A Bayesian Model for Forecasting the Choice of Candidate in a Presidential Election

Abstract

Opinion poll plays a prominent role as a source of information in different societ ies of the world and are permanent feature of contemporary politics. Such election surveys have several purposes, including forecasting elections outcomes and study ing the distribution of votes as they vary over geographic, demographic and political variables. Most of the opinion polls data are analyzed superficially using discrete or summary statistics. This work provides an in-depth statistical inference to these data. The goal of this thesis is to formulate and apply Bayesian model for comparing the two leading candidates in the presidential voting process. Although much of the media attentions during presidential election years focuses on polls tracking popular support for the major candidate, the vital role played by Bayesian statistical analysts in pre dicting elections incorporated in order to address forecasting election outcomes. We considered a Bayesian hierarchical model in predicting Kenya’s presidential elections outcomes based on pre-election polls collected at most four months prior to the 2007 and 2013 general elections taking into account the evolution of opinions during cam paigns Kenya’s presidential elections are predictable and we were able to come up with a powerful methodological option for predicting the outcome of Kenya’s presid ential elections that uses Bayesian estimation approach and incorporates polling data to account for the evolution of opinions during campaigns. The results show that the leading candidate in the polls will win the election if the observed pattern does not portray misclassification; otherwise, the race is too close to call if there is underlying uncertainty. In conclusion, the research obtained predictions which suffices to prove that the main novel points of the analysis - namely the use of representative areas, the Bayesian analysis of an appropriately chosen hierarchical model and the probabilistic classification of undecided vote in opinion polls – are certainly important points in the right direction.