Targeted Smooth Bayesian Causal Forests: An analysis of heterogeneous treatment effects for simultaneous versus interval medical abortion regimens over gestation

• DOI: 10.48550/arxiv.1905.09405
• Publication Date: 2019-01-01
• Authors: Jennifer E. Starling; Jared S. Murray; Patricia A. Lohr; Abigail R.A. Aiken; Carlos M. Carvalho; James G. Scott
• Keywords: Credible interval; Medical abortion

We introduce Targeted Smooth Bayesian Causal Forests (tsBCF), a nonparametric approach for estimating heterogeneous treatment effects which vary smoothly over single covariate in the observational data setting. The tsBCF method induces smoothness by parameterizing terminal tree nodes with smooth functions, and allows separate regularization of versus prognostic effect control covariates. Smoothing parameters can be chosen to reflect prior knowledge or tuned data-dependent way. use analyze new clinical protocol early medical abortion. Our aim is assess relative effectiveness simultaneous interval administration mifepristone misoprostol first nine weeks gestation. model reflects our expectation that varies gestation, but not necessarily other demonstrate performance on benchmarking experiments. Software available at https://github.com/jestarling/tsbcf/.