View article

This document accompanies the paper “Influencing Elections with Statistics: Targeting Voters with Logistic Regression Trees”(Rusch, Lee, Hornik, Jank, and Zeileis 2013). It contains a section where we conduct the analysis with a historic proxy, the voting behavior in 2000. For that situation the predictive accuracy is low. It further contains a section describing our efforts in predicting the voting behavior with neural networks, support vector machines, random forests, Bayesian additive regression trees and trees with boosting in the node models. They all perform practically equally or less well than the LORET models, with BART having a very slight edge over the other methods.