Гендерный разрыв в оплате труда: за рамками декомпозиции Оаксаки-Блайндера

Высшая школа экономики , 2022. № 8.

Lukyanova A.

The paper investigates how recent advances in the treatment effects literature can be applied to study gender pay inequality in Russia. The key parameter of interest is the mean unexplained gender wage gap. Using the RLMS HSE data for 2005–2020, we consider linear regression, Oaxaca-Blinder (OB) decomposition, and estimators based on propensity scores (PS) and machine learning algorithms. The latter group of methods provides more flexible and robust ways to model the impact of gender. PS-based estimators address the difficulties of comparing like with like. LASSO is used to select data-driven specifications of treatment and wage equations for all methods under consideration. In addition to the models with gender as a single treatment, we estimate the models with multivalued treatments interacting gender with relevant characteristics of female and male employment that reflect gender segregation and gender roles. We show that the samples of men and women are indeed extremely heterogeneous, and that heterogeneity has risen since 2005, mirroring an increase in gender segregation. Heterogeneity causes common support violations and aggravates the risks of model misspecification. Double robust estimators can tackle both problems. Unfortunately, a relatively small size of annual samples in the HSE RLMS imposes serious restrictions on the use of treatment-effects methods. With the HSE RLMS data, we did not obtain significant improvements in the estimates of the unexplained gender gap compared to the OB decomposition. However, since PS-based estimators have stronger causal interpretations, the similarity of the results enhances the credibility of the OB-estimates

Keywords: декомпозиция Оаксаки-Блайндера, гетерогенность, propensity score, индекс склонности, Oaxaca-Blinder decomposition, gender pay gap, гендерный разрыв в оплате труда, treatment effects, LASSO, common support, LASSO, оценка воздействия

Publication based on the results of: Structural Changes in the Labor Market: Results of the Last Decade (2022)