This paper presents an item-response theory parameterization of ideal points that unifies existing approaches to ideal point models while also extending them. For time-varying inference, the model permits ideal points to vary in a random walk, in a stationary autoregressive process, or in a semi-parametric Gaussian process. For missing data, the model implements a two-stage selection adjustment to account for non-ignorable missingness. In addition, the ideal point model is extended to handle new distributions, including continuous, positive-continuous and ordinal data. To enable modeling of datasets with mixed data (discrete and continuous), I incorporate joint modeling of different distributions. Finally, I also address ways of implementing Bayesian inference with big data sets, including variational inference and within-chain MCMC parallelization.
Haplotype and Missing Data Inference in Nuclear Families
