Topics Covered
Coefficient Collapse
Understand why significant predictors can suddenly vanish in a multiple regression when adding new variables.
Mediation Analysis
Decompose the total effect into direct and indirect pathways using the mediation framework.
Partial Regression
Visualize the unique variance of predictors after removing the linear influence of other variables.
Key Concepts
- Baron & Kenny's framework for causal mediation
- Direct vs. Indirect effects (ACME and ADE)
- Bootstrapping for significance testing of mediated paths
- Full vs. Partial mediation
Homework
- Fit a multiple regression and observe coefficient collapse
- Perform a formal mediation analysis using the
mediationpackage - Create path diagrams and partial regression plots to visualize your findings
Resources
- When Your Predictor Disappears: Mediation vs. Multiple Regression
- The amazing saga of the Chinese relative clause: A cautionary tale for tomorrow's psycholinguists
Parting Note: Bayesian is the next frontier 🚀