Online Program Home
My Program

Abstract Details

Activity Number: 403 - SPAAC Poster Competition
Type: Topic Contributed
Date/Time: Tuesday, July 30, 2019 : 2:00 PM to 3:50 PM
Sponsor: Social Statistics Section
Abstract #307158
Title: Poll-Based Bayesian Models to Predict United States Presidential Elections
Author(s): Brittany Alexander* and Leif Ellingson
Companies: and Texas Tech University
Keywords: political polling; conjugate priors; election prediction

A previous Bayesian model used to predict the 2008, 2012, and 2016 United States Presidential Elections using only poll data resulted in nearly identical electoral college predictions to FiveThirtyEight, and 95.329% relative accuracy to the FiveThirtyEight Polls Plus model in terms of root mean square error of the predictions of the two major candidates. The previous model used poll data from either another single similar state or national polls to create prior distributions and used the MLE estimators to fit the model. We present new models with minor differences are used on the same data used as the previous model. The new models now pool the polls together from other states in the regions and uses the pooled estimates as the prior instead of relying on poll data from one state. The new models compare the beta and Gaussian conjugate prior and use three different methods to reassign undecided voters, and either updates iteratively or pools the polls together and performs the calculation once. We also provide a variety of models to serve as comparison for the model’s accuracy such as a noninformative model, and polls only model.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program