This paper introduces Front-door Difference-in-Differences Estimators, an extension to front-door estimators that provides identification under one-sided noncompliance and exclusion restriction, even when relaxing parallel trends assumptions. It demonstrates that these methods can recover experimental benchmarks in a job training study, while also revealing small positive turnout effects from Florida's early in-person voting program in 2008—a finding that counters recent claims of negative impacts.
Data & Methods: Combines front-door estimators with difference-in-differences framework. Demonstrates applicability through two empirical examples: a job training study and an analysis of Florida's early in-person voting policy from 2008.
Key Findings: Successfully recovers benchmarks without control units; shows positive effects for early in-person voting, challenging prior negative estimates.
Why It Matters: Offers clearer identification strategies even when standard assumptions are relaxed. Provides an independent methodological approach to evaluate policies like early voting programs.
| Front-door Difference-in-Differences Estimators was authored by Adam Glynn and Konstantin Kashin. It was published by Wiley in AJPS in 2017. |
