Disproving Causal Relationships Using Observational Data

Economic theory is replete with causal hypotheses that are scarcely tested because economists are generally constrained to work with observational data. This article describes the use of causal inference methods for testing a hypothesis that one random variable causes another. Contingent on a sufficiently strong correspondence between the hypothesized cause and effect, an appropriately related third variable can be employed for such a test. The procedure is intuitive, and is easy to implement. The basic logic of the procedure naturally suggests strong and weak grounds for rejecting the hypothesized causal relationship. Monte Carlo results suggest that weakly-grounded rejections are unreliable for small samples, but reasonably reliable for large samples. Strongly-grounded rejections are highly reliable, even for small samples.