Mediation and Indirect Effects Learning Objectives Flashcards
(11 cards)
Explain the research questions that can be addressed using mediation analysis
Research questions that are asking what potential mediator variables are that explain a relationship - and how/why it explains it.
Explain the difference between mediation and moderation with respect to:
- The research question addressed by each
- The relationship between the main predictor and the second predictor (i.e., the mediator or moderator)
Moderation explores variables that change direct relationships between predictor and criterion (RQ: how does M change interaction), whereas mediation explores a predictor that can explain the relationship, through an indirect relationship (essentially the underlying process or mechanism that explains the predictor-criterion relationship).
Specify the paths (i.e., A, B, C, C’, AB) in a simple mediation model with one focal predictor, one mediator, and one criterion
- Draw these on a diagram depicting the mediation model
In a simple mediation model, Path A is the relationship between X (predictor) and Y (criterion). Path C is the relationship between X and Y without M included in the model. Path B is the relationship between M and Y, controlling for X. Path C’ is the relationship between X and Y, controlling for M. Path AB is the relationship between X and Y through M.
Identify which paths represent:
- The “total effect” of the focal predictor on the criterion
- The “direct effect” of the focal predictor on the criterion
- The “indirect effect” of the focal predictor on the criterion
Path C is the total effect of the focal predictor on the criterion (relationship between X and Y without M in the model). Path C’ is the direct effect of the focal predictor on the criterion (relationship between X and Y, controlling for M). Path AB is the indirect effect of the focal predictor on the criterion (the relationship between X and Y through M).
Explain the criteria that need to be met (e.g., which paths should be significant) in order for us to conclude significant mediation using the traditional approach
In order for us to conclude significant mediation, Path A, C, B all need to be significant. Additionally path C’ needs to be weaker than path C, and path AB needs to be significant.
Describe the three steps in testing for mediation, and for each step explain:
- What analysis we would use
- Which paths would be tested
- Which statistics we would need to interpret
- Running a bivariate regression to test Path A (relationship between X and M). The statistics needed to interpret would be Pearsons r, p-values, F value, t-value, B.
- Running a hierarchical multiple regression to test path C (X and Y relationship without M - done in step 1 of HMR), path B (M and Y relationship with X in the model - done in step 2 of HMR), and path C’ (X and Y relationship with M in model - done in step 2 of HMR). The statistics needed to interpret would be the same as those for conducting a HMR.
- Running either a sobel test or bootstrapping to test path AB (indirect relationship between X and Y via M). The statistics needed to interpret would be either z or coefficient of the indirect effect.
Explain the difference between partial mediation and full mediation
- Specify what evidence we would use to determine whether a significant mediation would be classified as “partial” or “full”
Partial mediation is when the relationship between X and Y is weaker but still significant with M in the model (Path C’ is still significant, but weaker than path C), whereas full mediation is when the relationship between X and Y is no longer significant with M in the model (path C’ is not significant).
Explain what the Sobel test and bootstrapping are used to test
The Sobel test and bootstrapping are used to test if path AB (the indirect relationship between the predictor and the criterion via the mediator) is significant.
Explain what 95% confidence intervals are conceptually, and explain how we tell if a confidence interval indicates a significant effect
Conceptually 95% CI are the range within the true mean sits. If the CI does not include zero, them it indicates a significant effect.
Explain what statistical suppression and what a suppressor variable does to the relationship between two variables once it is added to a model
Statistical suppression is when a mediator makes the relationship between the predictor and criterion stronger. There may be a mediator that hasn’t been considered which could produce the opposite relationship between X and Y to which you were expecting - the variable may be suppressing the expected relationship. By identifying and controlling for the suppressor variable it can account for the opposing pathway, and allow for alternative pathways to become stronger.
List the main assumptions of multiple regression analyses
Linearity, normality of errors (normal distribution of residuals around predicted values), homoscedasticity (the variance of the actual Y values around all predicted values are constant), independence of errors, lack of multicollinearity
