When the Winner Comes Third: Simulating Candidates' Winnability With Inaccurate Polls

When the Winner Comes Third: Simulating Candidates' Winnability With Inaccurate Polls

Abstract

We present a modelling approach to forecast elections in complex settings, characterized by a high number of candidates, sparse and inaccurate polls, and very short memory on the polling houses performance. Because of the complex multivariate dimensions of multiparty elections in Brazil for which the integral is difficult to compute, the model relies on Gibbs sampling methods to obtain approximately independent samples from a posterior distribution, which follows a Dirichlet probability distribution, to predict each candidate’s probability of advancing the runoff. Our model includes third-party candidates’ effect, undecided voters, political ads broadcasting, and house effects in the analysis. The 2012 local election of Sao Paulo city provides the raw data for the analysis.

Date
Location
Brasilia, Brazil
Avatar
Daniel Marcelino
Data/Political Scientist