Estimating Treatment Effects from Counts of Binary Outcomes: A Conditional Likelihood Estimator of Relative Risk

Abstract

This paper introduces the conditional likelihood estimator of relative risk (CLERR). The CLERR estimates the relative risk of an outcome analogously to the way the conditional logit estimates an odds ratio. Aside from the fact that relative risk is often the preferred measure of association, the CLERR has superior statistical properties, including both unbiasedness and effciency in small and large samples. The CLERR can be thought of as an exact matching approach which allows estimation of treatment effects for binary outcomes, without the need for structural assumptions about the in influence of confounding variables. We apply the CLERR to World Bank's Enterprise Survey data and show that firms with female owners are signifcantly more likely to export than similar firms with only male owners.

Keywords: conditional likelihood, relative risk, treatment effects, matching, female owners, exporting

JEL Classification: C13, C14, C21, C25, F14, J16