Dirichlet process mixtures and extensions
Michael J. Daniels, Antonio Linero, Jason A. Roy · 2023
We introduce Dirichlet process mixtures of distributions as a prior for unknown joint distributions, as well as various extensions such as the enriched Dirichlet process mixture model. We then show how these models can be used in missing data and causal inference problems, both to estimate arbitrary functionals of the distribution of the potential outcomes and as a tool for imputing missing outcomes and covariates. We also review posterior computations, some of which can be performed automatically in standard software such as Stan or JAGS.