A Causal Graphs - based approach for assessing gender disparities: an application to child health & nutrition in China

Most of gender statistical measures proposed in the last decades are in fact composite indicators, i.e. weighted linear combinations of basic statistics such as ratios, percentages etc. Composite indicators then involves several arbitrary choices -for instance the weighting/aggregating system, variables selection, standardization affecting both indexes transparency and interpretation. Furthermore gender inequality is a complex latent phenomenon, a collection of disparate and inter-linked issues that can be hardly caught in a single indicator. The development of statistical tools and ad hoc models is then required. The aim of this work is to explore the potential of graphical models as a language able to clearly represent the complex relationships among variables involved in the statistical measuring the gender disparities. In particular we will focus on causal graphs allowing to deep and interpret the causal mechanism that may originate gender gaps as well as to explore the effects of gender tailored policies. Causal models indeed provide transparent mathematical tools to formulate the assumptions underlying all causal inference, to translate them in term of joint distribution and to read off the conditional independences using the d-separation criterion (Pearl 2000). It is thus possible deriving causal effects in non-experimental studies, representing policies’ effects and interventions through the do operator, controlling confounders and interpreting counterfactuals. We show the potential of such models through an application to real data from China Health and Nutrition Survey 2011 ; in particular we explore the eventual existence of gender discrimination in children’ nutrition and health as possible indicator of preference for sons. The analysis takes in exam socio-demographic, economical as well as biological variables. Resorting to the PC algorithm and the IDA algorithm, we aim to learn the underlying causal structure and to estimate causal effect of siblings on children’ nutrition from observational data.