Chiburis, Richard C.Das, JishnuLokshin, Michael2012-03-192012-03-192011-03-01https://hdl.handle.net/10986/3368This paper presents asymptotic theory and Monte-Carlo simulations comparing maximum-likelihood bivariate probit and linear instrumental variables estimators of treatment effects in models with a binary endogenous treatment and binary outcome. The three main contributions of the paper are (a) clarifying the relationship between the Average Treatment Effect obtained in the bivariate probit model and the Local Average Treatment Effect estimated through linear IV; (b) comparing the mean-square error and the actual size and power of tests based on these estimators across a wide range of parameter values relative to the existing literature; and (c) assessing the performance of misspecification tests for bivariate probit models. The authors recommend two changes to common practices: bootstrapped confidence intervals for both estimators, and a score test to check goodness of fit for the bivariate probit model.CC BY 3.0 IGOASYMPTOTIC DISTRIBUTIONBOOTSTRAPCONFIDENCE INTERVALSCRITICAL VALUECRITICAL VALUESDEGREES OF FREEDOMDEVELOPMENT RESEARCHDISTRIBUTION FUNCTIONDUMMY VARIABLEECONOMETRICSENDOGENOUS REGRESSORSEQUATION SYSTEMERRORERROR TERMSESTIMATORSEXHIBITSGOODNESS OF FITHYPOTHESIS TESTINGINSTRUMENTAL VARIABLESINSTRUMENTAL VARIABLES ESTIMATIONKURTOSISLIMITED DEPENDENT VARIABLELIMITED DEPENDENT VARIABLESLOG-LIKELIHOOD FUNCTIONLOGISTIC REGRESSIONMATRIXMAXIMUM LIKELIHOODMEAN SQUARE0 HYPOTHESISNUMBER OF OBSERVATIONSPREDICTIONPROBABILITIESPROBABILITYPUBLIC SERVICESRANDOM VARIABLESREGRESSION MODELRESEARCH WORKING PAPERSRESEARCHERSSAMPLE SIZESCIENCESSIMULATIONSIMULATIONSSKEWNESSSMALL SAMPLESTANDARD DEVIATIONSTANDARD ERRORSSTANDARD NORMAL DISTRIBUTIONSTATASTATISTICAL SOFTWARESTRUCTURAL PARAMETERST-TESTSTECHNIQUESTEST STATISTICVALIDITYWorld Bank10.1596/1813-9450-5601