Week 2 - Mediation Flashcards
What is model specification (Box and Jenkins)
Choose correct variables to explain Y and set up statistical model to test predicted relationships
May have sequence of variables that lead to an outcome (mediation)
What is an indirect relationship?
The relationship only exists in the presence of a variable in the middle between X and Y (Mediator variable)
What is the causal steps approach (Baron and Kenny)
Provided basic structure to the logic of mediation
Problems with Causal steps process
Only worked on all continuous variables
Depended on the significance/lack of c pathway that depended on sample size
No variable/parameter directly mapped onto the indirect effect (Could not test significance of the mediator)
Mis-specified model by not adding a mediator (inflate the effect of the X variable b-weights)
Too many steps in this approach
Requirement for significant path c (prime) is false (Many other reasons for non-significance
Dependant on subjective judgement
Did not predict null well
Steps in the Baron and Kenny Approach
a path = Iv - M
b path = M -DV
c path = IV - DV
c (prime) path = IV - DV (Controlled for by mediator) (B-weight of DV residualised)
3 condition for Mediation (Baron and Kenny)
1) Iv is associated with the mediator
2) Mediator is associated with the DV
3) Direct association between IV and DV, will vanish/shrink in the presence of the mediator
Sobel Test - Problems
Attempted to test the indirect effect
Problems with this test - assumed normal distribution (Data not always normally distributed)
Low Power
Product term (ab)
Allow for direct testing of the indirect effect (mediator)
Can estimate the SE from it to get significance test
Can have enough power to reject the null
require no subjective judgement
What did Mackinnon contribute?
Reviewed different approaches
Many had either type 1 or 2 errors
Bootstrapping performed the best
What did bolger and quinn contribute?
First to use boot strapping
Preacher and Hayes - Bootstrapping
Refined bootstrapping method
Does not rest on assumption of normal distribution
Estimates SE by using a resampling technique to generate an empirical sampling distribution and derive a SE from it (with replacement)
Used to test significance of ab term (product term)
Always do 10000 at least! (Under more likely of a type 2 error)
What does mediation determine
Whether the path from the IV through the mediator to the DV is different from 0, than the direct association between the IV and DV (Reject the null if different)
