Fundamentals of Experimental Design in Cognitive Psychology

Hello and welcome to the course basics of experimental design for cognitive psychology. I am Dr. Arkwarma from the

department of cognitive science at ID Kpur. Uh we'll be starting the third week of the course and now it is that we

are going to get uh slightly in depth about experimental designs. try to understand uh the broad factors that are

related to experimental designs uh the underlying principles uh the types of designs variables this and that and and

a lot of details that we typically sort of avoid uh you know in in general discussions when we're talking about

experimental designs. So uh let us start. We'll give you first in this lecture a broad overview of what are the

uh you know what is the blueprint of an experimental design. What is a broad uh skeleton of this design and in the next

lectures of this week I will dive slightly deeper into the mechanics of experimental design and how we go

forward. Okay. So uh we have seen so far in the two weeks that psychology pursues the

scientific method in trying to understand the relationship between variables that affect human behavior. So

the idea is to understand what are the factors that trigger that drive human behavior. what variable? Say for

example, let us say uh does uh luminance of a given stimulus uh drive how long or how well will I remember it or for

example broadly speaking uh does empathy drive for example how much help uh a person will give to somebody else or not

and and broad and specific questions both. So that is basically what we uh you know uh are pursuing here under

psychology but we can basically pursue the kind of questions uh using different methodologies. So we can have

quantitative uh research designs. We have we can have qualitative research designs. You can talk about uh higher

order behavior. Say for example things like empathy or you can talk about very specific uh uh you know uh basic kinds

of behavior. Say for example how luminance uh dilates the pupil of our eyes and the consequences of pupil

dilation and say for example uh what are the uh you know what kind of things pupil dilation indexes. For example does

it index sphere? Does it infect more light? Things like that. All right. Now, uh there is also something that uh you

know we have seen so far. We've been talking about uh you know uh qualitative designs. We've been talking about in in

the quantitative section uh descriptive research designs and correlational research designs. The critical factor

that I think should have emerged so far is this idea of uh you know control and this idea of

uh chasing causality. All right. Now I'll I'll talk about control first. Uh what do we mean by control? When you try

and study a particular behavior, let us say uh when it's happening in the in in the wild for example through

naturalistic observations, you try and not control anything. So the idea is to not sanitize not sterilize the

environment and just observe the behavior as it is occurring. All right. So uh the players uh in the behavior the

environment everything is uh or actually nothing is disturbed. You do not give instructions to the players you do not

uh uh you know uh control the different settings in the environment. It's just like you've put your camera here. You

are let's say a participant or a nonparticipant observer. remember the previous lectures uh and you're

observing how things are going on. So that is basically the least degree of control that you want uh and while it

has its consequences while it has its uh you know let's say pros and benefits that you are being able to measure the

behavior as it is occurring the best form of observation that there can be but it brings in issues you are not sure

about why something is occurring. There could be any number of reasons. There could be uh variables 1 2 3 4 5 6 and

any number of variables that might be affecting the behavior. So when you try to develop explicatory models, when you

try to develop predictive models based on whatever you have observed, there is a lot of problem uh there there are a

lot of problems that emerge. You don't exactly know what has caused that behavior because there could be variable

A which has caused variable B. But you did not control for variable C, D and E. And they might also have had some

influence. Remember in correlational designs we talked about common causal variables. We talked about extraneous

variables and a bunch of these things. All right. So control is is something that is interesting. And from control if

the degree of control is less the degree of success with which we can make causal inferences is less as well. All right.

Now if you want to make causal inferences, if you want to understand why a given behavior is occurring, if

you want to understand why and how exactly uh you know variable one, variable one has caused variable two,

then you cannot but escape controlling for variables 2 3 4 and everything around it. You have to control the

situation. You have to control the variables. You have to basically create a situation that no common causality is

possible, no prior causation is possible. And in that sense that would allow you to infer causal behavior that

would allow you to infer causality and that is what is broadly the attempt in experimental research designs. All

right. So uh there is the scientific method there are quantitative and qualitative research methods. Within

that there is this notion of uh control variables controlling situ controlling the experimental conditions for example

and then the degree of causality that you can infer from a given research design. Okay. So that's the broad

outlook of what we've covered so far and where I think it gives you a good hint of where we are going from here on.

So these two factors control and causality as I was mentioning uh along with a specific research question uh

basically in govern the choice of what kind of uh you know method what kind of tool we will adopt for our

investigation. And we've seen basically that while descriptive and correlational research designs are excellent tools to

study human behavior, they have their own strengths and weaknesses. I'll not revisit them again. But at least we can

say uh that if we want to infer causality, I think the experimental research designs are the best equipped

and we'll talk about the reasons uh why they are uh you know best place to give that. We've already talked about uh uh

avoiding common causal variables. We've already talked about uh the effect of confounding or extraneous variables,

prior causation, reverse causation and and a bunch of these things. So we have some idea why this might be the case.

But let's let's go deep and try and understand how things happen. So let's let's look at some very basic

concepts. Experiments experiments typically require two things. An independent variable and a dependent

variable. Uh what is an independent variable? The independent variable is the one that you are manipulating to see

its effect on the dependent variable. Say for example, you can manipulate the color of the ink to uh test how well or

how much of an influence the color of a ink has on the legibility or the uh you know memorizability of a given text.

Let's say the text in both these uh uh you know uh ink colors is the exact same text. All right. Uh now what are the

other two things that I mentioned? legibility and memorizability. These are the dependent variables. So dependent

variable is basically the response measure of an experiment. It is basically on uh you know the variable on

which we are seeing the effect of the manipulation of the independent variable that the experimental has done. All

right. So basically two things independent variable the things that you will change. So you'll change the color

of the ink to see its effect on memorizability. So you'll see uh better memory, poorer memory uh in terms of

let's say number of correct responses uh you know the time taken the faster it is the better it is uh you know or say for

example the more time a person takes the poorer it can be considered. So those kind of things you are manipulating the

independent variable color of the ink you are basically getting an uh you know some response in terms of memorizability

uh reaction time accuracy any of these kinds of measures. Another important things what is what

did we do here we manipulated the color of the ink what were the levels let's say uh we talked about red green and

black now what are red green and black in this scheme the red green and black are basically the levels of the variable

okay so for example if I want to study let's say there is this study I'm a be a first year student I've been given this

assignment that okay go create an experiment so I'm uh doing that I want to create an experiment I want to study

the effect of color of the ink on memorization ility color thing is one concept memorizability is another

concept both of these con concepts will need operational definitions I will operationally define them uh in this

part I'll say oh color ink is again it's very basic uh the color in which the text is written so done uh at that I

have to at that point I have to decide the levels so what are the levels either I can take two levels it'll at least

have two presence absence for example in some cases or let's say in this case it'll be at least let's say blue or uh

red or we can take blue, red, green, black let's say four levels does not matter these will be the effect of the

different colors will be measured on memorizability. All right. So this is an important aspect of manipulation uh

which is the levels of a you know a given variable uh it has to be at least two. Say for example if you want to test

the effect of uh fatigue uh on uh driving performance for example let's stay let's say uh so uh fatigue can have

u obviously you can take a continuum of fatigue how tired you are 10% 20% 30% and so on so you can take a continuum

but you can also say that my criteria or threshold of people being tired is some variable uh and that uh the the

threshold of that variable is let's say 75%. So everybody who's tired less than 75% is not fatigued and everybody who's

tired more than 25% is fatigued. So you can also conceptualize the levels of the variable in presence absence presence of

fatigue absence of fatigue something like that and then you can measure it on the dependent variable say for example

driving performance number of red lights missed number of near misses number of uh the amount of brake ports applied and

so on. So you can basically see levels of variable and then how you are measuring it in the dependent variable.

Yes. Uh something that is important before we start carrying out uh you know these manipulations of variables and so

on uh is this idea of initial equivalence. So I I I'm probably going to be a little bit nonlinear in in this

and the next few lectures uh because I've sort of put together some of these concepts that are necessary. I am

visiting them here but I will be sort of uh you know elaborating them as I go forward as well. So initial equivalence

what is initial equalence? Remember we are talking about uh avoiding reverse causation. We were talking about the

fact that only the variable that we are manipulating should have the effect on the dependent variable. Only the

independent variable of interest should have an influence on the dependent variable and all the other variables

should be controlled. Say for example if uh variable V_sub_1 is the one that we are control that we are manipulating IV1

for example IV1 is the one that we are manipulating and we are seeing its result on DV1. Now there can be IV2,

IV3, IV4 and so on. There can be these these can be called other independent variables or you can call them uh you

know confounding or extraneous variables whatever you want to use uh you know as a term but the idea is that the

experimental situation is sanitized to the uh effect that there is nothing else that can cause the variation in the DV1

uh other than the IV1. So if there are uh you know uh environmental situations or if there are subject level uh uh you

know uh characteristics uh all of them uh basically will uh have to be controlled for or matched. Okay. So you

will basically try and see that okay if if you're talking about let's say uh let's let's introduce a variable here uh

for the sake of an example uh color of the ink memorizability. Now let us bring a variable called uh gender

in. So uh let's say the hypothesis is uh males will remember material written in red ink better uh and females will

remember materials written in blue ink better something like that. If you have this kind of a hypothesis. Now here if

you have something like that if you want to conduct an experiment uh that tests this first hypothesis males will

remember the material written in reading better then what you will want to do is uh let's say you have two levels of the

independent variable read inking and green ink. Both of these uh you know uh treatments uh will be given to two

groups but groups of males which are matched on all possible characteristics. Say for example same age education uh

educational uh same age educational background uh gender obviously I'm talking about just males because now

you're going to test uh for this one. So this is one and gender by the way is also a subject variable. This is a

subject level characteristic. All right. So if you bring in that kind of thing you would want to uh you know uh control

for this or match for this. We are in this case in this specific example we are matching the participants in both

these groups red ink and uh black ink are both uh gender matched. And that is how we are testing the hypothesis of

this uh idea that uh you know males will remember material written in red ink better. Okay. So this is broadly what

we're doing. Uh I'll come back to initial equivalence uh in in more detail later. I'll come back to the techniques

of creating initial equivalence also. Now broadly what are the things independent variable dependent variable

uh we talked about um hypothesis we talked about control initial equivalence and so on. Now what are the advantages

of doing experiments? We've talked about this uh earlier uh as well. But in an ideal experiment, we're just extending

what we were talking about. In an ideal experiment, no variables except the one being studied are allowed to influence

the changes in the dependent variable. Experiments do take care of, you know, to avoid the influence of extraneous or

confound variables as well as the effects of prior and common causation. Hence, experiments will provide us a

better way to understanding causal relationship between two variables. So the first and the foremost advantage of

experimental research designs is that it allows us to understand what what is called uh you know the causal

relationship between variables. That is one. Okay. Other is that uh experiments are typically uh you know economical and

it's it's a side advantage uh you know in my opinion but uh typically see if you're setting up a research design to

observe uh behavior in the wild. Okay. or say for example uh phenomena uh phenomenon occurring in uh you know

public places social situations and so on if you want to go there it it is a it's a logistical exercise because you

want to uh identify which places I am going to go to let's uh remember from the previous uh uh lectures that say for

example if somebody wants to study a busy street corner in Mumbai city or Delhi or in Kpur for example you have to

figure that place out take permissions etc etc. You have to set up your cameras and this and that everywhere to take uh

the uh you know uh observations come back. Uh the amount of data that you'll have generated is used. the amount of

useful data might not be that much uh uh you know in that whole thing because we did not control for any number of

variables and in that sense the exercise is much bigger and it is logistically and otherwise more expensive and in that

sense what we are saying is experiments are economical okay if I want to simply study this color of the ink and

memorizability uh you know phenomenon I can just take say for example uh you know uh a group of 20 males uh let's say

maybe two different groups of 20 males but still let's say I I am in a college where undergraduate students are uh you

know uh not that difficult to come by and I can just uh give them material written in uh you know uh red ink and

black ink small text maybe make printed on uh small uh you know uh postcards and I give them to read I take give them 5

minutes or 10 minutes or 15 minutes recall time and I ask them to repeat it that's all that is needed no expensive

equip equipment, no logistical permissions required. And and in that sense, you can see now that experiments

are economical to do. Economical because they strip off all the uh you know extra information, extra variables from the

measurement scenarios. Okay. So these are two interesting advantages of experiments. Now uh why do you need to

conduct experiments? What are the experiments actually trying to do? uh there is this idea of critical

experiments and there's this idea of exploratory experiments. So uh what are critical experiments? Critical

experiments typically are those uh that are driven from a theoretical standpoint. Okay. So for example uh

somebody has a theory about uh I I'll stick with this example just for continuity sake. Somebody has a theory

about the fact that the uh you know inh inherent properties of color or let's say of our visual system are more

aminable to reading in uh red uh than to reading in blue or green. And from that perspective I developed this hypothesis

or theory that anything that is written in red is easier remembered uh better read and faster recognized. Let's say I

have this idea. Now I have this theory. Let's say I call this theory XYZ theory of whatever and from this theory I can

come up with hypothesis. I can say okay this is uh you know from this particular standpoint let's say take the visual

system uh you know or the uh you know the retina in in uh consideration and say because of uh these rods and these

cones uh you know in in the retina uh it makes it easier for us to read in red than in uh green. Now I have these two

hypothesis. I will be able to uh compare them when I conduct the study as I just described. So uh critical experiments in

that sense are uh experiments that pit testable hypothesis about certain phenomena against each other and develop

potential databases of human behavior. So now whatever I do I collect data. I collect uh you know the amount of time

people took to let's say recognize memorize information written in red ink versus black ink and I collected data on

let's say 40 people 30 uh you know 30 males 30 females I I have that data now so I create that data whatever my

finding is if I if I get a null effect that there is no effect uh that there's no difference between reading and bluing

then also I create some data I I create that okay if a person reads such and such material in in this ink. Uh this is

what the performance indices look like. Or if I do find something, yes, red ink is easier to read than uh blue ink, then

I basically add something to the existing knowledge saying okay uh my theory is correct. Uh and my in in that

sense by extrapolation my theory or say for example this given theory about why things are easier to read in red might

be correct. I'm adding to some evidence that uh can be cited in support of this particular theory. So ideal experiments,

critical experiments, they typically try to compare two theories. So you can have another theory. Somebody can have a

theory that says no uh red is not the best uh you know red uh ink. It's probably uh yellow for reasons whatever.

Okay. So you have one theory that says red is good. Other theory says that yellow is good. You conduct an

experiment. You have red and yellow and blue and green inks and you get some data and then whatever the data does uh

you whatever the data you gather you analyze it and the data will either support red or green and in that sense

you'll have a theoretically uh you know driven hypo theoretically derived hypothesis which have been tested and

new data is created. I have another example here. For example, uh you know the complimentary

theory of literalization predicts that uh language and spatial functions are lateralized to the opposite hemisphere.

So for an individual uh people who are called typically lateralized, language is found in the left hemisphere, spatial

functions are found in the right hemisphere. There are atypically literalized people as well in whom

language is found in in a very small percentage of people. Language is found in the right hemisphere and uh special

functions in the left hemisphere. Now if you look at the typical and the atypically literalized people you get an

idea that okay if language is in this side then uh the uh special functions are on the other side. This is what is

called complimentarity. So there's this theory of literalization. The other theory says that these two functions are

statistically independent and is it does not matter where lang I mean basically uh the compliment complimentarity does

not hold. So it is possible for us technically to have both language and spatial functions in the same

hemisphere. All right. So that can be the statistical independence view of lateralization. Now uh a student say for

example uh let's say I am the student who has been given this task. I can conduct a within subjects and same

people uh you know study of literalization of a verbal task let's say word recognition or word naming and

a spatial task let's say symmetry recognition or something and I do this for all my participants and I basically

collect reaction times and so on okay now whatever I find let's say I find uh that for all of my participants so this

is what was here if I find that for all of my participants language and spatial functions were in opposite hemispheres.

These results in principle favor the complimentarity hypothesis of literalization. So I get some data. I

did an experiment. I wanted to compare the two theories statistical independence and complimentarity. My

data tells me that there is evidence for complimentarity. All right. Now are these results

definitive and final and do we stop doing research or lalization after that? No. Absolutely not. Uh there is uh you

know a lot of uh experiments on both sides and there is evidence sometimes favoring this one and that one and a lot

of times what happens is people you remember Popper said that we do not throw uh the throw away the theory at

the first falsification pursuate we sort of uh you know uh test more nuances perform different tests and then only we

junk a particular theory. So basically what will happen is that these are not the final results because the supporters

of these other theory uh will actually find interesting ways ingenious ways to discredit the unfavorable

interpretations that are found in my experiment and then they'll conduct more experiments uh they'll design them

slightly differently or probably in the same manner and try and find more uh ways to basically uh you know test the

given theory. Okay. So this is one kind of experiment. Critical or ideal or whatever you want to call it. The other

kind of experiments that are done are called exploratory experiment. Now as the name suggests exploratory

experiments are explorations. They're not typically theory theorydriven. So you don't have a theory driving the

hypothesis and driving basically uh you know already providing you a background of interpreting whatever you may find.

These are basically what if experiments. What if I do this? Let's say what if I manipulate the color of the street light

uh from uh you know yellow to something else maybe blue what will happen then uh or say for example from red to blue now

we will not have red and green uh as the go and the stop signals resp you know as the stop and the go signals respectively

I can have let's say I don't know maybe blue and pink or something like that okay so these are whatif experiments

there's no clear theory behind them. Uh they're basically think okay let's do this and let's find out what happens. A

lot of times uh when you're reading about particular phenomena when you're reading a lot of literature and when you

are observing uh you know things and the environment around you uh you might want to use those observations to create

experiments. A lot of times you might find by the way uh for the most part uh if there is a hypothesis that you come

up with it's always a good idea to perform a good literature search because the literature search will tell you that

maybe somebody has thought of this before. But let's keep that aside. A lot of times we don't do that and we say

okay let's do a what if experiment. Let's try this out and see what happens. These experiments as I said require no

prior theory and are typically formulated based on personal experience and observations and are therefore very

popular amongst uh you know new students you know bachelor or even uh you know lower levels. Some scientists obviously

they frown upon these experiments citing their inefficiency because they don't break new ground so to speak. Also if

nothing much happens in this kind of experiment there's nothing gained from these experiments. On the other hand, if

nothing much was happening in a critical experiment, that null result would also be useful. However, if a desired effect

emerges in this what if kind of experiment, we still don't have any theoretical grounding to explain why

this is say for example why do red uh why do blue lights work better for example than red lights as stop signal.

Suppose I did that experiment. I went out and I found uh that people actually stopped better uh when blue light was

given as compared to when they were stopping with red. Now uh if I did not start with a theory or if I do not start

with a hypothesis or a prediction, I do not have a way to explain these findings as well. I will go fishing so to speak.

I will go uh and try and find out reasons why this might have happened. And it's not an entirely uh you know

futile exercise but it is a decent uh uh you know uh it's it's a slightly radarless thing to uh you know initiate

but obviously you read more you uh get uh uh you know more information and then you can develop explanatory uh

speculations which can be sharpened uh from there on. Okay. Other kinds of experiments say for

example you can do direct replications. It's something uh that is uh you know rather a buzz in cognitive psychology

and cognitive science recently. There are a lot of classic experiments uh do not get replicated and u you know

there's a series of papers in the last 5 10 years which have said that there is a replication crisis in uh cognitive

science. So it might be a good idea to when you are setting up your first experiment just pick up a a good famous

established experiment and do a perform a replication. See whether things happen in exactly in the same way. Obviously

there are conditions for performing a replication. You have to try and appropriate the exact same experimental

conditions and exact same manipulations that were done in the original experiment. There are also extension

experiments. For example, if you uh you know extend the procedure and uh manipulation maybe by adding a level or

two and something and uh sometimes it yields new findings. Sometimes it helps to add on to the already existing

results. So uh that is the kind of experiments you can do exploratory uh critical experiments replications

extensions and so on. Now uh we were talking about this uh you know concept of independent variables

earlier. Uh that these are the ones that are controlled by the experimentter. Uh broadly they consist of things like uh

stimulus characteristics such as brightness, color etc. Since the experimental is uh you know deciding

their quality or quantity. Now they are typically selected why why do we select a given level of an of an independent

variable? they are typically selected as I was saying just now uh because the experimental believes that they will

influence the behavior of the individual participant. So ideally why would I select let's say uh you know red ink for

uh better legibility or better memorizability because I have a theory that tells me that red uh color is uh

you know uh recognized faster and if it is recognized faster it'll probably have uh slightly uh you know more elaborate

deeper perceptual representations and so on and so forth. So I know from the theory I have a belief before I have

done this experiment that red will be remembered better or recognized faster because of so and so. That is why I will

choose that uh particular uh you know uh independent variable. Obviously it might completely fail and no difference

emerges between red ink and blue ink and that will be referred to as the null result.

Now null results are also interesting when the manipulation does not work. So null results can be interpreted uh you

know for in independent variables null results can be interpreted in more than one way. Uh one way would be to uh you

know basically it is possible that the experimenttor has incorrectly guessed the importance of the independent

variable or the experimenttor has not carried out a valid manipulation of the independent variable. Let's let's uh

dive uh in and see. So if the hypothesis uh was that the let's say in this new example if the hypothesis was that the

presence of caffeine in coffee leads to alertness or you know vigilance in participants after they have had a cup

of coffee uh and in an experimental condition and individual is given coffee with predetermined amounts of caffeine

you can nowadays measure it uh and it does and it does not cause that kind of alertness that you're looking for. It

can be taken to mean two things. For example, uh you can either conclude that uh no caffeine does not cause alertness

or you can basically uh say that uh you know amount of caffeine in the coffee was not enough. So uh caffeine does

cause alertness but beyond a certain threshold. So a certain kind of uh you know concentration of caffeine in the

coffee must be there. Nowadays there are different kinds of decaf coffees also available. Okay. It is also possible

that prior to the experiment the participant had saturated the levels of caffeine in their system and hence the

addition of more caffeine was ineffective. So that is also possible. So basically it is it is there that if

you found a null result it could be because your uh manipulation at the level of the independent variable has

failed. It can fail in let's say we've seen two three different ways and that is something that the experimental will

need to be mindful about. Now coming to dependent variable as we just said it is the response measure of

an experiment is the participant's response to the experimental's manipulation of the environment or the

independent variable so to speak. Now in the previous example we talked about the degree of alertness caused by

caffeinated coffee versus decaf coffee versus let's say no coffee. Now what is the dependent variable here? Because we

wanted to measure alertness. So any measure for alertness will become your dependent variable. Okay, that I'm going

to you can basically say that I'm going to measure alertness by the uh response time to a probe stimulus. Okay, or you

can basically say I will measure alertness by the characteristic of uh the eyes eye blinks and so on and so

forth. So whatever you are using as an index for your critical behavior is your dependent variable. Okay. uh a very

interesting criterion for dependent variables uh you know for a good dependent variable is stability. Let's

let's uh look at this in some detail. Uh a dependent variable a good dependent variable is one that does not vary too

much by itself. Okay, it should be stable in that sense. So uh if the same experiment is replicated with the same

participant under the same experimental condition and levels of the independent variable the dependent variable should

give you the exact same values or near about the same values. If uh you measure in keeping all things constant, if you

measure the dependent variable one time, two time, three time, four time and five time and if it gives you different

values across the five times, then it is not a stable dependent variable and might not be a great choice for

measuring uh you know the effect of the IV manipulation. Also, inadequacies in the measurement of the dependent

variable might also lead to uh null results. Uh sometimes uh even say for example if it is a stable dependent

variable for example the most common case of null results could be that you're sampling from a very restricted

or narrow range of dependent variables. Uh okay uh sometimes there are flow effects. So remember we talking about

this fact that you were talking about uh you know we talking about alertness. Now uh if there is a very high level of

alertness and all the participants are typically alert uh then it is uh significantly high level uh everything

is at ceiling. So whether you give caffeine or you do not give caffeine will not make any change in the

dependent variable. Okay. So that are ceiling effects or if everybody is in a highly uh you know drowsy state and so

on and it's completely rock bottoming uh then also it does not give you a range of values. So if you are sampling the

dependent variable from either end of the continuum then you're not uh giving yourself a great chance of finding the

effect of the manipulation. In that sense the experiment might not uh uh run well and will give you null results.

Okay. So uh that again is something that we should talk about. Finally the third kind of variables that we can talk about

are control variables. What are control variables? Uh control variable is typically uh you know a variable that

must be held constant. Remember we talking about all other things being constant. So must be held constant

during an experiment so that it does not influence the dependent variable or interacts with the influence of the

independent variable. Okay. So the manipulation of the independent variable should be kept pure.

Let's say uh what are these control variables? Actually for anything the control variables for any kind of

experiment you can think of large number of control variables. Okay. For the reading legibility example, the

difficulty of text can be something. Let's say we control for it. Uh the IQ of the individual who's taking the text

uh who is reading the material can be uh you know uh variable you want to control for that. The gender could be something

you want to control for that. The mental emotional state can be something you want to control for that. So for any

given experiment you can think of 10,000 things that must be controlled uh for uh you know the IV manipulation to show its

effect. Now uh will we do that all the time? Will we be able to do that all the time? No. So there are two things. There

is controlling for something by matching and there is controlling for uh these uh variables by randomizing. Okay. Again

either you can match. So for example things like gender uh IQ uh educational qualification you can match uh things

like uh for example uh what did the person have in the breakfast? uh what kind of emotional state the person is

going through uh what kind of uh conversation the person had just before walking in the experimentalist room

these are the things that you cannot really go out and control for so and you know match in in that sense control by

matching so what will you do is you'll basically assume that uh I sampled 40 people that's why uh you know sampling

and uh a decent enough sample size is important because it takes away the statistical variation in the large

number of these uh you know uh control uh variables and in that sense provides you with a decent chance of measuring

the effect of IV manipulation on your DV. Okay. So uh yeah this is something that is extremely important. Now holding

these extraneous variables constant is obviously not the only way. There are statistical ways and we'll talk about

them when we uh go deeper into this. But uh do remember the fact that null results uh do emerge in an experiment if

there is insufficient control of these other extraneous factors because uh sometimes it could be these other

factors. Time of the day you were performing your experiment in a room uh that had too much uh background noise.

You were basically testing your participants for an attention experiment at the end of the day and everybody was

uh extremely tired or distracted or something like that. So uh these extraneous variables may have an effect

on your overall experiment if they are not thought about in detail. All right controlled for match thought about and

so on. So this uncontrolled uh variation uh of extraneous variables can sometimes actually obscure the uh or even inflate

the effect of the independent variable on the dependent uh on the DV as it is more uh you know common when and this is

something that remember happens in the naturalistic research designs and so on. Okay. So that is why uh if you want to

have a good idea of uh you know causality controlling these variables is extremely important. Okay. So uh there

are a few examples uh in Kowitz and colleagues's book. I've already given you some of my own. I'll take one or two

of uh them here. Let's say a pigeon is trained to peck a key. Now we have to talk a little bit about uh you know

measurement. How do you measure a particular variable? you know what is it that you basically measure something as

so we'll talk about this operational definitions for example a pigeon is trained to peck a key if a green light

is illuminated but not if a red light is on uh here correct pecs are rewarded by access to grain okay so uh how do we get

these variables so we've talked about independent variable dependent variable and you know extraneous variables let's

leave the extraneous variable for now what is the independent and dependent variable here so uh basically the number

correct PEX could be the dependent variables uh and say for example uh you know what the IV is the you know uh the

levels of the light that are coming red light and green light and so on. Okay. Uh other thing is a social psychologist

does an experiment to discover whether men or women give lower ratings of discomfort when six people are crowded

into a telephone booth. So who gives what kind of ratings? Remember men and women gender is is one uh ratings of

discomfort is basically the dependent variable and so on. And so let's let's see what are they saying. So they're

saying here uh independent variable gender of participant uh in this fourth example dependent variable ratings that

are given and control variables what kind of phone booth we're talking about very small. So six can anyways not get

in very large then six might be comfortably placed and so on. In this uh second one with the pigeon color of

light is the independent variable. dependent variable as I said number of keyps control variables could be how

hungry is how you know hungry has the pigeon been uh you know so that how motivated he is to you know peck at the

light to get food grains all right so this is one way uh to talk about these things now uh sometimes we can have more

than one independent variables as well and uh that probably uh gives us a chance of controlling the situation

better and measuring the outcome in in more detail let's say you know We're talking about

lateralization in the beginning. Let's say I want to study the lateralization of objects versus non-objects or tools

versus non-tools. In such a design, we'll have at least two dependent variables. We have similar type object

non-object the other one will be visual field. Where am I presenting in this? Because lateralization studies, you

present something in in one of the visual half fields and then you compare the performance in this visual half

field as compared to that visual half field. So, uh one is type of stimulus tool or object. uh the other is the

visual field, left visual field or right visual field. So uh again there are a lot of experiments with multiple uh

independent variables up to four and so on. So we we'll talk about them when we talk about uh other kinds of designs.

Also uh having more than than one independent variable is basically been found useful for a few reasons. Let's

talk about them. First, sometimes it is considered uh more efficient to conduct one experiment with say three

independent variables to conduct uh you know than to conduct three separate experiments. You know the the principle

of economy. You cannot be uh conducting uh any number of different experiments again too many experiments or you can

manipulate keep everything uh you know controlled and then manipulate the three variables within a given experiment.

Second, experimental control is sometime better since with a single experiment control variables, all of the things

that we talked about can be controlled all at once. Uh if you do experiment one, experiment two, experiment three,

you're not sure of what went wrong here, something else might have gone wrong here and something completely different

might have gone wrong here. Put all of them together, have three variables, control everything to the best of your

capability. That might be a better and more economic design. Generalization is more easy. uh and finally interactions.

It allows us to study interactions as well. Okay, there is an example that the Kanderitz and colleagues uh book have

this uh research article titled when God sanctions killing. So here what happened is that participants were made to read a

violent passage that uh either it was told to them that it was either came from the Bible or just from an ancient

scroll. uh and after that participants performed an additional task that allowed them to uh you know present loud

sounds to another subject another participant in the next room. Okay, they control these participants are

controlling the intensity of the sound and higher intensities were uh interpreted as being violent. So if uh I

have a choice to present uh you know some kind of loud white noise to my uh co-experimenttor co-uh co-articipant

sorry uh I can reduce it to being very comfortable or I can increase it to being extremely noise extremely hard to

sort of listen to. What is the dependent variable here? The dependent variable is basically the number of times the

participant selected this highest setting uh which is going to be most com uncomfortable for the co-articipant.

and in a set of 25 trials. So you can see here what is found. The results are interesting. Reading a

passage from the Bible produced greater aggression and subjects who believed in God acted more aggressively. You can see

here when people believe that the passage came from Bible, they are more aggressive. Uh when they believe in God

then they are also more aggressive. Uh but are these two different things or are they linked together? Let's talk

about them. That is where you see interaction. Now here you can see this is the uh you know figure from uh the

paper and here you can see something very interesting when basically you know there was no mention of God in the in

the first part here you can see because the passage was from an ancient scroll subjects who believed in God and

subjects who did not believe in God showed similar uh tendencies for aggression. However, when God had

sanctioned the violence, say for example, the passage is coming from Bible, uh, greater levels of aggression

were exhibited by those subjects who believed in God. So again, uh, this is a study by Bushman and colleagues. It's a

very sort of a loose interpretation, but the idea here is that you can now see the interaction of two variables. the

source of uh the uh you know the text the source of the passage as well as the person's individual's belief uh in uh

you know god and so on. Okay. So you can now see the interaction here. This is basically what is referred to as the

interaction effect. There are other things also. For example, if you remember the lateralization example, uh

we had two visual fields. We have two kinds of stimula. uh what we are actually looking for is what kind of

similar is processed in which visual field better and that is basically uh you know given to us by the interaction

plot. Say for example you find that tools are recognized faster in the uh right visual field and slightly slower

in the left visual field. That will tell us that the uh right uh sorry that the left hemisphere processes tools better

as compared to the right hemisphere. That kind of thing is is interactions. Okay. Uh sometimes you can have more

dependent variables as well. uh typically having more dependent variables is is useful because it

provides us with converging evidence of or converging indices of behavior. Okay. So typically you will see in most

studies reaction time and accuracy are together considered in some cases you can have reaction time accuracy and some

kind of tracking phenomena. You can have reaction time accuracy and some kind of bio feedback phenomena. You can have a

range of things here that will basically provide you uh you know a more uh significant uh idea of how did the

participants behavior vary in response to your manipulation in the independent variable. Okay. Now how do you come up

with these dependent variables? There are various ways. Let's take this example. This is from Stanganger's

introduction to psychology. Now you can see here aggression can be measured by the number of presses of a button that

administers shock to other student. It can also be uh measured in terms of number of seconds it's you know taken to

honk the horn at the car ahead uh after a stop light turns green. So there are different ways in which you can

operationalize your uh you know conceptual or dependent variable and basically uh in in some cases you'll

find that you can use more than uh you know two ways more than uh you know two three ways uh in the same experiment to

give you the best idea of how uh you know the dependent variable or say for example how the behavior is changing in

response to your independent variable manipulation. Okay. So that's broadly the overall uh you know uh scheme of

where experiments are. Uh in the next lecture I'll talk uh in in more detail about the mechanics of uh you know how

uh do we conduct experiments, what are the principles behind it and so on. Thank you.