economic project
Gender Differences in Returns on Education
I. Introduction
For a society that claims to value equality in the workplace, the gender gap in wages in America seems awfully persistent. This paper investigates the differences in wages between men and women at different levels of education using data from a sub sample of the Current Population Survey (2012). Such analysis will help reveal the nature of the gender gap, and may help identify the segments in which discrimination in the workforce may exist. Using linear regressions, I first confirm the wage gap in the data and that returns to education are positive. Next, I use interaction variables to illuminate gender differences on returns at the different levels of education (high school, bachelor’s, and master’s). Overall, I find that females see higher returns than men for completing high school and college, but not for graduate school.
II. Data
The data set consists of 999 observations of working individuals between the ages of 18 and 54:
The average age in the sample is 39.11 years old. On average, individuals made $16.92 an hour with a standard deviation of $9.80. The average highest grade completed, 13.28, shows that most graduated high school. 88% of the sample have high school diplomas, 24% hold a bachelor's degree, and 7.4% have completed at least a master’s. A majority was white (81.6%). 10% of the individuals were black, 9% were other races. 22.7% of the workers were parttime. Approximately half of the sample was female. The following histogram shows the distribution of education level:
Most of the data lies on the milestone years. The 12, 14, 16, and 16 areas represent high school diplomas, associate's, bachelor’s, and master’s degrees. However there is some ambiguity at the 14th grade level: these observations could be both associate’s degree holders or four year college dropouts.
III. Empirical Methodology
To compare gender differences in the returns on wages at different levels of education I run a linear regression on log wages:
The particular variables of interest are B9, B10, and B11. These interaction variables will show the additional percentage point increase or decrease in wages that females accrue at the different levels of education.
Because the distribution of wages is skewed right, I choose to use log wages, which are more normally distributed and thus may increase the goodness of fit. Based on prior research, I expect to see positive, though diminishing, returns to age. Thus, one would expect B1 to be positive and B2 to be negative. Income inequality between whites and blacks is well established in economic literature, so I expect B3 to be negative. B4 is also likely negative since many of the higher paying jobs would be full time. I expect a negative coefficient on the female variable, matching my hypothesis that the wage gap is present in the data. Lastly, the coefficients on the dummy variables for completion of high school, completion of a bachelor’s, and completion of a master’s are expected to be positive because higher education levels allow individuals to access higher wage positions.
There are some potential concerns with this methodology. First, there is inevitably a sample selection problem since we are only looking at the data of employed people. For example, if it were the case that being female lowered the probability of being employed due to discrimination, then the sample observations would only represent the females with a relatively high marginal productivity of labor. Thus, B5 might underestimate the true magnitude of the wage gap. Another possible concern is omitted variable bias. For example, living in a city is likely positively correlated with higher wages. Moreover, it may be the case that having a master’s degree is correlated with a higher probability of living in a city. These positive correlations would cause an upward bias on B8.
IV. Results
Column 1 shows the baseline specification which includes all the race dummy variables and does not include the interaction variable for female and education. All the variables are significant at the 5% level except for the race dummy variables for American Indian, Asian, Mixed, and Hispanic. All signs on the significant coefficients are consistent with the predictions discussed in the previous section. I conducted an F (4,986) test on the variables for American Indian, Asian, Mixed, and Hispanic and found that none had a statistically significant impact on wages, all else constant (p=.18). Thus, I decided to remove them from the regression in column 2. The adjusted R^2 for column 1 was .164.
In column 2, I add the interaction variables for females and education level. The adjusted R^2 slightly improved in this specification to .168. The interaction variables are interpreted as follows: females see an additional return of 5.6% compared to men for graduating high school, an additional 24% return compared to men for graduating college, and a negative 26% return compared to men for a master’s degree, all else equal. All the interaction variables are significant at the 5% level. The signs on the rest of the coefficients are consistent with the initial predictions, except for the coefficient on bachelor’s degrees. The statistical insignificance of the bachelor’s variable in this regression can be explained as a point in the education level where female wages catch up to the male wages (i.e. the wage gap closes). In fact, my regression predicts that at the bachelor’s degree level of education, female wages average about 8 percentage points higher than men. However, the gap reappears for graduate level jobs. At that point, the model predicts that men see about a 20% higher return than women, all else constant. The following graph shows how the wage gap is “pinched” for bachelor’s degree wages:
V. Conclusion
In theory, it is not surprising that a wage gap persists at lower levels of education. Jobs that do not require degrees tend to involve more manual labor, thus have positive returns on physical strength. According to my results, female wages catch up to male wages with a bachelor’s degree, but lag behind male wages at the graduate level. This may be evidence of gender discrimination for senior positions. All in all, these results lend insight into the nature of the gender gap: Since the gap closes with a bachelor’s degree, there does not seem to be evidence that women earn less doing the same jobs as men. It seems a more likely explanation for the the overall wage gap is that a disproportionate amount of men get hired for top paying positions. Further investigation could involve using linear probability models to test gender differences in the probability of being employed in senior executive positions.