Causal Inference and Understanding Causal Structure

Abstract

This thesis aims to show that explicit understanding of possible causal structures often aids in inferring the true causes from data. This is done by first understanding that causes are chains of counterfactual dependence. Insofar as experiments, active or natural are not perfect, data can easily support false counterfactuals. Even those tools especially designed to identify unbiased estimates, like instrumental variables, often fail. Causal structure explains the failure of these tools, but more importantly allows us to better identify which counterfactuals to reject or accept.