This is a five-section course as part of a two-course sequence in Research Methods in Psychology. This course deals with descriptive methods and the second course deals with experimental methods.
Lesson 5C – Ruling Out Confounding Variables
Loading...
來自 Georgia Institute of Technology 的課程
Descriptive Research Methods in Psychology
2 個評分
從本節課中
Module 5: Correlations
與講師見面
Dr. Anderson D. SmithRegents’ Professor Emeritus
School of Psychology
Hello, again.
In the last video we talked about the third variable problem,
the fact that there might be a third variable that we're not measuring that's
causing the effects we're looking at.
And that means that there's a confounded variable, and and
that we had to understand those confounded variables to truly look at relationships,
even with descriptive studies.
Now, the third variable problems can be either
where you have a variable that you're not measuring like, for
example, we're not measuring risk taking due to personality.
And maybe that's causing this third variable, is causing the effects between
aggressive behavior and playing violent video games.
And that's what we're going to look at today.
That can be also due to moderation.
There's a third variable that actually is there
that we need to understand in order to do that.
So, for example, Arthur gave you an example
in the last video of the relationship between adult aging and episodic memory.
And that was that maybe there's a health variable we're not measuring that actually
is produced by adult aging, and that's, then, causing episodic memory.
So rather than have a third variable that we didn't know about,
we have a third variable that actually needs to cause.
It didn't cause both, health doesn't cause adult aging.
Adult aging causes health.
And then health produces the episodic memory changes that we see.
So it's mediation or moderation problem.
So the question is is it age that produces the change in memory?
The biological factors associated with aging
cause the loss in memory performance with aging?
Or is it health?
And that's the question we want to answer.
And we can do that by actually looking at whether or not health accounts for
any of the age, memory relationship.
So, for example, I use a Venn diagram,
and this is research that I did in 1997 with my colleagues here at Georgia Tech.
And you find this big overlap between the variance due to aging and
the variance due to memory.
And we say that overlap, then, is the correlation.
It is the relationship between, the negative relationship between, aging and
episodic memory performance.
So now, if we include health into this relationship, how much of this age,
memory variance is accounted for?
When we do that, what we find is that health only accounts for
a very small amount of the age, memory relationship.
So health has an effect on memory.
Health has an effect on age, fairly large effects,
as you see by the overlap between the two variables.
But if you look at the overlap with health with the age, memory variance,
it actually accounts for a very small part of it.
So it can't be health that's moderating or
mediating the relationship between aging and memory.
It has to be something else.
And we'll talk more about moderation and mediation effects in the next video.
So multiple regression is a technique that can be used to help us understand
causal effects when we're looking at the correlations between variables.
So when you know that many variables contribute and predict the dependent
variable you're interested in, we have to use these multiple regression techniques.
Now, the general purpose of multiple regression, and the term was first used by
Pearson just at the turn of the century, is to learn more about the relationship
between several independent variables, or predictor variables, and
a dependent variable, or what's called the criterion variable.
So we have a correlation between a predictor variable and
a correlational variable.
One predicts the other, it is a correlation.
But we often have many predictor variables that we need to understand.
So we need to see what is the relationship with each predictor variable and
the criterion variable that we're measuring.
And that's called multiple regression analysis.
It's a technique for seeing what is the weight that's given to each predictor
variable, in terms of predicting the criterion variable.
And that, actually,
will give you a numerical value of the importance of each predictor variable.
And, also, we can use multiple regression to look at what happens
to the relationship when we remove a predictor variable.
So how much is left over, sort of like we did with the memory example.
So what you end up with in multiple regression is an equation where a value is
given to every predictor variable in terms of how much it influences the measurements
that we're taking in the dependent variable, or the criterion variable.
How much of memory variance is predicted by aging,
the biological processes of aging?
How much of the dependent variable memory is predicted by health?
So that's the equation that the multiple regression analysis gives us.
So Y, in this equation, Y is the criterion variable.
And the Y criterion variable is predicted by a whole bunch of
different predictor variables, that is, the Xs.
And then b is actually the weight in the equation for each predictor variable.
There's always some noise in the way the asymptote of the relationship's going to
be, and that's represented by b0.
But b1, b2, and b3 are the values that the multiple regression gives us for
each of the predictor variables in the equation.
So let's look at an example.
This is the example by Kliewer and colleagues on what are the effects of
violence, stress and social support on internalizing behavioral problems.
So you have measures of each one.
These case participants of children, 8 to 12-years-old.
They lived in high violence areas of the US.
And so our hypotheses are that violence and
stress will increase internalizing behaviors, and
social support will actually decrease internalizing behaviors.
And we'll look to see whether that's true by using multiple linear regression.
So our predictor variables are degree of witnessing violence.
They're living in a violent area.
Measures of life stress, measuring stress on the part of the children.
And then measuring social support, the one that's supposed to decrease internalizing.
And we measure all of those.
And then we have an outcome, or criterion variable,
which is internalizing behavior, things like depression, anxiety, withdrawal.
These are symptoms of internalizing the behavior,
bringing it into us rather than dealing with it.
And we use that by measuring the Child Behavior Checklist,
which is a measure of anxiety and depression in children.
So here are the results of the correlations.
We can see the correlation among the independent variables like,
for example, stress and social support.
And then we can look to see that, in fact, stress, when you look at amount of
violence witnessed, correlates with stress and it correlates with social support.
Current stress, though, has a negative relationship on social support.
Remember, that's supposed to decrease internalizing.
So we have these criterion variables.
Social support shows a negative effect.
But current stress and amount of violence witnessed shows a positive effect,
when we're measuring the amount of internalizing, anxiety, depression, and so
forth, using the Child Behavior Checklist.
The regression will actually give us an R square,
which is how much of the variance that we're totally accounting for.
In this case, in this particular multiple regression model, we’re accounting for
about 0.135 with a variance.
So when we measure current stress, amount of watching violence witnessed,
and measuring, then, internalizing, we’re only accounting for
about 13 to 14% of the variance.
That’s kind of a small R square, a small effect.
And here we should see the unstandardized standard coefficients, the betas.
So we have a beta of 0.201 for watching violence.
That's the weight of that particular variable, watching violence,
on the dependent variable,
on the criterion variable, which is internalizing with anxiety and depression.
So we have 0.201, with current stress, minus 0.247.
And with social support, we have a negative.
The greater he social support ,the less anxiety, depression internalizing we see.
And that's significant.
In fact, as you can see, the significant variables are all three of the criterion.
All three of the predictive variables are significant
in terms of predicting internalizing anxiety and depression.
So if you look at the equation, we put in our beta weights.
And then we get the amount of total variance that we're accounting for, Y,
which would be about 1.4 to 1.5.
So now we've talked about multiple regression,
where you can at least get some insight into causality by looking at correlations.
Next time, in the next video, we'll talk a little bit more about this moderation or
mediation effects.
Thank you.
[SOUND]
