploratory type
of research.

Depending upon the participation or nonparticipation of a

researcher/investigator/enumerator, the observation methods of
data collection can again be divided into participant and
nonparticipant methods. In a participant method of observation,
the observer (i.e., the researcher/the investigator/enumerator) acts
as a member of a group in which he/she shares his/her
[182]
experiences. Thus, in the participant method, the added advantage
is that the observer can extract information which were not
thought earlier but may be found suitable while discussing.
Nonparticipant observation method of data collection leads the
observer to record the feelings/experiences of others maybe
without disclosing the identity. As such, this method is also
known as disguised observation. In a disguised method of
observation, participants are free to exchange their experiences
without being cautious by the presence of an investigator. But, in
the process, the investigator may not be in a position to ask for
any additional explanation or experience arising out of discussion.
Both participant and nonparticipant observation methods are
applicable in social as well as experimental studies.

Observation methods of data collection may again be grouped

into controlled and uncontrolled observations. When observations
are recorded in accordance with the prearranged plans of the
observer, which involves experimental procedures, the
observation is known as controlled observation. On the other
hand, when the observation takes place in normal natural orders,
without requiring any pre-plan for the recording of the
observations, it is called an uncontrolled observation. In social
sciences, particularly in the studies of human behaviour,
uncontrolled observations provide natural and complete behaviour
of the human society. These are mostly applicable in an
exploratory type of a research study. Controlled observations
mostly take place in laboratory or field experiments.

In research, observations must be taken with utmost care as

personal bias could be encountered in the process. If the observer

[183]
is biased, she/he can introduce biased observations and may come
up with a wrong conclusion. For instance, sponsored research
from corporate firms or projects funded by sources with dubious
agenda may result in intentional biases. Personal bias of a lesser
degree may also occur when selecting plots for a particular
treatment or locating the plots. In field experiments, unintentional
biases can be eliminated by adopting the three principles of
experimentation—randomization, replication, and local control.
Interview method
This is one of the most common methods of data collection,
particularly in social and behavioural sciences. Oral
communication is the main theme behind such method. Different
questions, sometimes called stimuli are presented, to the
respondents to record their responses.

Interview is a personal form of research compared to a

questionnaire. For seeking opinions or impressions from a person
or group of persons, interview is an ideal choice. An interview is
a verbal interchange, often face to face, in which an interviewer
tries to elicit information, beliefs, or opinions from another
person. In interviews, the interviewer works directly with the
respondent. If required, the interviewer can probe further by
asking follow up questions. Interviews can be conducted in many
ways; it may be a structured interview, unstructured interview, or
semi-structured interview based on the interview schedule used
and the context. However, quantitative analysis is possible only
from the numerical data generated out of structured interviews. If
you need only qualitative information, then either semi-structured
or unstructured interviews can be adopted.

[184]
Structured interviews
In structured interviews, the interviewer uses a pre-prepared
questionnaire (an interview schedule) and conducts the interview
as such without much variation. A major advantage of structured
interview is that the quantitative data generated are amenable to
statistical analysis.
Unstructured interviews
This takes the form of a conversation between the informant and
the researcher. There is no standardized list of questions, and it is
a free flowing conversation in a natural setting. In an unstructured
way, it focuses on the respondent’s perception or opinion on
various issues. This is also called open-ended interviewing or in-
depth interviewing.
Semi-structured interviews
Semi-structured interview makes use of a guide list with some
broad questions or issues, which are to be discussed for possible
investigation during an interview. The list normally includes
broad and open-ended questions to be answered in a free flowing,
conversational, relaxed, and informal setting. The interviewer is
left free to rephrase the questions and to ask probing questions for
added details. Semi-structured interview is one of the main tools
used in ‘participatory learning and action’ (PLA), an umbrella
term for a variety of participatory approaches to development.
Interviews can broadly be classified as personal interview,
telephonic interview, and chatting.
Personal interview
In a personal interview method of data collection, a researcher or
an interviewer generally asks some questions and notes down the
respondent’s responses, while in a direct personal interview
[185]
method, an interviewer collects the information directly from the
respondent. This is comparatively easy, and data collection takes
place then and there. But, in some cases, it may not be possible to
contact directly the person concerned due to various reasons; in
such cases, indirect oral examinations can be conducted and of
course followed by cross-examinations from other persons who
have sufficient knowledge about the problem under investigation.
An indirect method of interview is mostly useful if respondents
are high profile in nature or are members of the commission and
committee appointed by the government for specific
investigations.

A personal interview can again be structured or unstructured. As

usual, a structured personal interview is concerned with the use of
a set of predetermined questions with standardized technique of
data recording. These are mostly useful in descriptive and
experimental type of research studies. An unstructured personal
interview method is characterized by a freestyle approach of an
investigator for getting the answers to questions as per situations
the investigator finds them suitable. In this process, the
investigator has a greater freedom to include and exclude some of
the answers and also to inquire some relevant supplementary
questions if he/she found them suitable. The unstructured
personal interview method depends greatly on the capability of an
interviewer for its success. In an exploratory type of research or a
formulative type of study, an unstructured personal interview is
mostly useful.

Interview is a method of learning the hidden views at the heart of

the respondent and, therefore, the skill and capability of an
interviewer greatly influence the outcome of an interview. In
[186]
many of the situations, a specific and skilled interviewer is
required to conduct the process. As such interviews can further be
classified into three categories: (a) focus interview, (b) clinical
interview, and (c) nondirective interviews.

In a focus interview method, an interviewer is free to design and

prepare a sequence of questions. Mostly this is a conversation-
based interview in which an interviewer concentrates in getting
information from the respondent about the subject in which the
respondent has enough experience. Clinical interview is
concerned with the recording of information about the feelings
and opinions of individuals about their experiences in their own
lives. However, in a nondirective interview method, information
on a particular aspect is recorded from the respondents. In this
method, the work of an interviewer is to stimulate the respondent
to go on talking about his/her feelings, beliefs, and experiences on
a particular aspect.
Criteria for better interviewing
 An interviewer should be selected carefully; only those
persons who are well acquainted and honest and have the
intelligence to capture the essence of the interview should be
selected for this purpose.
 A selected interviewer should be trained adequately so that
there would be no ambiguity arises in understanding the
questions and the expected responses.
 An interviewer should be well behaved, honest, sincere,
hardworking, and impartial. Interviewing is an art of
extracting the inner heart of the respondent.
 An interviewer must enjoy the confidence and faith of the
respondent.
[187]
  An interviewer’s approach must be friendly, courteous,
conversational, and unbiased.
 An interviewer should refrain from asking undesirable and
unwarranted questions.
Surveys
The survey method, a common tool for data collection in social
sciences for descriptive studies, does not involve direct
observation by a researcher. Rather, inferences about behaviour or
situations are made from data collected through interviews or
questionnaires. Surveys are highly useful when researchers are
interested in collecting data on characteristics of human
behaviour, which are often not amenable to direct observation.
Surveys are widely used in social sciences such as sociology,
psychology, economics, and management to assess attitudes,
opinions, or prevalence of something in different subjects. In
agricultural and medical sciences too, these are used especially
for those studies involving farmers and patients, respectively.
Cross-sectional studies and ex-post-facto studies generally use
survey methods for data collection. A limitation of a survey is that
it depends on a self-report mode of data collection, and therefore,
problems such as poor memory, non-cooperation, purposeful
cheating by way of providing wrong responses, vague responses,
or imprecise responses because of confusion in questions can
contribute to inaccuracies in collected data. Furthermore, as this
method is descriptive and not explanatory or analytical, it is
incapable of offering any insights into cause and effect
relationships. Many kinds of survey procedures including attitude
surveys, questionnaires, and interviews are used by researchers.

[188]
Explanatory research analyses and justifies the reason behind the
occurrence of particular phenomenon or association between the
variables. It fundamentally answers the “Why” type of questions.
It explains why a relationship, association, or interdependence
exist. Some of the examples of explanatory research are “Why is
continuous use fertilizer for some years creates a situation that the
land becomes progressively unproductive next time when no
fertilizer is applied to the crop”?
Attitude surveys
Attitudes are predispositions to react positively or negatively to
some social objects. In attitude surveys, also called opinion
surveys, standardized questionnaires indicating the degree of
favourability towards an object are used to measure attitudes or
opinions. It is mainly based on the individual’s agreement or
disagreement with a number of statements relevant to the attitude
object. An attitude scale usually consists of the belief component
of the theoretical attitude in a number of statements. Many
attitude scales like the Thurstone scale and the Likert method
were developed by researchers. The Thurstone technique of
scaled values which is a method to facilitate interval scale
measurement. The researcher selects a large number of evaluative
statements, usually 20 or more, about a particular group,
institution, idea, or practice. The statements should be
unambiguous and cover a full range of attitudes towards the topic;
and after the set of statements are made, a panel of judges are
asked to rate each statement, either agree or disagree, and finally
analysis is done. However, in surveys with questionnaires, the
Likert scale is the most widely used scale. A Likert item is a
statement, which the respondent is asked to evaluate according to

[189]
any kind of subjective or objective criteria. Often, the level of
agreement or disagreement is measured on 5- or 7-point scale
response levels. For example, if the 5-point scale is taken, 2 are
subjective degrees of affirmation, 2 are subjective degrees of
negation, and 1 neutral. The format of a typical 5-point scale
Likert item is 1 stands for strongly disagree, 2 for disagree, 3 for
neutral (neither agree nor disagree), 4 for agree, and 5 for strongly
agree.
Questionnaire
A questionnaire consists of a series of pre-determined questions
that can be self-administered, administered by mail, or asked by
interviewers. The information is collected usually through mail by
sending a proforma. The respondent is asked to fill it, and there is
no interaction between the investigator and the respondent.
Hence, simple questions only are included in questionnaires.
When the questionnaire is administered by an interviewer, it is
often called an interview schedule or simply schedule. Here,
investigators ask questions directly to the respondents using the
schedule. They frame a set of questions, and while asking
questions, they can make some variations in the mode of asking
questions.

Questionnaires are used in research based on a fundamental and

important assumption that the respondents are willing to give
truthful answers. They must also be able to give responses freely.
Interviews or questionnaires commonly include three types of
questions, namely closed questions or forced-choice questions,
open-ended questions, and scale items.

[190]
Forced choice items
Forced-choice items or closed items allow the respondent to
choose from two or more fixed alternatives. Most frequently used
are the dichotomous items, which offer two alternatives only, for
example: ‘yes/no’ or ‘agree/disagree’. Sometimes, a third
alternative is also given as ‘undecided’ or ‘don’t know’. The
alternatives offered should cover all the possibilities.
Open-ended items
The respondents are free to answer anything to the questions
offered to them. The answers may be a short note or short essay
depending on the nature of question. No restraints are imposed on
their answer to the question.
Scale items
As already mentioned, the scale is a set of verbal items to which
the respondent answers by indicating degrees of agreement or
disagreement. Individual responses are located on a scale of fixed
alternatives, for example: ‘strongly disagree’ to ‘strongly agree’.

Each and every questionnaire should reach the respondent with a

request from the researcher in a very humble approach clearly
mentioning the need for the study and the help needed from the
respondent. The size of the questionnaire is a matter of great
concern. If one gets a very lengthy questionnaire, he or she may
get discouraged to fill it. Therefore, the questions should be kept
as minimal as possible. Repetitions and unnecessary questions
should be avoided. Questions should be framed in a logical
sequence so that both the respondent and the researcher become
very much clear about each other with respect to the objective of
the study vis-a-vis the questions framed.

[191]
Clarity in the nature of questioning is the most important factor in
getting an actual response. The degree of clarity of the questions
asked and their wordings are the most important things.
Therefore, simplicity in language and communication is also the
art of framing a good questionnaire. Thus, a researcher should
communicate to the respondent in a very simple and sober manner
using simple language. Any vague or ambiguous or confusing
question should be avoided. Questions should be very short,
simple, and informative. One question should convey only one
message at a time. The researcher should try to avoid asking any
personal questions which may make the respondent hesitant in
answering those questions.

Questions related to personal income, income tax paid, sale tax

paid, and extramarital relationship should be avoided. Asking
personal information greatly influences the overall response of the
respondent. As such, personal questions, if at all required to be
asked, should be placed at the end of the questionnaire.
Instructions in filling out the questionnaire will provide an
additional advantage to the respondent and may encourage
him/her to respond positively to the request of the researcher. The
researcher should clearly clarify the code, groups, or other special
notations used in the questionnaire. A declaration on the part of
the researcher stating that the information provided by the
respondent should be kept confidential and should not be used for
other purposes other than the purpose mentioned in the study
without prior permission from the respondent may encourage the
respondents in providing feedback to the questionnaires.
For comprehensive and big studies, pretesting of questionnaire,
i.e. application of the questionnaire to a small group of
[192]
respondents before it is being used for the main study, is essential.
This is also known as pilot survey. There are several advantages
of a pilot survey as, based on the feedback from the pilot survey,
the researcher can modify or adjust, incorporate some new
questions, and delete some unwanted or nonresponsive questions.
Schedule method
The method of data collection through a schedule method is
almost similar to that method of data collection through
questionnaires with a slight difference. Under this method, an
enumerator or a researcher prepares a list of questions and takes it
to the respondents to fill them in during the process/time of
interviewing. While the questionnaires are being filled in by the
respondents themselves, the schedule is most likely to be filled in
by an enumerator or investigator. The added advantage of data
collection through schedule is that the enumerator or investigator
clarifies or explains the questions which seem to be difficult to
answer by the respondent. The success of this method of data
collection depends on the capability of the enumerator. The
enumerator should be trained in such a way so that he/she
becomes competent enough about the objective of the project and
his/her duties. Though this method is time and resource
consuming, a good, qualified, and capable enumerator can deliver
wonderful information even beyond the scope of the research
studies.
11.2.2. Secondary data
Secondary data are obtained from different sources like journals,
booklets, books, monograms, technical reports, newspapers,
Internet, etc. In any country, each and every state government and
central government along with their different establishments

[193]
publish different data and reports, which can be used as sources of
secondary data.
Prior to using any secondary data, a researcher should make sure
that the data are original and authentic, with very much care about
the originality or authenticity of the data. Thus, reliability,
accuracy, and adoptability of the data are the most important
features. Note that the sources of secondary data may be
published or unpublished.

Reports and documents prepared by different research scholars,

universities, and establishments also serve as good sources of
secondary data. Among the unpublished sources of information
are diaries, letters, biographies, autobiographies, and other
relevant unpublished documents.

11.2. Data measurement

Measurement is generally the process of assigning numbers to

objects or observations. The level of the measurement is a
function of the rules under which the numbers are assigned.
It is not difficult to assign numbers to features or characteristics
of properties of some objects, phenomena, etc. However, it is
relatively difficult to do this for other features or characteristics of
certain phenomena. For example, unlike measuring physical
weight such as yield, height, etc. or biological property such as
age, etc., which are easily and directly measured with some
standard unit of measurement, it is not straightforward to measure
intelligence, happiness, color, race, etc. Thus, a researcher can
expect high precision in measuring the length of maize cobs in
cm, but for abstract concepts, if measurement tools are not
[194]
standardized, the researcher will be less confident about the
precision of the measurement.
In agricultural research, we often measure height, weight, age,
length, etc. We also try to measure how good a crop variety or an
animal breed is, how good an improved heifer is, how good a
prototype of an improved farm implement is, how best the
adoption of a new technology is by farmers, etc.

In assigning numbers to objects or observations, there are two

considerations: (a) assigning numbers in respect of the property or
quantity of some object and (b) assigning numbers relative to
others.

The second aspect is relatively difficult. For example, measuring

social integrity, intelligence, varietal adjustments, etc., requires
close observations and attentions compared to measuring physical
quantities like height, weight, and age. In the measuring process,
one needs to have some scales. In fact, in literature, one can find
different scales of measurement; sometimes, these are also known
as levels of measurement. Mostly one can find nominal, ordinal,
interval, and ratio scale/levels of measurement.

Researchers need to measure several characteristics of the

variables under study. It is essential to have a numerical method
for describing observations. Four scales of measurement are
commonly used depending upon the nature of variables; namely
the nominal or classificatory scale, the ordinal or ranking scale,
the interval scale, and the ratio scale as described earlier.

[195]
 (a) Nominal scale
Nominal scale (also called classificatory scale or categorical
scale) is the most elementary method of quantification of
observation, and the least precise. This scaling is done for
nominal variables. A nominal scale describes differences between
characters by assigning them to categories. By using nominal
scale, you can classify animate beings, inanimate objects, or
events into a number of mutually exclusive categories, so that
each member of the subgroups has some characteristics in
common. A nominal scale is used when variables do not have a
natural order or ranking. You can include numbered or
unnumbered variables, but common survey examples include
gender, location, political party, pets, and so on. For example, the
variable ‘gender’ can have two categories, males or females; and
considering ‘educational qualification’, six categories can be
suggested, non-literate, literate, matriculate, graduate, post-
graduate, and doctorate.
For example, if a researcher wants to measure attendance of a
conference based on sex, i.e., as the ratio of male to female, the
researcher may tabulate those have attended the conference
according to sex. If the object in the domain appears to be male,
assign to “0” and if female assign to “1”. Similarly, we can record
a person’s marital status as 1, 2, 3 or 4, depending on whether the
person is single, married, widowed or divorced. The researcher
can as well record “Yes or No” answers to a question as “0” and
“1” (or as 1 and 2 or perhaps as 59 and 60). In this artificial or
nominal way, categorical data (qualitative or descriptive) can be
made into numerical data and if the various categories are thus
coded, they are referred to as numbers recorded as nominal data.
[196]
Nominal data are numerical in name only, because they do not
share any of the properties of the numbers we deal in ordinary
arithmetic. For example, if the researcher records marital status as
1, 2, 3, or 4 as stated above, he/she cannot write 4 > 2 or 3 < 4
and we cannot write 3 – 1 = 4 – 2, 1 + 3 = 4 or 4 ÷ 2 = 2.
Using nominal scale, you can only allot individuals to a category,
but cannot rank them. Sometimes, code values are given for
processing data, for example, 1 for female and 2 for male.
Individuals having a single value are alike and those having
different values are different. The labels tell us that the categories
are quantitatively different from each other (Table 7). However,
they have no quantitative significance, implying that they cannot
be added, subtracted, multiplied, or divided. Probably, the only
arithmetic operation possible with nominal scales is counting.
Table 7. Educational level of people in a village showing nominal scale (fictitious data)

Male Female Total

Non-literate 5 7 12

Literate 80 52 132

Matriculate 78 86 164

Graduate 46 42 88

Postgraduate 26 22 48

Doctorate 3 2 5

(b) Ordinal scale

In those situations, where one cannot do anything except set up
inequalities, such data are referred to as ordinal data. In other
words, an ordinal scale of measurement is a one-step upgradation
of a nominal scale. Ordinal scales place events in order, but the
[197]
intervals between two consecutive orders may not be equal.
Ranking and gradation are examples of ordinal scales. In an
ordinal scale of measurement, there is a sense of greater than or
less than, and events are categorized not only to provide numbers
but also to indicate the order of events. This scale has no absolute
zero point. Though the intervals are unequal, the ordinal scale
provides more information than the nominal scales by virtue of
their ordering properties of the events. For example, crop varieties
or improved heifers may be ranked or graded according to their
desirable performances. Thus, the top five varieties of a given
crop may have numbers assigned to them starting from 1 to 5,
with the highest yielding variety being assigned number 1. In
economic surveys, families may be categorized into poor, lower
middle class, upper middle class, higher income group, etc. Thus,
this grouping not only assigns particular group to a family
concerned, but also it describes its position relative to others. The
variables measured with ordinal scales can be put under statistical
calculations like median, percentile, rank correlation coefficient,
and chi-square test.

For instance, if one mineral can scratch another, it receives a

higher hardness number and on Mohs’ scale the numbers from 1
to 10 are assigned respectively to talc, gypsum, calcite, fluorite,
apatite, feldspar, quartz, topaz, sapphire and diamond. With these
numbers we can write 5 > 2 or 6 < 9 as apatite is harder than
gypsum and feldspar is softer than sapphire, but we cannot write
for example 10 – 9 = 5 – 4, because the difference in hardness
between diamond and sapphire is actually much greater than that
between apatite and fluorite. It would also be meaningless to say
that topaz is twice as hard as fluorite simply because their

[198]
respective hardness numbers on Mohs’ scale are 8 and 4. The
greater than symbol (i.e., >) in connection with ordinal data may
be used to designate “happier than” “preferred to” and so on.

In a similar way, if you want to study the level of infestation of a

disease in a crop, scores can be given by visually observing the
disease infestation and assessing its severity. The scores can be
arranged either in increasing or in decreasing order. If 0–5 is the
range of scores to be given for the level of intensity of infestation,
you can give score 0 for no infestation and 5 for the severest
infestation. The scores would be 0 = no infestation, 1= very slight
infestation, 2 = slight infestation, 3 = moderate infestation, 4 =
severe infestation, and 5 = very severe infestation. You can rank
the units as 5 > 4> 3 > 2> 1 > 0.
With a nominal scale, the researcher may only be able to indicate
that certain things differ as they fit into certain categories.
However, with an ordinal scale (also called ranking scale), one
may be able to assert the amount or degree of their differences.
For ordinal variables, ordinal scaling is done. Note that they will
have all the properties of a nominal scale; but in addition, it is
possible to rank the subgroups in a certain order. They can be
arranged either in ascending or descending ranks according to the
magnitude of variation, but actual differences between adjacent
ranks may not be equal, as they have no absolute values.
All the properties of nominal scale can be applied to ordinal
scores too, but not vice versa. For ordinal scale, frequency can be
identified. As in the case of nominal scores, in ordinal scale too,
individuals with the same scores are treated alike. After tabulating
and ranking the scores, these can be subjected to analysis.

[199]
Consider the example given in Table 8. In the example, ranks
were allotted based on the degree of adoption of integrated pest
management (IPM) in wheat by farmers. The lowest score of ‘0’
was given to ‘no adoption’ and the highest rank of ‘5’ was given
to ‘full adoption’.
Table 8. Scales of measurement showing ordinal scale
0 No adoption
1 Slight adoption
2 Partial adoption
3 Moderate adoption
4 Majority adoption
5 Full adoption

(c) Interval scale

The ordinal scale is an improvement over the nominal scale.
However, it still poses a problem. Since the actual differences
between adjacent ranks are not equal with ordinal scale, the
differences between two values do not have any specific meaning.
If a researcher intends to elucidate such differences, he/she should
opt for interval scales. Interval level scores are higher than ordinal
scores in such a way that the differences or intervals convey some
more meaning. All characters described in ordinal level scores are
described also in interval level scores. However, besides, an
interval scale uses a measurement unit, which enables the
responses to be placed at equally spaced intervals in relation to
the spread of the variable.
Thus, when in addition to setting up inequalities we can also form
differences, we refer to the data as interval data. Thus, interval
scales of measurement are further improved over an ordinal scale
of measurement. In this type of measurement, scale numbers are
[200]
assigned to objects or events which can be ordered like those of
ordinal scale with an added feature of equal distance between the
scale values.

For instance, the difference between 3 and 4 is same as the

difference between 10 and 11. Because the interval is same in
both cases, the differences between the characters are also the
same.

In general, interval scale is the combination of the nominal and

ordinal scales. But, in the interval scale, there is no true zero. But,
some variables do not have a universally constant zero. Interval
scale is applicable for interval variables coming under continuous
variables. Most quantitative measurements are suited to interval-
level measurement. The interval scale has a starting point and a
terminating point divided into equally spaced intervals. However,
the starting and terminating points and the number of intervals
between them are arbitrary and vary from scale to scale. An
interval scale can have an arbitrary zero, but it may not be
possible to determine what could be called an absolute zero.

For example, take into consideration the following temperature

readings (in degrees Fahrenheit): 48°, 53°, 60°, 85°, 100°, 116°
and 125°. In this case, one can write 100° > 60° or 85° < 125°
which simply means that 100° is warmer than 60° and that 85° is
cooler than 125°. One can also write, for example, 85° – 60° =
125° – 100°, since equal differences in temperature are equal
because the same amount of heat is required to increase the
temperature of an object from 60° to 85° or from 100° to 125°.
On the other hand, it would not mean much if we said that 116° is
twice as hot as 58°, even though 126° ÷ 58° = 2. To reveal this

[201]
fact, we have only to change to the centigrade scale, where the
first temperature becomes 5/9 (116 – 32) = 46.6°, the second
temperature becomes 5/9 (58 – 32) = 14.4° and the first figure is
now more than two times the second. This problem arises from
the fact that Fahrenheit and Centigrade scales both have arbitrary
origins (zeros) i.e., the number 0 of neither scale is indicative of
the absence of whatever quantity we are trying to measure.

(d) Ratio scale

When in addition to setting up inequalities and forming
differences we can also form quotients (i.e., when we can perform
all the customary operations of mathematics), we refer to such
data as ratio data. In this sense, ratio data includes all the usual
measurement (or determinations) of length, height, money
amounts, weight, volume, area, pressures etc.

A ratio scale, in addition to having equal interval properties of an

interval scale, has two additional features. It has a true zero,
meaning a fixed starting point. Therefore, it is possible to indicate
the complete absence of a property. Another feature is that the
numerals of the ratio scale have the qualities of real numbers and
can be added, subtracted, multiplied, and divided, and expressed
in ratio relationships. Measurement of height, weight, area,
income, and age are examples of this scale. Ratio scale is
applicable for ratio variables coming under continuous variables.

Thus, a ratio scale is also a type of an interval scale with an equal

interval between the consecutive scales along with the added
feature of having the true zero point on the scale. The demerit of
an interval scale of having no absolute zero point of measurement
is overcome in a ratio scale. Thus, the presence of zero point on
[202]
the scale benefits the scale in comparing two events, with their
respective positions. The ratio scales have wider acceptability and
use. Generally, almost all statistical tools are usable with the
variables measured in a ratio scale. Examples of a ratio scale
include variables that possess an absolute zero like age, weight,
and height.

The differences between nominal, ordinal, interval and ratio data

stated above may suggest the use of particular statistical
techniques. When data are measurable in units that are
interchangeable e.g., weights (by ratio scales), temperatures (by
interval scales), that data is parametric and can be subjected to
most types of statistical and mathematical processes. But when
data are measured in units which are not interchangeable, e.g.,
product preferences (by ordinal scales), the data is said to be non-
parametric and is susceptible only to a limited extent to
mathematical and statistical treatment. Therefore, a researcher has
to be aware about this aspect data while measuring characteristics
of situations or phenomena or of abstract concepts.
In general, in physical and natural sciences, variables are mostly
expressed in ratio scale. However, in behavioural sciences such as
sociology and psychology, researchers are generally limited to
describing variables in ordinal scale, nominal scale, or interval
scale justifying the use of nonparametric tests.
Some parametric tests used for interval data are the following:

T-test: The aim is to compare the means of the variables. It

requires two samples.
For example, it can help determine the difference between yields
of two different genotypes.
[203]
ANOVA: The aim is again for the comparison of means of the
data. It requires a minimum of three samples. For example, it can
help calculate the difference in the average yield of a genotype
from three tests.

Pearson’s correlation: The aim of this test is to assess the

correlation between the attributes of the data. It requires two
variables from the data.

Simple linear regression: The aim of this process is to determine

regression. It requires two variables from the data. For example, if
you want to generate a survey to analyse the relationship between
income and happiness, you can use this procedure to find if there
is a linear relationship between the two.
Use of interval scale
Questionnaires are often used to collect data meant for interval
scale. To get response from the target audience the choice of the
option must be properly generated. This will help assign a
numerical value to the data which will make the difference
between two variables equal and easy to measure.
Likert Scale
Likert-type scales (summated scales) (or) are developed by using
the item analysis approach. In his approach, a particular item is
evaluated on the basis of how well it discriminates between those
persons whose total score is high and those whose score is low.
Those items or statements that best meet this sort of
discrimination test are included in the final instrument.
The Likert-type- scales consist of several statements which
express either a favourable or unfavourable attitude a given
[204]
situation a respondent has to react. The respondent indicates his
reaction. Each response is given a numerical score, indicating its
favourableness or unfavourableness. The scores are then summed
with each statement in the instrument. This means the total score
represents the respondent’s position towards the situation. In a
Likert scale, the respondent is asked to respond to each of the
statements in terms of several degrees, usually five degrees of
agreement or disagreement. For example, when asked to express
his opinion how satisfied he/she is with the performance of a
released crop variety or a breed of cow, a farmer may respond in
any one of the following ways: (i) strongly agree, (ii) agree, (iii)
undecided, (iv) disagree, (v) strongly disagree.
You can use any kind of Likert scale to determine the value, by
denoting each attribute with a number. For example, on an agree-
disagree scale you can allot numbers as following:
Each question may measure a separate component of that overall
topic. For example, if you want to assess attitudes towards a new
crop variety or a new improved dairy cow, you can design a
Likert scale with a variety of questions that measure different
aspects of this topic.
Likert scales commonly have 5 or 7 items, and the items on each
end are called response anchors. The midpoint is often a neutral
item with positive items on one side and negative items on the
other. Each item is given a score from 1–5 or 1–7.
More items give you deeper insights but make it harder for
participants to decide on answers because there are more choices.
Fewer items mean you capture less detail, but the scale is more
user-friendly.

[205]
Types of items in Likert scales
You can measure a wide range of perceptions, motivations, and
intentions
Some of the most common types of items include:
Agreement: Strongly agree, Agree, Neither agree nor disagree,
Disagree, Strongly disagree
Quality: Very poor, Poor, Fair, Good, Excellent
Likelihood: Not at all likely, Somewhat likely, Extremely likely
Experience: Very negative, Somewhat negative, Neutral,
Somewhat positive, Very positive
15.4. Causes of errors in data measurement

For any research project to have meaningful research findings, the

measurements taken during the process should be precise and
accurate. Any failure or shortfall on this aspect may seriously
affect the research findings. As such, every researcher should
attempt to minimize the error of measurement. Mainly there are
four sources of errors in measurement: the respondent, the
situation, the measurer, and the instrument.

(a) The respondent

Reluctance on the part of a respondent may be due to the fact that
either he/she has tremendous negative feelings or may have very
little knowledge about the subject concerned, but very much
reluctant to express his/her ignorance. In addition to this, fatigue,
apathy, anxiety, etc., may cause problems in responding
inaccurately. That is why researchers strive to overcome biases
arising from the use of a single method or a single observer
through approaches like triangulation, i.e. an effort which enables

[206]
validation of data by using multiple methods, multiple observers
and multiple investigators.

(b) Situation
The situation during the time of collection of information plays a
great role in manifesting an effective research output. A situation
(strained/easy) during the time of data collection, particularly for
socioeconomic research, may enhance or distort information
needed for a research project. Sometimes, a respondent may feel
that anonymity is not assured and she/he may be at risk as a result
of which he/she may be very reluctant to provide genuine
information. The presence of somebody may also hinder or ease
out an information extraction process.

(c) The measurer

The behaviour and attitude of a measurer or a surveyor, and
interviewer are very important in getting reliable and valid
information. Particularly in social studies, the style of interactions
of interviewers with the respondents may encourage or discourage
the data gathering process. Faulty recording or tabulation may
also hamper the authenticity and creditability of the data.

(d) The instrument

Faulty or an unstandardized instrument may give rise to
inaccurate data. Similarly, in social and market studies, complex
languages beyond the comprehension of a respondent must be
avoided. A researcher must know how to take a measurement
correctly under the given social norms, resource and time.
Whatever the planning hypothesis may be, if the data gathered is

[207]
not up to the standard because of faulty instruments, it will have
some adverse effects on the output of a research program.

12.5. Criteria for good measurement scale

Validity, reliability, and practicality are the three major criteria

one should consider while evaluating any measurement tool.

Validity refers to the degree to which the instrument measures

what is supposed to be measured. Validity specifies the utility of a
measurement scale; it actually measures the true differences
among the values measured.

Reliability is measured in terms of stability and equivalence

aspects. Stability refers to the consistency of results under
repeated measurements, while equivalence aspects consider the
amount of error that may be introduced by the different
investigators or different samples of the items being studied.

Practicability is measured in terms of economy, convenience, and

interpretability. Economy refers to a trade-off between the budget
allocated for a particular project and the budget for an ideal
project. While convenience suggests the ease of handling an
instrument or measurement, the interpretability consideration is
supported or supplemented by detailed interaction for
administering the test. The key scoring points and evidence about
the reliability are the guide for interpreting results.

[208]
12. Organizing Data

Working with data starts with the scrutiny of data. This is the data
organizational stage, i.e. the practice of categorizing and
classifying data to make them more usable.

Thus, before analyzing data, the researchers need to put the data
into a standard format like the excel spreadsheet or other formats
in the given statistical package that can be used for the analysis
method. Consistency is highly required starting from this stage
particularly different people are entering data, for instance, on
different response characters are coded.
Researchers cannot rule out occurrence of errors no matter how
properly the experiment designed. At the data organization stage,
erroneous data points are identified, cleaned, edited or trimmed
out for effective and efficient analysis. Each and every data point
should be checked for deviation from expectations if, for instance,
anything is incomplete, not understandable, or out of range.
For describing and analysing the results, a basic understanding on
statistics and its application in agricultural and life sciences is
necessary. A number of statistical software have a built in system
purely meant for organizing the collected data by showing
outliers, estimating missing values, trimming outliers, etc. An
outlier is an observation in a data set which are far removed as an
unusually large or an unusually small value compared to the
others. If an outlier is a genuine result, it is important because it
might indicate an extreme of behavior of the process under study.
For this reason, all outliers must be examined carefully before
embarking on any formal analysis. Some erroneous data may be
[209]
corrected by referring back to the original data collection sheet or
logbook or the researcher may be compelled to re-collect the data
by running the same test again, for instance, by re-analyzing the
same sample if the data are originally from laboratory analysis.

Management of data that violate assumptions of statistical

analysis may also be treated at data organizational stage. In
normal cases, application of F-test in the analysis of variance is
valid only when some assumptions regarding the data are correct.
Thus, the analysis of variance and regression methods are based
on the assumptions of independence of errors, normal
distribution, homogeneity of the error variance, and the concept
of additivity of variance. Violation of these assumptions affects
the sensitivity of the F-tests in the analysis of variance.
Significance tests of mean comparisons are also affected; the
level of significance may be much different from what is assumed
and this can result in invalid conclusions if the assumptions are
not fulfilled.
The assumptions of independence of errors (experimental errors
are independent) is usually achieved by adequate randomization.
It is assumed that the error (residual) of an observation is not
related to that of another. The other assumption of ANOVA is the
normal distribution of error (revealed by frequency distribution).
A histogram of the residuals should always be obtained in order
to check whether the assumption of normality is seriously
violated. Computer packages can also be used to produce a
normal plot of the residuals. One can use a histogram of the
residuals or a normal plot of the residuals or a graph of the
residuals versus the fitted values.

[210]
The other assumption of the analysis of variance is homogeneity
of the variance. If this assumption is justified, then two or more
sample variances can be pooled and the test completed. This
pooling is only valid if the within-treatment variation is similar
for each treatment.
The last assumption of the analysis of variance is related to the
concept of additivity of variances (treatment effects and
environmental effects are additive). Validity of the analysis of
variance is also based on the assumption that the yields are made
of components which are added together.
As a case example, assumptions of analysis of variance
(ANOVA), particularly test of heterogeneity of error variance for
which several methods are available, may be considered. As the
first step towards remedial measure for handling variance
heterogeneity, a correct diagnosis of the specific type of variance
heterogeneity present in the data must be made before appropriate
remedial measures are selected as, for instance, a simplified
procedure is presented below.
First, the variance and mean across replications are computed for
each treatment; second a scatter diagram is plotted between the
mean and the variance values (note that the number of points in
the scatter diagram equals the number of treatments unless and
otherwise there are overlapping points); third the scatter diagram
is visually examined to identify the pattern of relationship, if any,
between the mean and the variance.
Figure 2 below depicts three possible outcomes of such an
examination. The first part of the Figure (left) shows
homogeneous variance as there is no functional relationship

[211]
between the mean and the variance (Figure 2a); the second part of
the Figure (middle) shows heterogeneous variance when there
was a strong functional relationship between the mean and the
variance (Figure 2b); and the third part of the Figure (right) shows
heterogeneous variance when there was no clear functional
relationship between the mean and the variance but simply
because of existence of some outliers (Figure 2c).
There is no need to go for data transformation for the data shown
in the first part of the Figure 2a (left) as it has a readily
homogeneous error variance but the data presented in the second
part of the Figure 2b (middle) need proper transformation as the
data has clearly heterogeneous error variance. However, for the
data shown in the last part of the Figure 2c (right), even if the data
has heterogeneous error variance, there may be no need to go for
data transformation as trimming of the limited outliers may
resolve the limitations due to error variance heterogeneity in such
particular cases.
Data transformations
If your data do not conform to the assumptions, which validate
the analysis of variance, it may be possible to transform them to a
new scale conforming to the assumptions. Analysis of variance is
taken up on the transformed data and inferences are made
accordingly with the transformed means. In this case, analysis of
variance carried out on the transformed data will be valid.
Presentation of results should include the untransformed and
transformed treatment means.
In general, transformations should be skipped unless they are
obviously necessary to avoid biased tests of significance. As the
analyses can be carried out very quickly on the computer, you are
[212]
advised to compare the results with and without transformations.
If alternative analyses lead to the same conclusions, it is best to
present the results based on the untransformed data. The
transformation chosen should be the one which most closely
satisfies the analysis of variance assumptions; this may not be the
one which gives you the results you are looking for!
Among the different types of transformation to make the data
corrected for analysis of variance, logarithmic transformation,
square root transformation, and angular transformation are widely
used.
Logarithmic transformation
This is most suitable for data where the standard deviation is
proportional to the mean or where the effects are multiplicative.
Examples include the number of insect pests observed per unit
area, number of egg masses per plant or unit area, number of
insects per plant and number of diseased plants per plot. Usually,
simple logarithm of the values (log X) is used. These data are
counts that frequently cover a wide range from single-digit
numbers to numbers in hundreds or thousands.

Thus, the number of plant insects, number of egg mass per unit
area, number of larvae per unit area, etc., are typical examples
wherein variance is proportional to the mean and logarithmic
transformation can be used effectively. The procedure is to take
simply the logarithm of each and every observation and carry out
the analysis of variance following the usual procedure with the
transformed data. However, if in the data set small values (less
than 10) are recorded, then instead of taking log(x), it will be
better to take log (x + 1), i.e. if the data contain zeros or numbers
closer to zeros, add one to each number before taking logs.
[213]
The final results or inference should be drawn on the basis of
transformed mean values and calculations made through
transformed data. However, while presenting the mean table, it
will be appropriate to recalculate the means by taking the antilog
of the transformed data. In practice, the treatment means are
calculated from the original data because of simplicity of
calculations, but statistically the procedure of converting
transformed mean to original form is more appropriate. If there is
a mismatch in the two procedures, then the procedure of
converting the transformed mean with the help of the antilog is
preferred over the other procedure.
Square root transformation
Data needing a square root transformation are typically number of
events per unit of time or space when the event is considered rare,
for example numbers of insects caught in a trap, the number of
death per unit time, the number of infested leaf per plant, or the
percentage of infestation (disease or pasts) in a plot (either 0–30%
or 70–100%).
Usually, square roots of the data are used (X½). However, if the
data set include small values (less than 10), especially when zeros
are present, use (x + 0.5)½ for transformation.
Some researchers recommend the square root transformation for
leaf area because it is measured in square units. Small increases in
length when the length is large are associated with large increases
in area. This transformation is also recommended for percentage
data in the range 0–30% or 70–100% (skewed to the left or right)
because for such data, the variances tend to be proportional to the
mean.

[214]
Arcsine transformation
Angular transformation is suitable for proportions, data obtained
from a count, and data expressed as fractions or percentages. For
arc sine transformation, usually a table prepared by Fisher and
Yates is used. However, for percentage data lying between 30 and
70 percent, no transformation is required. For percentage data
lying within the range of either 0–30 or 70–100 percent (but not
both), the square root transmission is most ideal. For percentage
data that do not follow the ranges noted, arc sine transmission
could be used.
Percentage data (e.g., percentage of carbohydrate, protein, and
sugar) which are not arising out of count data may not commonly
need transformation. Moreover, all the percentage data arising out
of count data need not be subjected to arcsine transformation
before analysis of variance:
(a) For percentage data ranging either between 0 and 30% or
70 and 100% but not both, the square root transformation
should be used.
(b) For percentage data ranging between 30 and 70%, no
transformation is required.
Percentage data which overlaps the above two situations should
only be put under arcsine transformation. Thus, a data set having
0 to more than 30%, less than 70–100%, and 0–100% should be
put under arcsine transformation. The values of 0% should be
1 1
substituted by 4𝑛 and the values of 100% by 100 – 4𝑛 , where n is
the number of counts on which percentages are worked out before
transforming the data following arcsine rule of transformation.
The essence of arcsine transformation is to convert percentage
data into angles measured in degrees, that is, to transform 0–
[215]
100% data into 0–90⸰ angles. The actual procedure is to convert
the percentage data into proportions and transform them it into
sin–1 √𝑃 , where p is the percentage data measured in proportions.
Ready-made tables are available for different percentage values
with their corresponding transformed values. However, in MS
Excel with the following functional form, percentage data can be
directly converted to arcsine transformed data as: degrees
(arcsine(sqrt(p/100))), where p is the percentage value. As a
general guide to deciding between a Log and a Square Root
Transformation, if the ratio of standard deviation to treatment
mean is constant, for data from a randomized complete block
design (CRD) so that treatments with large means have large
values, a log transformation is recommended. Check by making a
graph of standard deviation against treatment mean. If the points
lie approximately on a straight line, a log transformation is
appropriate. However, if it is the variance and not the standard
deviation which is proportional to the treatment mean, the square
root transformation is recommended.

[216]
 a b

Figures 2a–c. Different levels of variance heterogeneity as depicted from the relationship
between intra-treatment means and variances across replications

Pre-transformation of scoring scales

Scoring scales, foliar diseases scoring scales like 0–3, 0-5, 1-5, 0-
9 or 1-9 for instance, should not be directly subjected to ANOVA.
Scoring scales should be pre-transformed to percentage value as
described in Table 9 below and then percentage values should be

[217]
Arc Sine transformed (if error variance is heterogeneous) before
conducting analysis of variance.
Table 9. Pre-transformed percentage values for commonly used rating scales of disease severity
Rating scale Alternative scoring scale
0–3 0-–5 1–5 0–9 1–9
0 0 0 - 0 -
1 25 10 0 3 0
2 75 35 15 12 4
3 100 65 50 25 15
4 90 85 42 30
5 100 100 58 50
6 75 70
7 88 85
8 97 96
9 100 100

13. Analyzing Data

The data collected from primary or secondary sources at the

initial stage are known as raw data. Raw data have to be
processed and analysed as per the requirement of the research
objectives.

Data analysis, which is considered as the first step taken towards

data interpretation, is the process of ordering, categorizing,
manipulating, and summarizing data to obtain answers to the
research questions.
Researchers mostly use data analysis to assess variability among
and within treatments, to study interrelationship between
characters, and treatment by environment interaction and
treatment performance stability across spatial and temporal
variation. Just like data collection, the procedures to be followed
[218]
in data analysis also depend on the particular study. Data analysis
and interpretation should also follow the research question(s).
In general, there are three types of analysis based on the number
of characters handled simultaneously. The first type is Univariate
analysis which is done to determine how the test treatments vary
for a single trait at a time. The common types of univariate
analysis include measures of central tendency (arithmetic mean,
median, mode), measures of dispersion (range, standard
deviation, standard error and variance), analysis of variance
(ANOVA), and tests of significance.
The second type of analysis is Bivariate analysis. Bivariate
analysis involves the simultaneous analysis of two variables for
the purpose of determining the empirical relationship between
them. Examples include correlation, regression and analysis of
co-variance (ANCOVA). Bivariate analyses are conducted to
determine whether a statistical association exists between two
variables, the degree of association if one does exist, and whether
one variable may be predicted from another.
The third type is Multivariate statistical methods, or simply
multivariate methods, are statistical methods for the simultaneous
analyses of data on multiple variables, thereby examine their
simultaneous effect. As the name indicates, multivariate analysis
consists of a collection of methods that can be used when several
measurements are made on each individual or object in one or
more samples. Multivariate analysis is not a specific method but it
encompasses different techniques. Examples include multiple
correlation and multiple regression.

[219]
13.1. Univariate Analysis

Descriptive statistics
During the course of investigation, you may be accumulating
voluminous data, which need to be summarized and presented in
a clear accurate format for easy comprehension. For this purpose
of summarizing a data set, you make use of descriptive statistics
or deductive statistics. Using descriptive statistics, you can reduce
a set of numbers into simple values and describe the main features
of a data set quantitatively. These include calculation of measures
of central tendency such as mean, median, and mode; dispersion
measures such as standard deviation and range; association
measures such as correlation and regression; frequency
distribution tables; cross tabulation tables; and graphs and figures.
Normally, a combination of tables, graphs, charts, and a
discussion is used to summarize and present a data set. A major
drawback of descriptive statistical analysis is that it limits
generalization to the particular group of individuals only.
Therefore, no conclusions are possible beyond this group, and
you cannot assume any similarity to those outside the group.
There are three general types of descriptive statistics used to
describe data, the measures of central tendency, the measures of
spread or dispersion, and the measures of association.

Measures of central tendency are used to describe the central

position of a frequency distribution for a group of data. For
describing this central position, the mean, mode, and median are
generally used. Among these, the most commonly used measure
of central tendency is the arithmetic mean. If all the data fall in a

[220]
‘normal distribution’, the mean, the median, and the mode
coincide.

Measures of spread or dispersion help us to have some idea about

the variation of measured data around the mean. A number of
statistics are used to describe this spread including the range,
percentiles, quartiles, mean deviation, variance, and standard
deviation. Among these, standard deviation is the most popular
one.

Measures of association enable to see whether there is a relation

between two sets of observed data through the use of correlation
and regression techniques. The relationship between two paired
variables can be estimated using correlation. You can estimate
this even by using natural data generated without doing an
experiment. It provides a measure of the degree of association
between the variables or the goodness of fit of a prescribed
relationship of the data. Correlation can be negative or positive.
The extent of relationship can be quantitatively represented by the
coefficient of correlation.

A perfect positive correlation is +1.00, and a prefect negative

correlation gives a value of −1.00. On the other hand, regression
analysis describes the effect of one or more independent variables
on a single dependant variable by expressing it as a function of
the independent variable. The relationship may be simple, if only
one variable is involved, and multiple, if more than two variables
are involved. It may be linear, if the form of the relationship is
linear. Sometimes, the relationship may be curvilinear or
hyperbolic, that is nonlinear. Regression equations or models can
be derived and used to predict the effect of independent variables.

[221]
Other common descriptive statistical tools include percentages,
proportions, ratios, and rates. Frequency distribution tables, cross-
tabulation tables, graphs, and figures are also made according to
the type of observation and data collected.
Measures of central tendency
Measure of central tendency, namely, mean, median, and mode
provide central value or typical representative of a set of scores as
a whole. These measures give a characteristic of the whole group
by a single number. Then, it is possible to compare different sets
of scores to compare the achievement of any different groups.

Mean
Sample mean ( X ), a measure of center of the sample data
including the outliers commonly called the average, is considered
a good estimator for the population mean (). The sample mean,
in addition to giving information about the test sample, is also
important to make inferences (draw conclusions) about a
population, which is defined as the collection of all possible
observations of interest as it may not always possible to deal with
populations because of their too large size to study.
For example, if we want to know the average body weight of
goats in Ethiopia, it is practically impossible to measure each and
every goat in Ethiopia. Likewise, the average yield of the teff
crop in Ethiopia cannot be determined by weighing the whole
harvest. For practical reasons, we need to deal with a sample
where proper estimates of the population mean could be
generated. A sample is a representative group taken at random
from a population and the number of observations in the sample is
called the sample size. When a sample is properly taken, the

[222]
statistics from that sample can be applied to the population. The
mean shows that ½ the members of the population fall on either
side of an estimated value, i.e. the mean. Measured characteristics
of the sample are called statistics (e.g. sample mean) and
characteristics of the population are called parameters (e.g.
population mean). The arithmetic mean is the best-known
descriptive statistic as an estimator of population mean.
The arithmetic mean (denoted by X ) is mathematically defined
as the sum of all the observations in a sample divided by their
number as given in the following formula:

∑𝑋
Arithmetic mean( X ) = , where
𝑁

∑ = summation, X = an observation and N =

number of observation in that particular

sample.

Comparison of mean values could also be made using the 2-test,

t-test, and Z-test as discussed later.
Weighted mean
When we compute a simple arithmetic mean of a set of data, we
assume that all the observed values are of equal importance and
we give them equal weight in our calculation. In situations where
the numbers are not equally important, we can assign to each a
weight which is proportional to its relative importance and
calculate the weighted mean instead of the arithmetic mean. The
weighting procedure is used to find the mean when several sets of

[223]
data are combined. For instance, if some treatments are tested in
two replications and others are tested in three replications for any
reason in the same experiment, all other factors kept constant, the
weighted mean should be used in place of the arithmetic mean.
Suppose we have three sets of data consisting of n1, n2, and n3
observed values and having the means x1, x2, x3, respectively.
Then the mean for the combined data is the weighted average of
the individual means, the respective weights being the sample
sizes n1, n2, and n3.

Weighted mean ( X ) = n1x1+n2x2+n3x3

n1+n2+n3

Other measures of central tendency, less commonly applied, are

the median (for x's arranged in numerical order; if n is odd, the
median is the middle x; if n is even, the median is the average of
the two middle x's.) and the mode (the number in a given set of
numbers that appears most frequently). If two or more distinct x's
occur with equal frequencies, but none with greater frequency, the
set of x's may be said not to have a mode or to be bimodal, with
modes at the two most frequent x's, or trimodal, with modes at the
three most frequent x's.
Measures of dispersion (variability)
Knowledge of the central tendencies (i.e. mean, median and
mode) is not enough to properly describe a given sample or
population. These values give us merely an idea of the general
achievement of the group as a whole, and do not show how the
individual scores are spread out.
[224]
Therefore, normally, the mean should be expressed together with
the degree of variability to have a complete knowledge of how
close or how far the data is clustered around the center, i.e. the
degree that individual observations of a data set are dispersed or
'spread out' around their mean. When the dispersion is large, the
values are widely scattered; when it is small, they are tightly
clustered. There are mainly, four measures of variability or
dispersion. These are range, quartile deviation, average deviation,
and standard deviation. These measures of variability give the
degree of variability or dispersion and tell how the individual
scores are scattered or spread over distribution of the data.
Range
The range is the difference between the lowest value and the
highest value present in the observations in the sample or, in other
words, it is the limits between which variation is possible. The
range is simply calculated by first putting all the numbers in order
and then subtract (take away) the lowest number from the highest.
Variance
Variance is considered as statistical measure of the spread of each
number in the data set is from the mean. That means, the larger
the variance the larger is the deviation of the individual data from
the mean; a small variance indicates the data deviate little from
the mean. The variance of a set of n observations, x1 x2 … xn, is
the sum of squares of the difference between the observations and
their mean divided by one less than the number of sample. The
variance is mathematically expressed as the sum of the square of
the deviation of all values from the mean ( X ) divided by (n-1)
for a sample.

[225]
 ∑(𝐱− X ) 𝟐
S =
2
𝒏−𝟏

Standard deviation
Standard deviation, which is used as a measure of average
deviation from the mean designated by the symbol s when dealing
with a sample, is the square root of variance. The mean is usually
expressed together with standard deviation to have a complete
meaning as standard deviation is a measure of variability that is
more convenient in analysis of statistical data. Standard deviation
is considered the most important measure of statistical dispersion,
measuring how widely spread the values in the data set are. If
many data points are close to the mean, the standard deviation is
small; if many data points are far from the mean, then the
standard deviation is large. If all data values are equal, then the
standard deviation is zero. A useful property of standard deviation
is that, unlike variance, it is expressed in the same units as the
data.

𝑠 = √𝑠 2

For example, if S2 of a given sample for plant height is 2.5, then

the square root of 2.5 is 1.6, which means the measurements vary
plus or minus +/- 1.6 cm from the mean. Likewise, if the mean of
a given sample is 4.49 and the standard deviation is 1.55, the
mean should be expressed as, e.g. 4.49 ± 1.55. The meaning is, on
average, the range of values is from 3.14 to 6.24 (4.69 -1.55 =
3.14; 4.69 + 1.55 = 6.24). It should be noted that the range will
[226]
not be equal to the actual minimum and maximum values because
there are few samples. As the number of samples increases, the
range based on standard deviation will more or less be equal or
very close to the range of values of the actual data.
Standard error
The standard error (SE) is the measure of the difference between
sample mean ( X ) and the population mean (), or in other
words, standard error is a statistical term that measures the
accuracy with which a sample represents a population. The
standard error (SE) of a sample of sample size n is the sample's
standard deviation divided by √n.

𝑠
SE =
√𝑛

Take the example of a sample mean and the population mean. The
sample mean is considered as a probable estimator of the
population mean. However, different samples drawn from the
same population may have different values of the sample mean.
In such cases, researchers would be interested to know how far
their sample means are fluctuating. For this purpose, estimation of
standard error becomes useful. The standard error depends on
mainly two factors, the size of the sample, and the variations of
measurements in the sample indicated by the standard deviation.
For example, standard error of the mean (SEm) is the standard
deviation of all those sample means drawn from the population.
The standard error of the mean (SEm) is estimated by sample
standard deviation divided by the square root of the sample size
assuming statistical independence of the values in the sample.
[227]
The standard error should be clearly differentiated from the
standard deviation. The standard deviation is a measure of the
variability in the particular sample studied. At the same time, the
standard error is a measure of the uncertainty in a sample statistic.
The standard error, which depends on both the standard deviation
and the sample size, is recognition that a sample is unlikely to
determine the population value exactly.
Coefficient of Variation
Coefficient of variation gives an idea about the degree of
precision with which the treatments are compared. It is a good
indicator of the reliability of the research. It provides an idea
about the experimental error, which is obtained by dividing the
standard deviation with the mean and expressing it as a
percentage or proportion of the total. Coefficient of variation is
helpful in comparing the extent of variation between groups with
different means. A higher CV value means that the reliability of
the research is low. While doing an ANOVA, CV is obtained by
dividing the square root of error MS by the grand mean, and
multiplying it with 100.

In general, there are no hard and fast rules on the magnitude of

CV for reliability. However, experiences show that acceptable CV
ranges could vary based on the types of trait, crop, and trial.
Accordingly, for yields, the maximum CV values that are
considered reliable are 20% for field crops (teff, rice, maize,
soybean, wheat, etc.), 25% for horticultural crops (fruit crops and
leafy vegetables), 40% for crops grown under the soil (potato,
sweet potato, onion, radish, carrot, etc.). In general, however,
acceptable CV values may range from 6 to 8% for variety trials,

[228]
10 to 12% for fertilizer trials, and 13 to 15% for insecticide and
herbicide trials.

In livestock research, for traits related to beef cattle nutrition

(weight grain, feed intake, and feed conversion efficiency),
acceptable CV values may range from 4 to 20%. For traits related
dairy cattle nutrition, acceptable CV values may range from 8 to
25% while for traits related to poultry (commercial layers) the
values may range from 2 to 10%.
Inferential statistics
A researcher can make use of descriptive statistics or deductive
statistics for summarizing a data set. When he or she summarizes
and describes the data using descriptive statistical tools, he or she
may also want to examine whether the estimates made in the
study can be generalized and extrapolated beyond the samples
under study. It is also essential to ascertain whether the
differences or associations found can be explained by chance, or
whether there are any real differences. Inferential statistics or
inductive statistics offer the tools to answer these questions. The
tools of inferential statistics are used to make claims about the
populations to which the samples belong.

The main objective of using inferential statistics is to arrive at

conclusions about populations based on observations on
representative samples. We make use of observed data from
samples to make inferences about population parameters.

Representative samples are drawn from a population, and using

the sample statistics, population parameters are estimated.
However, remember that as sample data give only representative
figures, they cannot be perfect estimators of the population.
[229]
Different procedures are followed to make sure that the inferences
made are appropriate and rational.

The properties of populations such as the mean or standard

deviation are called parameters as they represent the whole
population. It is very difficult to have access to the whole
population in which one has interest. However, one may have
access to data from a fraction of the population, a sample. For
example, you might be interested to know the fruit yield from
mango trees in a particular region. As it is difficult to estimate
fruit yield data from all the trees in the region, you may attempt to
measure only a smaller sample of trees, say 200 trees selected at
random, to represent the larger population of all mango trees.
Samples also have properties such as the mean or standard
deviation, which are called statistics and not parameters. Using
the sample statistics, one generalizes the behaviour of populations
from which the samples are drawn. However, it is important that
the selected samples accurately represent the population avoiding
sampling errors to the maximum. In general, inferential statistics
involve two aspects, i.e. estimation of population parameters and
testing of statistical hypotheses.
Tests of significance

Chi-square (2) test

The chi-square test is a nonparametric test used for qualitative
data (especially of the nominal types) to test hypotheses
concerning the frequency distribution of one or more populations.
A Chi-square test for independence compares two variables in a
contingency table to see if they are related.

[230]
In agricultural research, chi-square test is mainly used for testing
the significance of proportions, analysis of attribute data, test for
homogeneity of variance, and test for goodness of fit. Chi-square
test is used widely in breeding experiments, for example: to test
the significance of deviation of an observed segregation from a
theoretical one, simultaneous testing of a number of questions
such as single factor ratios, linkage, and heterogeneity.
The null hypothesis testing for the Chi-square test of
independence is the same as it is for other tests like ANOVA
where a test statistic is computed and compared to a critical value.
Calculating the Chi-square statistic and comparing it against a
critical value from the Chi-square table allows the researcher to
assess whether the observed values are significantly different
from the expected values which some time are established facts
from previous study.

∑(O−E) 2
Chi-square (2) = , where ∑ = summation,
E

O = observed frequencies, and E = expected

frequencies.

The resulting quotients are then all added together to obtain a

single value of 2. The calculated value of 2 is compared with
the table value of 2 at n-1 degrees of freedom, where n is the
number of frequency classes. If the calculated value of 2 is
smaller than the table value, it indicates that the observed value
complies with the expected value and is no more significantly
different. On the other hand, if the calculated value of 2 is
[231]
greater than the critical/tabulated value, it can be concluded that
the observed value is different from the expected value with the
predetermined level of significance (P = 0.05 or 0.01). Thus, the
null hypothesis will be rejected and you accept the research
(alternate) hypothesis.
The t- and Z-tests
The Z-test and the t-test are popular tools to find whether an
observed difference between the means of two groups can be
considered significant. Here, the word ‘significant’ has a specific
meaning. For example, we know that 10 and 12 are not equal, and
the difference is 2. However, 10 and 12 may not be ‘significantly’
different, if the differences are due to chance. Our concern is how
to find out whether the differences are real (significant) and not
due to chance (non-significant). The researcher should have some
skepticism in this regard, especially when using statistics. We
assume that all results are chance results unless shown otherwise
(remember how a null hypothesis is written).

The z-test is generally used to ascertain the significance of

difference between the mean of a large sample and the
hypothesized mean of a population or to compare the difference
between the means of two independent large samples (when n is
greater than 30). The z-test is also utilized to compare the sample
proportion value to a theoretical value of population proportion or
to compare the difference in proportions of two independent
samples when n is large. For applying z-test accurately, the data
should follow normal distribution and standard deviation must be
known. The z-value is worked out and compared with its probable
value, obtained from normal table at a specified level of
significance (5% or 1%). For example, to compare the score of
[232]
students of two classes, we may use the z-test, if standard
deviation is known. Sometimes, we may want to compare the
mean yields of two wheat varieties, oil yields of two soybean
types, oleoresin content of two ginger types, staple length of two
cotton types, or milk yield of two breeds of cows. Our intention
might be to test whether the two samples can be considered as
having been drawn from the same population. If the samples are
large enough to calculate standard deviation, we can use the z-
test.

Note the differences and similarities between the z-test and the t-
test carefully. The z-test follows normal distribution while t-test
follows Student’s t-distribution. A t-test is appropriate when you
are handling small samples (n < 30) while a z-test is appropriate
when you are handling moderate to large samples (n > 30).

The distribution curves of small sample sizes do not strictly

follow the normal distribution and may become flatter at the
centre; called a t-distribution. The logic of the t-test is that if the
difference between two means is large, the variability among the
data is small. Similarly, if the sample size is reasonably large, the
probability is increased that the difference is not due to chance,
but due to a cause. We calculate t-value using a specific formula
considering the difference between the two means and the
variability among the data. The calculated t-value is then
compared with the t-value obtained from the table. If the
calculated value of t or z is smaller than the table value at n-1
degree of freedom (n = sample size), the difference between the
two means is considered non-significant and vice versa.

[233]
The critical values necessary for rejection of a null hypothesis are
higher for small samples at a given level of significance. Each
critical t-value for rejection is based upon appropriate degrees of
freedom. As the sample sizes increase, the critical t-values
necessary for rejection of a null hypothesis decrease and approach
the z-values of the normal probability tables.

If the calculated t value is greater than the table t-value, you have
to reject the null hypothesis at the chosen level of significance
and accept the research hypothesis and proceed further.
Conversely, if the calculated t value is less than the table t-value,
we have to accept the null hypothesis and reject the research
hypothesis. The t-test is often referred to as the Student’s t-test.

We have two commonly applied options. The first option is small

samples with paired observations for which we use the student t-
test given with the following formula:

X −
t= 𝑠 , where ( X ) = sample mean,  = population
√𝑛

mean (assuming that the sample mean is

compared with the population mean), s =

standard deviation of population and n = sample

size or number of observation.

Paired samples (also called dependent samples) are samples in

which natural or matched couplings occur. This generates a data
set in which each data point in one sample is uniquely paired to a
data point in the second sample.
[234]
This data is described as unpaired or independent when the sets of
data arise from separate/unrelated individuals. In that case,
Fisher’s t-test is used for small samples with unpaired
observations as given with the following formula:

X −
t = 1 1
, where ( X ) = mean of first sample mean,
𝑠√ +
𝑛1 𝑛2

= mean of second sample (assuming that one

sample mean is compared with another

unpaired sample mean), and n1 = size of first

sample, n2 = size of second sample, and s =

the combined standard deviation.

Analysis of variance
The t- and z-tests should not serve for comparing the means of
more than two samples coming from different treatment groups
that are normally distributed with a common variance. Although it
is possible to use a number of t- and z-tests to determine the
significance of the difference between more than two means, it
would involve several separate tests. For example, if you want to
compare 6 means, 2 at a time, it would involve 15 separate tests
(number of pair wise comparisons of N items = N (N − 1)/2).
An analysis of variance (ANOVA) should be used in order to
make a statistical test of whether or not
the means of several groups are equal. As the name indicates,
[235]
ANOVA uses measures of variance instead of means as in a t-test.
ANOVA, based on linear regression and general linear models,
also requires the same assumption with the t-test. It is a statistical
procedure which separates the total variation in a replicated
experiment into different components, mainly variance between
treatments (mean squares due to treatment, MST) and
variance within treatments (mean squares due to error, MSE),
thereby providing the basis for the test of significance or what is
called the F-ratio (the ratio of the two, F = MST/MSE) as a
decision tool to determine whether an observed difference is real
or just due to chance. ANOVA, in the example mentioned above,
for instance, makes it possible to determine whether the 6 means
differ significantly with a single test rather than 15 paired tests.
Experiments involving several treatments are usually subjected to
ANOVA.

Three types of models are used in the analysis of variance,

namely, fixed effects models, random effects models and mixed
effects models. The fixed model is used when we are dealing with
a specific population (when the treatments are fixed) with no
inference to some other population, whereas the random model is
used when we are dealing with random samples (when the
treatments are not fixed) from some parent population to which
we make inferences, and the mixed model being the combination
of the two (when some experimental factors are fixed and other
experimental factors random, with appropriately different
interpretations and analysis for the two types).
The real difference among these models in terms of analysis of
variance (ANOVA) is related to the way the F values are
calculated. Table 10 below illustrates how the computation of F
[236]
values varies, assuming different scenarios in terms of the models
to be followed, in a case example where two factors A and B are
factorially evaluated.
Table 10. Calculation of F values in a factorial evaluation of factors A and B with the assumption
of different models

Factor Model

Fixed effect Random effect Mixed model (A Mixed model (A

(Both A & B are (Both A & B are fixed; B random)- random; B fixed)-
fixed) – Model I random) - Model II Model III Model III

Factor A MSA/MSE MSA/MSAB MSA/MSAB MSA/MSE

Factor B MSB/MSE MSB/MSAB MSB/MSE MSB/MSAB
Factor A  B MSAB/MSE MSAB/MSE MSAB/MSE MSAB/MSE

While ANOVA per se tells that the means are different, it does
not tell which means are different from each other when we have
more than two treatments for comparison (but we only know that
the highest and the lowest means are significantly different). If
the null hypothesis is not rejected based on the result of ANOVA,
it is not necessary or appropriate to further deal with the data. If
the null hypothesis is rejected based on the result of the F test,
however, it is appropriate to further question which treatment(s)
caused the effect.
Tests of significance is based on probability statements that our
findings reflect the truth.
The major reason for using the concept of probability in statistical
testing is that we cannot be sure if our inferences, based on the
sample data, apply to the population, since our sample may
deviate from the population.

[237]
The probability that a statistical inference is caused by pure
chance is called the p-value. The p-value is compared with the
significance level (α), which represents the maximum level of risk
that we are willing to take that our inference is incorrect.
Generally, p-value of p ≤ 0.01 is believed to produce very strong,
a value of 0.01 < p ≤ 0.05 strong, a value of 0.05 < p ≤ 0.10
moderate and a value of p > 0.10 weak or none evidences against
the null hypothesis. For most statistical analysis in agriculture,
however, α level set at 0.05 may be sufficient for the purpose of
mean comparison, i.e. the test treatment means are different at
least 95% of the time. A p-value less than α = 0.05 means that we
have enough statistical evidence to reject the null hypothesis and,
thereby, indirectly accept the alternative hypothesis. However, for
medical sciences, which involve human health and wellbeing a
minimum level is at 0.001 is necessary.
Whatever the p-value, we have a number of range tests commonly
used with their own advantages and limitations for mean
comparison or mean separation based on the outputs of ANOVA
or the F test including the least significant differences (LSD),
Duncan’s multiple range test (DMRT), and the Tukey’s test. It
should be noted that any method of mean separation should be
applied if and only if the F test for the treatments is significant
(when the ANOVA has an outcome which fails to accept the null
hypothesis - post hoc tests).
The LSD provides valid test criteria in two situations; making
comparisons planned in advance of data collection (e.g. treatment
with control) and comparing adjacent ranked means (to choose a
winner) but LSD should not be used for all possible pair-wise
comparisons particularly when the pairs are more than six. The

[238]
difference between two means is declared significant at any
desired level of significance if it exceeds the LSD value using
different formula for different designs.
The other method of mean separation, the Duncan’s multiple
range test (DMRT), which is more useful than the LSD when
larger numbers of mean pairs are compared with each other.
DMRT is considered to be more stringent as compared to LSD as
the former requires larger differences between means compared to
the latter. For instance, the same mean differences found to be
significant by the LSD (which is lenient) maybe declared non-
significant when the DMRT is applied.
The Tukey honestly significant difference (HSD) test, like the
DMRT, is used for pairwise comparisons to determine if multiple
means are significantly different from each other in all possible
pairs based on a studentized range distribution (q). Tukey’s test
tends to have the most stringent results as compared to both the
LSD and DMRT.

13.2. Bivariate analysis

Correlation analysis
Researchers in Biology, including Agriculture often find
important to consider the simultaneous variation within a number
of different parameters. He or she may wish to assess
relationships between yield, environmental data, and yield
components. Studies like that, for example, can be used to
investigate whether an increase or decrease in one variable, such
as number of pods per plant, can be explained by change in
another such as plant population and light intensity. Study of

[239]
simultaneous change in two (or more) variables can be carried out
using techniques of regression and correlation.

Correlation is a situation when two variables, say X and Y, tend

to show inter-relatedness in some way of definite manner. The
coefficient of correlation, denoted as r, is the measure of how
closely the two characters are associated. If the two characters
increase in the same direction simultaneously, they are said to be
positively correlated; if one increases while other decreases, they
are said to be negatively correlated. A perfect positive correlation
between the two variables results in a coefficient of +1, a perfect
negative correlation in a coefficient of -1, and a total absence of
correlation in a coefficient of 0. The degree of correlation is
calculated as follows:

Covxy
𝑟=
√ x2 ∗ y2

Where r = correlation coefficient, cov(xy)= co-

variance of traits X and Y, x2 = variance of X

and y2 = variance of Y.

The level of significance of correlation coefficients is determined

from r table (correlation table) at n–2 degrees of freedom and at a
given probability level (p < 0.05 or p < 0.01) or it could be tested
with a t-statistic with n-2 degrees of freedom as:

[240]
 r√n−2
t=
√1−r2

If the value of r or t calculated is less than the corresponding

value of r or t tabulated, respectively, then there is no evidence of
correlation, i.e. the value of r or t is not significant.
Coefficient of determination, calculated as the square of the
correlation coefficient, also helps in better interpreting correlation
coefficients in that it indicates the proportionate relationship or
dependence of dependent variable on independent variable. For
example, if r = 0.9, then r2 = 0.81 showing 81% dependence of
the dependent variable on the independent variable.
Regression analysis
Regression, beyond association between variables, measures the
amount of change in the dependent variable (say Y) associated
with a unit change in the independent variable (say X) or it is a
measure of the strength of relationship between the two variables.
A regression equation is a mathematical relationship which can be
used to determine the expected value of a dependent (or response)
variable for a given value of a correlated independent (or
predictor) variable. The independent variable is generally denoted
by X and the dependent variable by Y. For example, a study of
the effect of change in fertilizer rate (X) on yield (Y) of a crop per
hectare of land is called the regression relationship of Y on X.
The results of a regression analysis provide information on the
average rate of change in yield of the crop per unit increase or
decrease in fertilizer rate. Regression equations can be used to

[241]
provide a succinct summary of large amounts of experimental and
observational data.

Unlike correlation, where the value of the correlation coefficient

does not change whether we treat both variables as dependent or
independent, we need to identify the dependent and the
independent variables causally depending on our theory of the
relationship.

The degree of linear regression between two characters is

calculated as follows:

Y = bX + a; where Y = the mean value of the

dependent variable, X = the mean value of the

independent variable, a = the constant value or

the intercept or the point where the line

crosses the Y axis, and b = the regression

coefficient or the slope of the line.

From the regression equation, it is possible to predict the value of

a dependent variable once the value of the independent variable is
given and vice versa.
Analysis of covariance (ANCOVA)
In statistics, it is often encountered that a variable that is not
supposed to be important as part of the research objective or
target may affect the dependent variable through its relation,
which reduces the precision of the result of the experiment. Such
[242]
a variable, which is called a covariate or a “concomitant” or
“confound” variable, needs to be included in the model to remove
its effects on the primary variable of the research interest. This
allows researchers to account for variations associated not with
treatment variables themselves, but with one or more covariates.

ANCOVA, which combines ANOVA with regression, adjusts the

linear effect of the covariate and improves the precision of the
results. The main difference between the two is that an
ANCOVA model includes a “covariate” which is correlated with
the dependent variable and means on the dependent variable are
adjusted due to the effects the covariate has on it.

Covariates can be used in many ANOVA based designs to answer

if mean differences or interactive effects likely to have occurred
by chance after scores have been adjusted on the dependent
variable because of the effect of the covariate. ANCOVA, which
combines regression analysis and analysis of variance (ANOVA),
controls for the effects of extraneous variable, called a covariate,
by partitioning out the variation attributed to this additional
variable. Extraneous variable or sometimes called “nuisance
variable” is any condition not part of a study but that could have
an effect on the study’s dependent variable. Covariance analysis
controls experimental error and adjust treatment means. Blocking
can reduce experimental error (patterned or unidirectional
variability) but blocking cannot cope with certain types of
variability (non-patterned variability).

Suppose you want to lay out an experiment in a coffee plantation

where the yield levels of the coffee trees differ considerably.
Using the initial yield as the covariate, the final yield after the
[243]
experiment can be adjusted to the values that would have been
attained had all experimental trees started with the same yield.
This means that you have to take some nuisance variables into
consideration to improve precision.

Another example is the layout of a field experiment in a land

where you suspect non-uniformity due to fertility or slope
variations. In such cases, you can collect nuisance variables from
a uniformity trial conducted over the experimental plots before
the layout of actual experimental trial. Then, the yield data from
the uniformity trial is used as an index of variation from plot to
plot because of factors such as soil heterogeneity. Even if you
adopt local control or blocking effectively, such plot-to-plot
variations will unduly reduce your precision. The observation
from uniformity trial can be utilized in estimating these variations
to reduce the error affecting the treatment comparisons.

Another common example is breeding trials, where yield

differences may be noticed due to the failure of some of the seeds
to germinate or subsequent death of some already germinated
plants. By recording plant numbers and yield from plots, this
effect can be accounted by conducting ANCOVA.

In livestock research, there are several applications of ANCOVA.

For example, in an experiment designed to test the effects of
plane of nutrition on the growth performance of animals, various
initial weights or ages at the start of the experiment will affect the
accuracy of the experimental results. It is then required to adjust
the final body weights for differences in initial weight or initial
age, which can be achieved by including the initial weight or
initial age as a covariate in the model.
[244]
Suppose that you have taken all the precautions to design and
conduct an experiment. Even then, observations for certain plots
might be lost or even affected by accidental causes such as pest
attack or damage by animals that it would not be proper to
consider these as normal experimental observations.

In this way, the researcher is better able to investigate the effects

of the primary independent variable. The ANCOVA F test
evaluates whether the population means on the dependent
variable, adjusted for differences on the covariate, differ across
levels of a factor. An ANCOVA will reduce Type II error, i.e. the
null hypothesis is not rejected when it is in fact false. The F-tests
associated with a standard ANOVA are computed by dividing the
MS for the main effect by the MS for error. If MS for error can
somehow be made smaller, then the calculated F will be larger,
and as a result, there is a better chance that null hypotheses will
be rejected.

As a requirement of ANCOVA, the covariate must not be affected

by the treatment being tested. Otherwise, the adjustment removes
both the variation due to experimental error and that due to
treatment effect. Just like ANOVA, ANCOVA also assumes
homogeneity of error variance, independence of error terms,
normality of error terms, the relationship is linear, and the
researcher must carefully select the covariate. In order for
ANCOVA to be effective, the covariate must be linearly related
to the dependent variable. In addition, the covariate must be
unaffected by other independent variables. By statistically
controlling for the variation attributed to the covariate, the
researcher increases the precision (accuracy) of the research by
reducing the error variance.
[245]
13.3. Multivariate analysis

Multivariate statistical methods, or simply multivariate methods,

are statistical methods for the simultaneous analyses of data on
multiple variables, thereby examine their simultaneous effect. As
the name indicates, multivariate analysis consists of a collection
of methods that can be used when several measurements are made
on each individual or object in one or more samples. Multivariate
analysis is not a specific method but it encompasses different
techniques. Examples include multiple correlation and multiple
regression.
Multiple correlation
A multiple correlation coefficient yields the maximum degree of
liner relationship that can be obtained between two or more
independent variables and a single dependent variable.
Multiple regression
Multiple regression, unlike simple linear regression that uses just
one explanatory variable, also known simply as multiple linear
regression, regression, is a statistical technique that uses several
explanatory variables to predict the outcome of a response
variable.
Multiple regression analysis is made with the goal of modeling
the linear relationship between the explanatory (independent)
variables and response (dependent) variables. That means, it is
conducted to determine how the dependent variable Y is related to
the independent variables x1, x2, . . . xn.

[246]
 Y = a + b 1 x 1 + b 2 x 2 +b 3 x 3

Where Y= dependent variable (what is being

predicted or explained), a = y

intercept (constant term), b1 = the slope

(Beta coefficient) for X 1, X1 = first

independent variable that is explaining the

variance in Y, b2 is the slope (Beta

coefficient) for X2, X2 = second independent

variable that is explaining the variance in Y,

b3 = the Slope (Beta coefficient) for X3, and

X3 = third independent variable that is

explaining the variance in Y.

13.4. The concept of parametric and non-parametric

statistics
There are two common tests of significance. These are parametric
and non-parametric tests. Parametric data are measured data, and
parametric statistics assume that all the data are ‘normally’
distributed. A parameter refers to a function of the population. In
the ‘bell shaped’ normal distribution curves, most populations
display the majority of average values at the central bell part of
the curve with extreme values tailing off at each end. If you take
[247]
the case of height of people, it could be seen that most people are
of average height, with a few extremely tall or short cases; that is,
heights of people are normally distributed. Although individual
values vary, the generality of this type of curves among
populations is so strong that the parameter is taken as a constant.

For applying parametric tests, three pre-conditions have to be

met: the observations must be independent; the samples must
have equal or nearly equal variances; and the variables involved
are expressed in interval and ratio scales. Parametric tests include
the z-test, the t-test, and the analysis of variance (ANOVA).

Non-parametric are also called distribution free tests. Such tests

are used when the nature of the distribution of the population is
not known or when the data are expressed as nominal or ordinal
measures. Presentation of variables in a non-parametric test are
usually done in rank order or discrete values. The discrete data
could represent such classifications as high, medium or low,
urban or rural or male or female, counting number in each
category. This means nonparametric data are either counted or
ranked. Thus, sometimes researchers may come across samples
and populations that do not behave in the form of a normal
Therefore, for applying nonparametric tests such as chi-square,
the stringent assumption of normally distributed populations is
not necessary. The Chi-square test is applied to discrete data
(discrete variables are those expressed in frequency counts). The
test is based upon the concept of independence the idea is that one
variable is not affected by, or related to another.

[248]
14. Illustrations of Analysis of Variance

To perform an ANOVA test, we need to compare two kinds of

variation, the variation between the treatment means, as well as
the variation within each of our treatment. We combine all of this
variation into a single statistic, called the F statistic, because it
uses the F-distribution. A few case examples are presented to
illustrate steps in analysis of variance for different experimental
designs and these examples are mostly taken from different
sources including text books.
Example 1. Analysis of variance for CRD
Assume that seven foliar and granular insecticides to control
brown plant hoppers and stem borers were tested in CRD with
four replications (the experimental units were assumed to be
homogeneous). The grain yields of rice resulting from the use of
the insecticides are presented below (Table 11).
Table 11. Grain yield (kg ha–1) of rice resulting from the use of different types and rates of
insecticides

Treatment RI RII RIII RIV Treatment Treatment

total (T) mean
Dol-Mix (1 kg) 2537 2069 2104 1797 8507 2127
Dol-Mix (2 kg) 3366 2591 2211 2544 10712 2678
DDT + -BHC 2536 2459 2827 2385 10207 2552
Azodrin 2387 2453 1556 2116 8512 2128
Dimecron-Boom 1997 1679 1649 1859 7184 1796
Dimecron-Knap 1796 1704 1904 1320 6724 1681
Control 1401 1516 1270 1077 5264 1316
Grand total (G) 57110
Grand mean 2040

[249]
Step 1. To calculate the different sources of variation of the
analysis of variance, the following step should follow:

𝐺2
 Correction factor (CF) = , where G is the grand total
𝑛
and n is the number of observation (i.e. n = r  t).
57,1102
CF = (4)(7)
 116,484,004

 Sum of square of total (TSS)  sum the squared values

of treatments in each plot and then minus the
correction term.
TSS = ∑x2i –CF
TSS = [(2,537)2 + (2,069)2 + … (1,270)2 + (1,077)2] –
116,484,004 = 7,577,412
 Sum of square of treatment (SST)  sum the squared
value of treatments, divide by the number of
replications and then minus the correction term.
∑T 2
SST  – CF
𝑟
(8,507)2 + (10,712)2 +… (5,264)2
SST  − 116,484 =
4
5,587,174
 Error sum of square (SSE)  Sum of square of total
minus sum of square of block minus sum of square of
treatment.
SSE = SS total - SST
SSE = 7,577,412 - 5,587,174 = 1,990,238

[250]
Step 2. To calculate the mean squares (MS) for each sources of
variation by dividing each sum of square (SS) by its
corresponding degree of freedom, the following step
should have followed:

SST
MST = 𝑡−1
5,587,174
MST = = 931,196
6
SSE
MSE = 𝑡(𝑟−1)
1,990,238
MSE = = 94,773
(7)(3)

Step 3. Construction of the ANOVA table (Table 12)

Table 12. Analysis of variance for grain yield of rice resulting from the use of different types and
rates of insecticides

Source of Degree of Sum of Mean Computed F Tabular F

variation freedom squares square (1%)

Treatment 6 5,587,174 931,196 9.83** 3.81

Exp. Error 21 1,990,238 94,773
Total 27 7,577,412

Step 4. Calculation of the coefficient of variation (CV%)

√𝑀𝑆𝐸
CV% = Grand mean  100
94,773
CV% = √2,040  100

CV% =15.1%

[251]
The CV, i.e. the experimental error standard deviation percentage
of the mean, indicates the degree of precision with which the
treatments are compared and is a good index of the reliability of
the experiment.
Note that the “F” value should only be computed when the error
degree of freedom is large enough for a reliable estimate of the
error variance. As a general guideline, some text books suggest
that the F value should be computed only when the error degree
of freedom is 12 or more.
In this case, there were highly significant differences among the
insecticide treatments as the computed F value of 9.83 is far
larger than the tabular F value of 3.81 at the 1% level of
significance. However, note that a non-significant F test in the
analysis of variance indicates not lack of difference among the
treatments but the failure of the experiment to detect the
difference because of either a small treatment difference or a very
large experimental error or both. Thus, whenever the F test is
non-significant, it is advisable to examine the size of the
experimental error and the numerical difference among treatment
means. If both values are large, the trial may be repeated and
efforts made to reduce the experimental error so that the
difference between treatments, if any, can be detected. On the
other hand, if both values are small, the difference among
treatments is probability too small to be detected.
Step 5. Mean separation
It should be noted that a significant F test tells only existence of
difference but does not specify the particular pair (or pairs) of
treatments that differ significantly. To obtain such information on

[252]
which treatment is superior over which, another procedure called
mean separation must be conducted.
There are different ways of mean separation but the most
common include the least significant differences (LSD), the
Duncan’s multiple range test (DMR) and the Tukey’s test.
The LSD provides a valid test criterion when making
comparisons planned in advance of data observed (e.g. treatment
with control and comparing adjacent ranked means (to choose a
winner) but LSD should not be used for all possible pair-wise
comparisons particularly when the number of treatment is large.
For the above experiment, for instance, the LSD value is
calculated as illustrated below:

LSD 0.05 = t √2 ∗ 𝑀𝑆𝐸/𝑟, where r is number of

replication, t is the tabulated value of t at the error
degree of freedom and the prescribed probability level.

LSD 0.05 = 2.08 √2 ∗ 94773/4

LSD 0.05 = 452.78

The result indicates that means of test treatments as far apart as

452.78 kg per hectare compared to the control represent a
significant difference.

[253]
Computation of the Duncan’s Multiple Range Test (DMR)
essentially follows the same procedure but instead of a t value we
use a Q value obtained from a table "the Studentized Range".
Assume that three treatments were tested in three replications and
ANOVA was conducted on the records as shown below (Tables
13a and 13b):
Tables 13a. Records from the test of the three treatments in three replications

Tables 13b. ANOVA Table of the three treatments tested in three replications

Computation of the DMRT goes as follows:

[254]
 DMRT0.05 = Q√2 ∗ 𝑀𝑆𝐸/𝑟

With three treatments and 6 degree of freedom of

error, the Q value from the Studentized Range table is
4.34 at  = 5%.

DMRT0.05 = 4.34√2 ∗ 8.78/3

DMRT0.05 = 10.50
 If we rank the treatments from highest to lowest mean, we
get the following order:
39.3
20.7
11.3
 Compare the highest mean (39.3) with the lowest (11.3)
mean: 39.3 – 11.3 = 28.
 The difference between the highest and lowest means is
greater than both the LSD (5.92) and DMRT (10.50)
values.
 So the highest mean differs significantly from the lowest.
 Compare the second highest mean (20.7) with the lowest
(11.3) mean:
 20.7 – 11.3 = 9.4.
 The difference between the second highest and lowest
means is greater than the LSD (5.92) value.

[255]
  So the second highest mean differs significantly from the
lowest.
 With DMRT we use a different Q value for the remaining
two treatments excluding the highest mean. The Q-value
will change to 3.46 as:

DMRT = 3.46 * 2.42 = 8.37

 The difference between the second highest and lowest

means, i.e. 9.4, is greater than the DMRT value (8.37)
 So the second highest mean differs significantly from that
of the lowest mean.
 If you have more means, continue the same way changing
only the Q -value as you go. Once you have a non-
significant result, you can stop at that point. Note that
DMRT has higher value than the LSD, showing that it is
more stringent.

Tukey’s test, also called Tukey's honest significance test (HSD),

unlike DMRT which generates independent critical values for
comparison of different pairs of means, this test calculates only
one critical but stringent value that can be used to evaluate
whether differences between any two pairs of means are
significant; as:

HSD = 𝑞√𝑀𝑆𝑤/𝑛𝐻, where q is readily obtained from the q table

based degree of freedom of error at K where K is the number of
[256]
treatment, MSw is mean square within groups and nH is harmonic
mean of the treatment means calculated as a geometric average of
the number of treatments in the group or as the reciprocal of the
arithmetic mean of the reciprocals of the given set of observations
(refer to the relevant sources on the calculation of harmonic
mean).

Assume in an analysis, for example, that MSw = 65, K = 4, Dfw

= 11 (df of error) and nH =3.69. In this case, q at Dfw = 11 and K
= 4 is q = 4.26 from the q table. Then:

HSD = 𝑞√𝑀𝑆𝑤/𝑛𝐻
HSD = 4.26√65/3.69 = 17.87

The interpretation of the meaning here is that means far apart by

17.87 units are significantly different from each other. In order to
avoid complication in manual analysis, however, it is advisable to
use computer packages for calculating HSD values.

Example 2. Analysis of variance for RCBD

Assume that a faba bean breeder wanted to compare yields of six
new varieties: Var 1, Var 2, Var 3, Var 4, Var 5 and Var 6, Var 5
and Var 6 being the standard and the local checks, respectively.
He run an experiment using randomized design replicated four
times. At harvest time he measured the yield on each plot. The
field plan and yields (t ha–1) were as follows (Table 14):

[257]
Table 14. Yield of the six varieties

Variety Yield (t ha–1) Total Mean

RI RII RIII RIV

Var 1 1.51 1.49 1.54 1.55 6.09 1.52

Var 2 0.60 0.90 0.74 0.66 2.90 0.72
Var 3 1.43 1.46 1.26 1.28 5.43 1.36

Var 4 1.30 1.20 1.33 1.26 5.09 1.27

Var 5 1.16 1.22 1.16 1.12 4.66 1.16
Var 6 0.98 0.90 0.76 0.82 3.46 0.86

Total 6.98 7.17 6.79 6.69 27.63

Grand mean 1.15

Step 1. To calculate the different sources of variation of the

analysis of variance, the following step should follow:

[258]
 𝐺2
 Correction factor (CF) = 𝑛
, where G is the grand total
and n is the number of observation (i.e. n = r  t).
27.632
CF = (4)(6)
 31.8090
 Sum of square of total (TSS)  sum the squared values
of treatments in each plot and then minus the
correction term.
TSS = ∑xi2 –CF
TSS = [(1.52)2 + (1.49) 2 + … (0.26) 2 + (0.82) 2] – 31.8090
= 1.9619
 Sum of square of block  sum the squared value of
block, divide by the number of treatments and then
minus correction term.
∑B 2
SSB  – CF
𝑡
[(6.98)2 + (7.17) 2 + (6.79) 2 + (6.69) 2] 2
SSB  – 31.8090 =
6
0.0226
 Sum of square of treatment (SST)  sum the squared
value of treatments, divide by the number of
replications and then minus the correction term.
∑T 2
SST  – CF
𝑟
(6.09)2 + (2.90)2 +… (3.46)2
SST  − 31.8090 =33.6446 –
4
30.8090  1.8356
 Error sum of square (SSE)  Sum of square of total
minus sum of square of block minus sum of square of
treatment.
SSE = SS total - SST
SSE = 1.9619-1.8356 – 0.0226  0.10378

[259]
Step 2. To calculate the different mean squares, the following
step should follow:

 Calculation of mean squares block (MSB)  sum of

square of block divided by degree freedom of block.
SSB
MSB = 𝑟−1
0.0226
MSB = = 0.0075
3
 Calculation of mean squares treatment (MST)  sum
of square of treatment divided by degree freedom
treatment.
SST
MST = 𝑡−1
1.8356
MST = = 0.3671
5
 Calculation of mean square error (MSE)  sum of
square of error divided by degree freedom error.
SSE
MSE = (𝑟−1)(𝑡−1)
SSE  0.10378/15  0.0069

Step 3. Calculation of F value for treatment

F  mean square of treatment divided by

mean square of error.
F  MST/MSE  0.3671/0.0069  53.06

Step 4. Construction of ANOVA table (Table 15)

Table 15. Analysis of variance

[260]
 Source of
variation df SS MS F
Block 3 0.0226 0.0075 1.09
Variety 5 1.8356 0.3671 53.1
Error 15 0.10378 0.0069
Total 23 1.9619

 Calculation of coefficient of variation

√𝑀𝑆𝐸
CV% = Grand mean  100
√0.0069
CV% =  100
1.15

CV% = 7.7%

Test of significance
If we look in the F table for the 5% and 1% F values using the
error degrees of freedom F tests larger than the F tabulated which
is 2.90 at 5% and 4.56 at 1%. Since 53.06 is greater than 4.56, the
difference among the mean yields of the six varieties are highly
significant. There is no significant difference between the blocks
since the F calculated (1.09) is far less that less than the F
tabulated (2.90).

[261]
Since the F test for the treatments is significant, we need to make
mean separation.

LSD 0.05 = t √2 ∗ 𝑀𝑆𝐸/𝑟

LSD 0.05 = 2.131√2 ∗ 0.0069/4
LSD 0.05 = 0.125 t/ha
LSD 0.01 = 2.947√2 ∗ 0.0069/4
LSD 0.01 = 0.173 t/ha

Application of the missing plot technique in RCBD

Missing values are almost commonly encountered in the process
of experimentation. For instance, the experimental animal may
die in the middle of the experiment and field plots maybe
damaged by animals. In such cases, we may be forced to apply
the “missing plot technique” in order to estimate the missing
value. The value for a missing plot can be estimated by using the
following formula:
(𝑟𝐵+𝑡𝑇−𝐺)
Yij = (𝑟−1)(𝑡−1)

Where r = number of replicates, t = number of

treatments, B = block total of block with missing value,
T = treatment total of treatment with missing value
and G = grand total of the experiment.

[262]
For example, assume that four treatments (A, B, C and D) were
tested in three replications and the record from treatment C in
block 2 is missing as indicated by the question mark in Table 16
below.
Table 16. Data from four treatments tested in 3 replications with a missing value

Treatment Records Treatment total

RI R II R III
A 9 8 7 24
B 11 13 12 36
C 3 ? 8 11
D 7 10 4 21
Block total 30 31 31 92

Step 1. The missing value (Yij) for treatment C in the second

block is estimated as follows:

(𝑟𝐵+𝑡𝑇−𝐺)
Yij = (𝑟−1)(𝑡−1)
(3∗31)+(4∗11)−92)
Yij = (3−1)(4−1)

Yij = 7.5

Step 2. Substitute the calculated value into the missing spot in the
data as indicated in Table 17 below.
Table 17. Data from four treatments tested in 3 replications with a filled missing value

Treatment Records Treatment total

RI R II R III
A 9 8 7 24
B 11 13 12 36
C 3 7.5 8 18.5
D 7 10 4 21
Block total 30 38.5 31 92

[263]
Step 3. Compute ANOVA following the right procedure

Source of df SS MS F
variation
Block 2 10.792 5.396 1.023ns
Treatment 3 60.063 20.021 3.795ns
Error 5 26.375 5.275
Total 10 97.229
Remember that you will lose one degree of freedom in error and
total for each missing value.
Points to remember about the missing value analysis:
 Use of the estimated value does not improve the analysis
or supply additional information.
 The error mean square calculated using the estimate of the
missing value is a minimum. Use of any other value but
the one calculated would result in a larger value.
 The mean calculated using the estimate of the missing
value is called a Least Square Mean.
Calculating LSD when you have one missing values:
When you have a missing plot, you need to calculate two (not
one) LSD values. The first LSD value is used to compare the
treatments with no missing values while the second is to compare
the treatment with the missing value with those having no missing
vlaue.
LSD value for comparing the treatments with no missing values is
calculated in the usual way as:

[264]
 LSD 0.05 = t √𝟐 ∗ 𝑴𝑺𝑬/𝒓
LSD 0.05 = 2.571√𝟐 ∗ 𝟓. 𝟐𝟕𝟓/𝟑
LSD 0.05 = 4.821

LSD value for comparing the treatment with the missing value
with those having no missing vlaue is calculated as:

2 t
𝐿𝑆𝐷 = t√𝑆 2 [ + ]
r r(r − 1)(t − 1)

2 4
𝐿𝑆𝐷 = 2.571√5.275[ + ]
3 3(3 − 1)(4 − 1)

𝐿𝑆𝐷 = 5.567

Note that the LSD value for comparing the treatment with the
missing value with those having no missing vlaue is more
stringent than the LSD value for comparing the treatments with
no missing values.
The steps become rather lengthy and the approach is a little bit
different when two missing values are calculated as demonstrated
in another example given below using the following hypothetical
data. Note that Treat 2 in block two and Treat 4 in block 1 have
one each missing values.

[265]
Block Treatment Total
Treat 1 Treat 2 Treat 3 Treat 4
1 3.1 3.3 3.6 X2 13.9
2 3.1 X1 3.4 4.0 13.9
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 6.5 10.6 8.2 41.8

Step 1. Temporarily estimate only one of the missing values by

using means keeping the other missing value as still
missing.
3.3+3.2 3.1+3.4 +4.0
X1 = ( + )/2
2 3

X1 = 3.375

Step 2. Substitute the temporary estimate into its proper place as

shown below and then estimate the second missing
value.
Block Treatment Total

Treat 1 Treat 2 Treat 3 Treat 4

1 3.1 3.3 3.6 X2 13.9

2 3.1 3.375 3.4 4.0 13.9
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 9.875 10.6 8.2 37.875

Step 3. Estimate the other missing value (X2) using the proper
missing plot technique.

[266]
 (𝑟𝐵+𝑡𝑇−𝐺)
X2 =
(𝑟−1)(𝑡−1)
(3∗10)+(4∗8.2)−37.875)
X2 =
(3−1)(4−1)

X2 = 4.15

Step 4. Substitute this value in place X2, put the place of X1

missing again as shown below and then estimate the
latter using the proper missing plot technique.
Block Treatment Total
Treat 1 Treat 2 Treat 3 Treat 4
1 3.1 3.3 3.6 4.15 14.15
2 3.1 X1 3.4 4.0 13.9
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 6.5 10.6 12.35 38.65

(𝑟𝐵+𝑡𝑇−𝐺)
X1 = (𝑟−1)(𝑡−1)
(3∗10.5)+(4∗6.5)−38.65)
X1 = (3−1)(4−1)

X1 = 3.14

Block Treatment Total

Treat 1 Treat 2 Treat 3 Treat 4
1 3.1 3.3 3.6 4.15 14.15
2 3.1 3.14 3.4 4.0 13.64
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 9.64 10.6 12.35 41.79

[267]
Step 5. Substitute this value in place X1, put the place of X2
missing again as shown below and then estimate the
latter using the proper missing plot technique.
Block Treatment Total

Treat 1 Treat 2 Treat 3 Treat 4

1 3.1 3.3 3.6 X2 10.00
2 3.1 3.14 3.4 4.0 13.64
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 9.64 10.6 8.2 37.64

(𝑟𝐵+𝑡𝑇−𝐺)
X2 = (𝑟−1)(𝑡−1)
(3∗10)+(4∗8.2)−37.64)
X2 = (3−1)(4−1)

X2 = 4.19

Step 6. Substitute this value in place X2, put the place of X1

missing again as shown below and then estimate the
latter using the proper missing plot technique.
Block Treatment Total
Treat 1 Treat 2 Treat 3 Treat 4
1 3.1 3.3 3.6 4.19 14.19
2 3.1 X1 3.4 4.0 10.5
3 3.0 3.2 3.6 4.2 14.0
Total 9.2 6.5 10.6 12.39 38.69

[268]
 (𝑟𝐵+𝑡𝑇−𝐺)
X1 = (𝑟−1)(𝑡−1)
(3∗10.5)+(4∗6.5)−38.69)
X1 = (3−1)(4−1)

X1 = 3.14

Step 7. Keep on doing the same eteps until you get more or less
constant values; fill in the final values and continue with
your ANOVA. Once you are done with your ANOVA,
the calculation of LSD value when two values are
missing is a little bit different from what we have seen
when one value is missing. It is advisable, in that case, to
use statistical packages like R or SAS to estimate LSD
than manually calculating it to rule out possibilities of
committing a mistake.

Example 3. Analysis of variance for Latin square design (LSD)

Case example 1. Animal research
Assume that an experiment was designed to determine the effect
of five diets (A, B, C, D and a control E) on liver cholesterol level
in sheep. Recognized sources of variation in these sheep were
body weight and age. The researcher selected a 5  5 Latin
Square, and randomly allotted the treatments (diets) to the rows
and columns as shown in Table 18 below, along with the results
of cholesterol levels from different diet treatments.

[269]
Table 18. Liver cholesterol level (mmol/L) recorded on sheep of different weight and age groups
which were fed with different deits.

Weight group Age group Row

I II III IV V
totals

1 A = 3.38 B = 3.37 D = 3.04 C = 3.27 E = 2.44 15.50

2 D = 3.70 E = 2.88 B = 3.35 A = 3.46 C = 3.34 16.73
3 C = 3.58 D = 3.56 A = 3.69 E = 2.67 B = 3.51 17.01
4 E = 3.32 A = 3.71 C = 3.74 B = 3.81 D = 3.41 17.99
5 B = 3.48 C = 3.91 E = 3.27 D = 3.74 A = 3.64 18.04
Column 17.46 17.43 17.09 16.95 16.34
totals
Grand total 85.27

Then compute treatment totals and treatment means as shown in Table 19 below.

Table 19. Summary of treatment total and mean.

Treatment A B C D E
Total 17.88 17.52 17.84 17.45 14.58
Mean 3.576 3.504 3.568 3.490 2.916
Grand mean 3.41

The ANOVA is computed as follows (Table 20):

Step 1. Calculation of sum of squares

[270]
 𝐺2
CF = 𝑟𝑐
85.272
CF = 25
CF = 290.84
TSS = X2 – CF
TSS = [(3.38)2 + (3.37)2 + … + (3.64)2] – 290.84
TSS = 2.97
𝑅 2
SSR = – CF
𝑡
(15.50)2 + (16.73)2 +⋯ (18.04)2
SSR = – 290.84
5
SSR = 0.87
𝐶 2
SSC = – CF
𝑡
(17.46)2 + (17.43)2 +⋯ (16.34)2
SSC = – 290.84
5
SSC = 0.17
𝑇 2
SST = – CF
𝑡
(17.88)2 + (17.52)2 +⋯ (14.58)2
SST = – 290.84
5
SST = 1.56
SSE = TSS –SSR-SSC-SST
SSE = 2.97-0.87-0.17-1.56
SSE = 0.37

Step 2. Calculation of mean squares

[271]
 𝑆𝑆𝑅
MSR = 𝑡−1
0.87
MSR = 4

MSR = 0.22
𝑆𝑆𝐶
MSC = 𝑡−1
0.17
MSC = 4

MSC = 0.04
𝑆𝑆𝑇
MST = 𝑡−1
1.56
MST = 4

MST = 0.39
SSE
MSE = (𝑡−1( (𝑡−2)
0.37
MSE = (4)(3)

MSE =0.03

Step 3. Calculation of F values

𝑀𝑆𝑅 0.22
F (rows) = 𝑀𝑆𝐸 = 0.03 = 7.12

𝑀𝑆𝐶 0.04
F (columns) = 𝑀𝑆𝐸 = 0.03 = 1.35

𝑀𝑆𝑇 0.39
F (treatments) = 𝑀𝑆𝐸 = 0.03 = 12.69

[272]
In crop research, one may not be that much interested in MSR and
MSC as this sources of variation block the variation in the
experimental unit so that block effects in different directions
would not be confounded with the treatment effects. But in
animal research, in this particular experiment for instance,
calculating F values for rows and columns tell how animals of
different age and weight group behave in response to different
deits.

Step 4. Construction of the ANOVA table

Table 20. Analysis oof variance for liver cholesterol level (mmol/L) recorded on sheep of different
weight and age groups which were fed with different deits.

Source of
variation df* SS MS F

Total k2-1=24 2.97

Rows k-1 = 4 0.87 0.22 Fr = 7.12**

Columns k-1 = 4 0.17 0.04 Fc = 1.35ns

Treatment k-1 = 4 1.56 0.39 Ft = 12.69**

Error (k-1)(k-2) = 0.37 0.03

(4)(3)= 12
*k = number of treatment or column or rows

[273]
 √𝑀𝑆𝐸
CV% =  100
Grand mean
√0.03
CV% =  100
3.41

CV% = 5.145%
LSD0.05 √2 ∗ MSE/r
LSD0.05 = 2.179√2 ∗ 0.0308/5
LSD0.05 = 0.241859

Many students, when they are asked “which of the diets is the
best?”, they answer as treatment “A” merly because this
treatment had the highest mean value. Note, in reality, that it
should not be the treatment that produced the highest mean value
that should be considered the best but rather it should be the
treatment with the lowest mean value as cholesterol level in meat
of animals is something that is undesirable, i.e. the lower is the
better.
The computed F value for column was non-significant indicating
that this experiment could have been conducted using a
randomized complete block design (RCBD) using the weight
groups as blocks (disregarding the age groups) instead of the
Latin Square design because the age groups were non-
significantly different.

[274]
Case example 2. Crop research
Assume that three hybrid maize varieties (A, B and C) were tested
with one standard check (D) using a 4  4 Latin Square design
and the following results were obtained with computed row totals,
column totals and treatment totals (Table 21).
Table 21. Yield of three test maize hybrids and a standard check from an experiment in Latin
Square design
Row Grain yield (t ha–1) Coloumn
Column I Column II Column III Column IV totals
1 B = 1.640 D = 1.210 C = 1.425 A = 1.345 5.620
2 C = 1.475 A = 1.185 D = 1.400 B = 1.290 5. 350
3 A = 1.670 C = 0.710 B = 1.665 D = 1.180 5.225
4 D = 1.565 B = 1.290 A = 1.655 C = 0.660 5.170
Column totals 6.350 4.395 6.145 4.475
Grand total (G) 21.365

Then compute treatment totals and treatment means as shown in

Table 22 below.
Table 22. Treatments, treatment total, treatment mean and grand mean for data in Table 19
above

Treatment Treatment total Treatment mean Grand mean

A 5.855 1.464
B 5.885 1.471 1.335
C 4.270 1.068
D 5.355 1.339

The ANOVA is computed as follows (Table 23):

To compute sum of squares due to rows, columns, treatment and
error, follow the following steps.

[275]
 𝐺2
CF = 𝑟𝑐
(21.365)2
CF = = 28.528952
16
TSS = X2 – CF
TSS = [(1.640)2 + (1.210)2 + … + (0.660)2] – 28.528952
TSS = 1.413923
𝑅 2
SSR = – CF
𝑡
(5.620)2 + (5.350)2 +(5.225)2 +(5.170)2
SSR = – 28.528952
4
SSR = 0.030154
𝐶 2
SSC = – CF
𝑡
(6.350)2 + (4.395)2 +(6.145)2 +(4.475)2
SSC = – 28.528952
4
SSC = 0.827342
𝑇 2
SST = – CF
𝑡
(5.855)2 + (5.885)2 +(4.270)2 +(5.355)2
SST = – 28.528952
4
SST
To = 0.426842
compute mean squares due to rows, columns, treatment and
SSE = TSS –SSR-SSC-SST
error, follow the following steps.
SSE = 1.413923-0.030154-0.827342-0.426842
SSE = 0.129585

[276]
 𝑆𝑆𝑅
MSR = 𝑡−1
0.030154
MSR = 3
MSR = 0.010051
𝑆𝑆𝐶
MSC = 𝑡−1
0.827342
MSC = 3
MSC = 0.275781
𝑆𝑆𝑇
MST =
𝑡−1
0.426842
MST = 3
MST = 0.142281
𝑆𝑆𝐸
MSE = (𝑡−1)(𝑡−2)
0.129585
MSE = (3)(2)
MSE = 0.021598

To compute F Value, follow the following step.

𝑀𝑆𝑇 0.142281
F (treatments) = 𝑀𝑆𝐸 = 0.021598 = 6.59

Note that, as mentioned above, F values due to rows and columns

need not be calculated as these two sources of varation are meant
to merely to block and partition out the effects of two directional
variation in the experimental unit, i.e. the experimental field.
Table 23. Analysis of variance (ANOVA) for yield of three test
maize hybrids and a standard check from an experiment in Latin
Square design

[277]
Source of df* SS MS F
variation

Total k2-1=15 1.413923

Rows k-1 = 3 0.030154 0.010051

Columns k-1 = 3 0.827342 0.275781

Treatment k-1 = 3 0.426842 0.142281 Ft = 6.59*

Error (k-1)(k-2) = 6 0.129585 0.021598

*k = number of treatment or column or rows

√𝑀𝑆𝐸
CV% =  100
Grand mean

√0.021598
CV% =  100
1.335

CV% = 11%

Interpretation of the meaning:

The F test tells that there were siginificant differences among the
varieties. However, it does not tell the specific pairs or group
of varieties that outperformed one another. Therefore, a mean
separation using the LSD test is needed as follows:

LSD0.05 √2 ∗ MSE/r
LSD0.05 = 2.447√2 ∗ 0.021598/4
LSD0.05 = 0.2543

[278]
 Case example 3. Crossover design (aimal research)
Assume that there are only a few cows (say eight) whose number
is not adequate for conducting a non-crossover trial or a parallel
design/trial to test the effect of two rations. In this case, it is
technically advisable to apply a crossover design where the ration
treatments are administrated to the same experimental units
(cows) but in a sequence, for example during different periods.
This approach is assumed to be cost-effective as the same limited
number of cows are subjected to different treatments at different
times again and again as having more number of cows for a non-
crossover trial or a parallel trial can be expensive. A random
sample of eight cows was taken to test the effect of two rations,
say A and B, on their milk yields. Out of these cows, four cows
were selected randomly. Each of these cows were given ration A
in their productive lactation period and ration B in their lean
(ppor) lactation period. Assume also that after an adequate
washout period or resting period between the periods of a
crossover/switch over study, each of the remaining four cows
were given ration B in their better lactation period and ration A in
their poorer lactation period.

The null hypothesis (H0) is: the effects of the two rations do not
differ on milk yield; or alternatively, the alternate or research
hypothesis (H1) is: the two rations have different effects on milk
yield. The layout and average milk yield per day (in kg) is gven
below (Table 24).

[279]
Table 24. Milk yield of eight cows fed on two rations (A and B) at their productive and lean
lactation periods in a crossover/switch over design

Lactation Cow No. Total

period 1 2 3 4 5 6 7 8
First B=25 A=22 B=15 A=20 B=15 B=15 A=14 A=20 147
Second A=15 B=11 A=15 B=15 A=15 A=12 B=6 B=14 103
Total 40 34 30 35 30 27 20 34 250

To compute the ANOVA table:

𝐺2
CF = 𝑟𝑐
(250)2
CF = = 3906.25
28
TSS = X2 – CF
TSS = [(25)2 + (23)2 + … + (14)2] – 3906.25
TSS = 315.75
𝐶 2
SSC = – CF
𝑡
(40)2 + (34)2 +(30)2 +⋯+(34)2
SSC = – 3906.25
2
SSC = 126.75
𝑅 2
SSR = – CF
𝑡
(147)2 + (103)2
SSR = – 3906.25
8
SSR = 121.00
𝑇 2
SST = – CF
𝑡
(134)2 + (116)2
SST = – 3906.25
8

SST = 20.25
SSE = TSS –SSC- SSR-SST
SSE = 315.75-126.75-121.00-20.25
SSE = 47.75

[280]
To compute mean squares due to rows (lactation period), columns
(cows), treatment and error, follow the following steps (Table 25).

𝑆𝑆𝑅
MSR = 𝑡−1
121
MSR = 1
MSR = 121
𝑆𝑆𝐶
MSC = 𝑡−1
126.75
MSC = 7
MSC = 18.10
𝑆𝑆𝑇
MST = 𝑡−1
20.25
MST = 1
MST = 20.25
𝑆𝑆𝐸
MSE = (𝑡−1)(𝑡−2)
47.75
MSE = 6
MSE = 7.95

Table 25. Analysis of variance (ANOVA) for milk yield of eight cows provided with two rations (A
and B) at their better and poorer lactation periods in a crossover/switch over
design
Source of df* SS MS F cal F tab
variation (5%)
Total nk-1=15

k-1 = 7 126.75 18.10 2.27 4.21

Period k-1 = 1 121.00 121.00 15.22 5.99

Ration k-1 = 1 20.25 20.25 2.59 5.99

Error (k-1)(k-2) = 6 47.75 7.95

*k = number of treatment or column or rows
[281]
Note that calculations of the CV(%) and LSD0.05 values are the
same as it is for the regular Latin square design given above. This
crossover design is, in principle, almost similar to the regular
Latin square design but, in this case, unlike the regular latin
square design, it can easily be used for small number of
treatments (less than 5). The F test tells that there was a
siginificant difference between the two lactation periods but not
between the two rations and among the eght cows.
Case example 4. Graeco-Latin square design (aimal research)
Assume that a reseacher is intending to determine the effects of
two factors, namely Lysine and Protein, on milk production of
cows. Assume also that there is negligible interaction between the
two factors. Graeco-Latin square design is used to assign a 7  7
combinations of the two factors (Lysine and Protein) to a period
and a cow. The seven levels of each factor include:
Lysine: A = 0%, B = 0.1%, C = 0.2%, D = 0.3%, E = 0.4%, F =
0.5% and G = 0.6%.
Protein: α = 2%, β = 4%, χ = 6%,  = 8%,  = 10%,  = 12% and
 14%.
Seven cows were selected at random for the experiment which is
to be carried out over seven three-month periods. The two Latin
squares are mutually orthogonal as illustrated by a 7  7 Graeco-
Latin square below. The data is tabulated below (Table 26).

[282]
Table 26. Data from seven cows (column) fed with different protein contents having different
lysine levels for over seven three-month periods (rows)
Column
Row Total
C1 C2 C3 C4 C5 C6 C7
R1 Aα=304 B=436 Cβ=350 D=504 Eχ=417 F=519 G=432 2962
R2 Bβ=381 C=505 Dχ=425 E=564 F=494 Gα=350 A=413 3132
R3 Cχ=432 D=566 E=479 Fα=357 G=461 Aβ=340 B=502 3137
R4 D=442 Eα=372 F=536 Gβ=366 A=495 Bχ=425 C=507 3143
R5 E=496 Fβ=449 G=493 Aχ=345 B=509 C=481 Dα=380 3153
R6 F=534 Gχ=421 A=452 B=427 Cα=346 D=478 Eβ=397 3055
R7 G=543 A=386 Bα=435 C=485 Dβ=406 E=554 Fχ=410 2919
Total 3132 2835 3170 3048 3128 3147 3041 21501

The model of ANOVA is: Yij(kl) =  + k + l + i + j + ij(kl),

where model Yij(kl) = the observation in ith row and the jth
column receiving the kth Latin treatment and the lth Greek
treatment;  = overall mean; k = the effect of the kth Latin
treatment; l = the lth Greek treatment; i = the effect of the ith
row; j = the effect of the jth column; and ij(kl) = random error (i
= 1-t, j = 1-t, k = 1-t and l = 1-t). The existence of interaction is
not assumed between row, column, Latin treatment, and Greek
treatment. Computation of sum of squares follows the following
steps (Table 27):

[283]
 𝐺2
CF = 𝑛
215012
CF = 47
CF = 9699665.327
TSS = X2 – CF
TSS = (3042 + 4362 + …+ 4102) – 9699665.327
TSS = 214461.76
Calculation of the sum of squares of rows:
𝑅2
SSR = 𝑡
– CF
(2962)2 + (3132)2 +⋯+ (2919)2
SSR = 7
–
9699665.327
SSR = 5831.959184
Calculation of the sum of squares of column:
𝐶2
SSC = 𝑡
– CF
(3132)2 + (2835)2 +(3170)2 +⋯+(3041)2
SSC = 7
– 9699665.327
SSC = 2124.24

In order to calculate the sum of squares of lysine or Latin letter

treatments (A, B, C, D, E, F and G), the data should be sorted
accordingly as tabulated below.
Latin letter
Observations
Treatments Total
A 304 413 340 495 345 452 386 2735
B 436 381 502 425 509 427 435 3115
C 350 505 432 507 481 346 485 3106
D 504 425 566 442 380 478 406 3201
E 417 564 479 372 496 397 554 3279
F 519 494 357 536 449 534 410 3299
G 432 350 461 366 493 421 543 3066
Total 2962 3132 3137 3143 3153 3055 3219 21801

[284]
 𝐿2
SSL = 𝑡
– CF
(2735)2 + (3115)2 +(3106)2 +⋯+(3066)2
SSL = 7
– 9699665.327
SSL = 30718.24

In order to calculate the sum of squares of Protein or Greek letter

treatments (α, β, χ, , ,  and ), the data should be sorted
accordingly as tabulated below.

Greek letter
Observations
Treatments Total
Α 304 350 357 372 380 346 435 2544
Β 350 381 340 366 449 397 406 2689
Χ 417 425 432 425 345 421 410 2875
432 494 479 442 481 427 386 3141
436 413 461 536 496 478 485 3305
504 505 502 495 493 534 554 3587
519 564 566 507 509 452 543 3660
Total 2962 3132 3137 3143 3153 3055 3219 21801

𝐺2
SSG = 𝑡
– CF
SSG =
2 2 2 2
(2544) + (2689) +(2875) +⋯+(3660)
7
– 9699665.327
SSG = 160242.8163

Calculation of the sum of squares of error:

[285]
SSE = TSS–SSR–SSC–SSL–SSG
SSE = 214461.76–5831.96–2124.24–30718.24–160242.82
SSE =15544.41

Table 27. Analysis of variance table (ANOVA)

Source of Df Sum of square Mean square F calculated

variation
Protein 6 160242.82 26707.14 41.23**
Lysine 6 30718.24 5119.71 7.9**
Cow 6 2124.24 354.04 0.55ns
Period 6 5831.96 971.99 1.5ns
Error 24 15544.41 647.68
Total 48 214461.76

It could be concluded that protein or Greek letter treatments (α, β,

χ, , ,  and ) and the lysine or Latin letter treatments (A, B, C,
D, E, F and G) were significantly different, whereas the seven
cows and periods were non-significantly different.
Example 4. Analysis of variance for factorial experiment (in
RCBD)
The basic background experimental design in factorial
arrangement of the treatments could be any one of the most
commonly used designs depending on the requirements stated for
the designs but a randomized complete block design (RCBD) is
used here for demonstration. Assume that three rice varieties (V1,
V2 and V3) were tested with five rates of nitrogen (N0, N1, N2,

[286]
N3 and N4) in RCBD with 4 replications and the following
results were obtained (Table 28):
Table 28. Grain yield of three rice varieties tested with five levels of nitrogen in an RCBD

Nitrogen level (kg ha– Grain yield (t ha–1) Treatment

1) total (t)
Rep I Rep II Rep III Rep IV
V1 = 95.388
N0 3.852 2.606 3.144 2.894 12.496
N1 4.788 4.936 4.562 4.608 18.894
N2 4.576 4.454 4.884 3.924 17.838
N3 6.034 5.276 5.906 5.652 22.868
N4 5.874 5.916 5.984 5.518 23.292
V2 = 100.840
N0 2.846 3.794 4.108 3.444 14.192
N1 4.956 5.128 4.150 4.990 19.224
N2 5.928 5.698 5.810 4.308 21.744
N3 5.664 5.362 6.458 5.474 22.958
N4 5.458 5.546 5.786 5.932 22.722
V3 = 101.162
N0 4.192 3.754 3.738 3.428 15.112
N1 5.250 4.582 4.896 4.286 19.014
N2 5.822 4.848 5.678 4.932 21.280
N3 5.888 5.524 6.042 4.756 22.210
N4 5.864 6.264 6.056 5.362 23.546
Rep total 76.992 73.688 77.202 69.508
Grand total (G) 297.390
Nitrogen treatment Nitrogen
total
N0 41.800
N1 57.137
N2 60.862
N3 68.036
N4 69.560

[287]
To construct ANOVA table, follow the following steps.

G2
CF = rab
(297.390)2
CF = = 1474.014
(4)(3)(5)
TSS = X2 – CF
TSS = [(3.852)2 + (2.606)2 + … + (5.362)2]
– 1474.014
TSS = 53.530
R2
SSR = – CF
ab
(76.992)2 +⋯+(69.502)2
SSR = – 1474.014
(3)(5)
SSR = 2.599
T2
SST = – CF
r
(12.496)2 +⋯+ (23.546)2
SST = – 1474.014
4
SST = 44.578
SSE = TSS-SSR-SST
SSE = 53.530-2.599-44.578
SSE = 6.353

[288]
To compute sum of square of the main factors (A and B) and their
interaction (AB), follow the following steps.

𝐴2
SSA = – CF
𝑟𝑏
(95.388)2 +(100.840)2 + (101.162)2
SSA = – 1474.014
(4)(5)
SSA = 1.052
𝐵2
SSB = – CF
𝑟𝑎
(41.800)2 +⋯+ (69.560)2
SSB = – 1474.014
(4)(3)
SSB = 41.234
SSAB = SST – SSA-SSB
SSAB = 44.578-1.052-41.234
SSAB= 2.292

Compute the mean squares of the main factors (A and B) and

their interaction (AB), by dividing the sum of square by the
corresponding degrees of freedom as:

[289]
 SSA
MSA = 𝑎−1
1.052
MSA = 2
MSA = 0.526
SSB
MSB = 𝑏−1
41.234
MSB =
4
MSB = 10.308
SSAB
MSAB =
(𝑎−1)(𝑏−1)
2.292
MSAB = (2)(4)
MSAB = 0.286
SSE
MSE =
(𝑟−1)(𝑎𝑏−1)
6.353
MSE = (3)(3)(5)−1
MSE = 0.151

Compute F values for each of the three factorial components as:

MSA
F(A) =
𝑀𝑆𝐸
0.526
F(A) = 0.151
F(A) = 3.48
MSA
F(B) = 𝑀𝑆𝐸
10.308
F(B) =
0.151
F(B) = 68.26
MSAB
F (AB) = 𝑀𝑆𝐸
10.308
F (AB) = 0.151
F (AB) = 0.286

[290]
Assemble the calculated values into ANOVA table as follows
(Table 29):
Table 29. Analysis of variance for 3 varieties tested with 5 nitrogen levels in a factorial
arrangement

Source of df SS MS F F tabulated F tabulated

Variation (5%) (1%)
Block 3 2.599 0.866 5.74** 2.83 4.29
Treatment 14 44.578 3.184 21.09** 1.94 2.54
Variety (A) 2 1.052 0.526 3.48* 3.22 5.15
Nitrogen (B) 4 41.234 10.308 68.26** 2.59 3.80
AB 8 2.292 0.286 1.89ns 2.17 2.96
Error 42 6.353 0.151
Total 59 53.530

The results of ANOVA showed that the main effects of both

factors A and B were significantly different at 5% probability
level. The variety V3 showed best performance over the others.
Levels of factor B also significantly differed at 1% probability
level, level N4 being biologically the best over the other levels.
However, varieties did not show significant differential response
to changes in fertilizer levels.

[291]
 √𝑀𝑆𝐸
CV% = Grand mean
 100
√0.151
CV% =  100
4.956
CV% = 7.8%
LSDV0.05 √2 ∗ MSE/rb
LSDV0.05 = 2.021√2 ∗ 0.151/20
LSDV0.05 = 0.06
LSDN0.05 √2 ∗ MSE/ra
LSDN0.05 = 2.021√2 ∗ 0.151/12
LSDN0.05 = 0.09

The variety by nitrogen level interaction effect was not significant

and, otherwise, the LSD value would have been calculated as
follows:

LSD(VN)0.05 √2 ∗ MSE/r

Example 5. Analysis of variance for split-plot experiment (in

RCBD)
Assume that an agronomist wanted to study the effect of three
irrigation treatments (I1 = none, I2 = once and I3 = twice) and three
nitrogen fertilizer treatments (N1 = none, N2 = 25 kg/ha and N3 =
50 kg/ha) on the yield of barley. Three irrigation treatments
require large plots, while the nitrogen fertilizer treatments, which
can be spread by hand, can be applied to small plots. He used

[292]
RCBD in split-plot arrangement replicated 4 times. The main plot
treatments were three irritation treatments and the sub-plot
treatments were three three nitrogen levels. The data collected
from the experiment is given below (Table 30):
Table 30. Replication-wise yield data (kg plot–1)

Replication Irrigation N rate, kg ha–1

0 25 50
I I1 15.5 17.5 20.8
I2 20.5 24.5 30.2
I3 15.6 18.2 18.5
II I1 18.9 20.2 24.5
I2 15.0 20.5 18.9
I3 16.0 15.8 18.3
III I1 12.9 14.5 13.5
I2 20.2 18.5 25.4
I3 15.9 20.5 22.5
IV I1 12.9 13.5 18.5
I2 13.5 17.5 14.9
I3 12.5 11.9 10.5
The following steps are followed to construct ANOVA.

Step 1. Calculate the replication total (R), and the grand total (G)
by first constructing a table for replication by irrigation totals as
shown below (Table 31):

[293]
Table 31. Replication  irrigation (RI) table of yield totals

Replication Irritation Replication total

I1 I2 I3 (R)
I 53.8 75.2 52.3 181.3
II 63.6 54.4 50.1 168.1
III 40.9 64.1 58.9 163.9
IV 44.9 45.9 34.9 125.7

Irritation totals (A) 203.2 239.6 196.2

Grand total (G) 639.0

Step 2. Calculate the irritation  nitrogen (IN) (Table 32):

Table 32. Irritation  nitrogen (IN) table for yield totals

Nitrogen Irritation Nitrogen totals

I1 I2 I3 (B)
N1 60.2 69.2 60.0 189.4
N2 65.7 81.0 66.4 213.1
N3 77.3 89.4 69.8 236.5

Step 3. Calculate sum of squares

[294]
 G2
CF =
rab
(639)2
CF =
(4)(3)(3)
CF = 11342.25
TSS = X2 – CF
TSS = [(15.5)2 + (20.5)2 + … + (10.5)2] – 11342.25
TSS = 637.97
R2
SSR = – CF
ab
(181.3)2 +(168.1)2 +(163.9)2 +(125.7)2 (168.9)2
SSR = – 11342.25
(3)(3)
SSR = 190.08
A2
SSA = – CF
rb
(203.2)2 +(239.6)2 + (196.2)2
SSA = – 11342.25
43
SSA = 90.487
(RA)2
Main plot error (SSE1) = – CF − SSR − SSA
b
(53.8)2 +(63.6)2 +⋯+ (34.9)2
SSE1 = – 11342.25 − 190.08 − 90.487
3
SSE1 = 174.103
B2
SSB = – CF
𝑟𝑎
(189.4)2 +(213.1)2 + (236.5)2
SSB = – 11342.25
4 3

SSB = 92.435
AB2
SSAB = – CF − SSA − SSB
𝑟
(60.2)2 +(65.7)2 +⋯+ (69.8)2
SSAB = – 11342.25 − 90.487 − 92.435
4

SSAB = 9.533
Sub-plot error (SSE2) = TSS-all other sum of squares
SSE2 = 637.97-(190.08+90.487+174.103+92.435+9.533)
SSE2 = 81.332

[295]
Step 4. Calculate mean squares for each source of variation by
dividing the sum of squares by its corresponding degrees of
freedom and compute the F value of each effect that needs to be
tested, by dividing each mean square by the error mean square to
construct the ANOVA table (Table 33).
SSR
MSR = 𝑟−1
190.08
MSR = 3
MSR = 63.360
SSA
MSA = 𝑎−1
90.487
MSA = 2
MSA = 45.243
SSB
MSB = 𝑏−1
92.435
MSB = 2
MSB = 46.218
SSAB
MSAB = (𝑎−1)(𝑏−1)
9.533
MSAB = 4
MSAB = 2.383
SSE
MSE1 = (𝑟−1)(𝑎𝑏−1)
174.103
MSE1 = 6
MSE1 = 29.017
SSE
MSE2 = (𝑟−1)(𝑎𝑏−1)
81.332
MSE2 = 18
MSE2 = 4518

[296]
Table 33. Table of ANOVA

Source of
variation df SS MS F*
Replication 3 190.08 63.360
Irrigation (A) 2 90.487 45.243 1.56ns
Error (E1) 6 174.103 29.017
Nitrogen (B) 2 92.435 46.218 10.23**
AB 4 9.533 2.383
Error (E2) 18 81.332 4.518
*ns = non- siginificant; ** = hghly siginificant

Interpretation of the results:

There was no significant difference among irrigation means.
However, yield was significantly affected by nitrogen levels. The
interaction betwween N rates and irrigation levels was also not
siginificant.

√𝑀𝑆𝐸1
CV(A)% = Grand mean  100
√29.017
CV(A)% = 17.75  100
CV(A)% = 30.31%
√𝑀𝑆𝐸2
CV(B)% = Grand mean  100
√4.518
CV(B)% = 17.75  100
CV(B)% = 11.98%

In split-plot experiments, mean separation using LSD follows a

different aproach for comparing different pairs of means in
different scenarios:

[297]
First, LSD value for comparison of two main plot treatments
is calculated as:

√2 ∗ MSE1/rb

Second, LSD value for comparison of two sub-plot

treatments is calculated as:

√2 ∗ MSE2/ra

Example 6. Analysis of variance for augmented design (in RCBD)

An augmented design is a design where a replicated check
experiment is augmented by un-replicated test entries. Assume
that 12 new varieties (v = 1-12) were evaluated along with three
standard check varieties (c = 3; A, B and C) using an augmented
design in RCBD replicated three times and the following results
were obtained (Table 34).
Table 34. Records from 12 new varieties and 3 check varieties tested in RCBD replicated 3 times

Block 1 Block 2 Block 3

A = 83 A = 79 A = 92
10 = 89 4 = 96 2 = 89
7 = 75 8 = 74 C = 87
B = 77 B =81 9 = 98
5 = 78 C = 81 B = 89
C = 78 3 = 70 6 = 82
11 = 82 12 = 92 1 = 79

[298]
Since the analysis of variance in augmented designs is based on
control/check treatments, records on these treatments across
blocks are sorted out from the above Table (Table 35) as shown in
Table 35 below.
Table 35. Performances of the check varieties as sorted out from Table 34 above

Check Block Total Mean

BI BII BIIII
A 83 79 92 254 84.67
B 77 81 89 247 82.33
C 78 81 87 246 82.00
Total 238 241 268 747 249.00

In augmented designs, it is difficult to maintain homogeneous

blocks when comparing so many genotypes and observations on
new un-replicated genotypes should be adjusted for field
heterogeneity (for block to block variation). Accordingly, the
block to block variation is adjusted based on records from the
replicated checks using the following formula:

rj = 1/c (Bj-M); where c is the number of controls, Bj is

the sum of all controls in j block and M is the sum of all
means.

For the above three blocks (r1-r3), the adjustment factors are
calculated as follows:

[299]
 rj = 1/c (Bj-M)
r1 = 1/3 (238-249) = -3.67
r2 = 1/3 (241-249) = -2.67
r3 = 1/3 (268-249) = 6.33
Note that the sum of r1, r2 and r3 should be zero.

Once the adjustment factors are generated for each block, then
adjustments are made on records from the un-replicated new test
entries as: adjusted record is equal to unadjusted record minus the
adjustment factor of that particular block (Table 36).
Table 37. Adjusted and unadjusted records from the test entries

Variety Block Unadjusted record Adjusted

record
1 3 79 72.67
2 3 89 82.67
3 2 70 72.67
4 2 96 98.62
5 1 78 81.67
6 3 82 75.67
7 1 75 78.67
8 2 74 76.67
9 3 98 91.67
10 1 89 92.67
11 1 82 85.67
12 2 92 94.67

Comparisons among the new genotypes and between new

genotypes and check varieties are made based on the adjusted
values of the new genotypes; not unadjusted values. The ANOVA
[300]
Table computed from records on control/check treatments sorted
from all over the blocks is shown below (Table 38).
Table 38. Analysis of variance table for the three checks tested in an augmented design

Source of variation df SS MS
Block 2 182.00
Checks 2 12.67
Error 4 23.33 5.83
Total 8 218

The calculation of CV of an augmented design is different from

what we have been doing so far:

CV% = (𝑐√𝑀𝑆𝐸/𝑚)  100

CV% = (3√𝑀𝑆𝐸/𝑚)  100
CV% = (3√5.83/249)  100
CV% = 2.91%
Calculation of LSD values:
LSD values to compare the checks:
LSD0.05 √2 ∗ MSE/b
LSD0.05 = 2.776√2 ∗ 5.83/3
LSD0.05 = 5.48
LSD values to compare the test varties wth the checks:
LSD0.05 √MSE (b + 1)(c + 1)/bc
5.83(3+1)(3+1)
LSD0.05 = 2.776√
3∗3

LSD0.05 = 8.95

[301]
Example 7. The incomplete block designs (ICBD)
Manually processing data from the incomplete block designs
(ICBD) is a little bit difficult and easily liable to mistakes unless
otherwise computer packages are used. However, just to
demonstrate the most important steps in analysis of incomplete
block design, assume that an experiment involving 9 rice varieties
was conducted in a 3  3 balanced lattice design replicated 4
times, with 3 incomlete blocks within replication and each block
with 3 plots (Table 39).
Table 39. The layout and grain yields of the rice varieties with
block total (B), replication total (R), and grand total (G).

[302]
Replication I Replication III

Block Block Block Block total

total (B) (B)

1 (7)=6.31 (5)=5.22 (3)=4.79 16.72 7 (1)=3.52 (9)=5.50 (5)=5.33 14.35

2 (6)=4.69 (8)=5.72 (1)=4.45 14.86 8 (4)=3.58 (8)=5.03 (3)=4.21 12.82

3 (2)=4.72 (4)=4.20 (9)=5.87 14.79 9 (7)=4.83 (2)=4.51 (6)=4.30 13.64

Replication total (R) 45.97 Replication total (R) 40.81

Replication II Replication IV

Block Block Block Block total

total (B) (B)

4 (9)=5.14 (3)=4.75 (6)=3.30 13.19 10 (5)=5.34 (4)=3.59 (6)=4.50 13.43

5 (8)=5.55 (5)=5.30 (2)=4.20 15.05 11 (3)=4.30 (2)=4.83 (1)=4.14 13.27

6 (7)=4.67 (1)=4.19 (4)=3.08 11.94 12 (9)=5.33 (8)=5.31 (7)=5.50 16.14

Replication total (R) 40.18 Replication total (R) 42.84

Grand total (G) 169.80

[303]
Before calculating the sum of squares, take the following actions
at a preliminary stage.
Step 1. Calculate the block totals (B), replication totals (R), and
grand totals (G) (Table 40).
Step 2. For each treatment, compute the sum of block totals over
all blocks in which the particular treatment appears. Denote this
value by Bt and also compute treatment totals (T). Check that Bt
= kG, where k is the block size (Table 40).
Step 3. For each treatment, compute W = kT-(k+1)Bt+G, check
that W = 0.
Table 40. Computations of the treatment totals (T), the sum of block totals over all blocks in which
the particular treatment appears (Bt) and W.

Treatment Treatment totals (T) Block totals (Bt) W=3T-4Bt+G

number
1 16.30 54.42 1.02
2 18.26 56.75 -2.42
3 18.05 55.60 1.55
4 14.45 52.98 1.23
5 21.19 59.15 -3.23
6 16.79 55.12 -0.31
7 21.31 58.04 1.57
8 21.61 58.87 -0.85
9 21.84 58.47 1.44
Total 169.80 509.40 0.00

[304]
Step 4. Compute the sum of squares and coonstruct the ANOVA
Table (Table 41).

𝐺2
CF = (𝑘 2 )(𝑘+1)
(169.80)2
CF = (9)(4)
(639)2
CF = 36
CF = 800.89
TSS = X2 – CF
TSS = [(6.31)2 + (5.22)2 + … + (5.50)2] – 800.89
TSS = 19.7838
R2
SSR = – CF
𝑘2
(45.97)2 +⋯+(42.84)2
SSR = – 800.89
9
SSR = 803.1550– 800.89
SSR = 2.2650
T2
SST = – CF
(𝑘+1)
(16.30)2 +⋯+(21.84)2
SST (unadjusted) = – 800.89
4
SST (unadjusted) = 815.6842– 800.89
SST (unadjusted) = 14.7942
W2
SSB (adjusted) = (𝑘 3 )(𝑘+1)
26.6022
SSB (adjusted) = 108
SSB (adjusted) = 0.2463
SSE (interblock) = TSS-SSR-SST(unadjusted)-SSB(adjusted)
SSE (interblock) = 19.7838-2.2650-14.7942-0.2463
SSE (interblock) = 2.4783

[305]
Table 41. ANOVA for a lattice design.

Source of variation df SS MS F
Replication (R) k=3 2.2650 -
Block (adjusted) (B) k2 -1 = 8 0.2463 0.0308
Treatment (unadjusted) (T) k2 -1 = 8 14.7942 1.8493 11.939
Intrablock error (E) (k-1) (k2 -1) = 16 2.4783 0.1549

Total k2 (k+1)-1= 35 19.7838

Interpretation of the results:

There was a significant difference among the treatments.

LSD0.05 √2 ∗ MSE/r
LSD0.05 = 2.120√2 ∗ 0.1549/4
LSD0.05 = 0.59

Coefficient of variation (CV)

Coefficient of variation is the ratio between square root of error
mean square (or standard deviation) and the grand mean. It is
computed to determine whether or not to accept the results of the
experiment as very high CV indicates something went wrong in
the conduct of the experiment such as improper lay-out, too much
variability in the area probably due to plot size, topography,
uneven or severe insect or disease population, severe weather
conditions, loss of samples due to animals, etc. When this
happens, the experiment is declared invalid and the results and
conclusions unacceptable.

[306]
15. Interpreting analytical results
Once data analysis is done, a proper interpretation of the results
should be made to acquire useful information for making
informed decisions. Data interpretation involves analyzing the
data first, taking the results of the data analysis, reviewing the
result through some predefined processes and making inferences
of the broader implication of the findings, and using them to
arrive at a sensible conclusion in terms of what the data are telling
including novelty and new insights provided by the information.
Interpretation of results should be compared not only against
targets set for the project but also against established standards.
The results should also be reviewed for consistency or
inconsistency of the findings with results of previous studies.
Interpretation of results may also include what key findings
would be of most interest to areas of the subject matter and what
need future attention as some studies may raise more questions
than they provide answers including unexpected results.
Generally, data interpretation may include aspects of establishing
explanatory concepts from the findings which can serve as a
guide for further researches and the effort put in place to establish
continuity in research through linking the results with previous
findings.

16. Possible Errors in Research

Any difference between the measurements a researcher makes
and the true value is an error. It is easy to obtain closely
reproducible results from experiments in ‘hard’ disciplines such
as physics and chemistry if the experimental techniques are sound
and the equipment is working correctly. In such disciplines, it is
[307]
relatively easy to keep the conditions under control during
experimentation, and chances of experimental errors are relatively
less; and in most cases, when you repeat the experiments you get
concurrent values. On the other hand, experimental errors are
more common in biological studies as the variables involved in
the study are human beings, animals, plants, soil, or climate.
Fortunately, the reality of such fluctuations in ‘soft’ disciplines
such as agricultural and veterinary sciences has been recognized
and measures to deal with them have been standardized by
statisticians.

Consider a simple experiment of comparing two cultivars of teff.

You assigned cv. A to Plot 1 and cv. B to plot 2. After the
harvest, you found that the yield of cv. A in plot 1 was higher
than cv. B in plot 2, and concluded that cv. A is superior to cv. B.
However, the conclusion may be entirely wrong because of a
false assumption! It is assumed that any difference between the
yields of the two plots is caused by the cultivars alone and
nothing else. In fact, the two plots may not be uniform and may
differed on many counts, for example, soil structure, texture,
water-holding capacity, pH, fertility gradient, and so on. There
might also be differences in pest and disease occurrence, bird
damage, etc. Therefore, the researcher should separate out
treatment differences from other sources of variation. This kind of
variation observed between experimental plots when they are
treated alike is called experimental error. These errors, caused
mainly by uncontrollable factors, must be accounted for while
reaching conclusions. Based on the magnitude of experimental
error, one can decide whether a difference observed in the
experiment is due to genuine reasons or just due to chance alone.

[308]
Therefore, plan the experiment in such a way as to have an
estimate of experimental error.

Experimental errors can hide the effect that the researcher is

trying to investigate. Errors should not be confused with personal
mistakes or faults, which can be rectified by taking appropriate
corrective measures. Personal mistakes can be eliminated largely
by checking one’s work frequently. Scientists should be careful to
observe and correct their own mistakes. Even if all the
precautions are taken, errors can creep in, and such errors are
natural variations, fluctuations, or deviations that are present in
the data or a set of observations.

Unavoidable errors are a great problem in both laboratory and

field experiments. The researcher should be aware of the possible
effects of errors on experimental data and consequent conclusions
made out of it. It is impossible to compare an experimental result
to a theoretical prediction, compare two experimental results, or
evaluate whether an apparent correlation is real without
knowledge of the possible errors. There should be some
mechanism to estimate the errors.

16.1. Random and systematic errors

Unrecognized variations in experiments can be of two types,
random errors and systematic errors. Random errors are produced
by a combination of uncontrollable and unknown variables. If
these errors occur as negative and positive deviations from true
values, and if they have a mean of zero, they do not normally bias
mean values. However, they may affect the accuracy of
measurements from replications. At the same time, if positive and
negative values of errors average to a nonzero value, such errors
[309]
are called systematic errors. Experimental units receiving a
treatment should show only random differences from units
receiving other treatments including the control. These units
should be allowed to respond independently of one another.
Random errors
Random errors are so called as these can occur due to chance
alone, which cannot be attributed to any specific cause. These
errors may happen quite unexpectedly despite all the precautions
taken. Because they are governed by probability, statistical
methods are used to estimate the error due to random variation.
Random errors can be averaged out by making many replicates or
repeat measurements. When the number of replicate
measurements is more, it permits us to predict the proper value
with greater confidence. An estimate of random error helps us to
settle on whether an observed difference is genuine or due to
chance alone. Therefore, every experiment should have
provisions to estimate errors due to random variations.
Most experiments tackle the issue of reducing errors using the
principles of experimentation—randomization, replication, and
local control—and by representative sampling. A representative
sample is a small division of the population, displaying the same
features as that of the original population. Non-representative
sampling is a frequent pitfall, which can be checked by taking
adequate precautions. If your sample is representative of the
overall population, it exhibits similar characteristics to any other
randomly chosen sample of the population, and your
generalization may have applicability to the behaviour of the
population as a whole.

[310]
Systematic errors
While conducting an experiment, we must make sure that there
are no systematic differences between experimental units
receiving a treatment from those receiving another treatment. At
times, systematic errors may be more complicated. A constant
systematic error affects accuracy of values, but it will not
influence the correctness of measurements. At the same time, a
variable systematic error affects both accuracy and correctness of
measurements. Systematic errors occur mainly because of two
reasons, experimental unit (plot) variations and faulty techniques.
It is assumed that any difference between the values of two
experimental units (plots) is caused by the treatments alone and
nothing else. In field experiments, chances of plot-to-plot
variations due to soil heterogeneity are a major problem. Soil
heterogeneity occurs mainly because of texture, fertility gradients,
pH variations, and erosion hazards. There might also be
differences in pest and disease occurrence, bird damage, etc.
These variations can also occur when the ‘plot’ consists of
animals or even human beings. Soil heterogeneity is tackled
largely by random arrangement of plots, replications, and suitable
layouts. Faulty techniques include inaccurate scales and
measures, faulty instruments, wrong methods, or personal bias.
For example, a frequent source of systematic errors is calibration
errors with equipment and instruments. If the calibration is faulty,
the values may be too high or low. This kind of errors can be
eliminated by the skill, care, and experience of the researchers.
Researchers well versed in research methodology and
instrumentation seldom commit this kind of mistakes.

[311]
16.2. Type I Error and Type II Error
In most research studies, we draw and examine a sample hoping
that it will reveal all the facts about the population accurately.
However, because a representative sample only is studied instead
of the original population, there is every chance of committing
mistakes in the decision for rejecting or accepting the null
hypothesis. Take the example of a null hypothesis, which says:
There is no significant decrease in grain yield when irrigation is
limited to critical stages compared to irrigation throughout the
growth period.

After analysing the data, you set an arbitrary value for the
rejection or acceptance of a null hypothesis based on some level
of significance; p is the probability that a difference or an
association as large as the one observed could have occurred by
chance alone. You reject the null hypothesis, if the p value is less
than the predetermined level of statistical significance, usually
expressed as a percentage. A result is considered statistically
significant, if the p value is less than 5% (p = 0.05) and is said to
be highly significant if p is less than 1% (p = 0.01).

In most field experiments, 5% (p = 0.05) level of significance is

often used as a standard for rejection of null hypothesis. In some
studies, that are related to health, researchers may even go for one
out of thousand or 1‰ (p = 0.001) level of significance. Assume
that the level of significance at which you are willing to reject or
accept this null hypothesis is at 5% (p = 0.05) in agriculture
which allows for a 5% probability that the outcome could have
occurred by chance. When you reject a null hypothesis at 5%
level of confidence, it indicates that a difference in means as large

[312]
as that between the experimental and control groups would have
occurred from sampling error in less than 5 out of 100 items of
the experiment. Conversely, the probability that the difference in
values is due to the treatment, instead of sampling error, is 95%.
The rejection of the null hypothesis while it is true is called a type
I error. At the same time, failing to reject the null hypothesis
while it is false is called a type II error. The chance of committing
a type I error, rejecting the null hypothesis when it is actually
true, is represented by α, and the chance of committing type II
error, proving an association when none exists, is generally
represented by β (Table 42).

The maximum probability of rejecting the null hypothesis when it

is actually true or the chance of committing type I error is known
as the level of significance. As narrated earlier, these probabilities
are generally taken as p = 0.05 or 0.01 (5% or 1%). It is essential
to ensure that the samples are true representative of the original
population, and there should be no chance for committing any
type of errors. The two types of errors are unavoidable, and it is
not possible to eliminate them simultaneously. When one type of
error decreases, the other type increases, and therefore, the best
option is to fix the probability of one of them and try to minimize
the other. As type I error is more serious, we usually fix the
probability of type I error and try to minimize the probability of
type II error.

A difference between sample and population is assumed genuine

and not due to chance mainly based on three criteria. First, based
on the quantum of the difference observed. It is reasonable to
expect that the larger the differences between the means, the more
likely it is due to treatments. Second, based on the degree of
[313]
variations in the values. If the values fluctuate widely with
extreme values, most probably the differences in means are due to
chance variations. And, thirdly, based on the sample size studied.
If the sample size is large, it is possible that the conclusion drawn
from it will reflect the parameters of the population. Therefore, it
is important to avoid small samples, as there are extreme or rare
sample means in the sampling distribution, and you are more
likely to get one of them in your experiment.
Table 42. Types of decision in statistical inference

Null hypothesis (H0) Decision taken

Reject H0 Accept H0
Incorrect decision
TRUE (type I error) Correct decision
Incorrect decision
FALSE Correct decision (type II error)

Based on the nature of alternative hypothesis, a test is one sided

(one tailed) or both sided (two tailed). For example, if we are to
test H0: μ = 50, then the test against alternative hypothesis (H1) μ
≠ 50 is both-sided or two-tailed test while the test H1: against
either H1: μ > 50 or μ < 50 is a one-sided or one-tailed test.

17. Popular Statistical Packages

The arrival of computers has revolutionized statistical work, and
presently, you can do the calculations quickly with ease. Before
finalizing a particular method of analysis, consult standard books
on statistics and understand more about common statistical tests
and their applications. Several types of software packages are also
available, both commercial and non-commercial, for analysing
the data. Some organizations have come with free statistical

[314]
software as a practical alternative to costly commercial packages.
Most free statistical software packages are easy to learn and use
by means of menu systems. A few popular packages (both priced
and free) are presented below.

SPSS (Statistical Package for the Social Science) is one of the

most popular statistical packages, which can perform highly
complex statistical data analysis with simple instructions. The
software was originally created for the management and statistical
analysis of data from social sciences, but now it is used in other
disciplines such as agriculture and health sciences as well for a
variety of analytical works. The SPSS software was launched in
1968 by SPSS Inc., but was later acquired by IBM in 2009.
Although the software was officially renamed as IBM SPSS
Statistics, it is still popular as SPSS. Statistical methods usable in
the software include inferential statistics such as ANOVA,
MANOVA, correlation, regression (linear, nonlinear, logistic,
loglinear, etc.), and nonparametric tests; descriptive statistics; and
several others such as probit analysis, cluster analysis, factor
analysis, geo spatial analysis, and simulation. In addition to
statistical analysis, it is highly suitable for management and
documentation of data.

SAS
SAS is an integrated system of software for solving many tasks.
The Statistical Analysis System, SAS was developed by SAS
Institute. The software can be used for data entry and
management, graphics designing, multivariate analyses, advanced
analytics, predictive analytics, business intelligence and decision

[315]
support, operations research and project management, and several
such applications.

R Software
R is a powerful, free and efficient software that is available freely
and used in different fields of research for data analysis and data
management. It is currently a free, open-source code and stable
and reliable software. Its application range from simple analysis
descriptive statistics (mean, variance, median etc.) to more
sophisticated data analysis such as genomic selection and spatial
analysis. Besides the R software has one of the best graphic-
producing packages. The other application of R is in data science
for data mining purposes from the big data and make usable for
the decision-makers. In agriculture, the application of R can go
from generating different experimental design randomization and
layout to the analysis of data based on different experimental
designs.
R software run in any operating system such as Windows, Unix
systems, and Mac that show how the software is versatile and the
code and its output can be read on different platforms. The R also
allows user to write their Package and distribute to other R users
in the world. R has an easily readable, precise, and
understandable help system that is integrated with the package or
you can access it from the internet. One of the best ways to get
help for R is R communities who provide information about R
and answer questions from different users:
[Link]
[Link]

[Link]

[316]
R connects with other languages such as Python, C++ etc.

MaxStat
MaxStat is a software package for statistical analysis specifically
designed for students and researchers who do not have much
knowledge in statistics. The MaxStat Lite version is offered free
to use for non-commercial purpose. This is useful for smaller
projects and theses. The full version, MaxStat Pro, offers analysis
of larger data sets. The Lite version features descriptive statistics,
t-tests, chi-square, ANOVA, nonparametric tests, linear
regression, correlation, and basic graphing functions. The Pro
version features advanced analyses, more types of graph, word-
processing functions, and options for formatting of results.
MaxStat supports over 100 commonly used statistical tests.

Stata
Stata is a statistical software package created in 1985 by
StataCorp for general use. It is specifically suited for researchers
working in the fields of economics, sociology, political science,
and health science. Data management, statistical analysis,
graphics, simulations, regression, and custom programming are
some of the highlights of Stata.

Microsoft Excel
Excel is a spreadsheet application available with ‘Microsoft
Office’ suite, useful for data management and analysis. With the
same basic features of all spreadsheet applications, Excel uses a
collection of cells arranged into rows and columns to organize
and manipulate data. It is useful to perform basic calculations and
statistical analysis. They can also display data as charts,
histograms, and line graphs.

[317]
CropStat
CropStat is a statistical package developed by International Rice
Research Institute (IRRI) for the management and analysis of data
from experiments. Although CropStat has been developed mainly
to cater to the needs of agricultural workers for analysing the data
from field experiments, some programmes can be used for
analysis of data from other disciplines too. Please note that the
much popular earlier version is not being used today because it
was originally written in the BASIC (Beginner’s All-purpose
Symbolic Instruction Code) language, which is not compatible
with ‘Windows’. You can download CropStat and some other
useful software from the IRRI website, [Link]
Similar to the above, you may find many freer or priced software.
R, Jamovi, JASP, PSPP are examples of open-source software,
and Minitab, Matlab, Systat, Statistix, AcaStat, Statwing,
Statistica, and XLStat are some proprietary products.

18. Field Layout for Agricultural Experiments (Case

Examples)
A designed experiment, by definition, is one for which the
researcher controls the specification of the treatments and the
method of assigning the experimental units to each treatment. As
indefinite number of layouts of experiments could be made in
many different ways depending on the type of experiment, it is
advisable to take two case examples one each for crop related and
animal related experiments using the common randomized
complete block design.
In crop related experiment, field layouts are based on the guiding
principle of what we call the famous Pythagoras theorem on the
[318]
algebraic relationship between the sides of a right angle triangle.
By this rule, square of the hypotenuse is equal to the sum of the
squares of the base and its height. Symbolically, this could be
represented as: b2 + l2 = h2, where h=hypogenous; b=base and
l=height of the triangle as shown below (Figure 3).

Figure 3. The guiding principle of a right angle triangle to construct an experimental field layout
The step-by-step procedure of layout:
a) Drive a wooden peg into the soil at point C;
b) Stretch a string from the peg at point C along the “b” or “l”
direction desired to be the borderline between the experiment
and the field;
c) Drive another peg at the point marked to be the end and tie the
string onto it very tightly;

[319]
d) Stretch out another string from the same peg at the origin
(point “C”) to the direction opposite to the borderline (along
the line “b” or “l” as the case may be);
e) In order to make this string form a right angle at point “C”
with the borderline, it has to pass through a definite point to
be determined using the above equation. For practical
purposes, this reference point can be determined by measuring
a distance of 4 m from “C” along the borderline (“l” for
example) and 3 m long the base, “b”; now by definition “h”
would be 5 m and mark the point of junction on the tape by
joining the “0” and 12 m (3+4+5) (Figure 4). Marking can be
done by setting a thin but straight piece of stick exactly at the
juncture;

Figure 4. Application of the Pythagoras theorem to construct an experimental field layout

[320]
f) Once the point of the reference is determined, adjust the
direction of the string along the base by stretching it well as
far as required so that it can only just touch the reference stick
(from the outer side of the experimental plot);
g) At the end of the required distance from the origin, drive a
peg into the soil and tie the string tightly;
h) Determine the diagonal opposite corner of the origin (point
“d”) either by directly measuring both the vertical and
horizontal sides or by the same 3, 4, and 5 principles. Drive a
peg into the soil at this point and extend the strings from the
pegs at the ends of the borderline and the base and tie both
strings tightly onto the peg;
i) Once the four corners of the experimental field are in place,
plots and blocks can be delimited using pegs according to the
plan drafted on the paper;
Blocking is done against the soil fertility gradient while the row
orientation is along the fertility gradient. Blocking is performed
such that the variation within a block is minimal and variation
between blocks is maximal as represented by different in color
intensity as shown in Figure 5 below. If the length of a plot is 4
m; width of a plot is 1.6 m (say 4 rows with 0.4 m between rows);
spacing between blocks is 2 m and spacing between plots is 0.6
m, then the layout of the experiment maybe made.

[321]
Figure 5. Field layout before treatments are assigned to plots

[322]
Then final stage of the layout would be to give consecutive plot
numbers for the ease of work, allot treatments to the plots at
random using lottery method (drawing ballot) or using the
random number table. Within each block, the treatments are
assigned at random so that a different treatment is applied to each
unit. That is, all treatments are observed within each block. The
defining feature of the randomized complete block design is that
each block sees each treatment exactly once.
Fertilizer and seed rates per plot are calculated based on the
recommendation on a hectare base.
Randomized complete block design (RCBD) is a design used
when the experimental units are not homogeneous and thus can be
allocated to groups of block such that the variation among blocks
is maximized while the variation within any particular block is
minimized.
When RCBD is employed in animal research, as the case holds
true for crop research, treatments (feed types for example) should
be compared under similar circumstances. For example, imagine
that the body weight varies among the experimental animals. If
the animals are not properly blocked in groups of similar initial
body weight, obviously, the effects of the feed types would be
“confounded” with the initial weight of the animals, i.e. the initial
weights of the animals are interfering with the conclusions about
the feed effects). As a result, if there is known variability in initial
weights of the experimental animals, it would be advisable to
block the animals thereby remove any advantage or disadvantage
of one treatment over the other because of the animal to which it
is allotted.

[323]
Individual animals may differ in many aspects for which they
must be blocked. In livestock research, animals are most often
grouped into blocks according to initial weight, body condition,
breed, sex, stage of lactation, litter size, etc. Note in this case that
block does not necessarily indicate physical grouping.
It is important that during the experiment all animals within a
block receive the same conditions in everything except
treatments. Example of a good blocking where five feed
treatments (A, B, C, D and E) have been randomized below to 4
different breeds of chicken (Figure 6).

[324]
Block 1
B C A E D

Block 2
C A D B E

Block 3

E C A D B

Block 4

A D C E B

Figure 6. Field layout after the 5 treatments are assigned to the experimental animals arranged in a randomized complete block design

[325]
 SECTION V. MANAGING AGRICULTURAL
RESEARCH IN ETHIOPIA

19. Managing Experiments

Agricultural productivity of biological commodities can be
basically improved by genetic manipulation organisms or by
altering the growing environments with the use of inputs like
better management and providing protection to suit the
organisms. Under the current conventional agricultural research
approach in Ethiopia, responsive technologies that can exploit
optimal resource bases are developed particularly under optimal
situation where climatic, edaphic and biological factors are
conducive for obtaining the genetic potential of a given commodity.
The ultimate goal of agricultural research in Ethiopia is increasing
crop and animal yields using best-bet combinations of genotype,
suitable physical and management environments, and any positive
synergic interaction between them. Under such situations, the
process in which the growing environment is manipulated to fit crop
or animal genotypes is encouraged instead of fitting the genotypes
to the prevailing growing environment.
The above option obviously does not apply under marginal natural
environments (e.g. drought prone environments) where resource-
poor farmers that cannot afford the cost of production inputs dwell.
Under marginal situations e, researchers recognize the unique
situation and challenges and try to fit animal or crop genotypes to
the bio-physical and socio-economic needs. Thus, in contrast to
technology development under optimal situations, the process in

[326]
which genotypes are made to fit the prevailing growth environment
is encouraged instead of the environment being altered to fit the
cultivars. Therefore, the genotype “fit-in” approach as an option has
been more appropriate under very marginal production conditions
where it is not feasible to develop and utilize resource-needy
technologies. Examples include failure of technology development
from among pure breeds of European livestock, particularly dairy
cows, under small-scale production because of inability of the
farmers to provide the required management level and poor
performance of the cows under the existing animal management
conditions. The genotype “fit-in” approach has also been a feasible
approach in crop technology development under abiotic stresses like
drought. In all cases, there is also a tendency to build on past
successes, as starting research works from the scratch each time
may take a longer and ultimately show lesser probability of success
to generate good results.
Research approaches in Ethiopian so far rely on conventional
methods which have tried to address the major research problems
facing farmers and pastoralists in Ethiopia. However, there have
been barriers to achieving the desired level of improvement using
the conventional approaches. Among the features of the
conventional approaches include less responsive and slow
technology generation/adaptation or product development process.
For effective utilization of the wealth of genetic resources available
in Ethiopia in research programs, however, forging the best way of
combining the conventional and cutting-edge biotechnological tools
deserve a special attention.

[327]
19.1. Working with Farmers

A successful on-farm intervention should be preceded with a

diagnostic survey to understand the farm situation (farm size,
productivity, production goal, etc.), priority problems, and likely
interventions. Diagnosis and characterization constitutes one of
the most important stages in agricultural research using systems
approach. Knowledge of the forms of production under study,
recognition of its components, characteristics and limiting factors
are important elements to plan research whose results are applied
to improve existing systems.
As an important step, the criteria for selecting experimental fields,
level of field preparation and management practices and choosing
of appropriate treatments including checks need to be clear. The
selection of fields should be jointly made with the full interest of
the farmer and the researcher. In this process, it is also advisable
to try to accommodate different interest groups in a proportional
way to avoid sampling bias as much as possible. A field selected
by a farmer based on his/her own criteria could be accepted or
rejected by the researcher if it does not qualify for reasons of
representativeness. Care must also be taken that the selected part
of any farmer’s field should be as homogeneous as possible in
order to avoid sampling bias.
Designing of any scientific study, be it on station or on farm, is in
principles the same, the aim being separation of the signal from
the noise. What differs is the way these principles need to be
applied in a given context. For on station studies, we may need
precision at theoretical level for contribution to scientific
advancement. The purpose could be not only to convince farmers
[328]
as of on farm applicability of the result but also to convince the
scientific community on the scientific value of the study. For on-
farm studies, however, we may need precision at practical level
for convincing farmers of the contribution of the results to
addressing their problems and technological advancement.
The other difference between on-station and on farm trials is that,
in on-station trials, all non-treatment factors are kept as uniform
as possible both within and between replicates at any
experimental site. Inter-replicate differences are not of direct
interest, except perhaps to assess the effectiveness of blocking.
Proper blocking, accounting for inter replicate differences that
reduces experimental error, allows unbiased and precise
estimation of treatment differences without being masked by
variable non-treatment factors across the experimental field.
Sampled farmers should represent the target population in that on-
farm trials aim at measuring technology performance under
representative farmers’ conditions and management. When the
target population of farmers exhibits considerable variations both
in biophysical and socioeconomic conditions and in management
practices, there may be a need to stratify the farmers into
meaningful categories to represent important biophysical (e.g. soil
type) and other farmer-related factors (e.g. gender). Performances
across diverse environments is often of major interest and is
measured by conducting replicated trials in each of the many
environmentally different sites. In on-farm trials, a trial at any
single farmer’s field could be considered as a single replicate of
treatments. In such trials, within field differences due to natural
factors like variability in soil fertility and plot size is usually
larger than in on-station trials.

[329]
All operations and management, not part of treatments, maybe
left to the farmer and different farmers/replicates even at a single
site (say a village) represent potentially distinct environments.
The way farmers manage their field may induce additional
differences between plot variations within any single farmer’s
field. For example, a farmer may weed part of his/her plot on a
given day and another part a few days or a week later. Farmer-
induced within field variations would further inflate error
variance, reducing precision and accuracy of estimates of
treatment differences.
Treatment effects are measured under as many different
environments as there are farmers/replicates at that site. Assessing
differences in treatment performance across different farm to farm
conditions is often of more interest than average performance
across farmers in order to identify treatments with consistent
performance, i.e. less G  E interactions. Objective and accurate
collection of relevant data on between and within field differences
in on-farm trials is the key to obtain quality results and draw
sound conclusions.
It is obvious that if a technology developed for a better
performance by researchers is unacceptable to farmers for some
reasons and is not adopted at the end of the day, all the resources
invested to the development of that technology will be wasted. It
is very important, therefore, that farmers be involved in problem
identification as well as selecting and testing processes of
technologies. Farmers may participate in different ways as
detailed below.
a) On-station trials (planned and executed entirely by
researchers with farmers’ inputs in design or in the form
[330]
 of comments/feedback on its outcome or relevance) (Type
0).
b) On-farm trials (planned and executed entirely by
researchers by taking account of farmers needs and
preferences) (Type 1).
c) On-farm trials (planned mainly by researchers or jointly
by both but executed entirely by farmers) (Type 2).
d) On-farm trials (planned and executed entirely by farmers
in their own way with the researcher merely observing)
(Type 3).
The farm to farm variation in productivity is tremendous because
of the inherent farm to farm variations and owning to the ability
of farmers to employ ‘best practices’ available to them.
Therefore, technological breakthrough in attaining high
productivity is achieved not only by introducing improved
technologies but also by improving the technical efficiency of
technology and knowledge adopters. In on-farm trials, therefore,
non-treatment factors, which reflect important components of
farmers’ normal crop management practices, should be observed
and analyzed to reveal how these non-treatment factors affect
performance of treatments. Data on between-field differences
relevant at farmer level could be used as explanatory variables to
assess how inter-farmer differences in technology performance
depend on differences in field and farmers’ conditions. At a plot
level, it may also be advisable to record data on relevant farmer-
related (and other) within-field variations for use as covariates in
data analysis to improve precision and accuracy.
Site replication is required to accommodate farm to farm variation
as many fields within a village or community have their own

[331]
peculiar features including difference in management practices.
Replication within each farmer’s field may or may not be required
depending on the stage and objectives of the experiment. When
the objective of the experiment is to develop farmer-preferred
technologies under the active participation of farmers, we may
need to estimate both the intra-farm and the inter-farm variability.
In that case, both site replication and within site replication is
required not only to partition out effects of site and effects within
site but also to assess effects of treatment by site (farmer)
interactions. In such cases, use of single factor trials with
experimental designs like a randomized complete block design
(RCBD) on a homogeneous part of the field, and keeping the
number of treatments manageable within the scope of the farmer
to avoid complication, is considered advisable. When the
experiment is a factorial type, again limiting the factors and factor
levels within bounds to avoid complication is required. There
should be adequate error degrees of freedom, say 10 or more, so
that a good estimate of the error variance would be possible.
The principle of experimental design for on-farm experiments is
more or less similar with on-station experiments but more
complex designs like incomplete block designs should be
avoided. The application of other designs like the split-plot or
strip-plot arrangements may complicate better understanding and
full participation of farmers.
When the objective of the experiment is to demonstrate “finished”
technologies already tested and proven to be superior through
prior experiments at different levels, however, the “one-farmer
one-replicate approach” or site replication alone may suffice for
the purpose. In this approach, we need precession only at practical

[332]
level (not at theoretical level). However, by its nature, the
approach keeps trial size manageable and the demonstration effect
and technology application could also be maximized because of
low level of complication. It should be noted that inter-farmer
variation is likely to be higher than intra-farmer variation which
also justifies single-replicate per farm approach, unless we wish
to estimate both intra and inter farmer variability with treatment
by site (farmer) interaction effects.
The number of site replications may vary based again on the
objectives of the experiment and variability of the
recommendation domain. Experiences show that the number of
differently treated plots should be limited to say not more than 6
for agronomic studies and 20–30 for variety tests where all
treatments should occur in every field. This is believed to enable
farmers to properly understand what is being tested and easily and
correctly assess treatments using their own criteria.
In addition to the regular designs, two types of trials, i.e. the
mother and the baby trial system is frequently applied at on-farm
level. The mother trial is fully replicated (Type 1 trial) and the
nearby baby trials are un-replicated (Type 2 trials). The mother
and baby trial is becoming a popular method to evaluate new
technologies under farmers’ own conditions. The usual way is to
have one replicated researcher managing a mother trial and
numerous farmers managing un-replicated baby trials within one
village. The mother may consist of two experiments: one
conducted under farmer’s management practices and the other
conducted with research specific inputs. On the other hand, the
baby trials comprise sub-sets of the mother trial in terms of
treatment and are located on many farmers’ fields. Treatments are

[333]
evaluated under real farmers’ conditions and create opportunities
for flow of information between farmers, between farmers and
researchers about the performance and appropriateness of the
technology under farmers’ conditions. It should be noted that the
check treatments including farmer’s preferred checks are essential
in the evaluation process. The farmer’s own crop or a neighbor’s
crop grown on an adjacent or nearby field may also be used as a
reference. Unless checks are included, it is not possible to tell if a
test treatment is better or worse than the ones that are currently
under promotion. From the baby trials, only analyzable farmers’
perceptions and acceptance of technologies are collected whereas
the actual statistically analyzable yield data are collected from
mother trials, carefully managed and monitored trials where all of
the entries are compared with each other. A matrix ranking of the
performance of the treatments can be done by a group of
interested farmers. There are only a few mother trials compared to
the number of baby trials. As a minimum, three mother trials may
be required in each village and these trials may be conducted in a
minimum of four villages. The three mother trials in a village are
grown by different farmers in different fields. The mother trial
may be repeated on the research station but baby trials are not
repeated on the research station. There are many more baby trials
than mother trials and, in order to avoid confusion, it is simpler
not to have any farmer growing both a mother and a baby trial.
The major differences between the mother and the baby trials are
summarized in Table 43 below.

[334]
Table 43. Summary of differences between mother and baby trials

Parameter of
comparison Mother Baby
Objective Obtain yield data Obtain farmers’ perception data (including
qualitative evaluation)
Number of trials Just a few Many
Replication Replicated (within site) Un-replicated (within site)
Supervision By researchers By farmers
Plot size Small Large
Treatments All Partial (fraction of mother trial)
Records Yield is recorded Yield is not recorded

Data on farmers’ perception (qualitative evaluation) are collected

from both the mother and the baby trials, analysed using different
tools like matrix ranking and interpreted. As mother trials maybe
replicated within site or between sites, a formal analysis of
variance could be done based on the model and the type of trial
(Type 0, Type 1, Type 2 or Type 3) as stated above. Results of
both the mother and the baby trials enable to understand how
technologies fit into farmers’ broader farming and livelihood
strategies.
In livestock research, on-farm evaluation is considered as an
intermediate stage between on-station level research and the
introduction of the improved system at farm level. The key
objectives are to verify the performance of a candidate technology
or practice that is new to the local environment, demonstrate a
technique that has been successful elsewhere, and collect
information on farmers’ attitude towards the technology or
practice.
Diagnostic survey in animal production system is essential for in-
depth understanding and prioritization of the problem in terms of

[335]
herd or flock demographics and productivity, the current
condition and productivity of pasture/range lands, and nutritional
and health constraints. Under situations, where nutritional
interventions are of high priority, it is worth giving attention to
the overriding factor, for example key nutritional factor or health
problem, the period/s most limiting to the animal, the class of
animal most affected (age, sex and physiological stage), the
resulting impact on herd productivity, and the cause–effect
relationships of the problem. Additionally, it is important to take a
note on assessment of the area for its suitability to on-farm
research) in terms of willingness, cooperative spirit of the farmers
and availability of facilities (feeding/drinking troughs, stores,
balances, etc.).
Unlike on-farm research on crops, on-farm research on livestock
is usually planned and managed by researchers when the
experimental animals, inputs, and services are provided by the
researcher because of the knowledge-intensive nature of the
discipline. To verify the performance of a new technology, for
instance, researcher-managed trials are usually preferred because
farmer-managed trials lack the necessary control to derive
accurate results.
19.2. Major Steps in Undertaking Agricultural Research in
Ethiopia
The Ethiopian Institute of Agricultural Research is a major body
that undertakes various agricultural research at the national level
supported by a number of prominent Centres located in various
parts of the country. In addition, the Institute coordinates
agricultural research activities that are run by Higher Learning
Institutions which have affiliations to agricultural education in the
[336]
country. Moreover, the Institute collaborates with Regional
Agricultural Research Institutes (RARIs) in the country and
engages in a number of joint research activities. The agricultural
research activities conducted by all of these entities are
coordinated under one loosely organized system known as
National Agricultural Research System (NARS).
Good agricultural research must contribute to the technology and
knowledge base of the difficult task underway to improve
livelihoods of farmers and benefit other various stakeholders,
thereby contributing to the overall development of the sector by
addressing the actual and potential agricultural problems.
Generally, good research involves a number of peculiar features
including: prioritization of the most important problems that
constrain production, productivity and product quality at farm
level, systematic planning with valid and reliable methods that
best suit the nature of a given study, built on appropriate data,
characterized by reproducibility, and replicability.

Conducting research requires organized steps each of which has

specific and or sometimes overlapping objectives.
All of the above-mentioned Institutions (i.e., members of NARS)
prepare research proposals before conducting research more or
less following a similar template developed by the EIAR.
However, each Institution may have its own slightly varied
version of template to write research proposals.
The major steps for conducting research are identifying the
research problem, reviewing literature, formulating the research
hypotheses, developing objectives, selecting the research design,
formulating the research methods, getting approval for the

[337]
research proposal (approval from competent authorities),
conducting the research work, collecting and analysing data,
interpreting the results, preparing the research report, presenting
the results in a scientific or other relevant fora, and publishing of
the report or the research paper.
Step 1. Problem identification
A research problem to be researched may be a statement
formulated about an area of concern, a condition to be improved,
a difficulty to be eliminated, or a troubling question that exists in
scholarly literature, in theory, or in practice that points to the need
for meaningful understanding and deliberate investigations.
Selecting and properly defining a research problem is the first and
foremost step in the agricultural research process. The problem
must be properly identified and defined for well-targeted
interventions. A properly selected and prioritized research
problem will provide guidance as to the techniques to be applied
to conduct the research, the type of data to be collected, the type
of analysis to be made, and general strategies to answer the
research questions. As a prerequisite for any research, a careful
formulation of the problem is the base for working out on the
research design and other subsequent steps involved in the
conduct of the research.
However, the researchable problem should not only be identified
but also prioritized based on different criteria. Agricultural
research problems may be defined as any difficulty or gap in
knowledge, technology, etc. Bridging this gap or solving the
problem may be required by farmers, ago-industries, exporters or
any users of other agricultural products to improve productivity,

[338]
efficiency, quality, income, etc. There is no purpose in repeating
research works done previously to achieve similar the same
results. Repeating the same previously done research for the same
result is what is called “re-inventing the wheel” and will have no
benefit, other than wasting resources.
Different approaches and techniques may be used to identify
research problems. These include surveys, stakeholder platforms,
literature survey, emerging issues, government development
plans, personal observations, etc.
Results of surveys
Surveys are important because they generate useful feedback on
gaps in agricultural production and productivity. Survey results
have led to better understanding of researchable problems of
different stakeholders in the agricultural value chain not only to
develop better targeted research agenda but also to make other
important decisions. Some sources ascertain that surveys
conducted with the intention of using the results to formulate a
research proposal that is reflective of the community’s input and
priorities can foster a greater level of trust and goodwill, thereby
increasing the relevance of the proposed project.
Not only social survey, but also information assembled from
biological and physical surveys (e.g. status of biotic and abiotic
stresses in a given area) could best serve in identifying
agricultural problems and designing different research agenda
under different conditions. Such surveys also provide insights into
the kind of technologies to be developed anew. They also help as
a decision tool to make readily available research interventions to
overcome certain agricultural problems.

[339]
Stakeholder platforms
The research system in Ethiopia has evolved through different
phases in terms of stakeholder participation, each phase with its
own advantages and disadvantages. At the initial stages, there was
largely a unidirectional top-down on-station based research
approach where there was little or no as such stakeholder
participation. As time went by, however, the linkage between
research, extension and farmers improved through different
approaches like farming systems research (FSR), an approach
implemented in the mid 1980’s. Another approach was the client-
oriented research approach, introduced in the late 1990's, which
attempted to change farmers’ participation from a consultative to
collaborative one. Recently, the agro-ecology based research
approach, which has enabled better targeting of the ultimate
production environments and recommendation domains, has been
practiced. The periodical improvement of the linkage among the
major stakeholders has obviously enabled agricultural researchers
to better understand the circumstances, needs, priorities, and
constraints of farmers, thereby enhancing research relevance.
The Ethiopian Institute of Agricultural Research has had a good
culture of rigorous internal evaluation of the research agenda at
different levels (peer-review, process or department level review,
center review, project or program review and corporate review).
This review stages have been helpful not only for technical
scrutiny but also for aligning the research projects with the
national development goals. Other criteria for accepting research
proposals include priorities, originality, gender sensitivity, and
resource requirement as well as for avoiding duplication of efforts
in relation to expected successes in terms of addressing
developmental questions.
[340]
Active engagement of the right stakeholders in problem
identification is critically important to comprehensively address
the whole value chains of a given commodity based on multi-
stakeholder needs and priorities. The interventions thereof could
also be adoptable when voices of multi-stakeholders are heard
and considered in this process. The engagement of
stakeholders not only in problem identification but also in search
of researchable solutions is essential as it provides ideas from
varying backgrounds with different viewpoints. In Ethiopia,
stakeholder platforms, particularly the Agricultural Research and
Development Partners Linkage Advisory Council (ARDPLACs at
Woreda, Zonal, Regional and National levels); Farmers Research
and Extension Groups (FRGs) and field days, for instance, have
served as a means of identifying researchable problems to a
certain extent.
Institutional research strategies
Any researchable problem to be identified must form part of the
research strategy documents developed by the Ethiopian Institute
of Agricultural Research, which, in turn, is aligned with the
national and sectoral economic drives of Ethiopia. In other words,
the problem to be identified should be sufficiently justified and to
be worth doing at least at the institutional level, as a priority by
any of the project beneficiaries including farmers, agro-industries,
exporters, and any end-product users.
Topics to be proposed need to be on problems of national
importance which is judged not only in terms of contribution to
food and nutrition security, export promotion/import substitution,
availing agro-industrial raw materials and job creation but also in
terms of the probability of success, scalability, gender sensitivity

[341]
and potential impact on the national development while
enhancing the natural resource base in the face of climate change.
Reading materials
During project plan preparation, a thorough review of pertinent
literature, i.e. the process of collecting, selecting and reading
reference materials (books, journals, reports, abstracts and others)
from previous studies, is of great help in selecting and defining a
research problem. Source materials reveal what information,
knowledge, and technology are ideally needed in the subject area;
what is available from previous research; what are the areas
understudied or gaps remaining; and knowledge and technology
to be generated anew as a rationale for further research.
Sometimes lack of consistency in results reported across prior
studies may justify further research to explore the basis of the
inconsistency. The research justification or the statement of the
'problem', which involves providing the reader with critical
background or contextual information that introduces the topic
area, must be presented in clear terms.
Reading pertinent literature helps not only in identifying a
researchable problem but also in ensuring that the identified
researchable problem is unique and prevents duplication of prior
works by others. A review of literature also provides theoretical
and methodological basis for the study; reveals evidences that the
problem really exists and is worth investigating; enables to
conceptualize a research problem, identifies and defines study
variables; describes the severity of the problem, and provides
lessons for data analysis and interpretation.

[342]
Emerging issues
Research projects need to be initiated to develop interventions
against compelling situations like emergence of biotic and abiotic
stresses. A good example in point for biotic stresses is the
emergence of a long list of crop pests including diseases, insects,
weeds and vertebrate pests. Likewise, a good example of abiotic
stresses is the recurrent drought which has been causing high
losses in crop and animal production in Ethiopia, which has been
exacerbated by climate change. The expansion of soil physical
and chemical degradation of soil erosion, acidity, salinity, etc. and
prevalence of climate change make another example of newly
emerging abiotic stresses.
While the familiar biotic and abiotic stresses themselves are
sufficient to complicate the task of agricultural production, the
emergence of new threats has worsened the situation and deserve
special research focus and priority. A number of such problems
are also evident in the livestock sub-sector. Project proposals that
are technically sound and have a strong policy-based thrust in
solving the emerged issues should be selected, presented, and
implemented.
Personal observations
Research problems can be identified by expert initiations
emanating from their own observations and previous experiences
based on real life experiences with farmers, agro-industries,
exporters and consumers. However, care must be taken that there
is no point in choosing a research topic of no interest to anyone
and, therefore, the principal investigator should prove the
significance of any research topic in terms of generating new
information, knowledge and technology. Whether the

[343]
information, knowledge, and technology generated thereof could
benefit the beneficiaries should be ensured through dialogue and
official endorsement in the project proposal review process.
When project proposals are based on personal observations, the
researchers need to somehow ensure that beneficiaries are
interested by involving them in the review process and
strengthening the proposal in the direction of their interests. Or
else, the requests for doing research on a topic should come from
the beneficiaries or development partners themselves.
Step 2. Prioritizing research problems
A priority is a statement about the importance or merit of selected
issues, actions, or circumstances. In some cases, you may have a
number of research problems. To state that “all priorities of are
equal” implies that you do not have a list of priorities at all.
Priorities have a rank-ordering character about them and you need
to address the top priority first, the second one next, and so on.
Thus, prioritization of research problems is a process of placing a
number of agricultural production problems in a rank order on their
importance or perceived significance as barriers to achieving the
desired level of productivity. Research problem prioritization is a
pressing issue because of resource scarcity and resource allocation
since addressing each and every problem is not affordable. The
research problems selected for the study must be prioritized based
on certain criteria including urgency of solving the problem,
comparative advantages of working over other similar projects and
even in terms of availability of the right staff with the right
experience and qualifications.

[344]
First and for most, the topic of research selected for proposal
writing should be within the scope of one or more of the national
development priority areas including food and nutrition security,
supplying raw materials for agro-industries, export promotion,
natural resource conservation, and job creation.
The probability of success needs to be justified as no research
project can be pursued with no probability of success. For a
research project to be accepted, not only should the project be
successful in terms of achieving its objectives and the likely
impacts, but also its level of risk of failure should be low. It is
also necessary to present convincing evidence that the risks of
investing in a project (potential negative consequences) are
smaller than the risks that will continue to occur in the absence of
solutions to the problems to be researched. Likewise, the ability
of the research organization (infrastructure, facility, and financial
capacities) hosting the project and its human power
(qualifications, training, and experience) to resolve the problems
needs to be clearly shown along with the affordability of the costs
involved and the time required to solve the problem. Interests of
research partners to cooperate may ensure the practicability of the
study.
From the scientific point of view, a subject which is overdone
should not be normally chosen to avoid investing to elicit
information, knowledge, and technologies that are already
available unless justified in terms of comparative advantages or
other vital aspects of the need for the research. A problem
preceded by a brief preliminary feasibility study may be better
justified for further in-depth study.

[345]
Step 3. Writing research proposal
A research plan or a research proposal may be defined as a
document that is typically written by a scientist which describes
an idea for investigating a certain topic, i.e. it is a document
summarizing prior literature and describing procedures to be used
for answering research questions. A research proposal is defined
also as a plan for a piece of scientific work written for obtaining
the means to conduct the proposed work. It is a mini-research
paper that has not reached the stage of actual implementation. A
scientific research proposal has two primary aim.
The first aim of a research proposal is to inform about the planned
work, its expected results, and the required efforts as well as to
convince decision makers that the work deserves to be pursued.
The second aim of a research proposal is convincing decision
makers that the work deserves to be pursued. These functions
determine the types of items that the proposal should contain, i.e.
description to inform and arguments to convince.
There are two types of research proposal. These are scientific
research proposal and development research proposal. Both are
more or less the same. A scientific research proposal is aimed
basically at the investigation of a situation or a phenomenon
through research and coming up with results that can be used for
addressing a given problem or problems. However, a
development research proposal is aimed at getting larger grants
for doing research-based activities to solve a societal problem of a
mega proportion and is somewhat more detailed in terms of the
methodology, input, expected output, monitoring and evaluation,
etc.

[346]
The current model of project planning in the Ethiopian
Agricultural Research Institute starts with formation of a team of
five to seven senior researchers from multidisciplinary
background in each program by the respective directors of the
research process. The team is tasked with designing new research
project proposals in different thematic areas every three years.
Team members constitute well experienced researchers who have
rich complementary competencies in different aspects of national
research and development.
Even if the procedure basically follows a top-down approach of
project proposal development, most of the implementing
researchers are represented by senior researchers as team
members and most of the researchers also contribute to
enhancement of the proposals at research program review levels.
This approach of project plan preparation is often criticized for
limited participation, less learning opportunity and limited
experience sharing and reduced scientific interaction particularly
for junior researchers.
The project proposals prepared in different thematic areas by
different research programs every three years is subjected to the
institutional review procedure and officially endorsed for
implementation. Each of the project proposals is then further
disaggregated to component activities which are presented to
review fora every year with all the necessary adjustments and
modifications based on contemporary situations. Inviting experts
(from inside and outside EIAR) from different background as
resource persons on the review fora has also become a norm.
A research proposal has outlines of the process from beginning to
end with the major components which include: Title, background
[347]
and rationale, objectives, research methods (detailed design,
collection and analysis of the data, interpretation and synthesis of
the results), work plan, financial requirement, project output,
logical framework matrix and beneficiaries and impact of the
project.
A research proposal should be written accurately, briefly, and
clearly. Writing a research proposal is both science and art. A
research proposal usually contains the following major sections:

Title
The title of a proposal is a very important section. It should be
written carefully with precision and clarity. See the section of
report writing for details.

Background and rationale

In this sub-section of the proposal, the researcher should provide
the background information necessary to understand the story and
put the planned work into context. In this section, it is important
to tell the reader clearly about the genesis of the problem and how
the research problem was identified and why it is deemed a
problem. In addition to briefly introducing the subject of the
study, the background should constitute the scene of the study, i.e.
why the study is really needed by revealing the ideally desired
level of information, knowledge and technology; the available
information, knowledge and technology from previous studies;
and the gaps to be bridged through further research. The back
ground section could be supported with references or it could be
written based on anecdotal information.

[348]
Statement of the Problem
This sub-section of deals with the main question that the proposed
work is aimed to address. By definition, a problem is a gap, a
discrepancy between the way things are and the way things
should be. It is the failure of the status quo to address the need. It
could be the cause of the discrepancy between actual and
potential yield of a crop. It could be existing conflict in research
results in the literature. It could be a neglected topic in the
literature. It could be also “real-life” problems found in the
workplace, at the home, the community, and so forth. It could be
also government concerns, or problems ear-tagged by
governments.

It is important to indicate the frequency and severity of the

problem. What would be the consequences of the problem if it
was not addressed? It is important to support these facts with
statistical information.

The researcher should not assume that everyone sees the problem
as clearly as he/she does. Therefore, he/she should argue the
problem sufficiently even if when it appears obvious. This is
because, even if the problem is obvious, reviewers want to know
how clearly the researcher can state it. In addition, the problem
should be significant i.e. not trivial or a repeat of previous work.
For simple problems that can be addressed through a quick fix,
for problems that have already been researched, there is no need
to do research. That will only waste resources.

It is an issue or concern that needs to be addressed. Furthermore,

the problems should be interesting to the researcher. The scope of
[349]
the problem should be manageable by the researcher. The
researcher should have the knowledge, time, and resources to
investigate the problem. The problem should also be the one that
can be investigated through collection and analysis of data. It is
also important that it is ethical to investigate the problem.
It is also important to reduce large questions to specific,
actionable strands. This means it is important to narrow down the
question to one or a few specific issues. The problem should be
delineated, in order to limit its scope for practical investigation
(time and funding). This is because resources are limited.
Therefore, the researchable problem should be time-bound.
Furthermore, it should not take unnecessarily high amounts of
budget.

Significance of the Study

This sub-section is an opportunity to consider the social or
political value of conducting the proposed research. The
researcher should tell ell the reader why investigating the problem
is important. The researcher must tell the reader the reasons why
and how s/he thinks the findings might change policies, theory or
practice. As a whole, this refers to the relevance of the study in
terms of academic contributions and practical use. The following
questions should be answered: What is the benefit of investigating
the problem? For example, the results of the study will add
knowledge to science. The researcher must tell the reader the
reasons why and how he or she thinks the findings might change
policies, theory or practice. In short he or she should indicate who
will benefit from the findings of the study and how. Farmers and
other beneficiaries such as government, researchers, national,

[350]
international agencies, and non-governmental organizations
should find the results useful in planning and decision making.
Purpose Statement (Objective, Research Questions, Hypothesis)
The purpose statement describes the aim of the study. It is a
statement of the central intent and questions for a study. A
proposal developer needs to convey the purpose of the study
clearly. The objectives of a study are the goals set to be attained
through the study.

Objectives should clearly specify what a researcher wants to

attain at the end of the study.

In an empirical or applied research, objectives set relationships

between independent and dependent variables or relationships or
associations among variables. An objective usually includes a
general objective and specific objectives. A general objective is a
statement about the overall thrust area of a study. A specific
objective indicates a particular topic a researcher wants to
investigate within the main framework the study. Thus, general
objectives are usually broad, scientifically far-reaching aspects of
a study. In some fields, general objectives are synonymous with
goals or aims. Typical general objective-oriented words include
advocate, analyse, appreciate, behave, develop, empower, enjoy,
extend, feel, illustrate, improve, integrate, internalize, know,
participate, promote, recommend, and understand. A specific
objective typically focuses on a particular question or hypothesis
and the methods needed and outputs expected to fulfil the aim
The objectives of a study are generally written in words starting
with “to determine,” “to find out,” “to ascertain,” “to measure,”
“to describe,” “to compare,” “to demonstrate,” etc.
[351]
The objective (purpose statement) is very broad, the research
questions narrow down our area of inquiry, and the hypotheses,
when necessary, focus on exactly the variables we want to test.
Research questions or hypotheses narrow the purpose statement to
predictions about what will be learned or questions to be
answered in the study.

Some authors only pose the question, whereas others do all three,
offering a question, framing it into a hypothesis, and then
describing specific research objectives. Each approach has its
place, but the question is the core of it all. To eliminate
redundancy, write only research questions or hypotheses—not
both—unless the hypotheses build on the research questions.
Choose the form based on tradition, recommendations from an
adviser, or whether past research indicates a prediction about
outcomes.

Hypothesis setting
Setting a hypothesis is an important step in experimental
approach. Thus, a hypothesis is used to test whether a relationship
exists between an independent and a dependent variable. It is s a
tentative explanation for an observation, certain behaviour,
phenomenon, or event, which has occurred or will occur. The
hypothesis gives an idea about the researcher’s expectations
pertaining to the relationship between the variables in the study.
In other words, a hypothesis is a probable answer to a specific
question, which describes what a researcher expects to happen in
his/her study in definite terms. A hypothesis may be either
correct or incorrect. A hypothesis is formulated after preliminary
observations or following a thorough review of related literature,
[352]
or both, but before the execution of the study. The researcher
collects enough data that either support the hypothesis or do not
support it. If the hypothesis works, the scientist can continue
further studies.

However, not all research studies have hypotheses. Sometimes, in

qualitative studies or in a study designed to be exploratory or
descriptive as in anatomy, ecology, or astronomy, researchers
may not be able to state hypotheses, since the purpose of the
study is to explore some particular area of interest. The results
obtained from such studies may be used to develop some specific
hypotheses or prediction for future research. However, research
questions can be set as an alternative to hypotheses.

A hypothesis adds clarity and specificity to the proposal. It

focusses on the research problem to be addressed; it directs the
researcher to the type of information or data to be collected or not
to be collected.

A researcher has to normally formulate two hypothesis

statements. One hypothesis is the one that the researcher assumes.
The second one is the hypothesis that shows all other probable
outcomes with respect to the relationship set out in the
hypothesis. For example, take two variables A and B. We can
assume two types of relationships between the two variables, one
positively and the other negatively. If we say that A and B are
related, we should also take into account the other possible
alternative that A and B are not related. The hypothesis the
researcher supports is called the research hypothesis (H1).
Example: Application of nitrogen fertilizer during the critical
stages of teff increases the grain yield of the crop.

[353]
After the variables are operationalised, the researcher proceeds to
formulate one or more hypotheses. For conducting the statistical
analysis, the negative hypothesis is formulated in case the
research (alternate) hypothesis is false. It is written as a negative
statement for testing purpose only, for example:

There is no increase in grain yield of teff when nitrogen fertilizer

is applied to the crop.

A well-formulated set of hypotheses leads directly to the

experiments and sampling program that form the basis for the
research. There are two primary kinds of hypothesis statements—
null and alternate.

Thus, a hypothesis should lend itself to constructing a set of

experiments that will prove it either true or false, and it should not
include words like ‘may’ ‘might’ or ‘could’ since these make the
statement impossible to falsify (i.e., show that it is untrue).

Newton’s statement that ‘objects attract each other by means of a

gravitational force’ is an example of a good hypothesis. It is
simple, clear and testable.

A hypothesis should not also contain words like ‘and’ and ‘or’
because these make it difficult to distinguish which parts of the
hypothesis you are testing.

There are two kinds of alternative hypotheses—non-directional

and directional. Therefore, in total there are really three primary
kinds of hypotheses. A null hypothesis predicts that there is no
significant difference between two groups with respect to the
variable being tested.

[354]
You write a null hypothesis as follows:

There is no significant difference between variety A and Variety

B.

A directional hypothesis relies on prior research to make a

prediction.
You write a directional hypothesis as follows:
Variety A performs better than variety B.

A non-directional hypothesis predicts a difference between two

groups, but does not predict what that specific difference will be.
You write a non-directional hypothesis as follows:
There is a difference between Variety A and variety B.
The statistical tests of the Null hypothesis may suggest its
rejection at a particular level of significance and degree of
freedom, as the case may be. The rejection of Null hypothesis
(H0) signals an automatic acceptance of the Alternative
hypothesis (H1).
A research hypothesis should:

• Flow logically from the discussion presented in the

statement of the problem and the review of literature.

• Be very specific in presenting what aspects of the research

topic you will be studying, and how.

The null hypothesis is necessary for statistical analysis. The null

hypothesis assumes that there is no difference between the mean
of the sample and the mean of the population, i.e. any difference
is only a random sampling error. Alternate hypothesis is a
[355]
statement about the population parameter or parameters, which
gives an alternative to the null hypothesis, within the range of
pertinent values of the parameter. Rejection of H0 is equivalent to
falsifying it and, therefore, provides support for the research
hypothesis (H1) as the only alternative.

A researcher can never prove a null hypothesis since any one

result that differs from the hypothetical result is sufficient to
disprove it. It can, therefore, be either disproved and rejected or
accepted within the limits of sampling errors. The data from the
experiments are subjected to statistical analysis. The probability
of occurrence of the null hypothesis is calculated. If the
probability is lower than the level of significance, you reject the
null hypothesis. The null hypothesis is tested for rejection under
the assumption that it is true, as distinct from the research or
alternate hypothesis. Rejection of null hypothesis suggests that
you have enough evidences to support your alternate or research
hypothesis.

The rejection or acceptance of a null hypothesis is based on some

level of significance. In most agricultural experiments, 5% (p =
0.05) level of significance is often used as a standard for
rejection. In certain cases, you may even go for 1% (p = 0.01)
level of significance. When you reject a null hypothesis at 5%
level of confidence, it indicates that a difference in means as large
as that between the experimental and control groups would have
resulted from sampling error in less than 5 out of 100 replications
of the experiment. In this case, the probability that the difference
is due to the experimental treatment rather than to sampling error
is 95%.

[356]
Literature Review
In some proposal, a literature review is necessary. The literature
review may include relevant theories or conceptual frameworks
(which may shape the research purpose statement and research
questions). It starts with brief historical background. It is mainly a
summary of the critical evaluation of previously completed work
which that is relevant to the present study. However, the
Literature Review should not be a place for a data dump where
one tells everything about the field. A researcher should be
selective. The literature review expands upon that knowledge and
provide the reader with a deeper understanding of the problem
area, prior research that has been conducted, and provides a basis
for a sound research methodology upon which to investigate the
problem, ideas to fill a research gap, the knowledge that the
researcher is an expert in the field.
In general, in the Literature Review, the following should be
noted: What's been done in this topic area to date? What are the
significant discoveries, key concepts, arguments, and/or theories
that scholars have put forward? Which are the important works?
On which particular areas of the topic has previous research
concentrated? Have there been developments over time? What
methodologies have been used? What prior qualitative studies
have contributed to our understanding of the topic? If no such
studies exist, it is important to point their absence. What will your
research add to prior quantitative studies? Are there any gaps in
the research? Are there areas that have not been looked at closely
yet, but which should be? Are there new ways of looking at the
topic? Are there improved methodologies for researching this
subject? What future directions should research in this subject
[357]
take? How will your research build on or depart from current and
previous research on the topic?
Materials and methods
The section of “Materials and Methods” focuses on the study
system, measurements, and data analysis. It is the most important
part of research since, without it, there could be no way for
researchers to agree on what is likely true result of the experiment
and what is not. Research methodology is vital because an
unreliable method produces unreliable results. The main purpose
of this section is often described to be enabling other researchers
to repeat the research. However, in fact, few researchers are likely
to repeat a piece of research work. Therefore, the practical
purpose of this section is to enable readers and reviews assess the
credibility of the data collection and analysis procedures.
This section should be organized logically, perhaps
chronologically by including necessary information, and omitting
unnecessary information as well as avoiding details of procedure
already published (just cite the original reference). The study
period and location (climate, soil, vegetation, cropping system,
etc.) is described under this topic.
Experimental materials
This is the sub-section of the Materials and Methods under which
the important study materials (study species, cell cultures, etc.)
are described. This section includes also population, sample, and
participants, data collection instruments, variables, etc.
For laboratory biological studies, complete taxonomic
information about the organisms is given under this sub-section:
genus and specific epithet as well as subspecies, strains, and so

[358]
on, if necessary. It is important to specify how the organism was
obtained. The composition, source, and quantities of chemical
substances, growth media, test solutions, and so on is also
described under this sub-topic. It is important to give crucial
details on such factors such as temperature conditions, pH,
photoperiod, duration of observation periods, sampling dates, and
criteria used to make for field studies, specify where and when the
work was carried out. Describe features of the study site relevant
to your research and include maps, drawings, or photographs
where necessary. If published information already exists on the
area, cite sources.
Treatments and experimental design
It is important to describe the treatment and how the experiment
will be laid out (design). Thus, the experimental subjects,
treatments, measurements, and statistical models, and design of
the planned experiment (randomization, replication, and blocking
structure) should be described here.
For surveys, the sampling procedures and observational methods
are described here, including sampling protocol of the surveys.
Methods of data collection and analysis are also described.
Experimental procedure
Under this sub-heading, the researcher should provide
information on how he/she will conduct the different activities of
the research and collect the data.
Statistical analysis
Under this sub-heading, the method of statistical analysis should
be described

[359]
The materials and methods section is usually written in the future
tenses.
Expected outcomes
Under this section, outcomes, which are the benefits, changes, or
effects that occur to the target population due to the research are
described. In other words, they are the desired changes in
peoples’ knowledge, skills, attitudes, or behaviours. Outcome
statements answer the question ‘‘so what?’’.
The main output of the proposed project, beyond publications,
should be described in terms of societal benefits and livelihood
impacts, which will be expected from the results of the
experiment. Many direct beneficiaries like farmers, industries,
exporters and indirect beneficiaries like consumers who will be
benefitted from purchase of the produce at a cheaper price may be
mentioned. Likewise, direct beneficiaries of a project to develop
drought-resistant varieties any crop may be farmers in dry areas
and other people who indirectly benefit from these new varieties
may include seed multipliers who can sell more seed to farmers.
The output could also be described in terms of deliverables like a
workshop, number of reports and trainees, a crop variety or
animal breed. For continuous projects like crop varietal
development or animal breeding, intermediate outputs may also
be presented as project milestones.
Work plan
It is important to develop a timetable (usually in table form or
flowchart), indicating the sequence of research phases and the
time that the researcher will probably need for each phase (Table
44).

[360]
Table 44. Work plan for a research undertaking

Activity May June July–Oct. Nov.–July

1. Site selection and land preparation X
2. Field preparation and planting X
3, Field management and data collection X
4. Laboratory analysis and data compilation X
5. Data analysis and write-up X
5. Submission of research report X

Budget
The budget should contain a detailed item-wise breakdown of the
funds requested for, along with a justification for each item (Table
45). The budget should be realistic, detailed, and accurate, listing
personnel costs, equipment, materials and supplies, travel,
workshops, publication, and any other significant expenses. Also
include the costs of preparing research reports and publishing
articles that will emerge from the research activities. Do not
include large amounts for contingencies or other vague purposes.
Exaggerated budget undermines funder confidence. On the other
hand, too little budget endangers execution of the project.
Therefore, each cost item should be justified.

[361]
Table 45. Budget arranged a quarterly basis for a research undertaking

Activities Unit Quantity per period Cost per unit (Birr) Cost per period (Birr)
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Land Preparation and Man-days 60 60 60 60 120 7200 7200 7200 7200

planting
Field management and person 60 60 60 60 200 12000 12000 12000 12000
data collection
Laboratory analysis and Man-days 2 2 2 2 20000 40000 40000 40000 40000
data compilation
Wages and salaries for Man-days 4 5 6 3 1000 1500 2000 2600 1600
support staff
Expenses for chemicals Pcs
Sub-total (Birr)
Sub-total (Birr)
Total (Birr) 60700 61200 61800 59200

[362]
References
A researcher needs to navigate the works of other authors to
extract information. This is because citations/ the reference
section of a proposal are very important for showing your depth
of research and field work; this section also shows your
exceptional effort that you put in developing a research proposal;
it also shows that you want to give credit to your fellow authors
and researchers for how their work helped you; it also provides
proof of your arguments in the research work; it also helps avoid
any kind of plagiarism. Therefore, citing the reference work has
to be done in a proper way. Improper or inadequate citations can
make your work plagiarized and blame you of copying, even
stealing others’ contents. Be consistent in using the four reference
systems, i.e., APA (American Psychological Association) style,
Vancouver referencing style, Harvard style, and Numbering
system.

In writing development research proposal, which is usually done

for seeking a grant for conducting a mega project usually for a
maximum of three years, a similar pattern of preparing a scientific
research proposal is followed. However, in this case, the sections
of purpose statement, methodology as well as monitoring and
evaluation are presented in more detail than in a scientific
research proposal.

For example, in a development research proposal, in the section

of purpose statement, target, the goals and objectives are defined
in a greater detail specificity. In the methodology section, results
to be attained, targets of the activities, input, output, outcome, etc.
are clearly and thoroughly articulated.
[363]
For example, in such a proposal, a target is the exact result that
you intend to achieve. e.g. 25% yield increment over the standard
variety; a goal is the end-result or outcome you want. It is what
you want to achieve. e.g. improved productivity; a purpose is the
reason you want to achieve that goal. It’s what motivates you to
continue. e.g. released variety. An objective, on the other hand, is
a step you must complete in order to make progress towards your
goal. e.g. compare mean performances of candidate varieties to
identify a better variety. Several ways can be used to signal the
research question (research objectives) such as: “to determine
whether …”, “the objective of this study was to …”, “this study
tested the hypothesis that …”, “this study was undertaken to …”,
“the objective of this study was, therefore, to …”, etc. Note that a
goal may hold a lot of objectives together in such a proposal.
In addition, in such a proposal, the monitoring and evaluation is
also very detailed. Thus, in addition to work plan and budget,
such proposals have also detailed log framework.
The Logical Framework Approach and the Matrix
The logical framework approach and the matrix are basic tools
used for project planning, monitoring, and evaluation at
institutional, national and international levels by various
institutions. It provides a handy summary for donors, managers,
scientists, and development workers for planning, implementing,
and evaluating a specific program or project.
The Logical Framework Approach (LFA)
The logical framework approach (LFA) is an analytical process
and a set of tools used to support objectives oriented planning and
management of development research project. It provides a set of
inter-locking concepts which are used as part of an iterative
[364]
process to aid structured and systematic analysis of a project or
programme idea. The LFA is an instrument for verification and
synthesis that helps to synthesize the different elements of a
project. It also helps to verify if the project elements have been
articulated in a logical manner and lays the foundation for
implementing monitoring &evaluation (M&E) system indicators
and means of verification. It is also an analytical presentation and
management process used for analysis of the existing situation
during project preparation, and establishing logical hierarchy of
means by which objectives will be reached. It also provides an
insight into how outputs and outcomes might best be monitored
and evaluated via identification of potential risks, presenting
summary of the project in a standard format, and monitoring and
reviewing projects during implementation.
A distinction is usually made between the Logical Frame
Approach (LFA) and the Logical Framework Matrix (LFM).
Along the project cycle management, the LFA involves
stakeholder analysis (analysis of interest groups); for research
planning the primary step is figuring out and understanding the
needs of stake holders or interest groups. This includes target
groups such as smallholder farmers, pastoralist/agro-pastoralists,
commercial farmers, agro-industries, farmers’ union, processors,
and organizations and individuals that affect or affected by the
research plan.
The stakeholder analysis is followed by problem analysis of
specific target group which would help to identify core problems
and guide the development of hierarchy of objectives and select a
preferred implementation strategy. The results of the existing
situation analysis are presented and farther analyzed through the

[365]
development of a LFM. When completed, the LFM provides a
summary of the project design. Along the project cycle
management, the log frame approach is used for the following
key stages (Table 46).
Table 46. Key stages of project cycle management and focus

No Stage Focus
1 Programing/planning • The situation at the national and sector level is analysed
to identify problems, constraints and opportunities which
a national project would address.
• Involves a review of socio-economic indicators of national
priorities with stakeholder and problem analyses.
• Represents the entire foundation of a research project or
program to be developed.
2 Identification • It is a stage at which the initial project proposal is
conceived and formulated.
• At this stage, it is worth checking whether the project
concept is relevant to priority local needs and consistent
with national policy priority.
3 Formulation • The main justification for the project, the description of
potential target groups and assumptions are more
important; key aspects of which include: situation
analysis/key assessments, project description and
management arrangement, and feasibility and
sustainability.
4 Implementation • Serves as basis for detailed operational planning.
including • The indicators and means of verification provide
monitoring and framework for a more detailed M&E plan for project
reporting management.
• The assumption provide basis for an operational risk
management plan.
• The results, indicators and means of verification
(activities, resources and costs) provide the framework
for preparing project progress reports.
5 Evaluation • Were the planned activities achieved? Will they be
sustained and what lessons have been learned?

[366]
The logical framework matrix
The logical framework matrix usually consists of four columns
and four rows (Table 47). The rows represent vertical logic while
the columns represent the horizontal logic. The key idea is to
show the interconnectivity, interdependence or the cause and
effect relationship between subsequent rows. On the vertical side,
if we begin from the lower level, i.e. activity leads to output,
output leads to purpose and purpose leads to goal and goal leads
to super goal. Any lower level serves as basis for the next stage
following a clear and logical pattern. Likewise, to ensure
attainment of the set targets, objectively verifiable indicators,
means of verification and important assumptions are articulated to
represent the horizontal logic for a particular issue in the vertical
logic. There is very strong interdependence/interconnectivity of
the focus areas in the horizontal logic as well.
The vertical logic of a logical framework consists of key
component including goal, purpose, output and activities. In some
cases, the fifth component above goal is put as super goal. Super
goal represents combined effect of several macro level goals,
including attainment of economic and social growth with
environmental sustenance as the basis for growth and
transformation in Ethiopia.

[367]
Table 47. The logical framework matrix

Log frame Narrative summary Objectively verifiable Means of Important assumptions

hierarchy indicators verification
Goal The higher order Means to verify an Sources of data to Important external factors necessary
objective accomplishment of a verify status of goal for sustaining objective in the long
goal level indicator run
Purpose The effect or impact of Means to verify an Sources of data to Important external factors needed to
the project accomplishment of a verify status of attain goal
purpose purpose level
indicators
Outputs The deliverable of the Means to verify an Sources of data to Important external factors needed to
project accomplishment of verify status of output attain purpose
output level indicators
Activities The main activities to Inputs: A summary of Sources of data to Important external factors needed to
accomplish project budget and verify status of attain outputs
resources activity level
indicators
 If we do these activities and these assumptions are valid we will achieve this output
 If these outputs are realized and these assumptions are valid we will achieve these objectives
 If these objectives are met and these assumptions are valid we will achieve the project purpose and contribute to the goal
 If this purpose is attained and these assumptions are valid we will contribute to the project goal

[368]
Thus growth and transformation can be taken as super goal while
attainment of food and nutritional security is a goal for the
agricultural sector. Arranged in cause-effect relationship, the
vertical logic consists of goal, purpose, outputs and activities.
The project purpose refers to the effect or change that the project
is expected to achieve and it describes, in verifiable and
measurable terms, the desired impact of the project once it is
completed. The operative word for a purpose is to “achieve”.
Generally, goal dictates the purpose while the purpose forms the
basis for objectives.
The objective of the project forms the basis for the activities and
it has to be (S-I-M-P-L-E or S-M-A-R-T), i.e. S= Specific, I=
Immediate, M=Measurable, P= Practical (real solution to a real
problem), L= Logical (systematic contribution to achieving the
project success), and E= Evaluate (how much has to occur for the
project to be effective). This implies that indicators need to be
specific, quantifiable (measurable both quantitatively and
qualitatively) and they must be attainable at reasonable cost using
appropriate data collection method, relevant to the management
information needs of the people who will use the data.
Additionally, an indicator needs to be collected and reported at
the right time to influence many management decisions. Example
of logical framework matrix is given Table 45. Suppose there is
an interest to develop comprehensive project for enhancing maize
production in the highlands of Ethiopia with key components of
breeding and genetics, agronomy, natural resource management
and capacity building, the example given in Table 48 is a typical
log frame matrix for the breeding component.

[369]
Table 48. Example of log frame matrix for enhancing maize production in the highlands of Ethiopia (Breeding component)

Log frame Narrative Objectively verifiable Means of Assumption

Hierarchy Summary indicators Verification

Goal Increasing maize Statistics of maize National statistics

production Production in the high land

Intermediate goal Farmers adopt high land Number of farmers and land Base line data before and No major and un-expected changes
maize technology used for high land maize after intervention in natural environment
production (Climate, pests and disease)

Purpose EIAR develop and Certified seed with an on-farm Baseline survey and project [Link] gives sufficient
disseminate location yield potential yield of 6 ton/ha mid-term report priority and support to high land
specific high land maize and desirable characteristics maize research
technology availed to 5% farmers in the pilot [Link] national research systems
learning district by 2027 and program put in place
3. EIAR’s training and technical
support program are effective

[370]
Output [Link] maize 1. Farmers have access to at 1.1. Variety release [Link] give sufficient priority
cultivars least one location specific document or registry and support to high land maize
[Link] crop improved high land maize variety 1.2. Results of location [Link] national research system
management and by 2027 specific on-farm trial [Link]’s training and technical support
cropping systems 2. Farmers have access to at 2. Location specific high land programs are effective
[Link] land least one location specific crop maize production package
management practices management best bet practice [Link] and
by 2027 implementation manual of
3. Farmers have access to at sustainable land
least one location specific management practices
sustainable land management
practice

Activity [Link] maize cultivar 1. Five collaborating centres [Link] release [Link] of seed company which
development develop and release at least 1 document/registry can avail certified seed to farmers
1.1. Obtain planting improved variety each by 2025. [Link] of on-farm trial [Link] weather
materials [Link] the end of 2027 a minimum 3. Report of field days [Link] and institutional stability
1.2. Plant test plots of 5% farmers in the pilot 4. Base line survey 4. Adequate budget
1.3. Collect field data learning district adopt improved 5. Project final report
1.3. Harvest and measure variety 6. Project impact study
yield 3. Use of Improved variety of report
1.4. Analyze and report highland maize up-scaled in 5 7. CSA report
results districts around the pilot learning
districts by the end of 2030

[371]
20. Proposal Review Process
Project proposals are evaluated at two levels in the National
Agricultural Research System (NARS). The first level is the peer
review processes where scientists working in the subject area are
assigned to review the scientific merits and technical soundness
of the research proposals, thereby providing constructive criticism
to improve the proposed research project. The purpose of the peer
review process is to make evidence-based decision in accepting
only proposals designed with valid procedures, credible, and
ensure that scarce resources are allocated to those projects with
the greatest chance of success. Peer review processes may also
ascertain that the objectives are adequate, clear, and attainable;
the methods are sound and the schedule workable; the proposed
study is within the scope of available researchers and support
staff; if the proposal duplicates areas where information is already
available; and if the project is fundable. If several proposals are
presented, there may be a need to select the ones with the highest
comparative priority for funding. Proposals recommended by peer
reviewers for further consideration are promoted to the next stage
of review fora.
The second level is to organize review fora at different levels.
The purpose at this level is to consider a number of focal areas
including alignment with the national development goals
(contribution to food and nutrition security, export earning, agro-
industrial raw material, job creation and natural resource use
sustainability), technical soundness (quality of the science and the
strength of the methodology), relevance of the proposed study
(value of the research in terms of solving the societal problem),
scalability and potential impact, probability of success, etc.
[372]
Multidisciplinary scientists from various institutions as resource
persons and different stakeholders from the beneficiary
organization are invited to contribute ideas from their own
perspective.
In the Ethiopian Institute of Agricultural Research, proposals are
annually evaluated at different stages including center based case
team, center based process, national program level, center level,
zonal and regional ADPLAC level, national case team level, and
corporate level. The proposal may be readily approved or
approved subject to some changes or may be completely rejected
with reasons for the rejection clearly spelled out at any one of the
stages.
When this step is completed effectively, then approval of the
research proposal is sought. This step is important because
without procuring funds, no research can be conducted. A
research proposal is approved based on the merit of the proposal
to solve the problem. When a research proposal is approved, the
researcher proceeds with conducting the research using
appropriate research methods.

[373]
 SECTION VI. REPORTING
RESEARCH RESULTS
21. Writing Research Reports
A research report is a document that describes what the researcher
has done, why he/she has done it, and the results obtained. Unlike
a research proposal, which is written before beginning the
research work, research reports are written after the reach work
has been completed.
Research reports constitute sections such as executive
summary/abstract, introduction/background, literature review,
research questions (hypothesis), objective, methodology
(materials and methods), findings, analysis, results, conclusion,
recommendations, and references.
Therefore, scientific papers that report experimental work are
often structured chronologically in five sections: title,
abstract/summary, introduction, materials and methods, results,
discussion, conclusion and references as detailed below.
Title
The first and important section of a research report is the title.
The title of a research report is the first section that informs the
reader about the subject of the report. Therefore, it should be not
only brief, concise, and understandable but also informative in
terms of describing the content and scope of the report. A
research report is written for a number of reasons including

[374]
sharing of original research work with fellow scientists, building
on current scientific knowledge, to get the credit for the discovery
or academic reputation to collaborate with others, to obtain
promotions and to fulfill requirements of donors. Research
communication documents may be research journals, research
reviews, conference proceedings, book chapters, annual reports,
posters, etc. However, it is advisable to focus on research journals
as journals are more advanced and skills acquired on writing
journal articles would obviously help to write for all other
research communications.
Although it is usually coined at the end of writing the research
report, the title is a very important section of the whole document.
The title of a research report should be clear and brief. It should
not be no more than 10 to 15 words (60–65 characters) and should
include the keywords to reference the classification of the
research subject. A title should contain no acronyms,
abbreviations, jargons, and formulas. Common names instead of
the Latin names of plants (and other living organisms) should be
used, especially for organisms whose common names are
familiar, well established. However, it is necessary to include
precise taxonomic information in the text of the paper.
The title must not contain abstract words such as that end in -tion,
cy, …ory, ment, etc.
It is rather important to use action words or infinitive+ ing forms
of a verb.
For example:
Incorrect: Development of new strains of bacteria under overnight
incubation in an agar medium

[375]
Correct: Developing new strains of bacteria by incubating
overnight in an agar medium
The other important aspect of the title is avoiding noun strings or
noun clusters (“noun trains”). It is better to use action verbs for
improved clarity. For example, instead of writing: “Carbon
sequestration determination of ecosystem services”. It is correct
to write: “Determining carbon sequestration of ecosystem
services”. One more example:
“Vegetable Produce Market Influence Factors in the Regions
Surrounding Addis Ababa”. “It is correct to rather re-write this
title as “Factors influencing Marketing of Vegetable Produce in
the regions Surrounding Addis Ababa”. However, it is okay to use
two noun strings in a title. For example: “Effect of nitrogen
application on barley yield”. However, using three or more than
three noun strings in a title is not acceptable. For example,
“Optimum nitrogen fertilizer rate determination for enhancing the
yield of ginger”. Instead, write this title as “Determining optimum
rates of nitrogen for enhancing the yield of ginger”.
In addition, the title should not contain low-impact or wastewords
such as: ‘‘Some notes on’’, ‘‘Observations on’’, ‘‘Investigations
on…’’, ‘‘Study of …’’, “Survey of…”, “Assessment of…”,
“Aspects of…”, "Comments on…”, "Inquiry Into…”, ‘'Analysis
of…”, “Development of.” For example, instead of writing a tile as
“Assessment of productivity of maize in western Ethiopia”,
simply, write “Productivity of maize in western Ethiopia. Instead
of writing “Survey of irrigation practices of farmers for potato
production in the eastern highlands of Ethiopia”, write “Irrigation
practices of farmers for potato production in the eastern highlands
of Ethiopia”.
Capitalizing a title
[376]
There are two ways to capitalize a title of a research report
correctly. These are “title case” and “sentence case”. In the “title
case”, a researcher should capitalize all major words, which are
all words except articles (a, an, the), prepositions
(e.g., on, in, of, at), and coordinating conjunctions (and, or, but,
and nor; also for, yet, and so when used as conjunctions).
Example: Effect of Fertilizer Application on the Yield of Selected
Barley Varieties in Central Ethiopia.
In the “sentence style”, a researcher should capitalize the first
letter of the first word of the title of his/her research report, and
then have all the other words in lower case. However, if the word
is a proper noun wherever it appears in the title, it should have an
initial capital letter. Example: Effect of fertilizer application on
the yield of selected barley varieties in central Ethiopia. However,
some journals and other publishing organizations may have their
own rule that prescribes writing all words of titles in capital
letters. In such cases authors must abide by the rules prescribed.
Types of title
There are four types of title. These include nominal (descriptive
or neutral title) title, e.g. “Effect of phosphorus application on
yield and yield components of teff in Central Highlands of
Ethiopia”; Compound (hanging) title, e.g. “Soil Fertility status in
smallholder farming system: a case study in Eastern Ethiopia”,
and question title, e.g., “Does balanced application of nitrogen,
phosphorus, and potassium risk the quality of malt barley?”;
declarative title.
In the declarative title, the main finding of the study itself is
described; it is believed to reduce the curiosity of the reader.
Declarative titles generally use past tense when describing the
results of a single investigation (‘Feeding cows on leaves of
[377]
legume trees increased protein content by 50%’) and present tense
when stating a universally accepted finding (‘Feeding cows on
leaves of legume trees increases protein content by 50%’).
Abstract
Abstract is a short form of a research report designed to give the
reader a complete, yet concise, understanding of a research
finding. Abstract is a mini-version of the research report (usually
a single paragraph) in that it is self-contained and concise,
explaining the whole work as briefly and clearly as possible. An
effective abstract will contain several key features including
problem statement, methods/procedure/approach, principal
results, and conclusions/implications without abbreviations (with
concise sentences using active voice).
The fact that most readers look at the abstract part (and only this
part) of a report makes it the most important section. Obviously,
the abstract must always be written last after the other sections
because it should be a stand-alone piece representing the whole
report with a clear message.
Generally, there are three different types of abstracts. These
include descriptive, informative and structured abstracts. First,
descriptive abstracts are usually used in the social sciences and
humanities, do not give specific information about methods and
results but may cover background, purpose, focus and an over
view of content. Second, informative abstracts are commonly
used in the sciences and present information on the background,
aim, methods, results, and conclusions. Third, structured abstracts
are essentially informative abstracts divided into a series of
headings (e.g., introduction, materials and methods, results and
discussion, and conclusion).
[378]
Keywords
Beneath the abstract, a limited number of keywords (mostly a
maximum of seven, but sometimes five to six) need to be listed as
a requirement by scientific or research journals. Keywords are
words that capture the essence of the paper, make the paper
searchable and ensure that the researcher gets more readers and
citations. This is because most electronic search engines,
databases, or journal websites will use these keywords to decide
whether and when to display your paper to interested readers.
Therefore, author(s) should include the most relevant keywords
that will help other authors find their paper – words help the
reader retrieve the article, words and phrases that suggest what
the topic is about, words and phrases that are closely related to the
topic, etc.
Keywords may include (when applicable) species, treatments and
the major response criteria. When listing keywords, words
appearing in the title should not be repeated as keywords because
such words are already considered keywords by search engines.
Common words such as plants, soils, potato, models, place,
people, etc., are too general to be of any value as keywords but
scientific names of organisms may help more. Thus, it is
recommended to list binomial names (scientific names of
organisms) here rather than in the title.
Introduction
Introduction is a part where you introduce the topic, describe the
background, establish research problem, specify your objective
and map out the paper. Introduction is part of the paper where you
convince the audience that your research result presented is

[379]
important, valid, and emphasis on why the study is needed or it
should set the scene of the study by presenting reasons for
conducting the research. It should clearly define the research
problem and what information, knowledge and technology are
ideally needed to overcome the problem, what is available from
previous research, and what are the areas understudied or gaps in
information, knowledge and technology gaps to be bridged as a
rationale for further research. The introductory part of a report
should refer to reliable literature as of the magnitude and causes
and consequences of the research problem, and its possible
solutions by including enough evidence to establish validity of the
expected outcome.
The scene of the study should be made clear to attract the reader's
attention in terms of what information is ideally required, what
information is already available, what information is missing,
what different opinion exist about the topic and which gaps are
going to be bridged by a particular study. Contents of an
introduction may include a brief introduction of study subjects, a
brief introduction of subject matter to be studied, the scene of the
study and its objectives.
The introduction should generally move from the broad big
picture that is well known then narrowing to the specific
knowledge gaps that are unknown and then finally end in research
questions which are important, specific and unresolved in a
“funnel-like organization”.
Materials and methods
This section explains clearly how the study was done by
chronologically describing the process undertaken to complete the
research. The materials and methods part is important in many
[380]
aspects particularly to ensure the credibility of the research,
ensure the reproducibility of the methods so that other research
may repeat the same research, and guide the readers on how the
research is conducted and organized. This section of your report
should be written in past tense as what was done in the past is
narrated following the research proposal.
The experimental materials used in the study such as subject
plants and animals (with scientific names), tools and equipment,
reagents and chemicals should be properly presented and
described in detail. It is also necessary to explain the methods or
how the experiment was conducted to ensure you meet the
research objectives. The methods constitute all the steps and
procedures followed in order to achieve the research objectives,
including the experimental design and data analysis with
reference to the original source. When well established methods
are used, the detailed description of each and every step may not
be needed at least for the sake of space economy but when new
methods that have not been cited anywhere are used, then the
details should be given.
Statistical analysis led to the valid conclusions of the data, clearly
showing the magnitudes, trends, patterns, models, and
relationships using quantitative data, should be presented in the
report as detailed in the initial proposal.
Results and discussion
The function of the results section is to objectively present key
results, without interpretation, in an orderly and logical sequence
using both text and illustrative materials like figures and tables.
The use of visual aids is believed to help the audience to better
conceptualize and understand the results. The results sections
[381]
qualitatively or quantitatively presents what was found from the
study. Results should better be presented in the same order as the
list of objectives in the introduction or they should follow orders
of headings and subheadings in the materials and methods.
Results should be stated in simple and clear language focusing on
the representative data based on statistical parameters like the
mean, standard error, standard deviation, etc. in a way that avoid
repetitiveness. Narrations should not entirely repeat the visual
aids like table and figure. Only data that related to subject of the
study as stated in the introduction should be presented and
properly referred to the visual aids in the text by number. Include
only tables (numerical values or text displayed in rows and
columns) and figures (any type of illustrations like chart, graph,
photograph, drawing maps, etc.) that are necessary, clear and
worth reproducing. Use past tense for results and present tense
when referring to visual aids.
The discussion section offers your interpretations and conclusions
about your findings. Discussion shows how do your results relate
to the objectives of the study, and how do they relate to the results
that might have been expected from background information and
motivation stated in the introduction as supported by evidence
based synthesis and analysis, evidence based evaluations,
interpretations and reasoning. The discussion should also explain
key limitations, discuss agreement or contrast with previously
published works.
Discussion, like any story, has also a beginning, middle and the
end. The beginning states answer and summarize main findings,
whereas the middle explains answer and how it fits in with
previous works and, finally, the end gives final conclusions in
[382]
terms of the significance of the study and need for follow up
endeavors.
Past tense should be used to summarize findings, with present
tense to interpret results and explain significance of the results.
The results and discussion part should be presented in the same
order as the methods. Likewise, the discussion section should be
clearly linked with the results.
In regard to the use of visual aids, tables usually show numerical
value or textual information and are almost always characterized
by a row-column structure (any type of illustration other than a
table is referred to as a figure). Tables should be numbered
sequentially and each table must be referred to in the text. Caption
title should be placed directly above the table itself and below or
next to the table number. Caption title shows what the table is
about. Note that footnotes (if any) should be placed directly
below the table to explain abbreviations, symbols, etc. Do not
forget to acknowledge the source of the table if it is not your own
original data.
A figure may be a chart, a graph, a photograph, a drawing, or any
other illustration or non-textual depiction. Capital title should be
place directly below the figure itself and below or next to the
figure number. A caption is a concise explanation of the figure;
i.e. a brief but descriptive phrase includes copyright information.
Use a capital F for Figure and sequential numbering (if you have
more than one Figure). When you use a figure that has been
adapted or copied directly from another source, you need to refer
to the original source. This reference appears as a caption
underneath the figure. Put a legend which explains the symbols
used in the figure. It should have the same kind and proportion of
[383]
lettering that appears in the rest of the figure. Place the legend
within the figure (it may already be there if you have copied the
graph from elsewhere). Ensure image clarity and label all
important parts of the figure. As a rule, number all figures with
Arabic numerals in the order in which they are first mentioned in
text, regardless of whether a more detailed discussion of the
figure occurs later in the paper.
With regards to the use of different types of figures, it is advisable
to use bar charts when discrete (not continuous) items are
compared; line graphs to show trends and relationships between
two factors; pie chart to show proportion of the whole taken by
various parts; and photographs and drawings to illustrate what is
being presented in the text.
Conclusions and recommendations
Conclusions are simply the interpretations given to the research
findings by the investigator whereas recommendations are the
follow up actions to be taken as suggested by the investigator
based on the conclusions. The conclusion section presents the
outcome of the work by interpreting the findings at a higher level
of consolidation than the discussion and by relating these findings
to the motivation stated in the introduction. Conclusions, like
discussion, restate the answer and then indicate the importance of
the study by stating implications, applications and suggested areas
of future research. Like discussion, a combination of tenses could
be used, i.e., past tense to summarize findings and present tense to
suggest the need for follow up studies.
It should be noted that good recommendations and conclusions do
not emanate from the researcher’s own opinion but they should be
based on logical analysis of the findings of the study. A good
[384]
conclusion avoids the audience from asking “so what”? It is
written focusing on the innovative findings as a home-take-away
message that the audience should retain after reading the report or
the paper.
Acknowledgements
Those who helped the author(s), granting funds, a laboratory or
institutions that supplied materials or services, and a person who,
one way or another, technically contributed in the process of
research and research publication, etc. should be acknowledged.
The writer should specifically mention the contribution(s) that
each organization or individual made to the research in the
acknowledgement.
References
The author should include a reference for all the sources of
information that are being used when writing. When you use
another person’s work in your own work, either by referring to
their ideas or by including a direct quotation, you must
acknowledge this in the text of your work. References are
important for a number of reasons including: supporting the
opinion of the author, adding credibility to the information
presented by the author, enabling the reader to trace the sources
referred, and avoiding possible accusation of plagiarism.

21.1. Styles of Scientific Writing

There are different styles of scientific writing. However, all
scientific writing must fulfil three principles. These are accuracy,
brevity, and clarity. Brevity is the quality of being brief.

[385]
Thus, accurate and shorter or brief writing tends to be clear,
which readers can understand without a struggle. Therefore,
writers should write with reader orientation, not self-orientation.
This means they should write with a concern that readers can
understand their writing unambiguously.

It is possible to identify some common text-bulging habits. There

are common and easy opportunities to improve brevity and
clarity. In this connection, in constructing sentences for scientific
writing, it is important to consider the following points: use of
simple words (avoid verbosity), use short sentences; use
commonly active voice instead of passive voice. This is because
an active sentence is clearer and the responsibility of the actor is
more clearly defined. Avoid using words and phrases that are
considered as jargon (i.e. terminologies from other fields that may
be difficult for readers to understand); avoid clichés (i.e.
expressions that are heavily overused, such as sayings and
exaggerations; also avoid use of gender-biased language such as
mankind instead of humankind and manpower instead of human
power. It is also important to avoid nominalizations. This means
that turning verbs into nouns makes writing both weaker and
longer. Thus, you can reduce length by not writing nouns. For
example, instead of writing “We did the measurement of plant
height using a ruler”, write, “We measured plant height using a
ruler”, which is shorter and clearer.

In general, there are two major approaches to achieve brevity and

clarity in writing: (1). Reduce content: if a section, a sentence, a
detail, a graphic, or a dataset is not necessary to the story you are
telling, it is good to take it out; (2) Reduce text: it is actually
better to count characters than words. This is because replacing
[386]
long words with short ones is just as helpful to the reader as
reducing word count (Eg. replace bring about with cause; utilize
with use; put forward with propose; consume with eat;
demonstrate with show; attempt with try, etc). On the whole,
every character of your text should justify its existence.

Paragraph construction is another important aspect in scientific

writing. A Paragraph, by definition, is a series of sentences that
are all related to a single topic organized in coherent manner.
When a paragraph is constructed in scientific writing, one should
not put different main ideas in a single paragraph and similar
ideas in different paragraphs unless otherwise justifiable. Each
paragraph needs to have a topic sentence or a single controlling
idea, which unifies the content and gives perspective to the
paragraph; define what the paragraph will argue for. It is often
good to put the topic sentence at the very beginning of the
paragraph. In some cases, however, it’s more effective to place
another sentence before the topic sentence—for example, a
sentence linking the current paragraph to the previous one, or one
providing background information. Note that it is not a must to
have topic sentence in a paragraph. For example, you might be
able to omit a topic sentence in a paragraph that narrates a series
of events like content of a manuscript.
Most paragraphs narrating, describing, comparing, contrasting, or
analyzing information have a three-part structure. Introduction:
the first section of a paragraph; should include the topic sentence
and any other sentences at the beginning of the paragraph that
give background information or provide a transition.

[387]
Body: follows the introduction; discusses the controlling idea,
using facts, arguments, analysis, examples, and other information.
Conclusion: This final section summarizes the connections
between the information discussed in the body of the paragraph
and the paragraph’s controlling idea.
Another important issue is paragraph coherence. In a coherent
paragraph, each sentence relates clearly to the topic sentence or
controlling idea. If a paragraph is coherent, each sentence flows
smoothly into the next without obvious shifts or jumps. Use
transition words or phrases between sentences and between
paragraphs to maintain coherence.

21.2. Writing some units of measurements and

punctuation
Use internationally accepted signs and symbols for units (SI
units). Do not use quintal (q), mile, etc. for example). Otherwise,
give equivalent SI units for traditional non-SI units (e.g., quintal,
which means 100 kg) such as kg or ton, etc.

All common names of organisms should be qualified at the first

mention by the full Latin name and authority. All pesticides
should be referred to by their common names; and genus and
species names should be italicized.

Writing numbers
Spell out single-digit whole numbers (one to nine) in text except
when followed by units of measurement and except when they are
numerals of 10 and above. The same holds true when using ‘‘-
fold’’ to describe a magnitude of response, e.g. sevenfold (one
word), not 7-fold; but 19-fold (not ‘‘nineteen-fold’’). However, in
[388]
a mixed series containing some numbers more than 10 and others
less, authors should use numerals throughout: 2 goats, 5 horses,
and 25 cows. Use a word at the beginning of a sentence, not a
numeral. Thus, do not start a sentence with ‘‘50% of the
seedlings’’; Write ‘‘Fifty percent of the seedlings’’…… In
addition, use numerals when units of measurements are included:
2 g (not two g), 5 m (not five m), etc. Write 12 333, instead of
twelve thousand-three-hundred-and-thirty-three. If at the
beginning of a sentence, fractions of numbers such as one-
1 1 2
fourth(4); one-half(2); two-thirds(3), etc should be expressed in
words. When expressed with decimal points (0.25, 0.50, 0.67),
these fractions should be treated just like any other number in
tables and text. In the case of large numbers, group five or more
digits in blocks of three, divided by single spaces or by using a
comma.

Examples: write 5000; 40 000; 70 000 000 with a space only or

with a comma 5000; 40,000; 70,000,000. Thus, five or more
digits should be written with a space or a comma before the last
three digits: e.g., 54 662 or 54,662; 432 765 or 432,765; 4 357
869 or 4,357,869, etc. Apply this consistently before the decimal
marker (in case a decimal number is used). Example, write 54
321.12345 or 54,321.12345.

Write four and less digit numbers without a space and a comma,
unless they form part of tabulation. Example, in text, four-digit
numbers (1000 to 9999) should be written without any space,
comma, or period (1267, 3492, etc.). However, in a table or
column with various values, four-or more digit numbers should be
written with a space or a comma before the last three digits.

[389]
Example, 3 000; 3 000 000; 30 000 000, etc.
Or 3,000; 3,000,000; 30,000,000, etc.
In a text, use a word to replace several zeros. Thus write 30
million instead of 30 000 000. Similarly, units should be adjusted
for numbers less than one to avoid a string of zeros after the
decimal point (25 mg instead of 0.025 g or 0.000 025 kg).

Writing symbolic punctuation marks

Dashes
In writing, there are three different symbol punctuation marks that
most people refer to as dashes. These are the hyphen (-), the en
dash (–), and the em dash (—).

Hyphen
The hyphen closely resembles the two other types of dashes.
However, these three punctuation marks are not used
interchangeably.
A hyphen resembles a small horizontal line (-). It is used to form
compound words.
Examples:
Use a hyphen to join two or more words together to form a
compound word. Examples, a two-week study, large-sized fruit,
Sister-in-law, left-handed, thirty-two, Editor-in-chief, eye-
opening (event, for example) or eye-opener, sub-Saharan etc and
in prefixes such as pre-Industrial era, anti-Progress, pro-Western,
etc
Use a hyphen after the unit when it is an adjective. Example, it is
a 12-month study. We had an agroforestry-based coffee
[390]
Production system. However, the system is agroforestry based (no
hyphen required here).

However, most (but not all) compound words that begin

with self use a hyphen, regardless what part of speech they are.
For example, self-esteem, self-pity, self-reflection, self-serving,
self-centred.

Hyphens are also used for writing words that state fractions of
numbers and compound numbers. Sentence examples, to bake a
cake, you need one-fifth of a tin of milk. Hyphens are also used
when writing out the numbers 21 through 99: twenty-five, forty-
seven, sixty-one, eighty-eight, fifty-two, thirty-three, one hundred
twenty-five, etc.

Hyphens are also used to divide words on line breaks. In certain

written works, such as books, newspapers or magazine articles, a
hyphen is often used to divide a word across two lines for
formatting reasons. However, a word is split after a complete
syllable. For example, we would write driv-er rather than d-river.

En dash (–)
The en dash is approximately the length of the letter n. The en
dash (–) is used to mark ranges (e.g Pages 23–29) or results are
discussed on pages 45–53; It is also used to express units in
exponential form (e.g. 50 kg ha–1) or as a minus sign 30 – 10 =
20. However, when the hyphen is used as a minus sign, there
should be space between the symbol and the numbers on both
sides. The en dash is also used with the meaning “to” in phrases
like “Addis Ababa–Adama Express Road.” Or the Addis Ababa–

[391]
Dire Dawa flight ... ... such as a Newton–Raphson algorithm ... ...
the DNA–enzyme interactions, etc.
However, most writers inappropriately us the hyphen to express
ranges, for example 56-60, or to express exponential units of
measurement such as kg ha-1 or as a minus sign as 30-20=10.
Use an en dash for such purposes. Use also an en dash to signify a
minus sign in a mathematical statement. In this case, use the en
dash (–) with a space between the en dash and each of the
numbers on both sides joined. Example, write Height = 30.7 m –
7.2 m = 23.5 m or Height = (30.7 – 7.2) m = 23.5 m.

Tus, use an en dash (–), not a hyphen (-), between two numbers in
a table or a figure, which usually indicates a range (i.e. 12–16
means 12 to 16). However, in a text, write as 12 to 16. A
negative exponential should be used instead of a slash (/) as in
a/b. Similarly, write values as 100 kg ha–1 (instead of 100 kg/ha)
or 20 ml L–1 (instead of 20 ml/L), 10 g m–2 (instead of 10 g/m2),
etc. Not that the minus negative symbol is not written as a
hyphen, but it should be en dash. While writing probability level
(P) with the ‘‘<”, “>’’, and “=” symbols, give space after those
symbols (P < 0.05 level, P > 0.05 and P = 0.05). There are some
SI units that are extensively used in science such as ‘‘y’’ or ‘‘yr’’
(for year; SI unit: ‘‘a’’ for annum), and (for ton; SI unit” t”: Mg
for megagram). It is common to use these non-SI units or the
corresponding SI units. Give equivalent SI units for traditional
non-SI units (e.g., quintal, which means 100 kg) such as kg or
ton, etc. Use parentheses
( ) as one way of referring to tables, figures and probability levels.
Example, application of the fertilizer had a significant (P < 0.05)

[392]
effect on the yield of the crop (Table 3). To avoid having round
brackets one within the other, use square bracket [ ]. The
numbers [six hundred (600) units] have been indicated in the
Table.

Em dash (—).
The longer em dash (—) is used to separate extra information or
mark a break in a sentence. The em dash is often used in places of
colon (), semi-colon (;), comma (,), and brackets. But, this use is
mostly done for stylistic purposes in literature other than
scientific literature. Examples, I have two excellent books — by
the way it was my brother who bought those books for me —
which I will read in preparation for my exams.
The en dash is sometimes also used in the same way as an em
dash; in this case, it takes a space on either side. Examples, I have
two excellent books – by the way it was my brother who bought
those books for me – which I will read in preparation for my
exams.

Both the em dash and the en dash can be inserted from symbol
menu of your personal computer. However, the hyphen can be
written directly by typing the hyphen key. It is also possible to
write the en dash by pressing the control key and then typing the
hyphen key.

[393]
 SECTION VII. RESEARCH ETHICS
22. Research Ethics
The term ethics was originated from a Greek word ethos, which
means character or custom or convention, or the spirit of
community. Ethics is defined as moral principles that govern a
person’s behavior or his/her conduct of activity or it is the branch
of philosophy that deals with morality.
Research ethics is the application of moral rules and professional
codes of conduct to the collection, analysis, reporting, and
publication of information about research subjects, in particular
active acceptance of subjects' right to privacy, confidentiality, and
informed consent. In short, research ethics is doing what is
morally and legally right in research.

Intellectual ethics that ensures honesty and truthfulness is vital to

prevent research misconduct. Any document that a researcher
prepares, presents, and publishes should be unique and credible.
Thus, researchers are “free” to conduct research they find
interesting, pursue new ideas and criticize old ones but
“intellectual freedom” is bounded by ethical standards. A
researcher should be capable to deliver, committed or dedicated to
serve the society, disciplined and respect research ethical values
during planning, execution and publication of research results.
One of such standards is that there is no purpose in investigations
that do not one way or another serve society.

[394]
22.1. General ethical principles that guide research
The general ethical principles that guide research include mainly
the following:
Honesty
Honesty is being good and truthful, which included not lying,
stealing, or cheating; not hiding the truth about someone or
something, not meant to deceive someone, showing or suggesting
a good and truthful character. Therefore, researchers should be
honest in reporting data, results, methods, and procedures, and
publication status. Researchers should not fabricate, falsify, or
misrepresent data. They should not deceive colleagues, research
sponsors, or the public.

Objectivity
Objectivity is striving to avoid bias or prejudices or subjective
decisions or evaluations in experimental design, data analysis,
data interpretation, peer review, personnel decisions, grant
writing, expert testimony, and other aspects of research where
objectivity is expected or required. In this connection, researchers
should avoid or minimize bias or self-deception. They should
disclose personal or financial interests that may affect research.
Integrity
Integrity is strict adherence to a moral code reflected in
transparent honesty and complete harmony in what one thinks,
does, says, and does. (no double standard).
Researchers should keep their promise and agreements; they
should act with sincerity, strive for consistency of thought and
action.

[395]
Carefulness
Carefulness is the responsibility to avoid careless errors and
negligence, carefully and critically examining one’s own work
and the work of one’s peers. It also implies that researchers
should keep good records of research activities, such as data
collection, research design, and correspondence with agencies or
journals.

Openness
Openness in the text of research ethics is sharing data, results,
ideas, tools, resources. It is also being open to criticism and new
ideas.
Transparency
Transparency in the contact of research ethics refers to the
responsibility of disclosing methods, materials, assumptions,
analyses, and other information needed to evaluate your research.

Accountability
Accountability in research ethics refers to the responsibility for
one’s part in research and being prepared to give an account (i.e.
an explanation or justification) of what one did on a research
project and why. It is taking responsibility for one’s actions,
decisions and their consequences.

Intellectual property
Intellectual property in the context of research ethics refers to
honouring patents, copyrights, trademarks, and other forms of
intellectual property. Researchers should not use unpublished
[396]
data, methods, or results without permission. They should give
proper acknowledgement or credit for all contributions to
research. Researchers should never plagiarize.

Confidentiality
Confidentiality refers to the requirement that information supplied
to you in confidence will not be disclosed directly to third parties.
It is the responsibility of pprotecting confidential
communications, such as papers or grants submitted for
publication, personnel records, trade secrets, patient records, etc.
Anonymity
Anonymity in the context of research ethics refers to taking steps
to ensure that what participants have said cannot be traced back to
them when the final report is produced. It also refers to the
responsibility of going to categories and store the information. It
also refers to the responsibility of how you are going to make sure
that the information is not easily accessible to anyone with
unscrupulous intentions.

Informed consent
This refers to the responsibility of the researcher that he/she must
inform participants of the research in advance in such a way that
the participants understand the various issues involved and that
might affect their decision to participate in an experiment. The
researcher should get informed consent from the participants
including, but not limited to, the purpose of the research, its
duration, potential risks and potential benefits to participants as
well as the probability of such risks and benefits.

[397]
Responsible publication
This refers to publishing in order to advance research and
scholarship, not to advance just your own career. It is also
avoiding wasteful and duplicative publication such as publishing
in predatory (non-peer-reviewed) journals.

Responsible mentoring
This refers to helping to educate, mentor, and advise colleagues,
students, etc. It also includes promoting the welfare mentees
under one’s supervision and allowing them to make their own
decisions.

Respect for colleagues

This refers to respecting your colleagues and treating them fairly.

Social responsibility
This means that researchers should conduct research not only to
advance their own career but also for the good of society as a
whole. Therefore, researchers should promote social good and
prevent or mitigate social or environmental harms through
research, public education, and advocacy.

Non-discrimination
This refers to avoiding discrimination against colleagues or
students on the basis of sex, ethnicity, clan, race, religion,
nationality, place of origin, or other factors not related to
scientific competence and integrity.

[398]
Competence
This refers to maintaining and improving your own professional
competence and expertise through lifelong education and
learning; it is important that researchers take steps to promote
competence in science as a whole.

Legality
This refers to knowing and obeying relevant laws and institutional
and governmental policies.

Animal Care
This refers to the responsibility of showing proper respect and
caring for animals when using them in research. Researchers
should not conduct unnecessary or poorly designed animal
experiments.

Protecting human subjects

When conducting research on human subjects, it is vital that
researchers minimize harms and risks and maximize benefits. It is
vital that they respect human dignity, privacy, and autonomy.
Researchers should also take special precautions with vulnerable
populations. They should also strive to distribute the benefits and
burdens of research fairly. If a researcher discovers a cure for a
particular disease, through his search on indigenous plants, a
reasonable part of the patent should go to the indigenous
community where the plants are located.

[399]
Conflict of interest
This refers to the responsibility for avoiding relationships that
could reasonably impair your professional performance or could
exploit or harm others.

Therefore, researchers should adhere to the abovementioned

principles of research ethics when conducting research as well as
when communication research results.
For this purpose, data collection and analysis should be done
honestly. Collect the data as accurately and carefully as possible
without any subjective bias on the part of the researchers.
Deliberate fraud and malpractices are condemnable. Fraud
involves deliberate deception and may take the form of
concoction of data or manipulation of data. A monitoring
mechanism should be there in the institution to oversee the works
of scientists and research students and to detect fraud and
malpractices.

Thus, any data generated through research, and documents

emanating thereof must be unique and credible. The contribution
of any written research paper or book or proceedings or research
report to science basically depends one meeting the required
quality standards of reporting by adhering to widely accepted
ethical norms by the writer. Manipulating, fabricating or
falsifying data, experiments, the research process, equipment,
procedures, observations, findings etc., dishonesty in authorship,
irresponsible publication, and plagiarism are the most common
unethical practices that often violate academic integrity and lead
to research misconduct. In addition, publishing in “fake” or

[400]
“predatory “or dubiously peer-reviewed journals or publishers has
also recently emerged as a research misconduct.

22.2. Research misconducts

The opposite of research ethics is research misconduct. Research

misconduct refers to any research work done in disregard to the
required research procedures and ethical principles. These are
practices that seriously deviate from those that are commonly
accepted within the scientific community for proposing,
conducting, or reporting research.

As the names imply, data fabrication, data falsification, and data

or process manipulation are research misconducts that often occur
in reporting research findings. These misconducts are extreme
forms of dishonesty and lack of disregard for doing research with
integrity and responsibility.

Research misconducts made by researchers mainly include

fabrication, falsification, plagiarism. They also include the
following especially in proposing, performing, and reviewing
research, or in reporting research result: publishing in “fake” or
“predatory” journals. Research misconduct also include conflict
of interest that may arise when a researcher has other goals
besides answering the study question, he or she has a conflict of
interest. Research misconduct also may occur in authorship.
In general, ethical issues in agricultural research misconduct are
mainly related to manipulating, fabricating, and falsifying data,
experiments, processes, interpretations, conclusions, etc. result in

[401]
issues of research misconduct, authorship, plagiarism, peer
review, publication in “fake” or predatory journals, partial or
duplicate publication and self-citation, etc.
22.2.1. Irresponsible publication
Irresponsible publication refers to the practice of submitting
results of the same study to more than one journal or publishing
more or less the same material in more than one journal.
Irresponsible publication also includes what is called “redundant
publication” (also described as ‘salami publishing’), which a
situation in which the result of one study is split into several parts
and submitted to two or more journals. Irresponsible publication
also included publishing in “fake” or “predatory” journals or
publishers, which has recently emerged as research misconduct
22.2.2. Misconduct in authorship
Authorship is the process of deciding whose names should be
listed on a research paper. All of the following three criteria
should be fulfilled to qualify as an author:

 Substantial contribution to the research

 Writing or critically revising all or part of the manuscript
 Approving the final version of the entire article
Authorship misconduct includes deliberately changing author
name order, deliberately removing names of authors, and
deliberately including names of authors that have not contributed
to the paper. Author name order is decided based on level of
contribution. Thus, the name of the person that made the most
contribution is mentioned first, followed by the one that made the
next most contribution, etc. If contributions are equal, names of
authors have to be mentioned alphabetically and it is better to
[402]
state that “all authors have equal contribution”. Three types of
authorship constitute publication misconduct. These are:

• "Ghost" authors, who contribute substantially but are not

acknowledged (often paid by commercial sponsors);
• "Guest" authors, who make no discernible contributions,
but are listed to help increase the chances of publication
(for lobbying the publication);
• "Gift" authors, whose contribution is based solely on a
tenuous (weak) affiliation with a study.
22.2.3. Peer review related ethical issues
Peer review is the process in which an author submits a written
manuscript to a journal for publication and the journal editor (as
leader of the peer review process) distributes the manuscript to
the experts (experts as reviewers); or it is the professional steps
from submission to publication of a manuscript. The ethical issues
in the peer review process are: first, confidentiality of the review
process, i.e. no person involved can disclose the information in
the manuscript. Secondly, protection of intellectual property right
associated with the manuscript. Confidentiality in the review
process is usually maintained with what is called “double-
blinding” or “Single-blinding”. Double blinding refers to a
situation where both the reviewer and the author do not identity
each other throughout the review process. Single-blinding is a
method of peer review where the authors do not know who the
reviewers are but the reviewers know who the authors are. To
facilitate this, authors may need to ensure that their manuscripts
are prepared in a way that does not give away to their identity.

[403]
22.2.4. Fabrication and falsification
Fabrication is faking up data or results and recording or reporting
them. Falsification is manipulating research material, equipment
or process or changing, omitting data such that research is not
accurately represented in research record.
22.2.5. Plagiarism
Plagiarism is an inappropriate act of claiming credit or ownership
of someone else’s ideas or unfair annexation of another’s work
and entitling it as your own. Plagiarism maybe a copypaste type,
paraphrasing type, translated plagiarism, artistic plagiarism, idea
plagiarism, style plagiarism, code plagiarism or misinformation of
references.
Several types of plagiarism are known in research. These include
the following:

Verbatim (word for word) or direct plagiarism

This form of plagiarism involves simply “copy and pasting”
someone’s exact words without changing a single thing. The
writer copies every single word from the source without use of
quotation mark, citation, acknowledgment, or attribution.
Paraphrasing plagiarism
This kind of plagiarism occurs when a person paraphrases a
source or multiple sources without crediting the original
author(s). In paraphrasing plagiarism, the wording or word order
of the original source may be changed—but the plagiarized text
may retain too many similarities to the original. This form
of plagiarism is especially common among researchers and
students.

[404]
Patchwork or mosaic plagiarism
Patchwork or mosaic plagiarism involves interweaving
paraphrasing of sources with some of the author’s own ideas or
arguments. Because the stolen and original material is mixed
together, it can be more difficult for a reader to recognize which
words or ideas have been copied from which sources.
Patchwork or mosaic plagiarism may also occur when an author
writer writes by fetching contents by copying from several
different sources and then modifies and stitches the relevant
contents from various documents, books, online sources, web
pages and repositories to make them fit together.
Collusion or sharing plagiarism
This kind of plagiarism involves two or more people working
together to pass off one person’s work as another’s, hence the
implication of collusion (“a secret agreement, especially for
fraudulent or treacherous purposes”). This type of plagiarism can
be common in schools or colleges, where students will share
essays or complete each other’s assignments. Even though a
person may have another’s consent to use their work, it is still
considered plagiarism because a person is trying to deceive
someone (their teacher or professor) by claiming another’s work
as one’s own.
Paid or hired plagiarism
Similar to collusion plagiarism, paid plagiarism involves paying
someone to create something and then passing it off as one’s own
work. Hired plagiarism can involve paying or hiring another
researcher or a freelance writer to complete the write up of a
research paper. Again, this type of plagiarism is becoming

[405]
common in schools or colleges where students pay money to get
cooked reports for submission as their own.
Improper citation and accidental plagiarism
Depending on the organization, poor or incorrect citations may be
treated the same as other types of plagiarism. e.g., Citing Gomez
and Gomez (1984) instead of Fisher (1890–1962) as the
originator of the ideas for ANOVA. Thus, this type of plagiarism
is more likely to be the case when it involves major errors such as
unattributed quotations.
Self-plagiarism or self-citation
Self-plagiarism is the use of one’s own previous work in another
new context without citing that it was used previously. Writing up
and submitting a manuscript for peer-review that has used the
same data or drawn the same conclusions as work that one has
already presented either in a previous publication or as a podium
or poster presentation at a conference also constitute self-
plagiarism.
Reusing the content of a literature review performed for
one dissertation or thesis directly within another dissertation or
thesis without referencing back to this is also self-plagiarism.
Readers should know that this was not the first use of the
material. Therefore, republished text work is considered as self-
plagiarism and the credit received from previous work is
considered as corrupt.
Related to self-plagiarism is self-citation. Self-citation is defined
as occurrences in which the citing and cited papers share at least
one author in common or as excessive reliance on research papers
from the same research lab or center.

[406]
Even if excessive self-citation can be considered rather unethical
and as a research misconduct, particularly unethical when
researchers cite their own paper at the expense of original
sources, a moderate amount of self-citation maybe inevitable
under certain conditions.
Sometimes self-citation maybe an indicative of a cohesive
research program in which authors refer to their previous relevant
works in order to enhance their subsequent contributions to
knowledge, particularly if one is working in a highly specialized
field where there is simply little other research effort.
To minimize or control self-citation as a research misconduct,
there is a need for confirmation when one is applying for
scientific positions, promotions or awards based on report of
“citation factor” without self-citations; report of “citation factor”
with self-citations; and ratio of the two “citation factors”.
Methods of avoiding plagiarism
There are several methods of avoiding plagiarism. These include
the following:
Direct quoting
Using a quotation-someone's exact words-can have a dramatic
and powerful effect in your paper, for you are displaying verbatim
(word for word). Special benefits of direct quoting include getting
direct support for our arguments, preserving the original expert
declaration, effective language, historical flavour, specific
example, controversial statement, material for analysis.
The following are examples of quotations done to preserve the
original expert declaration

[407]
 a) Example 1: “There are two kinds of people, those who do
the work and those who take the credit. Try to be in the
first group, there is less competition there”, Ghandi.
b) Example 2: “Injustice anywhere is injustice everywhere. If
you do not speak against oppression, you will soon be the
oppressed. If you sit unconcerned while others are
trampled on, it will soon be your turn. Until everyone is
safe, no one is. Injustice against others is a warning to
you. Fight it before it knocks you down”, Martin Luther
King.
However, quoting sources should be done with prudence. It is
important that researchers avoid quoting too often because too
many quotations in a paper will push the researcher’s idea your
ideas into the background rather than supporting or
complementing the paper being written.
It is also important to avoid quoting one source too many times.
There is a rule of thumb that one source should not be quoted
more than three times per 2500 words. A four- to six-lines
quotation might be okay, but a quotation of a ten or more lines
may be boring for readers to continue reading. Too many quotes
in a scientific writing may also put the research work into the
background and may clutter the writing with quotations.
Therefore, it is not common to use quotes in writing research
reports.
Paraphrasing
Another way of avoiding plagiarism is to paraphrase the source.
A paraphrase is a restatement or rewriting of a source in order to
present the source's idea or meaning without actually quoting the
source's words. A paraphrase should convert a source's words into
[408]
about the same number of your own words. Thus, a writer should
keep the paraphrase to about the same number of words as the
original, not omitting any significant features of the source
material. In paraphrasing, it is important to preserve the author's
original meaning and avoid taking the ideas out of context. What
the author stressed as the main points of the passage should be
kept as the main points of the paraphrase. The author's
conclusions, point of view, and even attitude should be preserved.
However, technical words should be maintained. For examples,
plants, animals, height, weight, flower, colour, etc. However,
paraphrasing alone will not free a researcher or writer from
plagiarism unless he/she cites the source appropriately.
Summarizing
Summarizing is the third method of avoiding plagiarism and
incorporating a source into your own writing is to summarize it.
Summarizing possesses a flexibility that allows for highly
creative uses of source material.
While a paraphrase must contain about the same number of words
as the source, a summary can be created with many fewer or only
somewhat fewer words.
A summary is a condensed restatement or rewriting of a source in
order to present the source's idea in a more focused or shorter way
than quoting or paraphrasing would allow. In other words, a
summary reduces a source's words into fewer of your own words
and is used for simplifying the source and eliminating extras such
as digressions, or explanations and keeping only the essential
parts of the discussion or the main argument.

[409]
Use of plagiarism checkers
There are a number of plagiarism checker software developed by
different companies and freely availed too users. Plagiarism
checking with the use of these software is basically based on
uploading the paper to be tested or entering the content into the
text box; scanning for plagiarism (checking all internet pages and
databases housing many million academic works); and receiving
plagiarism report which contains information about sections of
the scan that have been plagiarized and a list of sources.
Likewise, anyone can also check content from your own article to
make sure nobody else has been ripping you off!
What to cite and what not to cite
One requirement to avoid plagiarism is that the source must be
cited. Any source of information refers to any producer of
observation, knowledge, or experience other than you. The source
needs citation. However, in order to cite a reference appropriately,
it is important to understand the difference between common
knowledge, statements, facts, and opinions.
Common knowledge includes whatever an educated person is
expected to know. It represents the kind of general information
found in many sources and remembered by many people.
Common knowledge does not need to be cited. Example of
common knowledge: “Plants manufacture their own food using
sunlight, water, and carbon dioxide”; “certain types of trees in the
tropics and sub-tropics shed their leaves in the dry season”;
“ruminant animals cud their food”, etc.
Concerning opinions, if an author formulates a positive, negative
or neutral opinion based on a report or some other research work,
it is required to give due credit to the sources based on the author
[410]
that articulated his/her opinion. Thus, it is imperative to
understand and distinguish between facts, lesser known facts and
opinions during the process of citation.
In a number of sources, common knowledge can be found mixed
with interpretation, facts, analysis, and opinion. All such
commentary on common knowledge should be cited. For
example, if a source states that “installing the irrigation system is
an expensive project, it does not need citation because it is
common knowledge. However, if the source uses the word
uneconomical instead or in addition to expensive, that is
interpretation. Therefore, it would need a citation. It is not always
easy to know what common knowledge is and what is not. If you
come across a fact that you think may be common knowledge, but
you are not sure about it, it is good to follow the rule of
uncertainty to be on the safe side. Thus, it is better to cite
unnecessarily than to neglect citing something that should have
been cited. If there is a large amount of common knowledge in a
source, it is easier to summarize a single source and cite it than to
assemble the many pieces from various references.
Differences between reference list, bibliography, and citation
Reference, bibliography, and citation are often used
interchangeably. However, this is not quite right. A reference is
the list of the sources explicitly used in completing the write-up
of a manuscript, book, etc. It includes detailed descriptions of the
author, journal/proceeding/book, publisher, publication year,
volume, issue no., and page number (or date of download in case
of multimedia sources).
Bibliography, on the other hand, is a list of all the sources that
were used in writing a manuscript or any other document but may
[411]
have not been directly used in the contents of the paper. In other
words, it is the list of all the references published/unpublished
which have provided background information and better
understanding of the topic/concept but are not explicitly
mentioned in the manuscript.
Citation is an act or practice of mentioning/attributing/indicating a
specific source or quote material in the body of your paper.
Harvard Citation system, MLA (Modern Language Association)
citation system, Vancouver citation style, and American
Psychological Association (APA) citation style are common
styles or formats of citation.
23.2.5. Publishing in “predatory” journals
Publishing in the so-called “Predatory” journals (non-peer-
reviewed journals) has recently been recognized as a major
research misconduct in the world as a whole.
Before the advent of the Internet, disseminating journals was
possible only in print forms. This method of disseminating
publications was slow and costlier. The Internet has simplified
scientific communication and created the opportunity of
disseminating publications through the open access (OA) model
faster and more efficiently. As a result, open access publication
has increased visibility of scientific journals and other published
material. The advent of Internet has also reduced costs of printing
and dissemination.
Publishing scientific journals in both open access and print
models incurs costs of editorial decision making, peer review,
copy editing, production, archiving, website maintenance, etc.
For this purpose, journals or publishers use several payment

[412]
models for maintaining scientific publications. These include
publisher Pay Model through advertising and society
memberships, reader pay model through journal subscriptions and
per article fee, and author pay model through submission or
publication fees.
A lot of genuine OA journals and publishers that use author pay
model have proliferated with the advent of the Internet. However,
this opportunity has also contributed to the rise of for-profit,
scam-like, publishing industry that exploits the Open Access
model and engage in what is described as “predatory” practices.
Predatory refers specifically to a practice in which open access
publications exploit the author-pay model by setting up bogus
publishing operations and tricking authors into thinking that they
are legitimate scholarly publishing outlets” (Bornemann, 2013, p.
13). Predatory publishers trick unsuspecting authors with a
promise of rapid publication to submit their manuscripts to their
journals in exchange for payment of publication fee.
The major concerns with predatory journals or publishers include
that they accept manuscripts and publish them with little or no
peer review or quality control (Bohannon, 2013), resulting in poor
quality publications. They usually notify academics of article fees
only after papers are accepted (Stratford, 2012).
In fact, merely charging a fee for publishing a paper is not a
marker of a predatory practice: many reputable OA journals use
article-processing charges (APCs) to cover costs. Many
subscription-based journals also charge authors fees, sometimes
per page or illustration. However, predatory journals are primarily
fee-collecting operations with no genuine rigorous peer review.

[413]
Currently, it is important to realize that there are a lot of online
predatory publishers that trick authors into submitting their hard-
won research results to their “predatory” journals. These journals
are fake and live only to make money from unsuspecting authors
by publishing manuscripts in exchange for publication fee without
subjecting the papers to any peer-review processes. Therefore,
researchers should remain vigilant and protect themselves from
such scam or deception and publish their manuscripts only in
genuinely peer-reviewed journals.
There are two general ways of addressing unethical open access
publishing (predatory publishing). The first method is
Blacklisting. The second one is Whitelisting. Usually, there is a
fuzzy line between low-quality and predatory publishers. Thus,
whitelisting, or listing publishers and journals that have been
vetted and verified as satisfying certain standards, may be a better
solution than blacklisting. The central player in the whitelisting
movement is the Directory of Open Access Journals (DOAJ).
In addition to the above methods, authors can use different other
strategies to identify high quality open access journals for
submitting their manuscripts for publications.
There are strategies to identify and avoid submitting ones hard-
won research data to predatory journals and identify and using
only genuinely peer-reviewed journals. The major strategy for
avoiding predatory publications and identifying genuinely peer-
reviewed journals is using the following major criteria: the
journal is indexed by at least one legitimate abstracting, indexing,
and database services/agents (e.g. SCOPUS, DOAJ, PubMED,
Springer, SCI, Medline, Web of Science, RePEc, ProQuest,
EBSCO, Thomson Reuters, Taylor & Francis, BIOSIS, Agricola,
[414]
ScienceDirect, or any other proven legitimate abstracting,
indexing, and database services/agents) or in journals that are
listed in one or more of the Journal Citation Report (JCR),
Directory of Open Access Journals (DOAJ), Open Access
Scholarly Publishers Association (OASPA), Committee on
Publication Ethics (COPE), etc.; the journal has a genuine peer-
review process, assigning at least two reviewers of relevant
competence that complete a blind peer-review process per article;
the journal has a transparent publication process: clear author
instructions, ethical guidelines, and clear policies on copyright;
the journal clearly indicates journal editor(s), editorial staff and
review/editorial board members, with full names, correspondence
addresses including their institutional affiliation, who directly
respond to author queries; the journal or publisher has "contact
us" page with clearly identified detailed contact addresses: name
of institution and person, telephone numbers, email, P. O. Box,
street address, and other relevant information for direct contact;
the journal has well maintained website, with no dead links;
The second strategy through which researchers can ensure peer-
reviewed publication is submitting their manuscripts to journals
that have institutional affiliations (Homebase) in any country such
as a university or a research institution or a recognized scholarly
society.
The third strategy to use especially for local (Ethiopian) journals
is submitting manuscripts to journals in the country (no matter
what their affiliation or Homebase is) that are regularly (usually
every three years) revaluated and accredited for the genuineness
of their peer review system by Ministry Education.

[415]
Such Journals are professionally based and have definite peer-
review process although the rigour of the peer review process
may vary from journal to journal or from institution to institution.

23. Bibliography

Abebe Kirub. 2014. Essentials of Scientific Writing. Ethiopian Institute

of Agricultural Research (EIAR), Addis Ababa, Ethiopia
Bairagi, V. and Munot, M.V. eds., 2019. Research methodology: A
practical and scientific approach. CRC Press, pp 303.
Beall J. 2015. Beall’s List: Potential, possible, or probable predatory
scholarly open access publishers. Scholarly Open Access,
[Link] (retrieved May 29, 2015).
Bohannon, J. 2013. “Who’s Afraid of Peer Review?” In: Science
342.6154 (2013), pp. 60–65. DOI: 10.1126/science
Cabbolet, M.J.T.F. 2014. Scientific misconduct: three forms that
directly harm others as the modus operandi of Mill’s tyranny of
the prevailing opinion. Science and Engineer Ethics 20 (1): 41–
54.
Chambers, R. 2001. Direct matrix ranking (DMR) in Kenya and West
Bengal. PLA Notes CD-ROM 1988–2001
([Link]
pdfs/migrate/[Link]).
Chandra, S. 2004. Design and Analysis of Participatory On-Farm Trials.
International Crops Research Institute for the Semi-Arid Tropics
Patancheru, India
Clewer, A.G. and Scarisbrick, D.H., 2001. Practical statistics and
experimental design for plant and crop science. John Wiley &
Sons, p329.
Costley, C. and Fulton, J. eds., 2018. Methodologies for practice
research: Approaches for professional doctorates. Sage, pp 253.
[416]
Fentahun Mengistu and Abebe Kirub. 2014. Code of Research Ethics.
Ethiopian Institute of Agricultural Research (EIAR), Addis
Ababa, Ethiopia.
Gomez, A.A. 1984. Statistical Procedures for Agricultural Research.
2nd Edition John Wiley and Sons Inc., New York
Gonzalez, W.J., 2020. Methodological Prospects for Scientific
Research. Springer International Publishing, pp 247.
[Link]
s_of_Covariance_ANCOVA_and_How_to_Correctly_Report_its
_Results_in_Medical_Research [accessed Mar 17 2022].
Kaps, M. and Lamberson, W. R. 2004. Biostatistics for Animal Science.
CABI Publishing, CAB International, Wallingford Oxfordshire
OX10 8DE UK.
Kothari, C.R., 2004. Research methodology: Methods and techniques.
New Age International, pp 401.
Little T.M. and Hills F.J. 1978. Agricultural Experimentation: Design
and analysis. John Wiley and Sons. New York.
Margot Wehrmeijer. 2014. “Exposing the Predators. Methods to Stop
Predatory Journals,” Leiden University, 2014. [Link]
[Link]/handle/1887/28943. 29Jennings, C. 1999. Citation
data: the wrong impact? Neuro endocrinology letters, 20(1–2), 7.
Mukherjee, S.P., 2019. A guide to research methodology: An overview
of research problems, tasks and methods. CRC Press, pp 240.
Pruzan, P., 2016. Research methodology: the aims, practices and ethics
of science. Springer, 325.
Robert A. H. 2002. Using Sources Effectively: Strengthening Your
Writing and Avoiding Plagiarism, Routledge, Taylor and Francis,
Fourth Edition, pp146.
Sahu, P.K., 2013. Research methodology: A guide for researchers in
agricultural science, social science and other related fields. New
Delhi, Springer. DOI 10.1007/978-81-322-1020-7, 432p.

[417]
Schneider, C. 2000. Safeguarding good scientific practice: new
institutional approaches in Germany. Science and engineering
ethics,6 (1), 49–56.
Sheikh, F. A., Khan, M. N., Sofi, P. A., Dar, Z. A., Sofi, N. R. and
Bhat, M. A. 2017. Farmers’ Preference Ranking in Pole type of
Common Bean (Phaseolus vulgaris L.): Participatory Varietal
Selection. Int. J. Pure App. Biosci. 5 (1): 703-711.
Shoja, M.M., Arynchyna, A., Loukas, M., D'Antoni, A.V., Buerger,
S.M., Karl, M. and Tubbs, R.S. eds., 2019. A Guide to the
Scientific Career: Virtues, Communication, Research, and
Academic Writing. John Wiley & Sons, pp742.
Thomas, C.G., 2021. Research methodology and scientific writing.
Thrissur: Springer., 2nd Edn, [Link]
64865-7, p611.
Wehrmeijer, M. 2014. Exposing the predators. Methods to stop
predatory journals Publishing.

[418]
[419]