Difference-in-Differences Methods in Public Finance

Abstract

Recognizing that cross-sectional data are often insufficient to address the identification problems associated with estimating the effect of government taxation or spending, economists engaged in public finance research often utilize longitudinal data that span the period over which a policy change occurred. As economic data have proliferated over the last decade, uses of the difference-indifferences design and its variations have become more numerous. Nevertheless, published research that invokes difference-indifferences commonly fails to present evidence and reasoning that enable the reader to properly evaluate the causal claims under investigation. In this paper, we examine the threats to internal validity that exist when using difference-indifferences for causal inference and review variations of the design that can be used to address these threats. Next, we survey the public finance literature in order to examine the ways that these threats are addressed in practice. We conclude by proposing a number of recommendations for researchers to consider as they implement difference-indifferences as an empirical strategy.

Promise for Whom? “Free-College” Programs and Enrollments by Race and Gender Classifications at Public, 2-Year Colleges

“…As noted earlier, the construction of a comparison group is vital to estimating aggregate program impacts in a DiD framework (St. Clair & Cook, 2015).…”

The effects of the Summer All Out Foot Patrol Initiative in New York City: a difference-in-differences approach

“…The 121 st Precinct in Staten Island was summarily excluded because this area was not officially recognized as a precinct jurisdiction until the summer of 2013. 13 Empirical testing of common group trends is akin to a placebo treatment procedure whereby crime outcomes are regressed on interactions between a treatment indicator and a full series of − 1 dummies for months (St. Clair & Cook 2015). Coefficients on the interaction terms represent the conditional outcome distribution over time.…”

“…To proceed, we explicitly sought to model trend divergence in monthly shootings across treatment and control jurisdictions. Model 3 is an adaptation of the classical DiD specification adjusting for calendar effects in the months before the onset of each SAO initiative: of pre-intervention period dummies for each individual calendar month preceding program exposure(Ryan et al 2015;St. Clair & Cook 2015).…”

Taken for Granted? Effects of Loan-Reduction Initiatives on Student Borrowing, Admission Metrics, and Campus Diversity

“…The DD method relies on the parallel trends assumption, which states that the trends in outcomes for the treatment and comparison groups would have been the same in the post-adoption period in the absence of treatment. One method of assessing the plausibility of this assumption is through visual inspection of the average values of the outcome variables in the pre-treatment years (St. Clair and Cook 2015 ). Given the high concentration of policy adoptions in the 2007–2008 and 2008–2009 academic years, we can observe whether treatment and comparison institutions appear similar in earlier years.…”