Measuring Heterogeneous Effects of Environmental Policies Using Panel Data

Abstract

To measure the effects of environmental policies, researchers often combine panel data with two-way fixed effects models. This approach implicitly assumes that the distribution of the policy effect is constant across units and over time. Yet many environmental policies have effects that differ depending on the unit exposed to the policy and the period in which the policy is applied. In this setting we detail why the model parameters generally do not capture a useful measure of the effects. We then show that in a multiperiod setting, if the policy is applied in only one period, then the model parameters do capture a useful measure of the effects. In these settings, appropriate inference is based on cluster-robust standard errors. Because the resultant t-statistic may yield unreliable inference when clusters are heterogeneous, we present an appropriate measure of cluster heterogeneity and describe how the measure should be used to guide inference.