The role of work values and characteristics in the human capital investment of gays and lesbians

ABSTRACT

We show that educational outcomes of sexual minorities are consistent with efforts to mediate future discrimination. Gay men and lesbians obtain more years of schooling than heterosexual men and women, between 0.6 and 1.2 years. This difference is robust to controlling for observable characteristics for men but not women. Gay men and lesbian women also complete different college majors. Gay men are more likely to choose majors with lower levels of prejudice, higher levels of workplace independence, and occupations that emphasize relationships even though they pay less. Similarly, lesbian women choose majors with less prejudice and more workplace independence.

KEYWORDS:

• Human capital
• college major
• sexual orientation
• discrimination

JEL CODES:

• J1
• I21
• I23

Acknowledgements

The authors thank Sean Flaherty, Heath Henderson, Jessica Monnet and Leanne Roncolato for useful comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Black et al. (2000) utilize the 1998 through 1996 GSS-NSLS data to calculate educational obtainment.

2. Gay men are 11 percentage points more likely to have gone to college than married men and lesbian women are also 11 percentage points more likely to have attended college.

3. The feedback between performance in previous coursework and major choice implies that grade inflation may induce students to select into less rigorous majors, even if they are more adept at the more rigorous major (Ost 2010; Butcher, McEwan, and Weerapana 2014; Stinebrickner and Stinebrickner 2014; Sjoquist and Winters 2015). Equalizing the grading across majors has been shown to increase female participation in STEM fields, where grades are curved, at the expense of the Arts and Humanities, where there are no curves (Butcher, McEwan, and Weerapana 2014).

4. See Klawitter (2015) and Valfort (2017) for an overview of the literature on LGBT discrimination.

5. Unmarried partners exclude roommates, renters, and other household members who are not in a romantic relationship. We exclude respondents whose sex or relationship status was allocated by data administrators to avoid contaminating the sample, and any bias such contamination, would impose (Gates and Steinberger 2015). We refer to this sample as ‘gay’ and ‘lesbian’ but acknowledge it includes people who identify as bisexual as well.

6. In the model that we estimate for the main results, homosexuals and bisexuals are combined into a single group. Individuals who are mostly heterosexual are coded as heterosexual.

7. Respondents are asked ‘What about sexual relations between two adults of the same sex – do you think it is always wrong, almost always wrong, wrong only sometimes, or not wrong at all?’

8. The results are robust to utilizing a broader definition by counting respondents who answer ‘always wrong’ and ‘almost always wrong’ as prejudiced. These results are presented in Tables A14 and A15.

9. See National Center for O*NET Development for full details (National Center for O*Net Development 2015).

10. In preliminary analysis, we also investigated achievement, recognition, and working conditions. However, as Table 2 shows, these values exhibit significant correlation. We find that achievement, recognition, and working conditions are highly correlated with independence (Table A10). This is due in part to how O*NET calculates work values. Some of the components of one work value are highly related (but not always identical) to components of another work value. For example, autonomy is a component in the score for working conditions and authority in the workplace is a component of recognition. Of the four values that were highly collinear, we keep independence because there is a significant previous literature highlighting the role of independence in the wage gap for gay men (i.e. Tilcsik, Anteby, and Knight 2015; Martell 2018).

11. It is important to note that the O*NET data does not vary over time and therefore reflects a current view of work values by occupation. The levels of independence, support, and relationships in an occupation currently may be difference from the levels when an individual choose a college major.

12. The age of 18 is selected since this is the median age that most students enter college in the United States.

13. Specifically, we estimate the following regression. $Y_{gmt}=\alpha +{\text{β}}_{1}t+{\text{β}}_2(I_g\times t)+{\text{β}}_3(I_m\times t)+{\text{β}}_4(I_m\times I_g\times t)+{\theta }_m{I}_m+{\theta }_F{I}_g+{ϵ}_{gmt}$ Y is the average income of a graduate in major m in year t by sex. We express income in constant 1999 dollars. We include separate time trends for women ($I_g=1$) and major ($I_m=1$ if major graduated with a degree in major m). We predict the average income of male and female graduates with a degree in major m for each year between 1993 and 2015. Linear time trends interpolate averages for missing years.

14. See Table A2 for a more detailed comparison of LGB individuals and heterosexuals in the NHIS.

15. Hughes (2018) documented that there was a significant movement of gay men and lesbian women from STEM towards other majors during college. The data from the ACS reports completed majors. Therefore we cannot observe what major individuals started with.

16. From Table 1, we can see that top coding years of schooling shrinks the differences in years of schooling by sexual orientation. This may biases the estimates towards zero, meaning our results are a lower-bound.

17. These characteristics include responses to the following questions. ‘How often is the following statement is true of you? I am’: affectionate, conscientious, independent, sympathetic, moody, assertive, sensitive to the needs of others, reliable, understanding, jealous, forceful, compassionate, truthful, eager to soothe hurt feelings, secretive, willing to take risks, warm, adaptable, dominant, tender, conceited, willing to take a stand, am tactful, aggressive, gentle, conventional. ‘How often is the following statement is true of you? I’: am sensitive to the needs of others, have a strong personality, have leadership abilities, and love children.

18. We present coefficient estimates for full demographic characteristics in Table A4.

19. In Appendix Table A9, we show the years of schooling do not vary by cohabitation status for gay men and lesbian women after conditioning on demographics. We find no differences in schooling between LGB individuals who were married, living with a partner, or never married. LGB individuals who are separated, divorced, or widowed have education levels more similar to heterosexuals.

20. Results shown are for full-time workers. This pattern is robust to including part-time workers in the estimation sample. We also note that these significant estimates are present in our specification that may over control for preferences that may vary by sexual orientation by including controls for parenthood and current residence in urban areas. Thus, these estimates may be conservative.

21. The marginal effect combines the marginal effects of income and LGB*income. There is a positive effect of income for all men, but the effect is more negative for gay men. The net effect is a small, negative coefficient of $-0.05$. In addition to the potential income of a major, gay men and lesbians may have preferences over the safety or security of the income associated with a major. In additional results, available upon request, we find that the pattern of results discussed here is robust to specifications that also control for a proxy of the riskiness of potential income: the standard deviation of income within each major.

22. Throughout marginal effects calculated as $p_j(1-p_j){\text{β}}_p$, where $p_j$ is the percent of gay men in a major.

23. The marginal effect combines the marginal effects of income and LG*income. There is a positive effect of income for all men, but the effect is more negative for gay men. The net effect is a small, negative coefficient of $-0.05$.

24. There are two differences in the two approaches, which is to be expected since the OLS framework does not consider the covariance of major characteristics. In OLS specifications, the effect of relationships is no longer robust to the inclusion potentially endogenous industry controls and gays and lesbians are less likely to choose majors with higher levels of independence.