A lot of literature involving election predictions uses MRP, which often uses Bayesian methods, i.e., Bayesian logistic mixed effects regression. What if we allowed for non-linearities in our predictor through GAMs?
It is suprisingly effective! This project was done on Canadian election data, using the CES 2019 online survey and GSS data as census data for post-stratification.
This project was originally started as an assignment for STA304/1003 (assignment 2).
Please read our manuscript for more information as this is being written (to be finished at some point).