Regency EF
Using Poisson Binomial Models to Reveal Voter Preferences (304016)
*Evan Taylor Ragosa Rosenman, Stanford University
Keywords: ecological inference, generalized linear models, Poisson Binomial distribution, non-convex optimization, political science
We consider a problem of ecological inference, in which individual-level covariates are known, but labeled data is available only at the aggregate level. While our interest is in modeling voter preferences in elections, such problems are increasingly common in insurance, data privacy, and other fields. Our goal is to estimate the parameter vector beta in a logistic regression relating covariates to success probabilities. We pose the problem as maximizing the likelihood of a Poisson binomial, the distribution of the sum of independent but not identically distributed Bernoulli variables. We propose a computationally efficient method for fitting the coefficient vector, based on a Gaussian approximation. Using data on voters in Morris County, NJ, we demonstrate that this approach outperforms other ecological inference methods in predicting known outcome: whether an individual votes. We apply this technique to the 2016 presidential election, fitting a model to voters from the contested swing state of Pennsylvania. The model is predictive and learns intuitive associations.