Comparison of Alternative Propensity Score Approaches for Causal Inference from Observational Data

Sarah R. Brauner, University of Michigan

Demographers are growing increasingly concerned with the problems associated with making causal inferences from observational data. Statisticians and social scientists have both developed numerous approaches for circumventing the problem of the counterfactual, including measurement adjustments and alternative analytic strategies. In this paper we apply linear-model measurement adjustments and matching techniques, along with propensity score analysis, to studying the relationship between health services and family formation behavior in the developing country Nepal. We find that estimates of the effect of the availability of various maternal and child health services on the use of permanent contraceptives are highly sensitive to measurement and model choices. None of our measures were found to have significant effects across all of the models we present. However, availability of child vaccinations and oral rehydration therapy were found to have positive and significant effects on permanent contraceptive use in our most restrictive model.