e-Archivo (Carlos III University of Madrid) · 2012 · 99 citations · 12 references
Open access
Public HealthBiostatisticsCausal InferenceExperimental DesignCausal Effect EstimationTreatment EffectsParallel Trends AssumptionTreatment Effect EstimationTreatment Effect IdentificationTreatment Effects Estimation
The core assumption to identify the treatment effect in difference-in-differences estimators is the \nso-called Parallel Paths assumption, namely that the average change in outcome for the treated in \nthe absence of treatment equals the average change in outcome for the non-treated. We define a \nfamily of alternative Parallel assumptions and show for a number of frequently used empirical \nspecifications which parameters of the model identify the treatment effect under the alternative \nParallel assumptions. We further propose a fully flexible model which has two desirable features \nnot present in the usual econometric specifications implemented in applied research. First, it \nallows for flexible dynamics and for testing restrictions on these dynamics. Second, it does not \nimpose equivalence between alternative Parallel assumptions. We illustrate the usefulness of our \napproach by revising the results of several recent papers in which the difference-in-differences \ntechnique has been applied.The core assumption to identify the treatment effect in difference-in-differences estimators is the \nso-called Parallel Paths assumption, namely that the average change in outcome for the treated in \nthe absence of treatment equals the average change in outcome for the non-treated. We define a \nfamily of alternative Parallel assumptions and show for a number of frequently used empirical \nspecifications which parameters of the model identify the treatment effect under the alternative \nParallel assumptions. We further propose a fully flexible model which has two desirable features \nnot present in the usual econometric specifications implemented in applied research. First, it \nallows for flexible dynamics and for testing restrictions on these dynamics. Second, it does not \nimpose equivalence between alternative Parallel assumptions. We illustrate the usefulness of our \napproach by revising the results of several recent papers in which the difference-in-differences \ntechnique has been applied
12