Essays on the Gender Wage Gap and Intergenerational Mobility in the Presence of Sample Selection

Abstract

In this dissertation, I explore how the gender wage gap across the distribution and different indicators of intergenerational mobility in education are impacted by sample selection. In the first chapter, I analyze the gender pay gap across the wage distribution in Chile. I use quantile regression and correct for sample selection using a copula-based methodology. My results highlight the importance of heterogeneous effects and selective participation in gender pay gaps. If men's and women's employment rates were equal, the gap would be approximately 30 percentage points on average across all quantiles. However, the gap oscillates between 25 and 35 log points at the bottom half of the wage distribution but increases to approximately 50 log points in the upper quantiles, evidencing a "glass ceiling'' effect. Finally, I decompose the gap into "structural'' and "composition'' effects, concluding that it is explained mainly by differences in rewards for observable labor market characteristics and not by differences in the distribution of those characteristics. The second chapter (co-authored with Ercio Munoz) describes qregsel, a Stata module to implement a copula-based sample selection correction for quantile regression that was recently proposed. The command allows the user to model selection in quantile regressions using either a Gaussian or a one-dimensional Frank copula. We illustrate the use of qregsel with two examples. The third chapter (co-authored with Ercio Munoz) studies how the measurement of intergenerational mobility (IGM) in education (which requires linked information about children's and parents' educational attainment) is affected by sample selection. This sample selection emerges as several economies do not offer better data alternatives to estimate IGM than the use of coresident samples, which may yield biased estimates as coresidence is not random. In this line, a recently published paper concludes that the intergenerational correlation coefficient is less biased than the intergenerational regression coefficient as a measure of relative IGM, and researchers should move away from using the latter. We re-examine this claim. In addition, we use two data sources for 18 countries to provide evidence of the extent of coresidence bias on an extensive set of IGM indicators of absolute mobility, relative mobility, and movement. We show that there are indicators with varying coresidence bias going from less than 1% to more than 10%. Still, some mobility indicators with minimal bias produce high levels of re-ranking that make them uninformative to rank economies by the level of IGM. In contrast, other indicators with a large bias have more reliable rankings.