Basic research methods notes

Introduction to class, Nista vazno

Lekcija 1 Psychology as a science

Three fundamental features of science

Systematic empiricism, Empirical questions and public knowledge creation

Systematic empiricism - Carefully plan, analyse, record conduct observations of the

phenomenon under study

Take general claim, make it able to be examined empirically by making it more specific, take
groups, administer for example standard surveys and then analyse this data through proper
statistical methods

Depending on the hypothesis said and the examination of data it can be said that the
hypothesis is either supported or not supported

Empirical questions- Asking questions about the phenomenon that can be answered through
systematic empiricism

One of the most difficult aspects of psychological research is creating empirical questions that
are both interesting and empirically testable, as there are a lot of questions that simply can not
be answered empirically, at least for the time being.

Issues of specificity and issues of measurability

Does LOVE imply HAPPINESS

love- having friends, a partner, satisfactory sexual life

Happiness- being in a positive mood all day, less depressive symptoms, having more serotonin

This makes it measurable. All concepts involved in an empirical question must be measurable.

MEASURABILITY is strictly related with technology, example of 500 years ago not being able to
prove whether dead people can think, and we now have mri scanners

! for a concept to be measurable it must be specific, as specificity is an important, necessary but

not sufficient condition of measurability.
Note on empirical testability and childhood event memories research

If we for example find that good memories are often more recalled than bad memories we can
be tempted into conclude they are better remembered, however we cant rule out that for
example a person just has more happy memories or something of the sort. We need a reliable
MEASURE of the memories against people can be compared with (in memory of course)

Possible approaches (at least sum): Ask a parent to recollect positive and negative memories of
their childhood and compare this with the recollection of the person so that they may be
compared

2. initiate the study during childhood and ask parent to list diary of major events, question
individual as adult and compare

PUBLIC KNOWLEDGE: Science creates public knowledge, an ever-growing social process based
on the growth of knowledge across space and time

In regards to respect, psychology has an issue with respect as the other sciences that use the
empirical approach commonly examine phenomenon that are often not tangible and usually
not subjects of debate, which makes the average person hold them in a higher opinion while
psychology studies phenomenon closely intertwined with common sense and intuition

Reasons why common sense is worse

1 human reasoning is often driven not solely by logical thinking but often by content and
context specific aspects.

2 humans don’t tolerate uncertainty well, and as such they tend to classify things as true or
false, when often they can be untestable, uncertain or partially true

3 common sense and gut feeling is good enough for everyday situations (sometimes not even
those) but not for the complex functioning of the human mind

4 Human beings tend to look for results that support their prior belief, confirmation bias, which
makes it difficult to do revise knowledge and generate new knowledge

5 Empirical studies require adeptness of observing, memory and analysis that humans do not
naturally posses
6 tend to endorse opinions of what they deem to be experts

Anotha definition of scientific empirism

- Consists of careful planning, conducting, observing and analysing of the phenomena

under study, to a level that is cognitively challenging as it requires the surpassing of
some cognitive limits, which is why becoming a good scientist requires years of training

Scientists need
Skepticism- to stop during research and ponder possible alternatives and alternative
evidence, especially empirically collected
Scientific training- to conduct research with established and proper statistical methods,
and also consulting and knowing the scientific literature, taking years of training
High tolerance for uncertainty: sometimes it isn’t possible to prove or disprove a
hypothesis so we simply need to be able to say that its uncertain with the evidence
possessed at the moment

Research conducting

Generate research question

Design study (How will you collect data? From who? When where?)

Prepare materials (questionnaires, lab space etc)

Make sure your research is ethical

Contact participitants

Collect data

Analyse it, draw conclusions

Try to publish research

Research question generated from informal observation, practical problem, literature

Very important to consult literature even if question is generated from informal observation as
it could already have been covered

Question must be interesting (unanswered, fills a gap in literature) and feasible (realistic
planning with regards to money time and everything else)

Cumulative body of knowledge, constantly growing

Reviewing research literature, sources

Professional journals –

multidisciplinary or monodisciplinary, logical

main types of articles: metaanalysis, tutorials, empirical research report (one or more new
studies and results), review reports(report of the literature and can include propostition of new
approach or interpretation of literature and results gotten up till now)

according to some scholars gray literature can be used

poster presentations: usually by student or someone presented at conference because deemed

not worthy of a talk or just preliminary work

Abstract or conference proceedings: summaries of articles or small articles based on

presentations on conference

Preprint: Preliminary manuscript of study

Research reports conducted by a specific institution

Can be reviewed single blind, where reviewer anonymous and research not (can cause
problems of past relationship, other work, this and that), n double blind, bpoth anonymous
(possible problems not inspecting diligently or not having time or resources as it aint usually
paid)

Scholarly books on the matter

Profecionel zurnels

Often assessed through impact factor, indicates number of citations obtained by each article
published in the journal, divided into quartiles usually , q1-q2 best

• IF is sensitive to how many people work in a discipline: a) Comparisons of IF make sense only
for a specific discipline b) A journal IF should also not be used to evaluate single publications (or
the persons writing them)
Still use this shit.

Use google scholar n guide to research shit

First lecture done good night.

Lekcija 2 idemo bre da se razvaljanovic taj ispitanovic

Four basic principles of ethical research

1 weight risks against benefits

Dispersed over 3 actors, participants, scientific community and society

2 acting with trust and integrity, be honest with participants, don’t make unrealistic promises of
benefits, explain possible risk, duration etc, with this you increase the trust of participants in
the scientific community

Report findings truthfully and honestly , stick to principles of open science as much as possible

Open science

Yappa yappa make available to everyone, non open access more prestigious usually and
researchers no hav mony for publish, still in non oa journals can publish to research gate and
allat

3 Seek justice: award adequate compensation for participant, example of distributing lijek
efektivni
4 respect ppl rights and dignity, informed consent important document, should include
description, duration procedures, possible discomfort, adverse effects, ability to withdraw once
started experiment, foreseeable consequences of this, prospective research benefits, limits of
confidentiality and incentive for participation

IN a lot of studies degree of deception is included, which is controversial but this can be
remedied through a debriefing upon finishing study.

Some amount of participant privacy violation is unavoidable, except anonymous surveys, but
try to keep as confidential as possible

Ethics committee, senior researchers at institution that give pass or fail and can suggest
possible improvements for ethical reasons ofc

3 lekcija

Phenomenon- well established fact that was replicated in empirical research several times,
neeed to be replicated or else either its wrong or needs context inspection to define
discrepancy leading to clarification

Theory in psychology

Coherent explanation of one or more phenomena, usually goes beyond observable phenomena
in question and has abstract concepts, variables and other things, importance of parsimony, use
only as many concept as u need to explain it, generate hypothesis, hypodeductive method
where a hypothesis is either confirmed or disconfirmed, leading to the hypothesis either
supporting or weakening a theory, issue of induction, cant be sure, white n black swan example,
and when it is not supported it leads to either theory revision or examination of the hypothesis
testing

Paradigm

Coherent framework of how a scientific community conceptualizes, investigates and interprets

phenomena within a given domain

Fayerabend. Examine through history why a certain theory was accepted or unaccepted, due to
often reasons outside of science, historical periods n shiet

Hypothesis often just partially confirmed

Thomas Kuhn normal science, develop normal science, revolution that shifts paradigm, then
becomes normal science and cycle repeats

Toulmin, believes not revolution but evolution like Darwin

Focault, practically same shit as Fayerabend, examined through episteme of the epoch

Model, small and narrow theory of a phenomena under a certain theory

4 lekc

Population refers to the set of people constituting a group

Aim of research is to get info on a reference population, but studies often conducted on
population sample. Sampling is the choice of the sample of population

Colpa di feasibilita

Generalizing, u know, golden rule is that a good constructed sample is similar in all important
aspects to the population so it can be generalized unto iT HUSSAH

Samp[les don’t need to match population on unimportant parts

Sampling error, results cant be generalized onto population because it is BURGER. Bad sample

In election poll example of Roosevelt bad sampling method led to ts

Representativeness of sample most important

Sample size recommendation? Many variables that may affect should have bigger sample, and
logically less variables smaller sample

Constructing research literature good for getting the sample right

Two gr of sampling methods

Random methods:

Random

Completely random should in theory lead to generalizability but small sample can make this
fucked, mediate this by stratified sample which collects sample through different strata, where
strata are proportional to the population. These methods usually not used for practical reasons

Convenience sampling

Grab ppl easiest to you

Problem is participation bisas, when what makes the person select themselves into a group is
what distorts result, eg more psychology students tnering psychological research, diluting
answer from average person

Also theres the WEIRD acronym for usual population that is tested

Western, Educated, Industrialized, Rich, Democratic. It is so.

5 lekc

Variables nooo

Variable- sum that changes among people

Measurement- assign number to a variable in a meaningful waty

In case of psychological variables, usually interpreted as measurement of how much a person

has it, so psychological phenomena is usually regarded as measurable variable, like weight,
height n shit

Levels of variables, each second one has properties of one before but not the one after it

Categorical, no quantity just a category, diff values called diff levels of the variable

Must follow 2 principles

1 group members that are the same in measured variable must be placed in the same variable
level

2 all group members must be assigned a certain level of the variable

Ordinal, numbers don’t have metric meaning but are used to measure order of level or somn,
indicates whether someone has less or more of a variable

Ordinal must follow 2 principles aside from the 2 of the one before:

If one member has a larger amount of variable than other he is assigned a bigger number

If same level of variable, assigned to same number, level

Interval, quantity present but no natural zero, used to determine how much the different
individuals difer, need to know true ratio of differences for this and any numbers that
accurately express true ratio are acceptable
Linear transofmrations of an interval scale of measurement will also result in an interval scale of
measurement of that variable

Of any n1 n2 n3 yappa to transorm we get

N1- (a*n1)+b and so on for all numbers, where N1 is the transformed number, n1 is orginal
number, and a and are real numbers that a is not ZERO

Variable has meaningful 0 only when the 0 means the variable is absent

Limit of interval variables is that the number ratios aren’t meaningful, so one level isn’t that
many times bigger than this one etc

Psychological variables usually don’t have a meaningful 0 btw

Keypoints of interval scales are that 0s are arbitraty and don’t represent complete lack of
variable

Number ratios don’t represent ratios of difference among levels

Ratio scales, 0 is meaningful and differences in quantity are differences of ratio

Multicplicative transformation of a ratio scale of measurement leads to a ratio scale of

measurement for that variable

N1 is a x n1

Argument that psycho variables can at most be measured using ordinal scale, form a theoretical
perspective we should suspend judgement and wait for advancement but from a practical
standpoint we use mostly interval level scales for purpouses of statistical methods

Operational definition of a variable is how that variable can be measured, can be more than 1
operational definition

Lekc 6

Basics of JAAASP
Lekc 7 six seve

Descriptive statistics

Descriptive statistics is a set of methods used for summarizing and displaying data

Variables, three fundamental features, shape of distribution, variation of distribution and

central tendency of distribution
Concept of frequency and of distribution are closely intertwined with frequency being the
number of people of a certain level of the variable, while a distribution is the general term for
describing the frequencies of the levels of the variable in the population

Difference between frequency and measurement Is that measurement is assigning values to the
variable while the frequency is the number of people of a certain level of the variable in the
population sample

Shape of a distribution can be presented through histograms or frequency tables

In frequency table the scores are assigned from largest to smallest, and although categorical
variables can be displayed through frequency tables and histograms the shape of the
distribution is not able to be displayed with them

Can be used for 2 or more variable combinations in the frequency table

Histrograms used to assess shape of the distribution of the sample

Unimodal when one peak dragon ball

Bimodal when two peaks (vinland saga berserk)

Symmetrical, positively skewed and negatively skewed

Symmetrical, central peak, central values most often

Positively skewed, peak later, largest values most often

Negatively skewed peak earlier, smallest values most often

Frequency and probability

Probability is defined as the p being equal to the number of observations of a level I divided by
total number of observations pi equal ni divided by ntot

Normal distribution often seen, unimodal and symmetrical distribution but an unimodal and
symmetrical distribution is not always normal distribution

Most statistical models used in psychological research assume that the distribution is normal

The tests for normality are as follows, visual representation, statistical and numeric test

Histogram can be used in jasp along with the density line, with interval option setting the
intervals in the variable that is measured.

Numerical test, in jasp we can also quantitatively see the normality by selecting statistics and
then skewness coefficient

Skewness coefficient larger than 1 is positive, 0 is normal and smaller than -1 is negative.
Between -1 and 1 is considered a normal distribution

Statistical test

Shapiro Wilk normality test measures whether the distribution of the variable differs in a
statistically significant manner from the normal distribution. This is inferential statistics where
two hypothesis are displayed, n0 and n1 with n0 being the null hypothesis and n1 being the
alternative hypothesis

Data inferences made by reference population of sample, which are used to either discard null
and accept alternative or vice versa

P value, depends on context, here is probability of obtaining the observed distribution in our
sample , assuming the variable follows normal distribution in the reference population from
where it was inferred from

If less than 0.05 then h0 right, more h1 right

In here if it is not normal it is h0

If it isn’t it is h1

As it represents chance to get this

This test has its problems, with when the sample is small, rule of thumb less than 20 it can be
inaccurate and also when it is subject to sampling bias or a sampling error in general so it
doesn’t represent the population accurately

Central tendency of a distribution is where the most common values are distributed in the
variable, we have the most common 3 measures of it, mean mode and median

Mean is everything summed up divided by number of elements, mode is the most common
appearing value and median is the value at the middle

If uneven, median score is 2 middle components divided by 2

Mode is the only measure of central tendency that can be applied to categorical scales of
measurement

Variables measured on ordinal have median and mode but not mean, while ratio n interval
have it all

Mean is by far the most commonly used in ratio and interval scales but it is highly subject to
being affected by the distribution, while if the case of the distribution is symmetrical then the
mode mean and median are relatively the same

In case of strongly skewed distributions the median is preferred as it is not affected as the mean
is

The informativeness of a measure of central tendency depends on the dispersion of the

distribution

Since the c t measures depend on dispersion measures like the mean should be used with at
least one measure of dispersion

Dispersion deals with the variability of the data within the distribution

Three most important measures of distribution are standard deviation, interquartile range and
and range. Can only be applied to interval or ordinal scales of measurement
Simplest and least informative is range and it is the maximum and minimum value difference in
the distribution

The interquartile range describes the difference between the third quartile and the first
quartile, where the 3rd quartile is where below 75% of values fall under and 1st quartile is below
25% fall under. The median quartile is 50%, the 2nd

Percentile rank of a score measures which values fall under that %. We can calculate this by
organizing scores from lowest to largest, count how many of them are lower than it, divide
number by total and multiply by 100

In raw terms standard deviation is the mean distance between the values In a distribution and
the mean of the distribution

Divide xi where in the variable x an I score is there, take away from it the mean score of the
variable x, sve to u zagradi, kvadriraj zagradu, dodaj sve zajedno I podijeli s brojem ljudi u
uzorku

We also have variance which is standard deviation squared, same info as SD. Both should be
expressed with the units that correspond to them

Standard error is the standard deviation divided by the square root of the sample numerosity
Wtf

Part 2 deskriptivne jaoouuu

Outliers- problem where there is a sample member whose scores differ significantly from other
scores

This can fuck the data up, not obious by standard deviation or other numerical representations
of dispersion, cause the data dispersion might be relatively small but these outliers can skew
them. Best way to spot them is a histogram, through visual analysis

Outliers can be a significant problem because they strongly interfer with the mean, in some
extreme cases making it a lie!
What do we do with them? No rule, for example when we are sure that they result from
systematic bias, we can remove them, this is reasonable.

Example of this is if for example duhhhhhhhhh we are sure that the participant did not
understand the questions

We need to give validation for removal

Rule of thumb is to ensure in general that the main conclusions of a study are not substantially
dependent on the presence or lack of outliers

We may conduct the analysis both with and without them and if the results are robus they will
remain consistent regardless of their presence and absence.

We can describe the relative position of a score through the z score where from the score the
mean is subtracted and then the result of that is divided by the standard deviation of the scores

This is called the standardization of scores and the z score is called a standardized score

Z scores larger than 3 or smaller than -3 are considered outliers

Only for interval and RATIO scales this one yes…

We should when missing data occurs see the potential reasons for thei absence such as the
question being too complicated or overly intrusive. We can deal with this 2 way, 1 ignore, 2
replace wit mean 😊

Psychological research typically focuses on the relationship between the variables

Independent and dependent value with independent being the one that affects and dependent
the one being affected

Can be displayed through descriptive and or inferential statistics

We consider cases only involving quantitative, usually interval measurement scale as they are
most common for psychology

In graphs the independent variable is usually expressed on the horizontal axis and the
dependent on the vertical axis
Bar graphs, usually used for categorical variables or quantitative variables with a small number
of levels (2-10, usual rule of thumb)

Line graphs, equivalent to bar graphs, representing the mean n stnd err values for the
dependend variable at diff levels of the independent variable

In jasp the process for creating and computing graphs depends on the type of research design,
for now focus one 1 ind 1 dep variable

First, we see if ind variable has 2 levels or more than 2 levels. If 2 then t test if more then anova

We must also see if it is a between subject design or a within subject design with a between
subject design being that the participant experiences only one level of the independent variable
while the within subject designs all participants experience all levels of the independent
variable (within n between designs most common for one ind variable research)

Between subject design example is group as ind variable where group has levels like patient
control or yip yap and yop and each participant is in one of these values

Within subject design example is time where each participant is tested at one interval then the
next, then additional times if designed so

When we check for p to prove h0 we must note that it is only acceptable measure if the both
levels of independent variable are normally distributed which is easily checked in jasp

Bar n line graph limitations are that they provide no dispersion data and no visualization of
outliers

Some researchers think bar graphs should be banned

Raincloud plots work with each of the cases mentioned up till now and seem very niiice.. :)
Scatterplots applicable to between subject designs where the independent variable is
quantitative and has many levels

Effect size indices

Two common situation in which two diff types of effect size indices must be used

Situation 1: independent variable categorical and dependent variable quantitative

Situation 2: Independent quantitative with many possible values and dependent quantitative

Does bio sex affect in? compare mean scores, but we gotta check for dat dispersion. First case
std deviation difference is 2 something but its small, we see it visually and its aight and then 2 nd
case big standard deviation, like 13. This is a no no ☹

This example shows that for any numerical index of difference between groups the dispersion
must be taken into account for. Most common option for this is cohens d where the mean
scores that are subtracted are divided by the standard deviations of both the groups together.
• A Cohen’s d of 0.50 means that the two group means differ by half a pooled standard
deviation. A Cohen’s d of 1.20 means that they differ by 1.20 pooled standard deviations.
For the cases where we want effect size and the ind variable has more than 2 levels then the
equivalent of Cohens DICK is partial eta squared

Statistical relationship between two variables typically measured by correlation. Pearsons

correlation coefficient most often used ehre buddy 😊
Covariance is one of crucial components, calculated same as correlation coefficient but without
standard deviations, it shows how much these variables VARY TOGETHERRUUHHH

Tends to have a strong positive value when ind variable score higher than mean of x score
higher on dep variable mean of y and same for score low

Negative follows logically

Close to zero when these 4 are about equally distributed

Only valuable when the ind and dep variables are approximately normally distributed
Relationship between variables can also be quantified by r squared, which is just the absolute
correlational strength between them as it ranges from 0 to 1 as opposed to -1 to 1
Strongly affected by outliers so pearsons r should be presented with a scatterplot as well to
exclude that scenario

Lecture 8

Statistical relationships and causation

When there is a statistical relationship found between variables this could mean that there is a
causal relaitionship between them as in the ind variable causes the shift in the dep variable, but
this does not HAVE to be the case.

Directionality and confounding variable problem

Directionality problem

The choice of which variable influences which (ind and dep) is usually based on the researchers
hypothesis which is usually based on a definite theory

But how do we know which influences which? The answer is an experiment

In general its solved in the way of experimenting where the independent variable can be
manipulated

In some cases the directionality problem can remain partially or totally unresolved for reasons
of ethical nature.

In other cases the problem doesn’t exist because switching the roles of the variables makes no
sense, example of iq causing differences in bio sex

A confounding variable problem is that there exists a third variable that partially or completely
explains the relationship between the independent and dependent variable
In the case of bio sex influencing driving skills the confounding variable of driving experience
would be called a mediator, it mediates because It doesn’t affect biological sex, while the
atmospheric pressure listed above would be labeled a redundant relationship and a redundant
confounding variable

The answer to this again is experimenting, specifically where the experimenter tries to control
as many confounding variables as they can.

Lecture 9

Experimental research

Experimetns r useful for testing the causal relationship between variables. This is crucial for
testing hypothesis and theories

Characterized by 2 main features:

The experimenter manipulates the independent variable

The experimenter controls extraneous variables (any variables not ind or dep)
Manipulate and control have specific meanings in experimenting, as manipulating means
systematically changing the levels of a variable in the experiment while controlling a variable
means preventing it from changing systematically in the experiment

The manipulation of the ind variable is meant to solve the directionality problem

The point of controlling extraneous variables is that they solve the confounding variable
problem.

If we control the extraneous variables, no change in dep variable, then manipulate the ind
variable and the dep variable changes we can say that the changes of the dep variable are due
to the effects of the ind variable

Why isn’t systematic observation?

Diffusion of responsibility phenomenon

Possible levels of the manipulated independent variable are also called the conditions of the
experiment

Implications of active manipulation of the independent variable

In some cases the independent variable can only be manipulated actively in an indirect manner.
Example of mood. This is because the manipulation could be mediated by other factors

In these cases a manipulation check is conducted where the independent variable is tested
before and after the manipulation attempt

In some cases for practical or ethical reasons the independent variable cannot be manipulated,
ex of child illness. This doesn’t mean that the relationship isn’t able to be studied, it just means
it must be done through non experimental designs

Controlling the extraneous variables also differentiates experimental from nonexperimental

designs, the important ones specifically, which are all the variables that are not the
independent or dependent variables but may have an effect on the dep variable. We try to
prevent it from changing systematically.

Can be divided into 2 broad categories: Personal and Situational variables

If there is an important extraneous variable that changes systematically during the experiment,
this becomes a confounding variable

How to control extraneous variables

There is usually a large number of extraneous variables, for example the comfort of the
participant seems to be an important situational variable, however the situational extraneous
variables usually constitute a minor problem for researchers as they do not typically change
systematically during the experiment

Observable and measurable extraneous variables can also be controlled very easily

The process of sitributing the effect of a measurable extraneous variable is called the
counterbalancing of that variable

When an important extraneous variable is measurable, an alternative is keeping it constant for

the duration for the experiment

However this may lead to an impediment of the generalizability of the experiment results

Some of these variables are unmeasurable though, and they can constitute a bigger problem,
how do we solve this? Ramdagadam dagadigadigadam

Randomization implies the random assignment of the participant to the conditions of the
experiment, and if the sample is large enough this ensures that the distribution of personal
extraneous variables is approximately the same across all the conditions

How big does the sample need to be?

No rule, a few factors

Expected extraneous variables, larger size

Stronger effect of ectraneous variables expected, larger size

Stronger effect of ind variable on dep variable, smaller size

No of conditions, larger size

Consulting research literature helps this

Exp design p2
Two general measures of the overall quality of the psychological study are internal validity and
external validity

Internal validity

A study considered high in internal validity means that the design ensures that the change is the
dependent variable is CAUSED by the independent variable’s manipulation

The highest level of this can be achieved when it is concluded that confounding variables had
no effect on the dependent variable

External validity

A study considered high in external validity is one whose results can be generalized to people
outside of the sample of the experiment and to contexts outside of the one the study was
conducted in. For psychology this is very important as we seek to understand general laws of
human psychological functioniong

As a general rule the study has high external validity when the sample that the researcher
chose closely resembles the population onto which the researcher wishes to generalize the
results on

This is ofc regulated by the representativeness of the sample

Regarding situations studied, psycho exp are often criticized by the need for controlling of
extraneous variables and the manipulation of the independent variable, causing the conditions
to appear artificial, not as they are irl.

In many cases there are key diffs between this artificial setting and real situations

Computers and other measurement devices are used in the majority of psychological
experiments, they allow for great control of extraneous variables, the implemention of
standardized, automated procedures for data collecting, and to be very precise in measuring
the patients response.

They increase internal validity, however they reduce external validity as there is a difference
between how people behave in front of a computer and in real situations

Most researchers prioritize internal at the cost of external validity

Between subject designs are widely used in clinical research. Example of taking sample of
agoraphobic people and assigning them randomly to one of 3 treatment options

Internal validity of this type of research can be high provided the experimenter controls for
extraneous variables effectively
Often used to determine whether a treatment, intervention or training works

Often there is a group that receives the treatment, and then the control group which doesn’t.
this is called a randomized clinical trial, it is experimentu… -_-

The basic logic behind it is that simply seeing a difference before and after in the dep variable
isn’t sufficient to declare the treatment effective
We should do an effect size estimation on deez.

Lets say the treatment is effective, control group gets it. Lets say its not, both groups get
alternative treatment. Simplifies ethical concerns.

Placebo effect very important here

Expectation to improve leads to improvement

To avoid difference of ppl in treatment group getting placebo cause they know they getting
treated and the control group not getting it, we introduce the placebo control group. Self
explanatory

Another option is removing the control condition entirely and replacing it with th best available
alternative treatment. For one placebo reduced, secondly if an effective alternative exists, its
only important whether this new one is better or not

Again we mention within and between subject designs

Main advantage of within subject designs is that there is a larger ability to control extraneous
variables as all participants are tested in all the conditions

However it may also have disadvantages compared to between subject designs

Primary disadvantage carryover effect, being tested in one condition may affect the behaviour
while being tested in the next condition

At least 3 types

Practice effect – participants perform better in next conditions cause they had a chance to
practice it. Effect can be particularly strong in difficult tasks or response time tasks

An artifact is a result that is not caused by the phenomenon under study but by the arbitrary
choices of the experimenter. If not controlled for the practice effect could become an artifact.
Type of variable that an artifact could be is a situational confounding variable

Fatigue effect- opposite of the practice effect, may perform task worse due to fatigue or
boredom, especially significant in long and or fatiguing tasks. Especially in studies where
response time is measured and in which the number of trials is very large
Context effect- Effect in which the previous conditions influence the response of the successive
conditions through their experiences. For example if a chud were to be showed beforehand,
the perceived attractiveness of the current normie would be increased, while it may decrease if
a chad was shown beforehand. This type of context effect is also called the contrast effect

Possible solution counterbalancing, giving different participants different order of conditions to

be tested in, making the carryover effects be equally distributed amongst them

Stroop effect

With counterbalancing the carryover effects cancel each other

Read up on this shit man I don’t get it at all

An alternative to counterbalancing is the simultaneous between subject design, where the

conditions are assigned randomly n intermixrd
Due to significant confounding variables problem influence on the internal validity of the study,
within subject designs are usually preferred to the between subject designs

Sometimes between better such as we cant test treatment with withinsubject, and cant ask
them to recall memories as they already recalled them in a positive mood, in a negative mood
they will most likely recall the same memories

Further issues of external and internal validity

Experimenter expectancy effect

Refers to any systematic variation of the dependent variable which is due to the experimenters
expectancy to how participants should behave.

Might unconsciously give clearer advice to group they expect to do better

Standardization can reduce this, but the blind experimenter seems ideal where the
experimenter doesn’t know the research question

This effect decreases both internal and external validity of the experiment

Hawthorne effect, related to the effect where peoples behaviour improves as a reaction to the
fact that they are being tested

This also causes problems for external and internal validity of the experiment

John Henry effect- individuals who are aware that are at a disadvantage might perform with
more effort in an effort to close the gap between them and the other group

Lekcija 10 Complex research designs

In psychology we usually measure unobservable variables such as intelligence, memory capacity

personality traits. We call these constructs

They are called constructs because they are not obvious or selfevident, but they are
«constructed» upon a conventional conceptual definition.

Constructs are unobservable both because they are mostly internal but also because they refer
to tendencies

For example extroverted personality is a tendency that shows itself

The operational definition of a construct is how the construct should be measured. Three
categories

Self report measures- report own thoughts, feelings, actions. Widely used in clinical psychology

Behavioural measures: some aspects of an individuals behaviours are measured, directly

observed and recorded. Large variety of measures

Physiological measures: recording the activity on a physiological level ofone or more of the
participants body parts

Large number of constructs can be measured in multiple ways, and sometimes they are as this
can potentially deepen our understanding of the relationship of the ind and dep variable (s in
this case)

In one experiment the experimenter may measure more than one construct. Example of the
professors experiment of expectancy and perceived weight
Factorial designs

The most common approach to measuring multiple variables is a factorial design, where every
level of one independent value is combined with each level of the other indepent variable/s to
produce all possible combinations. Also the ind variables are called factors

Each cell represents one condition, with each condition benig a combination of the levels of the
ind variable (factors)

Number of ind variables and number of levels of these variables can be represented by
multiplying each no of levels in the first factor to the n factor

In practice usually not more than 3 factors

In between subject factorial deisgns each subject is tested in one condition.

The results of studies of factorial design are usually represented with line, bar and raincloud
plots. Usually added prefix factorial to graph

Jasp optimal factorial line plot representatiton

In within subject factorial designs each participant is tested in each condition.

Easier to control personal extraneous variables

Carryover effects likely to occur, so counterbalancing is necessary

In mixed factorial designs at least one factor is between and one is within subject

Rule like rule

Then each participant tested in withinsubject factor

As long as at least one ind variable is manipulated its considered an experiment

In some factorial designs not all the independent variables (factors) are directly manipulated by
the researchers. In other words, some factorial designs include nonmanipulated factors.

Gotta revise this shit when I sleep or when I get fresh

In factorial designs two kind fo results can be of interest

Main effects and interaction effects

A main effect refers to the individual influence of a specific factor on the dependent variable

There is one in each study

Interaction effects occur when the effect of one factor on the dependent variable is influenced
by level of another factor

In a factorial line plot, non-parallel lines indicate an interaction between the factor on the
horizontal axis and the factor represented by the different lines.

In factorial designs, interaction effects are often more interesting and more informative than
main effects.
Read ts again

Lekcija sledeca ako sznam

Features of nonexperimental research

Nonexperimental research may lack:

Manipulation of the independent variable OR Random assignment of participants to conditions

(or to order of presentation of conditions) OR Both

Four types of nonexperimental research

Single variable research

Typically focuses on how many individuals of a population exhibit a certain psychological

characteristic or engage in a certain behaviour

Don’t manipulate ind var, don’t cnt extr var, not experimental
Example is epidemiological study, which estimate prevalence of a specific disease within a
population

CORRELATIONAL RESEARCH

Focuses on the statistical relationship between variabbles

Directionality and confounding variable problem

Correlational research can be a preliminary step to research, as in once a relationship is found

then research can begin

In other cases it’s the only feasible approach

Example if the relationship isn’t believed to be causal

Distinction between ind and dep variable becomes much less relevant

Example 2 The relationship between the two measured variables is thought to be causal, but
researchers cannot manipulate the independent variable because it is impossible, impractical,
or unethical.

Correlational designs commonly used in clinical neuropsychology

which studies the effects of brain damage on cognitive functions, such as thought, emotions,
and behavior.

Some third variables that may affect the dependent variable can be controlled to some extent
using complex correlational designs, where potential confounding variables are treated as
independent variables (e.g., “the influence of the extent of parietal lobe damage AND drug use
on logical skills”).

The relationship between brain damage and cognitive functions can only be studied by means
of correlational research. • However, the more general relationship between brain and
behaviour can also be studied using experimental designs…

Correlation and correlational research are not the same thing

Again mentioning the use of correlational designs to be a preliminary step to determine
whether to conduct eperimetn orn not

Naturalistic observation is when the observer seeks to collect data I nthe environment of the
behaviour in which it occurs

They do not interact with the observed element

It’s a type of correlational research because it lacks manipulation of the independent variable

Widely used in ethological and anthropological research

It maximizes external validity

Quasiexperimental research

Involves manipulation of independent variable but does not involve random putting individuals
in conditions

Remember that random assignment of participants to conditions or to the order of conditions

allows researchers to control personal and situational extraneous variables, preventing them
from becoming confounding variables.

No directionality problem but confounding variable problem yes

Note however that: 1) Experimental research has high internal validity only when it is properly
conducted, including effective random assignment of participants to conditions or orders and
an effective manipulation of the independent variable.

2) Quasi-experimental research (and sometimes correlational research) may have relatively

high internal validity when there are no obvious confounding variables.

4 types

Non equivalent group designs are between subject designs where the individuals are not
randomly assigned to conditions, which is the step that allows researchers to prevent
systematic personal extraneous variable interference

Pretest-posttest designs- mainly used for evaluating effectiveness of a treatment, with

dependent variable measured once when treatment isn’t administered and once when it is

In clinical research another possible omission of the results is spontaneous remission where
physiological and psychological issues can get better over time by themselves

Randomized clinical trials are essential to test the effectiveness of a treatment

A treatment is judged to be effective only when the patients in the treatment condition
improve more than the patients in the control condition (no treatment and/or placebo control
conditions).
Pretest psottest designs widely used in educational research to test effectiveness of new
education programs

Sometimes ppl that only differ significantly to the mean are selected but sometimes this is not
concrete as regression to mean is possible
Interrupted time series designs are a variant of pretest-posttest designs

• Interrupted time-series designs are a variant of pretest-posttest designs. A time series is a set
of measurements taken at intervals over a period of time. In an interrupted time series-design,
a time series is interrupted by a treatment. • For example, suppose that the dependent variable
is the number of aggressive acts taken by a group of inmates. • Suppose that the DV is
measured once a month for 6 months. Then, after 6 months, the inmates get a 3-month
psychotherapy for the reduction of aggressive acts. The number of aggressive acts is then
measured once a month for other 3 months.

Combination designs combine elements of nonequivalent groups deisgns and elements of

pretest-posttest designs. A «treatment group» and a «control group» are both given a pretest
and a posttest. • The only difference between combination designs and randomized clinical
trials is that the former lack random assignment of participants to conditions.

Because history, maturation, and regression to the mean are likely to be similar for the students
in both schools, any observed difference between the improvements of the two groups of
students is probably due to the effects of treatment. • However, this type of design does not
completely prevent the possibility of confounding variables, for two reasons: 1) Something
could occur at one of the schools but not the other (e.g., a student drug overdose), so students
at the first school would be affected by it while students at the other school would not. In other
words, there might be different “histories” for the two schools. 2) Similar to the case of
nonequivalent groups designs, the treatment group and the control group may differ in some
important extraneous variables that the researchers could not control.

The problem of internal validity in quasi-experimental research: a summary The internal validity
of quasi-experimental designs should be evaluated “case-by-case”. Researchers can enhance
the internal validity of quasiexperimental research designs by controlling factors that may
potentially lower internal validity, such as spontaneous remission, history, and maturation.