B0937
Title: Gradient boosting for Dirichlet regression: Impact of protests on election results
Authors: Elisabeth Bergherr - Georg-August-Univerität Göttingen (Germany) [presenting]
Tobias Hepp - Friedrich-Alexander-Universitaet Erlangen-Nuernberg (Germany)
Michael Balzer - Georg-August-Universitaet Goettingen (Germany)
Swen Hutter - FU Berlin (Germany)
Abstract: Different sociological and economic influences are modeled on the outcome of elections during and after the recession in the 2000s. An outcome in percentages, which sum up to 100 percent, is best modeled with a Dirichlet regression. The model is implemented in a gradient-boosting environment since there are many candidate variables to consider. The implementation of this model, however, is less straightforward than for other, simpler, GLM-type outcomes. The approach as well as a simulation study as proof of concept will be presented.