3: Mediating effects Flashcards
What is the rationale of mediation?
- Although an IV may have an effect on the DV, this effect is sometimes indirect; mediated through another variable.
- Whether there is a mediating variable relationship depends on theory (i.e. is it justified to expect that the effect is mediated?)
- In general, the mediating variable helps us understand the process through which the independent variable produces the outcome.
What are linear models?
Simplified models/representations of reality, contains errors but can still be interpreted
With what variables is the indirect effect stated for mediation?
a*b
With what variable is the direct effect stated for mediation?
c’
How is the total effect stated for mediation?
c = c’ + ab
What is full mediation?
When the indirect effect is significant (both a and b), but the direct effect is not significant (c´).
What is partial mediation?
When the indirect effect is significant (both a and b), and the direct effect is significant but has lower effect size than when the mediating variable is not in the model.
What are the two procedures to investigate mediating effects?
- Baron and Kenny.
- Preacher and Hayes.
What are the four steps for investigating conditions for mediation as outlined by Baron and Kenny?
- Regress Y on X.
Show that the independent variable significantly predicts the dependent variable. - Regress M on X.
Show that the independent variable significantly predicts the mediator. - Regress Y on M and X.
Show that the mediator significantly predicts the dependent variable while controlling for the independent variable. - Check the significance of X on Y in step 3.
If the independent variable becomes no longer significant in step 3, there is full mediation.
If the independent variable is still significant (but has a lower effect size) in step 3, there is partial mediation.
What are the three steps for investigating conditions for mediation as outlined by Preacher and Hayes?
- Regress Y on M and X.
Show that the mediator significantly predicts the dependent variable while controlling for the independent variable. - Regress M on X.
Show that the independent variable significantly predicts the mediator. - Check the significance of the mediated effect using bootstrapping.
Bootstrapping = sampling your sample (creating the sample distribution).
What is the Sobel test?
The Sobel test calculates a z-score. We can use it to compare the effect in a model with and without mediation.
The mediating effect is significant when p-value < 0.05, which is when z-score >1.96
What are some problems with the Sobel test?
- Depends on distributional assumptions.
- The distribution of effect is normal only with a large sample size.
- The resulting p-value is not correct
Proposed solution: Using a non-parametric approach - bootstrapping.
What is ACME and ADE?
From Preacher and Hayes method.
ACME: Average Causal Mediation Effect. Indirect effect = ab
ADE: Average Direct Effect. Direct effect = c’
What is bootstrapping?
Using our sample to create more samples.
What is a limitation of using Baron/Kenny and the Sobel test for mediation?
They assume normality of the distribution of the indirect effect. It has been shown that this assumption is rarely met in practice. Hence, our p-value will probably be too small (too significant).