Partial Identification in Econometrics

Abstract

Econometricians long thought of identification as a binary event: a parameter is either identified or not. Empirical researchers combined available data with assumptions that yield point identification, and reported point estimates of parameters. Yet there is enormous scope for fruitful inference using weaker and more credible assumptions that partially identify parameters. Until recently, study of partial identification was rare and fragmented. However, a coherent body of research took shape in the 1990s and has grown rapidly. This research has yielded new approaches to inference with missing outcome data, analysis of treatment response, and other important problems of empirical research.KeywordsCounterfactualsDiscrete response analysisErrors in variablesGini coefficientIdentification regionLaw of Total ProbabilityNonparametric methodsParametric predictionPartial identification in econometricsReverse regressionSamplingStatistical inferenceTreatment responseJEL ClassificationsC14