On the Implications of Essential Heterogeneity for Estimating Causal Impacts Using Social Experiments

Ravallion, Martin

关键词: COUNTERFACTUAL;    DEVELOPMENT RESEARCH;    DISEASE;    ECONOMETRICS;    ESTIMATORS;   

DOI  :  10.1596/1813-9450-5804
RP-ID  :  WPS5804

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

Randomized control trials are sometimesused to estimate the aggregate benefit from some policy orprogram. To address the potential bias from selectivetake-up, the randomization is used as an instrumentalvariable for treatment status. Does this (popular) method ofimpact evaluation help reduce the bias when take-up dependson unobserved gains from take up? Such "essentialheterogeneity" is known to invalidate the instrumentalvariable estimator of mean causal impact, though one stillobtains another parameter of interest, namely mean impactamongst those treated. However, if essential heterogeneityis the only problem then the naïve (ordinary least squares)estimator also delivers this parameter; there is no gainfrom using randomization as an instrumental variable. Onallowing the heterogeneity to also alter counterfactualoutcomes, the instrumental variable estimator may well bemore biased for mean impact than the naïve estimator.Examples are given for various stylized programs, includinga training program that attenuates the gains from higherlatent ability, an insurance program that compensates forlosses from unobserved risky behavior and a microcreditscheme that attenuates the gains from access to othersources of credit. Practitioners need to think carefullyabout the likely behavioral responses to social experimentsin each context.