Estimation of time-varying average treatment effects using panel data when unobserved fixed effects affect potential outcomes differently

Abstract

This paper proposes a new approach to estimate the time-varying average treatment effect using panel data to control for unobserved fixed effects. The approach allows identifying the average treatment effect on the entire population, even if the fixed effects affect potential outcomes under treatment and no treatment differently, which can cause heterogeneity in treatment effects among unobserved characteristics. Note that a popularly used standard difference-in-differences approach can only identify the average treatment effect on the treated. Moreover, the proposed approach allows time-varying treatment effects. The approach exploits panel data with a specific structure in which the treatment exposure expands to the entire population over time. I apply the proposed approach to estimate the effect of the introduction of electronic voting technology for the reduction of residual votes in Brazilian elections.