Chapter 11: Gender Gap in Earnings: Methods & Evidence

The family Gap

Direct impact of family responsibilities on the wage gap includes measures of family and marital status as an earning regression along the with the traditional measures of human capital for both men and women. Marriage increased mens earnings by 11%, women by 4% Children had no effect on men's earnings, women earnings decreased by 10% for one child 20% for two or more children Human capital differences have shrunk and will continue to shrink, differences in value of education and work experience to men and women have also shrunk and will continue to shrink. This leaves marriage and children factors that disproportionately affect women and their earnings.

Regression analysis of earnings difference

Earnings(i) = α+β1(years educ(i)) +β2(years work exp(i))+...δ(female(i)) + μ(i) Notation: i subscript = these variables vary by subject/case α and β = regression coefficients μ = error term ; represents the effect of other factors that affect Y but are unobserved in the regression δ = the dummy variable as a regression coefficient to indicate sex. This measures the impact on earnings of being female/male, controlling for all other explanatory variables used in the regression (like yrs of edu. and work experience). If δ<0 and statistically significant, women are paid less than men with the same skills. Changes to the estimated values of δ at diff points in time inform changes in the gender earnings gap.

Linear pay and non linear pay

Goldin➔ gender differences within occupations are more important than gender differences between occupations. Women tend towards occupations with greater job flexibility which typically have lower wages linear pay➔ earnings and hours are directly correlated (one can pick where the other left off easily, suitable for part time work) nonlinear pay➔ earnings increase more than proportionately with hours worked, penalty for part time work is large (jobs that require face to face interactions, tasks are not easily communicated from worker to worker - one can't pick up where the other left off) jobs with non linear pay such as law and business have a significantly higher wage gap than jobs with linear pay like pharmacy.

Convergence in the 1980s

Increase in YRFT median earnings ratio. On the "explained" side➔ the gender gap in average work experience fell substantially between 1979 and 1988 from 7.5 years to 4.6 years.reflecting the more continuous nature of women's labor force participation. on the "unexplained" side➔ The wage structure changed in the 1980s narrowing some of the rewards differences between men and women that favored men. This resulted a convergence of both skills and returns to skills.

Work Histories and Human Capital (1970s)

Mary Corcoran and Greg Duncan examined gender wage differences as of the mid 1970s. chart on page 295 Bottom line of the study is that only 44% of the wag gap was explained for white women and 33% for black women, even thought the researches used an extensive set of explanatory variables. The researcher concluded that "the wage advantages enjoyed by white men cannot be explained solely or even primarily by superior qualification or more attachment to the labor force." (no shit)

Regressions Earning equations

Men: Yim=βmXim +μim Women: Yif = βfXif + μif β^m is the estimated effect of X on Y for men β^f is the estimated effect of X on Y for women

natural study: women in symphony orchestra

Switch to blind auditions increased the proportion of women among new hires by 30%

Limitations of the Oaxaca Decomposition

The burden of proof is on the explanatory variables. Not testing directly for discrimination but rather looking for indirect evidence in of wage differences. It is possible to include too many X variables, especially variables that are the product of discrimination themselves It is also possible to have too few X variables so that the skills of men and women are not being accurately and appropriately compared. ➔ omitted variable bias.

1960-1980 Gender Earnings Trends

The gap in educational attainment was eliminated and the gap in work experience that favored men expanded The fact that working women had a higher avg edu. level than working men does NOT mean women are more educated than men; it just means that the labor force was more selective w women than men. Women's earning did not fall relative to men's earnings; instead they stayed relatively constant. So the rewards for women's skills must have increased over time relative to the rewards for men's skills to offset the decline in women's avg skills compared to men's.

how does regression over estimate discrimination

Too few X variables could overestimate discrimination. if a researcher leaves out some important variable in which men greatly outscore women and which is important in deterring earnings for both men and women, omitting that variable might cause the researcher to overestimate discrimination and underestimate differences in skills. It can always be argued that an important determinant of earnings is missing from the regression equation and that this determinant is responsible for the difference in male and female earnings. For example, suppose our regression shows that men earn more than women with the same levels of education, years of work experience, occupation, and location. But further suppose that the difference in men's and women's annual income is due totally to the choice of major in college. Since we don't have a "major in college" variable in the regression, we aren't comparing men and women who chose the same major, so we will wrongly attribute the effect of the major to discrimination.

how does regression underestimate discrimination

Too many x's can mask discrimination Suppose that we could measure the jobs that men and women held so that we were then comparing wages for men and women doing the exact same job. We might well find that there were little or no earnings differences between men and women, and thus gender earnings differences is due to differences in skills (explained). but suppose discrimination operated to influence the jobs that men and women got, so that, for example many women ended up in jobs that were not particularly good relative to their skills (think women becoming nurses and men becoming doctors) controlling too much, or controlling for the job in that case.

Audit or correspondence study

economic experiment in which testers are sent out in matched pairs. The testers are alike as possible "on paper" and differ only in the interested variable, i.e. gender, race, etc. these have found that in high end restaurants men are 5x more likely to get an offer than women. this could be either employer discrimination or customer discrimination london study➔ women were less likely to be called back for engineering positions but men were less likely to be called back in mixed/female occupations Australia study➔ found that men were much less likely to receive call backs for jobs that were overwhelmingly female. audit studies➔ useful but they only examine call back rates not whether the individual is hired or the salary that is offered. they are not appropriate for measuring discrimination at higher level of the corporate hierarchy or among jobs filled though personal networks.

The Gender Wage Gap

ȳm-ȳf, the difference in average earnings ȳm-ȳf= β^m*x̅m - β^f*x̅f this suggests that the average earnings for men and women could differ either because X differs or because β differs. Another way to say that is because the average level of labor market skills differ (human capital explanation) or because the market value of those skills differs (discrimination)

Oaxaca Decomposition

ȳm-ȳf=[β^m(x̅m-x̅f)]+ [x̅f(β^m-β^f)] The "explained" portion--The first term in the brackets is the degree difference in average levels of human capital multiplied by, β^m, the value of X for men. This is the dollar amount of the gender wage gap that can be explained by difference in average skills. This would be there even if there was no labor market discrimination The "unexplained" portion--the second term is the difference in the market value of skills for men and women multiplied b the average skill level of women. This is the difference in monetary return for the same average skills. This is the result of discrimination