WP 2025-02 Causal Inferences Over Unobservables

AUTHORS: Glenn W. Harrison

ABSTRACT: Causal inference in economics involves inference about unobserved, latent relationships between variables, even when based on evidence from controlled experiments. In economics, most of the interesting structure defining causal hypotheses also reference unobserved, latent variables, and thought experiments defining and using those variables. One type of thought experiment defines crucial concepts in economics that are not directly observable. Another type of thought experiment posits counterfactuals, and uses economic theory to generate predictions of effects. We focus on preferences, beliefs and welfare, which play central roles in the counter-factual thought experiments of economics, and also rely on thought experiments to define them. Measurement of unobservables proceeds by devising controlled experiments or econometric inferences that match the definitions from thought experiments. In practice, economists have played “fast and loose” with the measurement of unobservable variables, even when paying lip service to the thought experiment used to defined the concept being measured. We characterize the nature of these problems in applied economics research.