Introduction to survey research

This chapter focuses on the key steps involved in conducting a survey, which is a common method
used in research. These steps are part of a larger process that includes planning, data collection,
and analysis.
1. Identify the Issue: Start by identifying the topic or problem you want to study. This could
involve reviewing existing literature and theories related to the subject.
2. Formulate Research Questions: Once you have a clear understanding of the topic, formulate
specific research questions that you want to answer. These questions guide your survey.
3. Decide on Survey Design: Determine if a survey is the best way to gather data for your
research. If not, consider alternative methods.
4. Define Population and Sample Design: Decide which population or group of people is
relevant to your study. Then, choose a sample design, which outlines how you will select
participants from that population.
5. Choose Data Collection Method: Decide how you will administer the survey. Common
methods include face-to-face interviews, telephone interviews, postal surveys, email surveys,
or online surveys.
6. Develop Survey Instrument: Create the survey questions or interview schedule that will
collect the data. This could be structured questions for interviews or questionnaires for
respondents to fill out.
7. Review and Pilot: Review the questions to ensure they are clear and relevant. Pilot test the
survey with a small group to identify any issues or problems.
8. Finalize Questionnaire: Make any necessary revisions based on the pilot test and finalize the
survey instrument.
9. Sampling from Population: Use the chosen sample design to select participants from the
population you identified.
10. Administer Survey: Distribute the survey to the selected participants using the chosen method
of administration.
11. Follow Up: Follow up with participants who haven't responded to encourage participation and
improve response rates.
12. Data Entry and Analysis: Transform the completed surveys into computer-readable data and
enter them into a statistical analysis program. Analyze the data to draw conclusions.
13. Interpret Findings: Interpret the results of the analysis to answer the research questions and
understand what they mean.
14. Consider Implications: Consider the implications of your findings for the research questions
and broader implications for the field of study.
Introduction to sampling
Sampling is a crucial aspect of research, especially in quantitative studies like surveys. Imagine
you want to understand the attitudes, behaviors, or backgrounds of all the students in your
university, but there are too many of them to survey individually. Sampling allows you to select a
smaller group of students from the entire population for your study.
However, not just any sample will do. You need a representative sample that accurately reflects
the characteristics of the entire student population. For example, if you randomly pick students
who happen to pass by a certain spot on campus or who are part of your course, your sample might
not represent all students equally. Some students may be missed, and your results could be biased
or not truly reflective of the whole population.
To ensure your sample is representative, you need to avoid bias. Bias can occur if your sampling
method is not random or if your sample frame (the list of all potential participants) is incomplete
or inaccurate. Additionally, non-response from some selected individuals can also introduce bias
if those who participate differ significantly from those who don't.
To minimize bias and achieve a representative sample, researchers often use random sampling
methods, where every member of the population has an equal chance of being selected. This helps
ensure that the sample accurately reflects the population's characteristics, allowing for more
reliable research findings.
Sampling error
Imagine you have a population of 200 people, and you want to take a sample of 50 to study whether
people watch soap operas or not. Let's say the population is evenly split between those who do and
those who don't. In an ideal scenario, if you randomly select 50 people for your sample, you'd
expect 25 to watch soaps and 25 not to.
However, even with random sampling, there can be some error. For example, you might end up
with 26 people who watch soaps and 24 who don't. This small difference is called sampling error.
It's natural and can happen even with good sampling methods.
But sometimes, the sampling error can be larger. For instance, you might have 28 people who don't
watch soaps and only 22 who do. This is a more serious error because it doesn't reflect the true
proportions in the population accurately.
In the worst-case scenario, you might end up with 35 people who don't watch soaps and only 15
who do. This is a significant over-representation of one group in your sample, and it can lead to
misleading conclusions.
While probability sampling methods can't completely eliminate sampling error, they are more
reliable than non-probability methods. They help keep the error in check and allow researchers to
make more accurate inferences about the population based on the sample.
Types of probability sample
Simple random sample
This is like drawing names out of a hat. Each student at the university has an equal chance of being
selected for the study. Imagine you have 9,000 full-time students, and you want to interview 450
of them. You assign each student a number, then randomly select 450 different numbers. Those
students corresponding to the chosen numbers become your sample. It's fair because everyone has
an equal chance of being picked.
Systematic sample
Instead of randomly picking students, you pick every nth student from a list. For example, if you
want to interview 450 students out of 9,000, you'd pick every 20th student. So, you'd start at a
random point in the list and then select every 20th student from there. This is easier than random
sampling but still ensures a representative sample.
Stratified random sampling
This method involves dividing the population into subgroups (strata) based on certain
characteristics, like faculties at the university. You then randomly select students from each
subgroup in proportion to their representation in the population. For instance, if there are 1,800
humanities students and 2,200 engineering students, you'd make sure your sample reflects this
ratio. This ensures that each subgroup is represented fairly in the sample.
Multi-stage cluster sampling
This is useful when your population is widely dispersed, like if you want to study students from
universities across a whole country. Instead of randomly selecting individual students, you first
randomly select clusters, like universities, and then randomly select students from within those
clusters. For example, you might randomly select ten universities and then interview 500 students
from each of those ten universities. This saves time and resources compared to trying to reach
every single student individually.
These methods help ensure that your sample accurately represents the larger population you're
studying, which is crucial for making valid conclusions about things like alcohol consumption
among university students.
The qualities of a probability sample
1. Sample Mean and Population Mean: Imagine you want to know how much alcohol
university students drink on average. You survey 450 students and find that, on average, they
consumed 9.7 units of alcohol in the past week. This number is your sample mean, but you
want to know if it represents the whole student population.
2. Sampling Error: If you took multiple random samples from the same population, each
sample's mean would vary slightly. This variation forms a bell-shaped curve called a normal
distribution. Most sample means will cluster around the population mean, but some will be
higher or lower due to chance. This variation from the population mean is called sampling
error.
3. Standard Error of the Mean: It's a measure of how much a sample mean is likely to differ
from the population mean. It tells us how spread out the sample means are around the
population mean.
4. Confidence Interval: This tells us how confident we are that the population mean falls within
a certain range of the sample mean. Typically, researchers use a 95% confidence level, which
means there's a 95% chance that the population mean lies within a certain range around the
sample mean.
5. Precision of Sampling Methods: Stratified sampling reduces the standard error of the mean
because it accounts for variations between different groups (like faculties at a university). This
makes the estimate of the population mean more precise. However, cluster sampling without
stratification increases the standard error because it may overlook important variations between
clusters (like different universities' drinking cultures).
In simple terms, probability sampling helps us estimate the average behavior of a population based
on a sample, but there's always some uncertainty due to chance. Different sampling methods affect
this uncertainty, with stratified sampling providing more precise estimates compared to cluster
sampling without stratification.
Sample size
Sample Size Decisions: People who teach research methods often get asked how big a sample
should be. Unfortunately, there's no one-size-fits-all answer. It depends on many factors, like time,
money, and how precise you need to be.
Absolute and relative sample size
Absolute vs. Relative Size: Contrary to what you might think, it's not about the size of the sample
compared to the population. Whether you sample 1,000 people from the UK or 1,000 people from
the USA, it's about having enough individuals in your sample to be reliable. Bigger samples tend
to give more precise results, but size alone doesn't guarantee precision.
Tolerance for Sampling Error: When deciding on sample size, think about how much error you
can tolerate. If you need really precise results, you'll need a bigger sample. But it's not just about
reducing sampling error – there are other errors in research too, so precision isn't the only
consideration.
Flexibility in Precision Goals: Researchers don't usually set a specific level of precision they
want to achieve because it's hard to predict. Instead, they aim for a balance between precision and
practicality, considering various factors like the type of research instrument they're using.
In simple terms, deciding on sample size is about finding a balance between practical constraints
like time and money, and the need for reliable results. Bigger samples generally give better results,
but there's more to it than just size.
Time and cost
when it comes to choosing a sample size for research, time and cost are big factors to consider.
Sure, having a larger sample usually means more accurate results because there's less chance of
error. But here's the thing: the biggest improvements in accuracy happen when you go from a really
small sample to a moderate size, like from 50 to 100 or 150. After that, as you keep increasing the
sample size, the gains in accuracy aren't as big. It's like hitting a point of diminishing returns. So,
at some stage, usually around 1,000, you start to see less bang for your buck. Going for even larger
samples becomes less efficient because the extra precision you get isn't worth the extra time and
money it takes. So, researchers have to balance between getting more precise results and not
breaking the bank or spending too much time.
Non-response
Non-response is a big deal in research. It happens when some people in a sample don't want to
participate in a study. So, even if you want to talk to 450 students, for example, you might need to
contact more like 540 or 550, assuming about 20% won't respond. This problem seems to be getting
worse over time, with more people saying no to surveys. Researchers try hard to get more people
to respond, like sending follow-up surveys or even including a little treat like chocolate. But even
with these efforts, response rates can vary a lot depending on who you're surveying. For example,
surveys focusing on top managers might get lower response rates than ones targeting regular
employees. So, while it's important to try to get as many responses as possible, researchers also
need to be honest about the limitations of low response rates and work on ways to deal with any
biases it might cause in their findings.
Heterogeneity of the population
Another thing to think about is how different or similar the people you want to study are. If they're
really different from each other, like in a whole country, you'll need a bigger sample to make sure
you cover all those differences. But if they're pretty similar, like students or people in the same
job, you can get away with a smaller sample because there's not as much variation to account for.
Basically, the more varied the group you're studying, the bigger your sample needs to be.

Kind of analysis
Lastly, researchers need to think about what type of analysis they plan to do. For example, if they
want to use a contingency table, which shows how two variables are related, they'll need to make
sure they have enough data in each category to draw meaningful conclusions.
For instance, let's say researchers are studying social class among cohabiting couples where both
partners work. They want to see if there's a correlation between the partners' social classes. To do
this, they divide social class into categories and compare them for each partner. If they have a
small sample size, they might not have enough couples in each social class category to get reliable
results.
So, depending on the complexity of the analysis and the number of categories involved, researchers
might need a larger sample size to ensure they have enough data for meaningful analysis. If they
have too few cases in each category, their results might not be very trustworthy.
Types of non-probability sampling
Non-probability sampling is when researchers choose participants in a way that isn't based on
random selection. There are different types of non-probability sampling, including convenience
sampling, snowball sampling, and quota sampling.
Convenience sampling
Convenience sampling means researchers pick participants who are easy to reach or readily
available. For example, if a teacher wants to know what other teachers think about school
leadership, they might ask their own students who are teachers. However, this method has
limitations because the people chosen might not represent all teachers—they're just easy to access.
But sometimes convenience sampling can be useful. For instance, if a researcher is testing a new
survey about leadership preferences among teachers, it's okay to start with a convenience sample
to see if the survey questions work well. It's also common in certain fields, like organizational
studies, because it's easier and cheaper than other methods.
Snowball sampling
Snowball sampling is when a researcher starts with a small group of people relevant to their study
and then asks them to help find more participants. It's like rolling a snowball downhill—it starts
small but gets bigger as it picks up more snow. For example, if a researcher wants to study
marijuana users, they might interview a few users first and then ask those users to introduce them
to other users they know. This method is often used in qualitative research, where the focus is on
understanding people's experiences and perspectives rather than trying to get a representative
sample of the entire population. While snowball sampling may not give a representative sample,
it can still be useful for exploring relationships and connections between people.
Quota sampling
Quota sampling is a method used in research, especially in commercial studies like market research
or political polling. It aims to create a sample that reflects the population in terms of certain
characteristics like gender, age, or region. Unlike random sampling, where people are chosen
randomly, in quota sampling, interviewers select participants based on specific quotas set for each
category. For example, they might need to interview a certain number of women aged 25-34 from
a particular area. Interviewers approach people who fit these quotas until they've reached their
targets. Quota sampling is faster and cheaper than random sampling but can introduce biases
because interviewers choose who to approach based on their own judgment. Despite its drawbacks,
it's often used when speed is important or when conducting preliminary research. However, it's
important to note that quota sampling can result in biases, just like other sampling methods.
Limits to generalization
Even when a study uses random sampling, its findings can only be applied to the specific
population it was taken from. For example, if a study looks at alcohol consumption among students
at one university, the findings can't be generalized to all students at other universities. Factors like
the availability of bars, campus culture, or the types of students at that university can affect the
results. Similarly, findings from a study conducted in a certain area, like Oxford, may not apply to
other places. Additionally, findings might be specific to a certain time period, and things may have
changed since then. For instance, a study on student finances conducted in 1980 wouldn't
necessarily reflect the financial habits of students today, especially considering changes in
financial aid systems. So, it's important to be cautious about applying study findings too broadly
without considering these factors.
Error in survey research
When we talk about "error" in a research study, we're referring to factors that can affect the
accuracy of the findings. There are four main types of error:
1. Sampling Error: This happens because it's very rare to get a sample that perfectly represents
the entire population, even when using random sampling.
2. Sampling-Related Error: This is a type of error connected to the sampling process. It includes
things like having an incomplete or inaccurate list of the population you're studying, or when
some people in the sample don't respond to the survey.
3. Data-Collection Error: This error occurs during the data collection phase of the research. It
includes mistakes like poorly worded survey questions, errors in interviewing techniques, or
problems in how the research instruments are used.
4. Data-Processing Error: This type of error happens when there are mistakes in managing and
processing the collected data. It includes errors in coding the answers or mishandling the data
during analysis.
The last two types of error are more about how the research is conducted and how the data is
handled, rather than the sampling process itself. To minimize these errors, researchers need to take
careful steps during the research process, which will be discussed in later chapters.