A1081
Title: Closed form Bayesian inferences for binary logistic regression with applications to American voter turnout
Authors: Kevin Dayaratna - The Heritage Foundation (United States) [presenting]
Jesse Crosson - Purdue University (United States)
Chandler Hubbard - University of Wyoming (United States)
Abstract: Understanding the factors influencing voter turnout is a fundamentally important question in public policy and political science research. Bayesian logistic regression models are useful for incorporating individual-level heterogeneity to answer these and many other questions. When these questions involve incorporating individual-level heterogeneity for large data sets that include many demographic and ethnic subgroups, however, standard Markov Chain Monte Carlo (MCMC) sampling methods to estimate such models can be quite slow and impractical to perform in a reasonable amount of time. An innovative closed form Empirical Bayesian approach is presented that is significantly faster than MCMC methods, thus enabling the estimation of voter turnout models previously considered computationally infeasible. The results shed light on factors impacting voter turnout data in the 2000, 2004, and 2008 presidential elections. It is concluded with a discussion of these factors and the associated policy implications. It is emphasized, however, that although the application is to the social sciences, this approach is fully generalizable to the myriads of other fields involving statistical models with binary dependent variables and high-dimensional parameter spaces as well.
