Essential Guide to One-Way Experimental Designs in Cognitive Psychology

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

department of cognitive sciences ID Kpur and we are now in week five of the course. This week we will discuss

various aspects of experimental research designs. We'll talk about oneway design, special designs. We'll also probably

talk about issues that must be taken into account when we are designing experiments such as reliability,

validity and so on. So let's begin. In this lecture, I going I'm going to talk to you about oneway experimental

designs. Let's let's talk about that. Now we have seen so far that the main aim of psychology uh is to study

behavior and its causes. All right. Now the different methodologies if you look at the qualitative and the quantitative

or even within the quantitative if you look at the descriptive the correlational and the experimental

research studies these basically vary in their ability to get closer to the cause behind a particular behavior. The goal

is always to be to be getting as closer to the cause behind behavior as possible. So the task that we are trying

to do here and we will discuss this in terms of experimental research is to estimate for example how the changes in

the independent variable are producing corresponding changes in the dependent variable. So you can pick up any

independent variable say for example the hours of sleep and productivity or say for example uh the amount of caffeine in

a person's body and overall uh you know arousal levels you can take any number of independent variables and you are

varying them and basically what you want to study is how does this variation that you producing in the independent

variable is causing changes in the dependent variable whichever dependent measure you take say you can take

performance you can take arousal you can take any number of things. Now the deal here is that you want to be as sure as

possible that it is only the variation in the independent variable that you are creating which is causing the changes in

the dependent variable. It is possible and we have seen that in correlational research that while you are sort of you

know placing a bet waging uh you know this on your independent variable that you are varying but there are common

causal variables there are spurious variables possible that are actually affecting both that are actually

affecting the independent variable as well as the dependent variable. there are also other factors that are probably

affecting the dependent variable but uh you know uh in addition to the effects that are being caused by your

independent variable. So these are some of the things that we try to take care of when we design good experiments. Let

us try and see uh where it goes. So causality basically is the main aim as I said. Now Stanganger emphasizes that

before you sort of get into you know this business of determining causality you have to figure out that there are or

you have to pay attention to three very important factors. All right the first is association. Now even before we can

sort of arrive at a causal relationship between two variables let us say amount of caffeine in a person's body and the

overall level of arousal that a person is experiencing. There must be some association inferred between them that

see for example if these are not related at all obviously there is no causality that you are eventually going to find.

So typically we start looking for causality when we think that there is this variable A which is in some way

associated with or affecting the variable B. There is something that is going on between these two variables

which we want to study better through our experimental designs. All right. So for example the uh you know the

experiment uh we have been talking about on and off in this uh course is this one uh by Sanger that say for example

watching violent television programs causes aggressive behavior in children. Now if why should we look for a causal

relationship between them? We should basically we will be interested in looking for a causal relationship

between watching violent television programs and aggressive play in children if there is some relation that we have

seen or observed in the past. So if there is a positive correlation, if it we have observed that indeed watching

violent TV programs has increased uh you know the amount of aggressive play that children are engaging in then obviously

we will sort of zoom in and we'll try to figure out whether these two variables are causally related or not. Also we

have to be careful of the fact that a lot of times the causal relationships that we are talking about or the

relationship between the variables that we are talking about are not definitive but probabilistic. It is not going to

happen that every time a child watches a violent video game it engages in aggressive play. It may happen. There is

a higher probability of the fact that when the child watches uh violent video games or violent television programs,

they may engage with a higher probability than otherwise they may engage in aggressive play. All right.

The example that Sanger has taken here say for example cigarette smoking causes lung cancer. Now the thing is a lot of

people do smoke cigarettes. All right. And obviously not each one of them is afflicted with lung cancer but there is

a higher probability these people are if you look at you know health websites and data they say uh doing certain uh things

say for example smoking cigarettes places you at a higher risk for contracting lung cancer. It does not say

definitively oh you smoked the cigarette you're going to have lung cancer. All right. So the nature of causal

relationships also has to be slightly uh understood. Now temporal priority when you want to

infer a causal relationship it is important that the variation in the independent variable has happened first

and after this variation that you have created you have measured the effects on the dependent variable. All right. So if

event A occurs before event B then it is possible that A could be causing B. Say for example you take aspirin and your

headache goes away. It is possible that taking aspirin has made your headache go away. But say for example if your

headache went away already and then you took aspirin obviously it will be difficult to infer that uh taking

aspirin had anything to do with your headache in the first place. All right. So in the same way just continuing with

the previous example if children view a violent television show before displaying aggressive behavior it is

possible that viewing that violent television program may have caused such behavior. But if the children were you

know uh displaying aggressive behavior beforehand, they were already engaging in aggressive uh behavior even before

they were exposed to violent television programs then it will become difficult to infer such a causal relationship

between the two. So two things temporal priority and the you know the idea that we were talking about earlier the

association and temporal priority. The third thing that you want to talk about is is controlling common causal

variables. So to be able to make causal statements you have to be sure the experimentter needs to be relatively uh

sure and should be able to rule out the influence of common causal variables. Remember I said say for example you're

talking about uh you know caffeine and arousal level that a person is experiencing. Now there could be other

common causal variables such as the time of the day such as the amount of rest that a person has already had or the

amount of rest the person has not had. All right. So there could be in in a bunch of cases when you are looking for

causal relationships between variables through your experiments, there could be common causal variables that are

probably uh you know adding on uh to the effect of your independent variable manipulation and that is something that

one should be extremely careful about. Uh in an in an ideal experimental setting, you would want to rule out all

other possible explanations that could uh account for the variance in your dependent variable.

All right. So the idea is that the experimental needs to be sure that the effects of the dependent variable are

solely and only due to the experimental manipulation in the independent variable. So these are the three things

association, temporal priority and control of common causal variables that have to be attended that have to be

taken care of when you are uh looking for causal relationships between variables. Now let us talk about one-way

experimental designs. Now one way experimental design is the simplest experiment that you can uh do. You are

basically having one dependent variable and one independent variable. You are varying this one independent variable uh

in different ways. Sometimes it is just the presence or absence. Sometimes it can be two level. Let's say at level

five and level 10 something like that. Remember we've had a bit of this discussion when we are discussing uh you

know experimental designs uh in in the previous weeks. Okay. Now, so you have one independent variable, one dependent

variable and you basically vary the independent variable let us say at two levels or just with presence or absence

let's say presence of caffeine or absence of caffeine or I can have an alternative hypothesis presence of 50 mg

caffeine and 100 mg caffeine. So I can talk about both of these things. Now what does this level of caffeine

presence absence or 50 or 100 do to let us say the arousal levels or the sleepiness levels of an individual that

is broadly what let's say we want to study now uh to understand these in more detail let us take an experiment you

know the ones tanganger has taken in his book and we will discuss various aspects of these uh you know the experiment

going forward. So what happens here? 20 fourth grade boys and girls were made to watch a sequence of five cartoons that

had been selected by a panel of experts and deemed to be extremely violent. Another 20 children similar grade boys

and girls watched a series of nonviolent cartoons. All right? So for example, maybe uh one set of children watched

something similar to uh you know a Marvel uh comic uh cartoon or a DC comic cartoon and the others watched something

like a Peppa Pig or something like that. Okay, maybe just uh imagine uh that scenario. Now after these children uh

have been exposed to the you know to the cartoons violent and nonviolent, they are taken to a play area where they are

allowed to play with toys. Okay. So there is a play area and there are lots of soft toys this that which is present

there and a team of observers behind a particular glass is basically who are not aware by the way. So these team of

observers do not know which children have seen violent uh video games or violent cartoons and which children have

not. Now they are basically coding for the aggressiveness of play of these children. So there are 40 children all

mixed in the play area. There are these observers who do not know which children have watched what uh video uh and they

are basically noting how aggressively these kids are playing. All right. So the research hypothesis let us say that

uh we can have at the beginning is that the children who had viewed the violent cartoons would play more aggressively

than the children who had not viewed or who had viewed the nonviolent cartoons. All right? So somebody who's uh seen say

for example Snow White or Peppa Pig or something like that will play less aggressively than somebody who's seen

you know a Batman or a a Superman kind of a thing. So this is basically the the schematic.

You can see participants are randomly assigned to the two conditions. So initial state boys and girls fourth

grade maybe other factors all are uh you know uh sort of matched and initial equivalence has been created. Uh this is

the experimental treatment. Half of the group watches violent cartoons, half of the group watches nonviolent cartoons.

Now the dependent variable is aggressive play. This is what we want to measure. And basically what we want to determine

as a result of our experimental manipulation is how much aggressive play is here compared to how much aggressive

play is here. Basically calculating the difference. That is broadly what we are going to do

now to ensure that the independent variable has occurred prior to the dependent variables. In this

experimental design, the independent variables are manipulated. Okay, they are manipulated before uh the dependent

variable is measured. So in the current experiment you can see uh if you remember just I showed you the schematic

the independent variable was the type of cartoons that the children were made to watch. They watch the type of cartoons

first and only after they have watched the type of cartoons their aggressive behavior or their aggressive play is

coded for. So uh we can refer to uh this independent variable as the cartoon type to indicate that the manipulation

involved children viewing either violent or nonviolent cartoons. These different aspects of the independent variable are

called levels. See, we can have presence absence or we can have level one uh you know we can have degrees of that but

we'll talk about it at some point. Now those different levels of the independent variable are referred to as

levels. That is violent cartoon is one level, nonviolent cartoon is another level and these can also be referred to

as conditions. So there are two conditions in our experiment. In one condition the kids watch violent

cartoons. In another condition the kids watch nonviolent cartoons. Now remember we talked about initial

equivalence. Look at this. We talked about initial equivalence. Even before the independent variable can be

manipulated even before the treatment is in play. We have to create initial equivalence. There should not. So the

the participant should not enter the experiment already with differences. You are assuming as an experimental that

before you uh create the experimental manipulation uh the participant sample that you have is equivalent to each

other in all respects. Okay. So typically you'll see in in behavioral experiments what we do is we take care

of handedness uh vision we take of prior prior experiences uh you know tiredness fatigue and so on. So let us see what uh

can be taken uh you know uh care of here. So while the experimental manipulation ensures that the

independent variable has occurred prior to the measurement of the dependent variable, it allows uh also the

researcher to rule out the possibility of common causal variables. Variables that could have caused both the

independent and dependent variable. So when you create this temporal priority, you take care of that common causal

variable factor. Now in experimental designs, the influence of common causal variables is eliminated through the

creation of initial equivalence. If say for example you have created two groups of participants both of which are

equivalent and matching to each other in all respects say for example let's say just for case of simplicity uh although

we just mentioned that the sample has both boys and girls if I were to do a simpler experiment I would say 20 boys

20 boys all fourth grade let's say the same age similar familial background similar parented disciplinary style uh

and any similar uh you know u I don't access to resources this that so any number of things. So it becomes the

prerogative of the experimentter to think of any number of possibilities any number of possible variables that can

have an effect on your dependent variable and then rule them out match the uh participants match the two

participant groups on the same uh you know on each of these criteria. So equivalence can be created that's

what we are talking about equivalence can be created either through using different but equivalent groups. So you

can have one group let's say group one watches violent video games group two watches nonviolent video games. You can

have that or you can have the same people you uh going through the two conditions. So that basically uh you

know reduces your burden of matching because it's the same participant who first watched a violent video game went

for aggressive play after some time watches a nonviolent video game and went for aggressive play which is being

measured. Now how do you create initial equivalence? One of the ways for

creating initial equivalence is uh assuming random or it's basically implementing random assignment to

conditions. Now the principle is very simple. There can be any number of variables and there can be a bunch of

variables that say for example the experimentter cannot even think of which might let's say have a potential effect

on the dependent measure. So what does the experimentter do? Experimenttor assumes that let us say there are

variables 1 2 3 4 5 up till n and there are variables in this group also 1 2 3 4 5 up till n. There are these two groups.

Let us say that on one variable this group is higher on the other. this group is higher on one variable. This group is

lower the other this one is lower. So there are these variations that are possible between the samples to account

for or to basically take care of circumvent this variation. What can be easily done is you can basically start

randomly assigning these participants to the two conditions. So sometimes what we do is uh you know participants turning

up at uh even number sequence are assigned to condition A. Odd number sequence are assigned to condition B and

you can go with that. Now what does this achieve for the experimental? It basically says that say for example this

group varied on a particular variable conf particular extraneous confounding variable to a certain level. You would

assume that by random assignment this group would also have some level of that variable and this and this group both

having those that variable cancels out the effect and does not uh you know create potential uh you know

difficulties on or potential uh influences on your dependent variable. It basically preserves that okay given

any number of random variables 1 2 3 4 5 6 if you have random assignment the idea is that they will both they will each

cancel each other out and then the only effect that you will be observed on the dependent variable will be of the

independent variable that you are manipulating that is that is the broad idea I'll read through the slides in a

bit all right so in a between participants experimental design the researcher would compare the scores of

the dependent variable between different groups so there is group A, group B. However, the participants in the groups

are activated even before the manipulation occurs. I just said we talked about creating initial

equivalence. The most common method of creating initial equivalence is random assignment to condition. All right, just

just as I was describing. Now, random assignment involves the researcher determining separately for each

participant which level of the independent variable he or she will be treated to. So for example participant

comes you have to decide whether he goes to the violent condition block or the nonviolent condition block. The

researcher can do this by say for example just randomly as I said odd number even number thing you flip a coin

uh you do that you basically pick numbers 1 2 3 4 5 6 and you say 3 7th and 19th will go to this condition

others will go to this condition you can choose any mode of bringing random selection here. So what is this doing? I

will just repeat myself so that it is clear. Random assignment involves the researcher to uh you know separate these

simple random samples of participants to be in each of the level of the independent variables or each of these

conditions that we were talking about. And since the samples are drawn from the same population, the researcher can be

relatively confident that before the manipulation occurs, the participants in the two conditions are actually on

average equivalent in every respect except for the differences in the condition. All right? So when you do

random assignments, half of the participants go for watching the violent video game, half of the participants go

for watching the nonviolent video game. Basically what you're achieving is that prior to that everything else is equated

for because you just randomly put people here and there and whatever uh other variations might have might have been

they canceled each other out. Now in the current example coming back to this experiment as the children were

randomly assigned to conditions those who are going to view the violent cartoons will on average be be

equivalent to those who are going to view the nonviolent cartoons in terms of every possible variable that the

researcher could have thought about. Okay, hormones, parental disciplines, soic status, any anything that you want

to uh talk about. Now random assignment to condition ensures that on average uh you know the score of these variables

will be the same for the participants in each of the two conditions. Now this is something that you have to take care

about is that although random assignment does not guarantee exact equivalence, it reduces the likelihood of differences.

See if you want to sort of really uh pinpoint and zoom and control for each uh variable uh one of the ways could be

to actually match the two groups on each variable that you think is important. Okay. So what you can do is you match on

gender, match on age, match on soio economic status, match on parental disciplinary style, match on any number

of things that you can think about. Now the problem is that it typically when we are doing experiments there are

potentially a large number of experiments that are there. Say for example the mood in which the

participant came to the experiment lab. Uh the morning the kind of morning that the participant has had. Say for example

he woke up uh to a pleasant morning or just woke up feeling cringy and feeling sort of uh you know not very happy. Now

the thing is since the list of these variables is too large and you cannot possibly match on every conceivable

variable that is basically why the researchers go for random assignment. They assume that

okay if somebody has gotten on a very bad day in the morning somebody probably have gotten on a you know much

pleasanter day and the these effects will cancel each other out something like that. All right. So the likelihood

of chance differences between the two conditions is reduced even further and it basically increases as the sample

size increase. The power and number of measurements increase. Now how do you select the dependent

variable? Remember we've talked about this fact that you have to have an operational definition and you have to

basically uh be able to measure something. But what dependent variable do I go with? Okay. So uh typically

experiments have one or more measure dependent variables which are designed to assess the state of the participants

after the experimental manipulation has occurred. So after you've shown them uh the violent video games or violent uh

cartoons, you want to measure how much how has aggression changed. It will depend upon your operational definition

of aggression. How do you want to measure it and accordingly you will basically uh you know just code for

that. Say for example the number of time a punches a kid punches a Bobo doll that has been done a number of time a kid you

know pushes other fellow kids or let's say a number of time a kid say for example uh throws a toy here and there

you can you can think of any number of things okay so current example let's come back the current example dependent

variable is a behavioral measure of aggression but any of the many types of measures could serve as the dependent

variable as I was saying you can take anything that you think approach appropriately measures aggression.

Obviously the number of time the kid pets the Bobo doll or say for example you know kids have this thing of uh

taking the uh doll uh you know in their arms obviously you know that those don't account for aggression and you will not

choose them any and every measure of aggression as long as you're sure it has construct validity we'll talk about

validity at some point you can choose that all right now the research hypothesis in in an experimental design

is that after the manipulation has occurred the mean scores on the dependent variable across the two groups

will be significantly different from each other. Okay. So let's go back. What we are trying to say here is that after

the manipulation has occurred. See you started with the initial equivalence you uh created a difference by manipulating

the independent variable and then you're measuring the aggression score. The idea here is that the mean aggression score

here will be significantly different than the mean aggression score here. That is basically what we are trying to

achieve. Okay. If the experimental does observe significant differences between the

conditions, then he or she can conclude that the manipulation basically has worked and the manipulation has caused

these differences because even because prior to the manipulation everything else was equal. The participants would

have received equivalent scores on any measure of aggression that you would have. All right? So the only difference

between the participants in these different conditions would be that they experienced a different level of the

experimental manipulation. So they were all exactly the same bunch. Some of them watched the violent cartoon, some of

them watched the nonviolent cartoon. And now whatever difference we are seeing in aggression scores is because they

watched the kind of cartoon they did. Uh there is also number of levels. So for example, your independent variable

has can have levels. So let's talk about that. Now experiments can differ in both the number of levels of the independent

variable and the type of the manipulation that you have used. Okay. So in a simplest experiment designs

typically you have only two levels. You have either presence absence or you have two degrees. So sometimes as I am saying

sometimes it could just be the presence of a variable or absence of it. So viewing violent cartoons versus viewing

nonviolent cartoons. You can also have viewing a violent cartoon and not viewing anything at all. All right. In

such a case, the level in which the situation of interest is created because you wanted to measure the effect of

watching violent cartoons. So, watching violent cartoons is treated as your experimental condition and watching

nonviolent cartoons is treated as your condition of uh control. So, it's it's your control condition. So, there is an

experimental condition, there's a control condition, experimental condition where the manipulation is

occurring. A control condition which can be used as a baseline. Remember, we've discussed this earlier as well.

you have to add control conditions so that it provides some reference against which thing is you know against which

aggression is measured. So there can be different kinds of control uh and the experimenters uh need to decide about

which kind of control they want to use. All right. So the control conditions normally designed to be the same as the

experimental condition in all respects. Just the watching of the cartoon uh the type of cartoon is different. That is

why a better control condition is not absence of watching cartoons but it is the some cartoon here also people are

seeing cartoons here they are seeing violent cartoons here they are seeing nonviolent cartoons.

So in the current example for example for example the children in the control condition watched uh violent rather than

nonviolent cartoons. In another instance, it could be possible to let some children uh watch 10 violent

cartoons and other children watching five violent cartoons. You can play with the degrees as well. In in this case,

five can be treated as a baseline condition. 10 can be treated as an experimental condition. Here, watching

violent cartoons is treated as an experimental condition. Watching nonviolent cardons is treated as a

control condition. You can obviously add more levels. And it is important sometimes I'll tell you why. So uh

typically which uh you know with experimental designs which have only two levels it may be useful for testing some

kinds of hypothesis. So presence absence kinds of things. Okay. But it probably creates some limitations. What kind of

limitations? Remember we are talking about estimating a curve. If you have only two points it'll be difficult to

know the nature of the curve. But if there are strategically sampled points across the overall curve then you are in

a better position to depict the nature of the curve. remember that discussion we've had. So, uh, now we are seeing it

in action. So, in current research, children behaved more aggressively, if the children behaved more aggressively

after viewing the violent cartoons than they did after viewing nonviolent cartoons, the experimenters would be

able to conclude that nonviolent cartoons decreased. In principle, nonviolence cartoons

decreased aggression rather than the violent cartoons increased aggression. That is possible. It could be reasoned

that children who watched the nonviolent cartoons got bored and were just too tired to play aggressively. All right.

So a suitable control condition could be one where the children watched no cartoons at all and then the children's

behavior in the two uh cartoon viewing conditions could be compared with reference to the nonviolent carto with

reference to not watching cartoons at all. So that becomes the baseline and then these two become the two degrees of

the experimental conditions. Okay. So uh that is one of the ways you can create a baseline and you can basically measure

it. Sometimes it is also possible that the variable has nonlinear relationship. It does not automatically increase.

Remember we've talked about uh of the relationship between performance and arousal. I don't know whether yeah so

anxiety and performance look at this. Sometimes the relationship between the independent variables and the dependent

variable could be uh nonlinear. You can see here till a certain point till moderate anxiety you have peak

performance levels. If the anxiety is too low you don't really care about performance and performance will suffer.

And if if you're too anxious about your performance then also you can if you're too anxious about how you're going to do

then also the performance will suffer. So sometimes you will see that the relationship between two variables here

anxiety and performance is curvy linear and uh an experimental design that basically uses only two levels let's say

high anxiety and low anxiety. Now look at this here if you use only high anxiety and low anxiety uh you know uh

samples then basically you will see you have a lower chance or you have probably no chance of getting a difference

between these two because in both cases the performance is going to be low. So in these cases what is better is to have

multiple strategic levels of the independent variable. Let's talk about that. So uh a limitation of experiments

with only two levels is that in cases where the manipulation varies with the strength of the independent variable, it

is difficult to draw conclusions about the pattern of the relationship between the IV and the DV. For example, as we

just saw, some relations could be curvy linear such that increases in the independent variable causes increases in

the dependent variable up to some point but causes decreases at some other points that we saw. Remember, say as

anxiety increases from low to moderate levels, performance will increase. But when it uh increases more from moderate

to even higher levels, the performance is bound to decrease because the person is extremely anxious, nervous and will

start making mistakes and will not be able to give the best of his performance. So a two-level experiment

could conclude that anxiety improved performance that or that it decreased performance or that it did not change

performance at all if you have just sampled at two points in this curve. So depending upon a better so experimenters

will be in a better space to be able to figure out what is happening here depending upon which specific levels of

anxiety were sampled or were basically created as conditions in the experiment. Okay. So uh experimental that used three

or more levels of the independent variable would be better able to demonstrate the relationship between

anxiety and performance. That is why even in one way experiments it's not a bad idea to have multiple levels of the

dependent variable of the independent variable sorry. Now uh so far we have been talking about

between group kind of studies. We've been talking about group one going to experimental uh condition, group two

going to control condition. Obviously it creates issues in terms of matching the two groups in absolutely all respects.

So an alternative could be that we ask these participants to do uh you know to go through uh what is called a repeated

measures assign. So the same group of people watch a violent video game or a violent cartoon and then their

aggressive behavior is measured for and then after some time the same group of people watches a nonviolent cartoon and

then their aggression is measured for. So in that sense your job in comes in terms of matching the two groups of

participant reduces. You can also work with a lower number of participants, it becomes more economical and more

logistically feasible. Okay. So let's let's talk about it. Yeah. So while between uh subject uh designs may build

upon a random assignment of participants to different conditions of an experiment, uh you know a better

alternative uh with better control would be if the same participants are put through both the experimental and the

control conditions. uh when equivalence is created in this manners this kind of design is referred

to as a within participants or a repeated measures design as the differences across the different levels

are measured more than once. Okay, once you're measuring with the experimental condition, other time you're measuring

with the control condition. We can test broadly the same research hypothesis from our current example as we were

testing in the between participants design. Look at this here. Now it's the uh same

participant. Now participants are randomly assigned uh first to violent cartoon condition then they go through

this entire thing then they are you know they come back and they are assigned to nonviolent cartoons watching and then

they are sort of going through this whole thing. See uh aggressive dependent where will you you first uh made them

watch violent cartoons and measured their aggressive play then after a gap you basically uh making them watch

non-violent cartoons then measuring aggressive play. Now across conditions you're not measuring. You're basically

comparing the differences between aggressive aggression scores after the participant had watched violent cartoons

and after the same participant had watched nonviolent cartoons. This is a possibility here. Let's let's look at

how does this happen. Now so this obviously as it says it creates control. It basically makes the

experiment more powerful. So let's talk about some of the advantages of these kinds of designs. First as I just

mentioned is increased statistical power. A major advantage of uh repeated measures designs is that they have

greater statistical power than the between participants design. All right? Because the number of measurement

decreases. We'll talk about that. Now, a child in our experiment, say for example, you know, who happens to be in

a particularly bad mood on the day of the experiment. Uh and assume that this negative mood state increases the aggre

her aggressive play. Let's say this is a child who came to our experiment in a bad mood, woke up on the wrong side of

the bed, something happened and then comes through the experiment is already aversive and then sort of goes through

the entire goes through the you know watching the violent video game and then aggression being measured. It probably

could have happened that the two factors have added uh together and there is this uh you know uh confounding of the

dependent measure that is happening. Now, as a child could have been assigned to either the violent cartoon condition

or the nonviolent cartoon condition in a between participants design, the mean aggression in whichever group the child

had been assigned to would have increased. The researchers however will not know that. They will have no way of

knowing that the child's mood state influenced his or her aggressive play. On the other hand, in a repeated

measures design, however, the child's aggressive play after viewing the violent cartoons can be compared to his

or her aggressive play after viewing the nonviolent cartoons. The same child which uh you know with the same level of

residual aggression or you know prior aggression. In this case, although the bad mood might have increased the

aggressive play, it would be expected to increase it in the same manner in both conditions. So your dependent variable

gets less affected. Also there is obviously the economy of participants. So uh another advantage of

these repeated measures designs is that while say for example you needed 20 uh in the violent condition and 20 in the

nonviolent condition here you can just do it 20 participants. So you don't need 40 you just need 20 participants.

There are also some disadvantages of the repeated measures design. Say for example there are carryover effects. So

a problem could be that sometimes when you have the same participant going through the two conditions they have

memories they they are affected by this you know treatment experimental treatment that you've done and in that

sense uh viewing the violent cartoons again or say for example viewing violent cartoons leaves a lasting impression.

Now you tested this person once and then after an hour half an hour 2 hours you tested the person again. Now the first

time when the person watched violent cartoons he collects some aggression residual aggression from here as well

and then it goes through the non violent cartoon condition. Now there if this person had not watched this the

aggression levels could have been different but now because the person has watched the violent cartoon just couple

of hours ago it might be sort of inflating the value on on that condition. So it is possible that there

are carryover effects between conditions when you are doing a repeated measures design.

So the second yeah that's what I was saying. So the second measure of aggression may be influenced by both the

non-violent cartoons and the violent cartoons that have been watched earlier. When effects of one level of

manipulation are still present uh on the dependent measures uh when the dependent measure is assessed for another level of

manipulation. This is called a carryover effect. So just for your definition sake carryovers can happen but also what can

happen is practice and fatigue you learn the task better. Suppose I'm giving you let's say a word recognition task. I've

ch I've manipulated frequency as my independent variable. Now what will broadly happen and this is seen in a lot

in a number of mega studies where they have 10,000 words and many blocks and so on is that by the uh you know second

third fourth fifth block maybe 20th or 30th block if a larger experiment is there the overall reaction times of the

participants increases uh uh decreases and accuracy increases because they are getting more and more familiar and

practiced with the task. So that can happen. So these are practice effects. Participants will get better if they

have done the task again again for a given period of time. Also what can happen is that the participant just gets

tired on a on an afternoon after lunch. You get the participant to your lab. The participant is doing pressing buttons

giving you responses. Uh you have a longish block. Uh you have let's say a 40-minute experiment. It is possible

that the participant will get tired. And if you take the parts again to another experiment or another block another

condition of the dependent variable the uh fatigue kicks in and it starts affecting your measurement.

So uh if the dependent measure involves uh you know the assessment of physical skills such as typing onto a computer

and individual might become fatigued tired uh and perform more poorly over time. So as the time increases in this

case scores on the dependent variable would have changed over time from a factor that is unrelated to the

experimental manipulation. So this can cause this can basically work as a confound in your uh you know experiment.

A possible solution to these uh carryover practice and fatigue effects is to increase the time gap between the

uh measurement of the two dependent variable conditions. So it could be possible that children are made to uh

view the violent cartoons on one day uh and then be observed for aggressive behavior and then after a uh gap of a

week or maybe more they are brought back to view the nonviolent cartoons and then their aggressive behavior is measured.

So you create this time gap it basically takes care of both fatigue carryover and practice effects.

All right. Now you can also do counterbalancing. You can basically create order effects also. A possible

tactical method is uh to uh you know is you do that you do counterbalancing. So for example one uh person watches a

violent cartoon first then the nonviolent cartoon. So AB and uh you know next time when he comes he watches

the nonviolent cartoon first then the violent cartoon. So you can create these kinds of things also. So it in basically

counterbalancing uh technically it involves you know arranging the order in which the conditions of a repeated

measures experiment uh are experienced by the participants so that each condition occurs equally often in each

position in each position. For instance in the current experiment it could it is possible that the uh you

know the conditions could be arranged such that one half of the children uh view the violent. Yeah. So this is just

what I explained. Now such an arrangement AAB kind of design a abba kind of design would ensure that the

carryover from watching either type of either you watch the violent cartoon first or you watch the nonviolent

cartoon first it basically sort of will uh you know get cancel out again it'll sort of get average out uh some

participants watch the violent cartoon first you measured them some participants watch the nonviolent

cartoons first and when you sort of combine and you create take their means uh these effects would have canceled

each other out. So this basically allows the researcher to estimate the effects of carryover by comparing the scores on

the dependent variables uh for participants who were in the two different orders.

In a repeated measure design with more than two levels, there are several possible orders. Uh so typically you can

basically also create a Latin squares kind of a design where you're you know presenting each condition but sampling

two or three levels of each. So there are different ways of counterbalancing. Now uh when do you decide to use a

repeated measure design? This is the last point that I wanted to talk about. Now obviously there is a possibility of

carryover practice and fatigue effects. Uh but uh the advantage is that repeated measures designs uh do provide you the

best kind of control. All right. So there are cases where and it is also possible that sometimes it is difficult

to use repeated measure design. For example, there can be cases where using the repeated measures design is simply

out of the question. For instance, when the participants uh you know are able to guess uh which experimental condition

they are in and what is it that a experimentter wants to measure. So when these kind of things are there then

obviously you would uh avoid uh this and the better control you'll probably say okay let me have two groups of

participants. One group goes through one manipulation the other group goes through the other manipulation. Also

sometimes the counter balancing cannot be done effectively as something that occurs in the level one uh will always

let's say you have a simple task and the participant learned the task in the first uh uh you know condition. Now even

if you call the participant after a day a week or a month the participant has learned how that task works and now you

cannot counterbalance. So it is basically the experimental who has to be careful about these things that whether

uh you know my experimental question or my manipulation is more uh aminable to a repeated measures design or a between

participants design. All right. So this is another decision that you have to make. Okay. In conclusion, the problems

caused by a repeated measures design do not occur all you know do do not occur that frequently. Say for example when

you're doing simple reaction time kind of measures unobtrusive behavioral measures then uh you know there are

lesser practice carryover and fatigue effects then you can do your uh you know repeat measure designs and that is why

these are very popular they present a very uh you know reasonable alternative to standard between participant designs

uh in case where carryovers are likely to minimally occur. All right. So I'll close here in this uh lecture if uh you

know we've talked about oneway experiment designs. We've talked about uh both between participants and

repeated measures design. Uh in the next lecture I will take this forward and we'll talk about factorial designs.

Thank you.