SpaCE: The Spatial Confounding Environment

SpaCE: The Spatial Confounding Environment is a benchmarking dataset for causal inference incorporating spatial structure. In particular, SpaCE datasets contain real confounder and exposure/treatment data inspired by environmental health studies. The synthetic outcome and counterfactual are generated according to recommended practices for causal evaluation by mimicking the real outcome data distribution learned with machine learning and neural network methods. Spatial confounding is achieved by masking influential confounders in the learned model.