This workshop will cover Kitagawa and the DiNardo, LeMieux, Fortin (Econometrica, 1996) reweighting decompositions using examples from Demographic and Health Survey Data. Such decompositions are best suited to research questions of the form: Group A differs from Group B in the distribution/mean/other statistic of variable X. How much of that difference in X remains if we match/hold constant/account for the difference in variable(s) Y (and Z)? Kitagawa decompositions allow the researcher to transparently decompose a difference in means into two components; DLF decompositions allow researchers to decompose differences in a wider set of statistics (such as distributions) and to account for a larger number of explanatory variables. We will contrast the methods with the Oaxaca-Blinder decomposition but not cover the latter in depth.
The workshop will discuss several examples in which the presenters and co-authors have applied nonparametric decomposition methods to Demographic and Health Survey data to understand differences in health outcomes. Among other examples, we will show that differences in infant mortality can across social groups in India can be explained by sanitation exposure, and that differences in neonatal mortality between babies born in different facility types in India cannot be explained by factors related to the risk profile of the birth. We will present Stata .do files for the examples we discuss and make them available to participants. We will invite participants to share their own research questions to consider these methods’ suitability for answering those questions.