Extensions of Rubin's Causal Model for a Latent-Class Treatment Variable: An analysis of the effects of employers' work-life balance policies on women's income attainment in Japan

Author Name YAMAGUCHI Kazuo  (Visiting Fellow, RIETI)
Creation Date/NO. July 2015 15-E-090
Download / Links


Combining inverse-probability weighting based on propensity scores and a semiparametric outcome model with a latent-class variable as an intervening variable, this paper introduces extensions of Rubin's causal model for the case where the treatment variable is a latent-class variable with indicators. Although the paper first introduces a method for the analysis of cross-sectional survey data, some extensions of the method for panel survey data analysis are also described. The method is especially useful when we have a set of mutually related categorical variables to characterize a specific latent characteristic of social contexts such as firms, schools, or neighborhoods, and when the latent characteristic is hypothesized to affect an individual-level outcome and we need to control for selection bias of people in different social contexts.

An application, which is based on data for employees and employers collected in 2009 by the Research Institute of Economy, Trade and Industry in Japan, focuses on the effect of employers' work-life balance policies on female regular white-collar employees' income. Six dichotomous indicators of policies are employed. Variables tested for possible confounding factors include individual human-capital and labor-hour variables and some firm-level exogenous variables. The analytical results show that although a certain portion of the effects of employer's work-life balance policies on income are explained as a result of such a selection bias that firms of larger size have a higher prevalence rate of those policies and at the same time higher average income for employees, the positive effects of those policies on income among female employees still remain significant after the elimination of the selection bias.