Primary Submission Category: Causal Inference and SUTVA/Consistencies Violations
Controlling for spatial confounding and spatial interference in causal inference: Model selection advice from a computational experiment
Authors: Tyler Hoffman, Peter Kedron,
Presenting Author: Tyler Hoffman*
Working with spatial data raises the possibility of encountering unique issues in causal inference—namely, spatial confounding and spatial interference. A blossoming literature on spatial causal inference is growing to address these issues, largely via regression adjustments in existing causal models. This research analyzes the usage of spatial causal models under a priori knowledge and a priori ignorance of dependence structures in a spatial dataset. We test whether spatial causal models accurately capture treatment effects in the presence of spatial confounding and interference by fitting these models on various spatial data scenarios. Based on the results of these experiments, we develop practical workflow guidelines based on the relative performance of spatial and nonspatial causal models across data scenarios. All experiments use Bayesian estimation techniques for additional uncertainty quantification. In parallel, we build a Python software package of spatial causal models and data simulators to facilitate the widespread use of these models and to enable reproduction of this work.