Generalized Ideal Point Models for Noisy Dynamic Measures in the Social Sciences

This paper presents `idealstan`, a measurement framework that addresses long-standing issues in the estimation of dynamic ideal points by incorporating diverse time series processes, mixed outcomes, computational advances, and adjustments for missing data. The model employs Stan, a robust Markov Chain Monte Carlo sampler, to permit parallelization of Bayesian inference and to resolve issues of convergence for high-dimensional time-series. The model is compared to existing dynamic ideal point models using Monte Carlo simulations to show that `idealstan` is both faster and more robust than existing approaches. The model is then applied to the challenge of estimating monthly trajectories from 1990 to 2018 for Congresspersons' ideal points while accounting for selection into votes.