Mediation Flashcards
What does mediation investigate?
How X = Y
What is the mechanism through which X influences Y
What is the total effect and why is it problematic?
Total effect is c - X = Y
Poverty predicts lower IQ
Total effect ignores how X is related to X, mediation investigates how this occurs
How can poverty predict test scores?
Because poverty leads to greater cognitive strain (M) which in turn leads to lower test scores
Cognitive strain is the mechanism which poverty affects it
Indirectly predicts lower test scores via increasing cognitive strain
What has the indirect effect got to be related too?
Both the predictor and the outcome
A change in the predictor changes the mediator, which in turn changes the outcome
M is the underlying mechanism or process explaining how X and Y are related
When can you use mediation analyses?
Anything you can analyse with regression or ANOVAs you can analyse using mediation analyses - imposes a direction and sequence to the relationship - just need a theory
Self report questionnaires - stress, well being etc - theory determines which is X, M an Y. Useful when investigating the process through which objective or observations are related to outcomes. Because usually, objective cannot affect outcomes direct e.g., poverty doesn’t affect well being directly. or income, education, rank, gender etc
Experimental research - investigates how your experimental manipulation influences the DV
X = IV
Y = DV
M = process through which X changes Y
Why do you need a theory?
To suggest the direction between the variables X = m = Y - directional pathways between the variables, investigates the predictors in your theory. Analysis won’t tell you the direction
What do you want to do to the total effect?
Unpack it - split it up into its different properties
Consists of direct effect and indirect effect
What path is the total effect on Y?
Path C
What path is the indirect effect?
Path AB
X on M = a
M on Y = b
What path is the direct effect?
c’
How can you work out the total effect?
C = ab + c' c = poverty on test score a = poverty on cog strain b = cog strain on test score c = poverty on test score, accounting for indirect effect
What is step 1?
Does the predictor predict the outcome?
Total effect of X on Y - relationship we want to explain
Simple regression, total effect = C
Look at unstandardised coefficients
What is the indirect effect a product off?
a and b together
What is the direct effect a product of?
an indirect effect = ab
direct effect = c’
What is step 2: path a?
Does the predictor predict the moderator
Simple regression: X - M
Look at poverty when DV = strain
What is step 3a: path b?
Does the mediator predict the outcome, controlling for the predictor?
Multiple regression: M - Y controlling for X
Have to control for X, Y and M, could be correlated simply because the X predicts them both without a direct relationship between M and Y
look at strain on score
What is step 3b: direct effect?
Once we have accounted for the indirect effect, does the predictor still predict the outcome?
multiple regression X=Y controlled for M
How do we know if there is mediation? The old way
Using P values
full mediation: if the direct effect reduced to non significance, the effect of X on Y goes entirely through M
partial mediation: if the direct effect is reduced, but still significant, theres both an indirect and direct effect of X on Y
Sobel Test - gives you a P value for the estimate of the indirect effect (ab estimate)
No longer used because of the all or nothing conclusions drawn from significance tests - a change from p = .049 to p = 0.051 does not imply full mediation
How do we tell if there is mediation? The new way
Confidence intervals
Bootstrapped confidence intervals of the indirect effect
If the 95% CI does not include zero, then the indirect effect is treated as significant - gives an indication of the accuracy of effect
Effect size of indirect effect
What happens if the confidence intervals don’t cross zero?
There is a significant mediation
How can we explain this effect?
Poverty predicts lower test scores because poverty predicts increased cognitive strain, which predicts lower test scores
Cognitive strain mediates the relationship between poverty and test scores
There is a significant indirect effect from poverty to test scores via cognitive strain
Indirect effect = -.03, 95% CIs ()
What are the problems?
Same warnings as aggression - assumptions (homogeneity of variance doesn’t matter, run estimates without it)
Only get out what you put in
Direction and causality - based on your theory, cannot infer causality
Can you use it with categories?
Yes - dichotomous variables with two groups, treat them as you would in regression
b = difference in the outcome between the two categories
males = 0
females = 1
b = change in outcome for females compared to males
control = 0
experiment = 1
b = change in outcome fro experimental group compared to control group
How can the effects cancel it?
Some paths can be positive and some negative
for example, positive indirect effect (experience of prejudice on group solidarity on well being) and negative direct effect (experience of prejudice on well being) but can still test mediation onv non significant total effects