A Bayesian Prediction for Undecided Voters

• Activity Number: 215
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Jai W. Choi*+ and Balgobin Nandram
• Companies: Centers for Disease Control and Prevention and Worcester Polytechnic Institute
• Keywords: correlation ; missing data ; multinomial-Dirichlet ; nonignorable ; sampling-based ; two-way categorical table

Abstract

Typically data from election polls or heath surveys are presented in two-way categorical tables. For example, in the 1998 Ohio polls for governor, the first category represents the candidates, Fisher, Talft, other, and undecided, and the second category represents the current status, i.e., likely to vote, not likely to vote and undecided, of the voters. There is a substantial number of undecided voters. Bayesian prediction allows to allocate the undecided voters to the three candidates and permits modeling different patterns of missingness, and a multinomial-Dirichlet model estimates the cell probabilities which can help to predict the winner. In 1998, Ohio Polls were conducted in January, April, and October. We extend our models to accommodate a correlation structure across these three polls. Markov chain Monte Carlo methods are used to fit the models, and a simulation study is used to assess the predictive power of the models.