Sexual orientation and labour supply

Abstract

This is the first study of the effects of sexual orientation on labour supply. After discussing various reasons to expect that labour supply could differ by sexual orientation, the 2001 Current Population Survey is used to test for differences in both labour supply and labour market status (part-time, full-time, and not employed). It is found that gay men supply less labour than married and unmarried heterosexual men. With regard to women, it is found that lesbians supply more labour and are more likely to be employed full-time than either married or unmarried heterosexual women. Extensive controls are used and links to the discrimination literature are found.

Acknowledgements

The authors thank Karen Conway, Ju-Chin Huang and participants of the Economics Seminar Series at the University of New Hampshire for helpful comments.

Notes

¹ We focus on the link between discrimination and unemployment duration or part-time employment choices due to the fact that pure wage discrimination could increase or decrease labour supply because of the counter-acting forces of income and substitution effects.

² Another CDC (2001) study finds that an average of 26% of gay men are HIV positive among patients at 23 STD clinics in 13 separate metropolitan areas in the USA. This compares with 2.3% of all non-gay male patients at those clinics. These data ended in 1994. A report cited in the August 10, 2000 edition of the San Francisco Chronicle estimates that between 30 and 32% of San Francisco's gay males are HIV positive.

³ Pencavel (1986) discusses this model in great detail.

⁴ The 34 hours per week dividing line corresponds with most of the literature, and is consistent with the federal government's definition of full versus part-time work.

⁵ Powell (1998), Baslevent (2002), and Bonnie (2003) specify the employment status as an ordered probit model. The argument to use an ordered probit model is based upon the ad hoc assumption that there is a ‘natural’ ordering among the three choices (not work, part-time, and full-time). This issue is still open in the literature. Given the lack of empirical support for advocating either the ordered probit model or the MNL and the fact that the ordered probit model requires the estimation of addition parameters, we use the MNL model. As a test for robustness, we also run an ordered probit model, not reported in the paper, and our main results are not affected by the choice of the econometric model.

⁶ The hypothesis of unit variance is highly restrictive, but it has been largely used in the empirical research because under other circumstances it would be only possible to estimate the parameters scaled by the variance of the disturbances.

⁷ The results of the econometric analysis, discussed below, did not change either when individuals younger than 30 years where excluded from the sample or when individuals.

⁸ Conway (1997) provides a useful discussion of the rationale for eliminating older workers from the study of labour supply.

⁹ Household income less individual's personal income.

¹⁰ In a model not reported in the paper, which does not control for the interaction term between Gay and joblessness, the estimates support that gay men are less likely to choose full-time jobs and more likely to choose part-time employment than are both heterosexual married and unmarried men.

¹¹ The difference between gay men and unmarried heterosexual men who cohabitate with women is significant at the 10% level.

¹² The differences are statistically significant at 99% of confidence.

¹³ The null hypothesis that lesbians who have dependents are as likely as partnered heterosexual women to work full-time is not rejected at the standard levels of significance.

¹⁴ The coefficients of state fixed effects were not included in the tests because first, they are not the coefficients of interest and, second, they were adding too much noise in the test.

¹⁵ McFadden (1984, p. 1414) points out that ‘empirical experience is that the MNL model is relatively robust, as measured by goodness of fit or prediction accuracy, in many cases where the IIA property is theoretically implausible.’

¹⁶ We estimated a probit model whose dependent variable is binary (0 = not work, 1 = work). The set of explanatory variables includes predicted wage, number of dependents, non-wage income, and net household income. We predicted the wages of individuals who were not working by running a standard regression of wage determination.

¹⁷ A linear model was also estimated, but we find no significant differences between the linear and the log-linear models.

¹⁸ We reject the null β _lesbian = β _{partnered non-married} in favour of the hypothesis that β _lesbian > β _{partnered non-married} at 95% of confidence (t-test = 2.02).

¹⁹ We also tested a cross-term between unmarried and joblessness. This variable was not significant and is not reported.

²⁰ We cannot reject the null hypothesis that heterosexual married women and black lesbians supply an equal amount of labour at the standard levels of significance.

²¹ This difference is significant at the 1% level (t-test = 4.30).