Understanding Correlational Research: Limitations and Causal Insights

Hello and welcome to the course basics of experimental design for cognitive psychologists. I am Dr. Arakma from uh

IIT Kpur. This is the second week where we are getting a bit of an overview of the different kinds of research designs

uh somewhat superficially before we dive deep into the experimental designs. Last lecture I was talking to you about

the very basics of the correlational design and we basically figured out that the you know it is possible that the

variables of interest are related to each other. They are co-varying. The nature of the relationship how however

can be different. It could be positively linear. It could be negatively linear. It could be just independent where they

are not related to each other or it could be a curvy linear kind of a relationship. So this is some of the

things that we talked about. We also talked a little bit about multiple regressions and uh co correlation

coefficients and so on. Now there are a few things about the correlational research designs that I will also take

up in today's lecture before we move on to experimental research designs in the next week.

So how useful are these uh you know correlational uh data? How useful are these correlational research? Now uh one

thing is is sure and we sort of started with that understanding that while correlations can provide us with a very

good idea about how the two variables are related to each other and that they covary uh the strength and the direction

etc. we get to know from the coefficient the magnitude the sign and so on. There are also certain shortcomings of this

kind of research. For one this cannot be used to draw causal conclusions. Okay. uh this is something very important

which is why uh you know experimental research is extremely important because we cannot really control for anything.

We cannot really so if you remember when we were talking about uh you know multiple regressions and when we were

talking about the fact that there can be three or four or five predicted variables for one outcome variable uh

that's basically why causality is very difficult. Okay. uh although uh you know multiple uh regression etc they can they

are statistical tools to control for other variables while looking for the contribution of uh your interest

variable. Uh we have not really gone out and controlled for certain variables. We have not really gone out and created

what is called an initial equivalence and matched the two uh groups on certain characteristics so that uh you know the

effects that we are observing are only because of changes in a variable of interest. So that is that is not there

that is not what is done in correlational research. Okay. So correlations cannot be used to draw uh

conclusions about the causal relationships among the measured variables. We know that that's something

I just repeated. Hence even if the research hypothesis indicates that a predictor variable can be used to

estimate the outcome variable. The causal relationship is not clear and the causal relationship is not what is being

argued for. How do we interpret? We talked about the R and you know R square and those kinds

of things. How do you make sense of this data? Okay. So Sanger this takes this example where a researcher seeks to

investigate the effects of viewing violent uh television on the propensity to involve in aggressive play in

children. So he's basically saying when children watch violent TV uh that is what causes them to behave violently in

the school playground. Something like that. So he goes out and collects data. uh data is collected from a sample of

fourth grade students. Uh two measures, a measure of how uh you know aggressively each child plays on the

school school uh uh playground and how much violent television shows. You know earlier I I remember growing up there

were a lot of uh rather violent uh you know cartoons that were there. you know they they were the fox fox kids uh so

Spider-Man they were uh you know slot cats there were there were bunch of these things and obviously while there

are cartoons and the uh the violence is sort of uh you know uh controlled there but people children used to like uh you

know WWE and those kinds of things. So people do watch movies for example uh were also violent not as violent as

today but was also violent uh you know in their own right. So how much violent TV a child is watching does it affect

his playground behavior? Does it basically make the child more aggressive uh competitive maybe uh hurtful to a

certain extent? This is basically what we want to check. When they collected this this data they

carried out correlations and they find that yes there is a significant positive correlation between the two measured

variables. So you can think of this as a positive linear relationship. All right. Now although this positive correlation

appears to support the researcher's hypothesis because there are there are uh you know alternate ways to explain

the correlation. It cannot be taken that this is causing that behavior. Okay. Uh the correlation is positive and the

relationship seems positively linear. There are alternative hypothesis that are there as well. And because there

there are alternative hypothesis, it automatically suggests that this was not co this is not a causal

relationship. So for example, children might be playing aggressively because something has happened. Maybe their

parents have a extremely discipline extremely strict disciplining style. Maybe because there are other things.

Maybe there are say for example by temperament uh uh you know aggressive. Maybe it could be because say for

example uh you know aggression is rewarded in the school uh paradigm something like that you can think of

anything. So there could also be in that sense or actually one of the interesting concepts

is that it there could be reverse causation. So what is probably happening is because they like playing

aggressively because there is competition because they want to sort of win and be on top that might be causing

uh their propensity of liking violent TV programs and cartoons and so on. So one interesting thing is that there could be

uh cases of reverse causation also in the correlations that we are observing. So it could be possible that children

who behave aggressively at the school develop residual excitement uh which leads them to watch violent television

shows at home. Okay, this is one. Although the possibility that aggressive plays uh you know cause the increased

viewing of violent television rather than vice versa may seem less likely to you there is and we cannot rule it

because we cannot rule this out. It can be referred to as reverse causation. Look at this violent TV causing

aggressive play. It could also be possible that aggressive play is calling causing violent TV. So at this point you

have two hypothesis. You have the same data. The same data can be used to you know talk about these two hypothesis.

So this is an example of what is called reciprocal hypo causation. Both variables are causing each other to a

certain extent. Okay. So this one here is reverse causation. Aggressive play is causing people to watch violent TV. This

scenario here is also possible which is basically both variables are causing each other to a certain extent which is

what we are calling as reciprocal causation. Now uh it could also be that there is a

third variable which we have not considered. We have basically considered uh viewing violent TV. We have

considered aggressive play. There could be a third variable that is commonly affecting these two. Okay. So there

could be a common causal variable as well which is leading to such a pattern of results. Okay. So common causal

variables uh you know that are not part of the original research hypothesis can also produce these kind of effects can

also have a uh strong influence on your results and must be really cared for. Okay. So in our example it could be as I

was just mentioning a moment ago that a potential common causal variable is the discipline style of the parents. For

instance, parents uh who use a harsh discipline style may produce children who both like to watch violent

television and like to play aggressively in the school playground in comparison to children who have a relatively less

harsh discipline style. Okay, so it is probably the discipline style that is leading to both of these outcomes. But

we've not factored this in. So we do not know whether this is happening or not. We have to look for other ways to sort

of uh figure out the causal relationship or maybe put all three of them in a uh you know correlation matrix and try and

understand what is happening. Multiple regression for example. Okay. So in this case TV viewing and aggressive play

would be positively correlated even though neither one actually cause the other. So the data can in this sense if

you're not careful can mislead you. And a lot of times you will see that people you will will pick will pick one or two

of the correlational coefficients and they'll make an entire story out of it. you have to always try and present the

full picture. So it's also possible in there could be another scenario where the predictor and

the outcome variables are both caused by a common causal variable. This observed relationship when this is happening

because of a common causal variable is set to be a spurious relationship. Okay, is a relationship that is spurious uh

because neither of the two variables are actually causing each other. It's the third variable that is having an

influence on both. In this case, if the effects of the common causal variable are taken away or controlled for, the

relationship between the predictor and outcome variables might disappear altogether because the third variable is

controlled for. In our example, the one that we were just talking about, the relationship between aggression and TV

view viewing might be spurious because if we it were to if we were to control for the effects of parents disciplinary

style, the relationship between aggressive television viewing and aggressive behavior might just vanish.

It's not there. All right. So here if a common causal variable is operating this would lead to a very different

interpretation of the data and especially if you are conscious of a common causal variable. A lot of times

we miss these things in our research designs. Okay. Uh you have to be very careful of not doing that. Okay. So the

identification of the true cause of the relationship would also leads to a very different plan to reduce aggressive

behavior and basically collect data in a particularly informed way. Now when one reads about correlational

research designs, one should keep in mind the possibility of reverse causation, reciprocal causation,

spurious relationships and so on. And this is something that is interesting because a lot of times correlational

research is reported as uh you know demonstrating causality without any mention being made of the possibility of

reverse coration common causal variable and so on. So informed students must be very careful of these interpretational

problems. That's basically what I was trying to say in the last slide. What are what are the other problems? Common

causal variable, reciprocal causation, reverse causation, all of that. There are also sometimes extraneous variables.

Okay. So, while common causal variables are obviously quite problematic in the interpreting the correlations and can

cause spurious relationships, correlational research designs sometimes also encounter other variables that were

not in the original research design. You were not talking about them. Okay? For example, how aggressively a child plays

at school is may probably be caused by uh the discipline disciplinary style of the child's teacher and not the parent.

So it could be the teacher who's making this inference. So the TV watching at home is is not creating the problem.

It's this extraneous variable. So if you're picking for example in this case, if you're picking for this comparison

children uh from different classrooms whose class teachers are different and have different behavioral profiles,

maybe that is causing this effect. You've controlled for parental uh style maybe getting three or two or three

children from the same household but these children are in different classrooms. So you've controlled for

parental design style but not for teacher dis style maybe that is you know uh uh you know leaking into the results

that you're observing. So these variables that are possible that are basically playing a role in predicting

the outcome variable but that do not cause the predictor variable basically you know can are defined as extraneous

variables. So this distinction between extraneous variables and common causal variables is

very important and we should keep this in mind because they can lead to substantially different interpretations

and a different story altogether. Extraneous variables may reduce the likelihood of finding a significant

correlation between the predictor variable and the outcome variable because they cause changes in the

outcome variable. So extraneous variables are actually you know a bit of a problem. However, because they do not

cause changes in the predictor variable, extraneous variables cannot really produce a spurious correlation. They are

affecting only one side of the thing. They are not changing the predictor variable. They are only changing the

outcome variable. Look at this here. So uh we have the predictor variable viewing violent TV. We have the outcome

variable aggressive play. We know that there is possibility of reciprocal causation that is why the double-sided

arrow. Here there is an extraneous variable that we did not account for teachers disciplinary style. This is not

a common causal variable. It is not affecting both. It is just an extraneous variable because it is affecting outside

from outside just the outcome variable. Okay. This is a best way to distinguish between a common causal variable and an

extraneous variable and must be kept in mind while you know interpreting the results of the correlation.

There are also sometimes possibility of what is called mediating variables. Okay. So sometimes while you're talking

about variable uh you know A and variable C. How is variable Affecting variable C? That is the uh you know

relationship of interest for you. It is possible that the relationship of A and C is being mediated by a B. Okay. So

there is is a possibility that another type of variable uh can be important in correlational research design and is

extremely relevant for gaining a full understanding of the relationship between these two is called a mediating

variable. Okay. So for example one might expect that the level of arousal of the child might also mediate the

relationship between uh you know viewing violent material uh on television and displaying aggressive behavior. So say

for example a child is watching violent TV but the child by temperament because that's what I I said temperament earlier

by temperament is having low arousal all right is not extremely excitable and so on. So if the child is not extremely

excitable maybe he will not translate into aggressive play. So while viewing violent TV is going to have an effect on

aggressive play. The degree to which that relationship will manifest might probably be determined by the mediating

variable that is arousal. All right. So understanding mediating variables is also important.

Our example let us look at this is something that I already mentioned. It could be there could be any number of

mediating variables that you are seeing. Say for example violent TV aggressive play fewer inhibitions. Children who

have a lot of inhibitions might not engage in aggressive play at all. Children who have fewer inhibitions

might engage in uh you know aggressive play. They want to compete. They want to push. They want to come first. They

might they're not afraid of doing sometimes mean things. and so on. So inhibitions also matter a lot.

Similarly, violence related ideas, if they're exposed to violence related ideas, if they are coming up with those

kind of ideas. So you know sometimes there are naughty kids in the in a you know colony that will keep coming up

with innovative uh slightly mischievous maybe in some cases violent ideas. Okay, how do I trouble this guy? Oh, maybe

I'll put a cracker in his uh you know uh in the exhaust pipe of his car or exhaust pipe of his bike or something

like that. Okay. So, uh violence related ideas if there are extremely you know if the child has watched violent TV he's

not having these violent related ideas may not translate into aggressive play but if he's having too many of these

violent ideas this relationship might be magnified to a certain extent. All right. So uh mediating variables are

uh important because they can explain why this relationship between two uh variables is occurring. So viewing

violent TV and aggressive play. There are other examples also that we can take. Say for example, more study time,

greater retention of material, better task performance. If somebody has an extremely good memory, if they have

extremely good comprehending power, uh and then they are having more study time. So for example, uh one of my

friends has a very good memory. Uh they are they spend let's say uh you know 6 to 8 hours uh studying. Uh if they do

that and they have very good memory, they're obviously going to do extremely well in their studies. Sometimes what

happens is that there's you know uh greater retention of material is there so somebody might have a lesser study

time and in that sense or say for example very good uh retention of material is there the study time might

not have that much effect on so even if you're studying for a small time it may basically lead to better task

performance by itself. So this capacity of retaining material can moderate the relationship between the amount of study

time and the uh quality of task performance. Similarly, failure on on a task, low

self-esteem, less interest in the task. Okay. If a person has high, you know, uh self-esteem and sort of suffers failure

in the task, he will probably try and take more interest in the task and maybe work harder or the other way around less

low self-esteem. Oh, I'm not going to be able to do this. I failed the task anyways. The interest anyways sort of,

you know, dips down further and further. So, that is also possible. So you can see there are different ways in which

these mediating variables can actually mediate or moderate the relationship between the variables of interest. So

viewing violent TV, aggressive play, failure on a task, less interest in the task. They are all they are all

potentially can be mediated by these variables that we were talking about. Now we have been saying descriptive

research designs don't talk about causality. Cannot they're not equipped to talk about causality. Correlational

research designs cannot talk about causal causality. There are reasons we've talked about the fact common

causal variable, reverse causation, reciprocal causation, mediating variables, this and that. Uh and because

you're not, you know, you're collecting data out in the open, you are not controlling for any number of things. So

obviously talking about causality becomes difficult. But it's not possible to conduct

experimental research all the time across all kinds of uh you know uh target groups and so on. So you'll see

that correlational research is actually the next uh most popular type of research design that is used. So

obviously descriptive research is done at the largest level. Uh the next uh it seems to me at least I might be wrong.

Uh the next most popular research design is the correlational research design. Okay. So what can we do uh to optimize

the you know the correlational research design to try and interpret causality some kind of good estimate of causality

maybe not causality established but something very close to that let's look at some ways one is in one of these

research ideas uh given by Eron and colleagues the idea was that we want to establish this so what they did was they

conducted a longitudinal study okay so uh the same individuals who are measured more than one time uh and the time

period was kept such that the you know the changes in the variables of interest

could actually happen. Okay. So they wanted to measure Eron and colleagues violent television viewing and

aggressive play in a group of children who were 8 years old. So first time they measured this

took out a correlation then they waited for almost 10 years to measure this again when the children were turn to uh

18 years old. Okay. So same measurement at 8 years old and 18 years old. 10 years is is a lot of time. All right.

The resulting data was two sets of correlations and each measured at a different time period. Let's see what

happens. What basically people would do here is that they will carry out what is called a path analysis. They'll

typically sort of uh you know analyze this through some kind of uh multiple regression that and that will assess the

relationship among various variables that are in play here and they basically will say over time the role of what

variable has increased and decreased. Let's let's see this in a figure. Look at this. This is the path analysis of

eron and colleagues study 1972. You can see TV violence causing aggressive play. Aggressive play calling causing TV

violence. You can see the relationship at when the kids were just 8 years old and the uh correlations when the kid

were 18 years old. Okay. So is aggressive play causing TV violence uh as a at the age of 821

at the age of 8 and in the second measure point uh 01. So it sort of reduces uh over time that does not

remain the main factor. Similarly, when children are almost adults, uh there's no direct correlation. It's minus.05.

So, it's a very weak negative correlation between aggressive play and TV violence. So this kind of a path

diagram tells us over time what the are the effects between these different variables and it basically allows us to

see what kind of coalitions will survive this test of time and in that sense you you can see that this causality that we

were very confident about you know positive linear relation that earlier studies had found sort of gets tested

and in in that sense gets undervalued. It says oh this is not really the main most important thing.

Other two things are say for example you can try and control for common causal variables that's why you can say okay if

there is a you know disciplinary style of parents if there is disciplinary style of teachers something like that

you can so when you're designing your study you have to be very careful you have to be very aware of the possible

causal variables common causal variables that might there that might be there and control for them. Similarly, when you

are designing these studies, you have to also uh you know assess the role of mediating variables. What kind of

variables would play the part? What kind of variables are there? So, the better idea is that to factor in that variable

in your design, find out a way of measuring it, find out a way of controlling for it. If you uh you know

want a better resolution and once you've done both of these things control for possible common causal variables and uh

control for the role of mediating variables that is where you are giving yourself or you are optimizing the

chances of in some sense optimizing the chances of uh determining not really determining but getting an causal

estimate just based on your correlational studies. Okay. Okay, so there are different ways, there are

structural equation modeling, there are several ways in which people use correlational research and they try and

get as close to causal reasoning as possible. But again, the gold standard is experimental research which we will

basically talk about from the next week, next lecture. Thank you so much.