Decomposition of Inequality Among Time-Constant Variables By Counterfactual Mediation Modeling: Evidence from the Gender Wage Gap in Japan
This paper links conventional decomposition methods such as DiNardo-Fortin-Lemieux (DFL) based on propensity-score weighting and counterfactual mediation modeling together by introducing sequential ignorability assumption discussed in Imai et al. (Statistical Science 2010). The reason for linking these two methods is that conventional decomposition methods typically control for post-birth variables that lie on the causal pathway from gender or race (which are basically randomly assigned at birth) to wage but neglect the potential endogeneity that may arise from this approach. Moreover, we never directly test the assumptions that lie in conventional decomposition methods and mediation modeling. Based on the newer literature on counterfactual mediation modeling, this paper therefore shows more attractive identifying assumptions and the sensitivity of the results to different sets of assumptions.
This paper also aims to bring time-constant variables back to the center of causal analysis in social inequality studies. The analysis focuses on the decomposition of the gender wage gap in Japan. Empirical results show that explained components with four mediators (education, occupation, employment status, and position) account for around 50% of the gender wage gap in hourly wage.