Generative Process of Wage Gap: Analyses of Nation-Wide Employer-Employee Matched Data

Wednesday, 18 July 2018
Koji TAKAHASHI, Japan Institute for Labour Policy and Training, Japan
This poster aims to introduce a new method to analyze wage gap in the labor market. Wage is an important component that frames social stratification since most individuals are employed by firms. In principle, wage is determined within each firm although decisions at the industry-level or national-level may be in effect in some countries. However, neoclassical economists have ignored each firm’s role in wage determination and created a wage function that assumes the labor market as a single existence; such an assumption is unsuitable for a country like Japan where wage determination is highly de-centralized. Analyses of the generative process of wage gap have to consider two wage determination levels: which firm they are working for, and how much the firm pays them. Based on this, we analyzed Japan’s nation-wide employer-employee matched data (“General Survey on Diversified Types of Employment” in 2014) that contains: (a) independent variables such as gender, age, education, occupation, employment type, and tenure, (b) per-hour wage, and (c) number that specifies the firm. We considered that: (1) coefficients of the wage function estimated by the pooled OLS show the overall wage gap in the labor market, (2) those estimated by fixed-effect model show the wage gap within each firm, and (3) the opportunity to work for high-wage firms is calculated by subtracting (2) from (1). Results indicated that gender affects wage primarily within firms. Conversely, education affects both the opportunity to work for high-wage firms and how much employees are paid within each firm. In addition, these findings were consistent with a previous study that employed the same survey in 2010. This endorses the reliability of this method, and the contrast in the generative process of wage gap by gender and education suggests that different research approaches and policy measures are required to handle each variable.