2017/02/27 by Jason Roy, Kirsten J. Lum, Roy, Jason +9 · 3 citations
Computer Science · Mathematics · #Bayesian Methods and Mixture Models #FOS: Computer and information sciences #Methodology (stat.ME) #Statistical Methods and Bayesian Inference #Statistical Methods and Inference
paper · pdf · doi:10.48550/arxiv.1702.08496
openalex publication_date 2017/02/27 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
We propose a general Bayesian nonparametric (BNP) approach to causal\ninference in the point treatment setting. The joint distribution of the\nobserved data (outcome, treatment, and confounders) is modeled using an\nenriched Dirichlet process. The combination of the observed data model and\ncausal assumptions allows us to identify any type of causal effect -\ndifferences, ratios, or quantile effects, either marginally or for\nsubpopulations of interest. The proposed BNP model is well-suited for causal\ninference problems, as it does not require parametric assumptions about the\ndistribution of confounders and naturally leads to a computationally efficient\nGibbs sampling algorithm. By flexibly modeling the joint distribution, we are\nalso able to impute (via data augmentation) values for missing covariates\nwithin the algorithm under an assumption of ignorable missingness, obviating\nthe need to create separate imputed data sets. This approach for imputing the\nmissing covariates has the additional advantage of guaranteeing congeniality\nbetween the imputation model and the analysis model, and because we use a BNP\napproach, parametric models are avoided for imputation. The performance of the\nmethod is assessed using simulation studies. The method is applied to data from\na cohort study of human immunodeficiency virus/hepatitis C virus co-infected\npatients.\n