Mediation Regression Analysis Flashcards
What is a mediator?
A variable that explains a portion of the association between Y and X1
What is mediation?
A hypothesised causal mechanism by which one variable affects another
In a mediation model, what is c’?
The direct effect of X1 on Y
In a mediation model, how do you calculate the indirect (mediated) effect?
A x B
(effect of X1 in M times effect of M on Y)
What is the total effect (C)?
Direct + indirect effect
c’ + (axb)
What are Baron and Kenny’s four steps for determining mediation?
- Test path c (X1 –> Y)
- Test path a (X1 –> M)
- Test path b (M –> Y, controlling for X1)
- Test path c’ (X1 –> Y, controlling for M)
If β3 (path c’) from the multiple linear regression is approaching zero (p>0.05), what can you conclude?
Complete mediation - there’s no effect of X1 on Y when controlling for M
The mediator substantially reduces the effect of X1 on Y
If β3 (path c’) from the multiple linear regression is significantly different from zero (p<0.05), what can you conclude?
Partial mediation - there is still an effect of X1 on Y when controlling for M
If there is partial mediation, c’ will be _____________ than c
Smaller
Which is Baron and Kenny’s steps are essential for establishing mediation?
Steps 2 and 3 - estimates indirect effect
What do newer methods recommend testing to establish mediation?
Only the indirect effect (paths a and b)
What is the Sobel test?
▪️Normal theory approach to testing indirect effect
▪️H0: a x b = 0
▪️Based on approximate z-statistic
When do you reject the null hypothesis of the Sobel test?
If Z in absolute value is greater than 1.96
When does the Sobel test work best and why?
In large samples because skewness is reduced
What are the limitation of the Sobel test?
▪️Based on normal approximation but sampling distribution of ab is highly skewed
▪️Large values much more variable
▪️Lowers statistical power
What is the non-parametric Sobel test?
▪️Tests indirect effect via bootstrapping
▪️Better alternative that imposes no distributional assumptions
What is bootstrapping?
▪️Takes large no. of samples with replacement from dataset
▪️Estimates indirect effect for each sample
▪️Forms non-parametric sampling distribution of indirect effect with estimates
▪️From this, estimates confidence interval for ab
If the bootstrap confidence interval doesn’t not contain zero, what can you conclude?
The indirect effect is significant
What is the recommended method for testing indirect effects?
Bias-corrected bootstrap
How do you calculate the Z statistic for the Sobel test?
ab ÷ SEab
If the Z statistic is greater than 1.96, what can you conclude?
The indirect effect IS significantly different from zero
(outside of 95% CI)
Complete mediation is determined when…
The path c’ is not significantly different from 0