bayesian moderation analysis
hello everyone,
I have a question about the Bayesian moderation analysis in JASP (https://jasp-stats.org/2020/03/12/mediation-and-moderation-analysis-in-jasp/). In various analysis (with different predictors), the analysis gives out anectodal evidence for the null model (with predictors X and Y) in comparison to the regression model with interaction term (X, Y and X*Y as predictors). However, in various cases the R2 value is higher for the model including the interaction term than it is for the null model. As the BF favours H0 over H1, but the model including the interaction term explains more variance, I am unsure which one I have to use to detect if there is a moderation effect or not. Could someone explain this to me?
Thank you very much in advance! :)
Comments
Dear lemakru,
The more complex model (H1) will always have a higher R2 than the simple model. But the R2 indicates the best-possible, cherry-picked fit, whereas the BF is based on a comparison of predictive performance.
EJ