The Married Men’s Wage Premium and Discrimination: 
Evidence from California and Texas

Amy B. Schmidt
Saint Anselm College

Introduction
     It has been well established in the literature that married men earn between 10 and 40% more than comparable single men. Several theories have been supported empirically: 1) unobservable factors that make men desirable spouses are correlated with market production and men with these characteristics are more likely to be selected into marriage; 2) marriage makes men more productive; 3) employers discriminate in favor of married men or against single men. This paper’s focus is on the last of these.  Specifically, the relationship between the size of men’s marriage premium and sexual orientation variables for the individual and Metropolitan Statistical Area (MSA) is investigated. The primary finding is that in both Texas and California, the marriage premium for men over 40 years is between 3 and 4 percentage points greater for every additional 1 percentage point increase of single men in same sex unmarried partnerships (SSUPs) in the MSA. In San Francisco, the effect accounts for over 40% of the marriage premium. There is no significant relationship between the marriage premium and the percentage of single men in SSUPs for men under 40. The results also show that in both states, the greater the percentage of single men in SSUPs in the MSA, the higher the earnings for all men. Being in an SSUP is associated with higher earnings relative to other single men in California, but there is no such significant relationship in Texas. 
     An additional finding is the strong correlation between the reported percentages of single men SSUPs and the percentages of self-identified gay men in California MSAs in the California Health Interview Study (CHIS).¹ This finding is significant because it indicates that the Census’ unmarried partnership data is a reliable proxy for the percentage of gay men in MSAs. This may have implications in other areas of research. The marriage market literature may be one area where this data could be of use.
Literature Review
     The explanations given for the marriage premium are not mutually exclusive, and evidence has been found for all. The selection hypothesis has been supported by Kashinsky (2004), who uses the National Longitudinal Survey of Youth (NLSY) and the Current Population Survey (CPS) to find that married men have higher earnings growth in periods before marriage, compared to those who do not marry. Twins data is used to confirm support for the selection hypothesis. Using the Panel Study of Income Dynamics (PSID), Nakosteen and Zimmer (1987) find that men with high unexplained incomes are more likely to be married in the following time period. Cornwell and Rupert (1995) and Loh (1996) find no evidence of productivity differences using fixed effects models. Hersch and Stratton (2000) use data on household responsibilities and attribute up to half of the premium to selection.² Dougherty (2006) finds evidence that marriage is a signal of maturity and that the marriage premium begins to accrue 5 years before marriage and continues to increase until it levels off at about 20% a few years after marriage.
     On the other hand, Antonovics and Town (2004) use different twins’ data and find no evidence to support the selection hypothesis. Similarly, Ginther and Zavodny (2000) look at those who had children shortly after marriage (shotgun weddings) and those who did not. They estimate that at most 10% of the marriage premium is due to selection.
     Kenny (1983) was the first to find support for the productivity hypothesis. Using the Coleman Rossi Life History Study, he determines that men are more likely to engage in on the job training after marriage than before. Gray (1977), Chun and Lee (2001) and Bartlett and Callahan (1984) all find that specialization within the family (Becker, 1981) is responsible for a large portion of the marriage premium.³ Mehay and Bowman (2005) use Navy personnel data and find that married men receive significantly better performance reviews and are more likely to be promoted. Controlling for selection into marriage does not alter their findings. Like Korenman and Neumark (1991), who use NLSYM data, they find that longer marriage duration results in a higher marriage premium. Akerlof (1998), like Dougherty (2006), finds that marriage is related to maturity, but he concludes that marriage itself results in maturity, thereby crediting marriage with an increase in productivity.
     There has been much less investigation regarding the discrimination hypothesis. Hill (1979) uses PSID ninth round, which includes detailed labor market information. She finds that workers with greater family obligations get higher wages. For men she finds that even after controlling for years of experience, job tenure, hours worked and other detailed labor market variables, the marriage premium remains stable at between 25-30%. She suggests that one possible reason for her findings is discrimination by paternalistic employers. Zimmer (2006) hypothesizes that if employers believe that marriage is an indicator of stability or maturity, then they might pay for those traits in the form of higher wages to married men. He finds evidence that there is statistical discrimination favoring those who are married where men have less than a high school education. He also finds evidence for the selection hypothesis.
     Carpenter (2007) suggests that marriage is a signal of heterosexuality and that the marriage premium may be due to discrimination against gay men. Using the California Health Interview Survey (CHIS) on nearly 10,000 men, he finds indirect evidence to support this hypothesis. He estimates the earnings equation, with the log of earnings the dependent variable, 7 times—once for each city, including an interaction term for that city with married. He estimates it an 8th time with interaction terms for all cities entered simultaneously. The only city that shows a significant result is Riverside, the city in his sample with the lowest percentage of gay men. He also investigates the impact of homosexuality on the marriage premium three additional ways by determining whether the marriage premium is: 1) increasing with the percentage of gay men in an occupation; 2) increasing in age; and 3) affected by local sexual orientation anti-discrimination laws.  He finds some evidence for the 1st, strong evidence for the 2nd, and no evidence for the 3rd. Frank (2007) uses a sample from the British university system and also finds some evidence to support Carpenter’s findings, especially for the finding regarding age.
Theory
     That discrimination against homosexual men results in a marriage premium can be demonstrated by applying Becker’s (1971) theory of employer discrimination. Figure 1 shows that assuming the taste for discrimination is constant, the greater the supply of gay men (NG), holding the supply of straight men constant (NS), the greater the wage differential between the two groups. Becker (1971, p.44) states that, “As long as tastes are not homogeneous, a larger relative supply of Nnonwhite must be associated with a larger Median Discrimination Coefficient.” In the empirical section of the book he adds, “If numbers of the same factor have different tastes for discrimination an increase in the relative supply of nonwhites increases the equilibrium market discrimination against them, even if all tastes remain fixed.” Of course he was writing about discrimination against nonwhites, but the theory can easily be applied to gay men.
     What makes discrimination against gay men different from discrimination against people of color or women is that it is not easy to distinguish gay from straight single men by appearance. Therefore straight single men are also discriminated against. However, so long as the employer can make an estimate of the proportion of single men who are gay, it will be the proportion of gay men that determines the wage differential, not the proportion of single men.^{4 5}  The probability that single men are gay becomes greater the older the men are. The marriage premium for older men should be greater than for younger men. For this reason, regression equations are estimated separately for men under 40 years old and men 40 years old and over.
The Data and Empirical Model
     Both California and Texas had the largest populations of same sex couples in 2000; California had 92,138 same sex households and Texas had 49,912.⁶ Los Angeles, San Francisco and Dallas were among the cities with the highest share of SSUPs in the US, ranked 2nd, 3rd and 8th, respectively. In addition, San Francisco and Austin were the 1st and 7th ranked cities in the concentration of SSUPs (NGLT, 2002). Yet, Texas and California had very different public policies regarding anti-gay employment discrimination in 2000. At that time California was one of only 11 states and the District of Columbia that banned anti-gay discrimination in both private and public sector employment. There were an additional 10 states that banned anti-gay discrimination in the public sector only. Texas had no law banning discrimination in either sector. In addition, 13 California cities and 3 counties had their own anti-gay discrimination laws, all but one of which predated the statewide law (Human Rights Campaign, 2001). In Texas, only Austin and Fort Worth had such laws.⁷ 
     If there is evidence in both of these very different states that sexual orientation variables affect the marriage premium, it is likely that the effect is countrywide.  Carpenter (2007) found no evidence that local antidiscrimination laws are effective.  Similar results regarding the sexual orientation variables for these two states will be an indication that statewide antidiscrimination laws are ineffective.
     The earnings equation that is estimated for white men in each state is:
Ln(Yi) = α + βXi + γZi + δMARRIEDi + η(MARRIEDi x Zi) +  ε
Where Y is wage and salary earnings, X contains the human capital variables and 8 occupational categorical variables for the individual.⁸  Z contains same sex variables for both the individual and the MSA where he resides. MARRIED is a qualitative variable that takes the value 1 when the individual is married or separated and 0 otherwise. MARRIED x Z is an interaction term where the marriage dummy is multiplied by the percentage of single men in SSUPs in the MSA. A positive coefficient on the interaction term indicates that the marriage premium is higher when the percentage of men in SSUPs is higher. 
     This paper uses the 2000 Census 5% Public Use Micro Sample (PUMS) for white men, which allows identification of men in SSUPs, but not the sexual orientation of those not cohabiting. Another shortcoming of this dataset is that not all Texas MSAs are included. Because of small population sizes and the need to maintain confidentiality, their data is not available using the 5% sample.⁹
     Beginning in 1990, the Census began including the category “Unmarried Partner” when asking about the relationship to the householder. The number of same sex couples choosing this category was very low in 1990. In anticipation of the 2000 census, the National Gay and Lesbian Taskforce (NGLT) began a large public awareness campaign to encourage gays and lesbians living together as same sex partners to check that box so that they would be counted. As a result, the number of same sex unmarried partnerships recorded by the census increased four fold (NGLT (2002)) to about 600,000. Both the NGLT (2002) and Badgett and Rogers (2003) warn that those who choose the unmarried partner category are not representative of gays and lesbians as a group and that results based on unmarried partners should not be generalized. In this study, the percentage of single men that are in same sex unmarried partnerships (%SSUP) is used as a proxy for the percent of gay men in the single male population. It is not required that those in SSUPs are representative of all gay men. The underlying assumption that is crucial is that the ranking of metro areas by SSUPs is the same as the ranking by percentage of gay men. Badgett and Rogers (2003) find evidence of a significant undercount of SSUPs in two commissioned surveys, but there is no reason to believe that the undercounts vary dramatically from city to city.¹⁰  Columns three and four of Table 1a show sample sizes and percentages of men in SSUPs in each California MSA used in this study. The percentages range from .72% in Yuba City to 4.56% in San Francisco.^{11 12} There is no way to prove that the rankings by percentages of SSUPs and gay men are the same. It is reassuring, though, and an important finding, that the correlation coefficient for the Census’ percent in SSUPs and Carpenter’s percentage of gay men is .984 for the seven cities in both of our samples. Unfortunately there is no similar dataset for comparison for Texas.
     The percentages are smaller for the Texas MSAs, but the ratio of the city with the lowest proportion, Bryan-College Station, to the city with the highest, Dallas, is about the same as Yuba City is to San Francisco. There is somewhat less variation among the Texas MSAs with a coefficient of variation of 25.73%, while California’s is 39.69%.
     For the purpose of better estimating the wage equation for males attached to the labor force, the sample is trimmed. The final sample only includes men who are between 18 and 64 years old, usually work at least 35 hours a week, are currently employed, have no work disability, are not self-employed, are not currently serving in the military, can speak English well, and have no missing data. There are 124850 white men in the California sample and 71121 in the Texas sample.
     Table 2 provides the definitions, means and standard deviations, in parentheses, of the variables used in the earnings equation by state and age group. In both states older men are much more likely to be married, earn higher salaries, and are nearly 3 times more likely to be veterans. Older workers are also more educated, with a higher percentage in both states graduating from college and attending graduate school. This is not surprising, since the younger group includes men in their late teens and early twenties, who may not have had an opportunity for higher education. Both age groups live in MSAs with the same proportion of single men in SSUPs and both have the same probability of being in SSUPs themselves¹³.
     Comparing the full samples for both states we see that Texans are more likely to be married and be veterans than Californians. On the other hand, Californians have higher earnings, live in more densely populated MSAs and are 23% more likely to live in an SSUP. There is not a large difference in the average age of the men in the two states or in the percentage with a college degree or more.
     Tables 3a and 3b show means and standard deviations, in parentheses, by marital status and state for both age groups.  % Never Married is not included in the regressions, but it has been included in this table to confirm that marriage is a stronger signal of sexual orientation in older men. The table confirms that indeed there is only a very small difference in the % Never Married between men in SSUPs and other single men in both states among the younger group. Among the older group, the difference is quite large. Approximately 30 percentage points separate the men in SSUPs from other singles in California, while there is a 25 percentage point difference in Texas. Note also, that 28% and 45% of the men living in SSUPs in 2000 were previously in heterosexual marriages. 
     Not surprisingly, men in SSUPs are more likely to live in cities with a higher percentage of SSUPs than the other groups. They also live in the most densely populated MSAs, have the highest level of education and an average income nearly as high as married men, which is consistent with other studies.¹⁴
     Married men in both states and for both age groups earn the highest salaries, although just barely for young men in California, and are most likely to be veterans.
Regression Results
     OLS is used to generate the results in Tables 4a and 4b. Two equations are estimated 6 times - once for the full sample, once for men under 40 years old and once for men 40 and over in both California and Texas. The dependent variable is the natural log of wage and salary income, ln(EARN). The first is a typical wage equation that has been widely used in the marriage premium literature. The results are as expected. The marriage premium, measured by the coefficient on MARRIED, ranges between 20 and 25%. In both California and Texas younger men have a smaller marriage premium than older men. This is consistent with both the productivity hypothesis and the discrimination hypothesis. Married men in the older group will be married more years on average than men in the younger group.¹⁵ If the marriage premium is higher in the older group for this reason, this is evidence of marriage enhancing productivity. But Table 3a and 3b show never married men are significantly more likely to be in SSUPs when they are over 40. If the premium is higher in the older group, because marriage is a more valued signal of heterosexuality for older men, this is evidence of discrimination.¹⁶
     The omitted education category is less than a high school graduate. The coefficients on the dummies get progressively larger, which is expected. The population density variable is included to capture higher land values, which are correlated with higher wages.     
     The second equation for each group includes SSUP variables. They indicate whether an individual is in a same sex unmarried partnership (SSUP), the percentage of the single men in the metro area who are in SSUPs (%SSUP), and the interaction term %SSUP x MARRIED. In Texas, there is no significant difference in earnings between men who are in SSUPs and men who are not. In California, men in SSUPs earn 7.7% more than men than who are not in SSUPs. This result appears to indicate that there is no discrimination against gay men. There are two possible alternative explanations, however. The same selection and/or specialization effects that explain part of the marriage premium may also increase earnings of SSUPs, but to a lesser extent. Also, Badgett and Rogers (2003) find that those GLBT individuals that correctly checked the “unmarried partner” box are more likely to have higher incomes than those in similar living situations who checked “roommates”. This would also account for the positive sign on SSUP.
     The %SSUP in the MSA is included because gay men are not randomly allocated across MSAs, but tend to live in large, high price, high income cities. The coefficient is positive and significant in each regression equation.  A one percentage point increase in %SSUP is associated with an increase in incomes as little as an additional 2.5% (older men in Texas) and as much as 12% (young men in Texas).
     The most interesting results are the change in the coefficient on MARRIED, measuring the marriage premium, and the coefficient on %SSUP x MARRIED. First, the coefficient on MARRIED for each group is smaller, whether or not the interaction term is significant, for each regression where SSUP variables are included. Using the full sample, the coefficient in California falls from .235 to .216, a decline of 8.1% and in Texas it falls 10.5%. But the greatest effect is for older men in the two states.  In California, MARRIED’s coefficient falls by 23.8% and in Texas by 19.3%. The size of the coefficient on the interaction term is significant in both states only for older men. A one percentage point increase in single men in SSUPs increases the premium by about 3 percentage points. This result is consistent with a portion of the marriage premium being due to statistical discrimination against older single men. Comparing the MSAs with the smallest and largest percentages of single men in SSUPs, these coefficients translate to marriage premiums of 21% in Yuba City vs. 36.8% in San Francisco and 22.4% in Bryan-College Station vs. 28.5% in Dallas. In San Francisco, 42% of the marriage premium is due to the interaction term, indicating discrimination, while 11.4% is due to the interaction term in Yuba City. In Texas, the percentages are 4.9% for Bryan-College Station and 25% for Dallas.
Conclusion
     California and Texas had very different public policies regarding discrimination against gays in employment in both the public and private sectors in 2000, when the Census data was collected. The differences have continued to the present day. Yet, regressions for both states show very similar results regarding discrimination as a possible factor in explaining the earnings premium enjoyed by married men. Including SSUP variables for individuals and the MSAs in which they live, reduces the coefficient on the marriage dummy variable. In addition, for men 40 years old and over each one point increase in the percentage of single men living in same sex unmarried partnerships adds over 3 percentage points to the marriage premium. For men younger than 40, there is no such effect. This is consistent with the discrimination hypothesis, since the proportion of men in and out of same sex partnerships have approximately the same percentage of never married men under 40, but the percentage of never married men is much higher among men in SSUPs in the 40 and over group. These results may also indicate that laws banning anti-gay discrimination may be circumvented by firms relying on the signal of marriage when determining salaries and promotions.
     Another finding in this paper is the high correlation coefficient between the percent of men in same sex unmarried partnerships in the Census micro sample in California MSAs with the percentage of self identified gay men in the California Health History Survey. This suggests that the percentage of men in SSUPs can be used as a proxy for the percentage of gay men in MSAs.          
          Further study could expand the sample to include MSAs throughout the US. This would allow investigation into additional questions: Does the marriage premium vary across regions and is that correlated with the distribution of men in SSUPs? Does the result found here, that a statewide law banning anti-gay discrimination in the workplace has no effect on the impact of the percentage of men in SSUPS on the marriage premium, continue to hold?  In addition, searching other statewide datasets could allow for more evidence on the relationship between %SSUP and the %gay in MSAs and/or counties.

Notes

1. Reported in Carpenter (2007).
2. The other half they attribute to discrimination in favor of married men or productivity differences.
3. Gray (1977) noted that the decline in the marriage premium occurred at the same time as the increase in married women’s labor force participation. He concluded that the reduction of specialization within the family was the reason for the decline in the premium. However, Cohen (2002) found that once one removes cohabitating men from single men, one finds that the marriage premium actually has not decreased very much at all.
4. The correlation coefficient between the proportion in Same Sex Unmarried Partnerships and proportion single for the California  MSAs in the sample is .486, using data from the 2000 Census Summary 3 File.
5. Single men are divorced, widowed or never married. Separated men are considered married.
6. 49614 were male households in California and 21,740 male in Texas. New York was 3rd with 46,490 SSUP households.
7. There were other differences in 2000 as well. Texas was one of only 13 states that still had a sodomy law. In California 16 cities and 2 counties provided for domestic partnership registration, and California became the second state to create a statewide registry in 1999. Only Travis County Texas provided for a domestic partnership registry, however. In 1999 the state of California became the 4th to offer health insurance benefits to domestic partners. 27 local governments also provided these benefits. No governments in Texas provided them (HRC, 2001).
8. The categories are: Agriculture, Office, Service, Transportation, Production, Construction, Professional and Managerial/Executive. Sales is the omitted category. In the interest of space the results on the occupational dummies is not reported in the regression tables.
9. In many MSAs, the variable indicating whether an individual lives in the central city is not reported, again for confidentiality reasons. This is why no variable indicating central city residence is included.
10. They do find differences between regions, but California and Texas are in the same region.
11. The numbers of unmarried partnerships that are male-male, female-female and male-female by MSA are available at the census website using Summary File 3. The number of men in each metro area is available by marital status and age at the same site. The percentage of single men over 15 years old who are in SSUPs is taken from this source, not from the 5% PUMS. The SF3 figures use the 1 in 6 sample, which consists of all those who completed the long form. The correlation between the SF3 figures and those over 15 in the 5% PUMS is .95.
12. The percentage of male-male unmarried partnerships out of all unmarried partnerships was also collected. The correlation coefficient between this variable and the percentage of single men in SSUPs is also .95.
13. The SSUP variable is calculated as follows: for each state, all men in the sample who are over 15 years old living in MSA households (as opposed to group quarters) are identified. Those who are the head of households and those living in that same household who identify their relationship with the head of household as an unmarried partner are both assigned 1 for the variable SSUP. All other men are assigned 0.
14. Klawitter and Flatt (1998); Carpenter (2005, 2007).
15. There is no marriage duration variable reported in the Census data.
16. † Like the causes of the premium, these forces are not mutually exclusive. The results on %SSUP x MARRIED below provide additional evidence for the discrimination hypothesis.
    

References
Akerlof, George (1998). “Men Without Children,” The Economic Journal, 108 (March), 287-309.
Antonovics, Kate and Robert Town (2004). “Are All the Good Men Married?  Uncovering the Sources of the Marital Wage
     Premium,” AEA Papers and Proceedings. 94(2), 317-321.
Badgett, M.V. Lee, and M. Rogers (2003). Left Out of the Count: Missing Same-Sex Couples in Census 2000. Amherst, MA:
     Institute for Gay and Lesbian Strategic Studies.
Bartlett, Robin and Charles Callahan, III (1984) “Wage Determination and Marital Status: Another Look,” Industrial Relations.
     23(1), 90-96.
Becker, Gary S. (1981). A Treatise on the Family. Cambridge, MA: Harvard University Press.
Becker, Gary S. (1971). The Economics of Discrimination, 2nd ed. University of Chicago Press.
Carpenter, Christopher (2007). “Do Straight Men “Come Out” at Work Too?  The Heterosexual Male Marriage Premium and
     Discrimination Against Gay Men,”  Lee Badgett and Jeff Frank (Eds.), Sexual Orientation Discrimination. London:
     Routledge, 75-92.
Carpenter, Christopher (2005). “Self-Reported Sexual Orientation and Earnings:  Evidence from California”, Industrial and
     Labor Relations Review. 58(2), 258-273.
Carpenter, Christopher (2004). “New Evidence on Gay and Lesbian Household Incomes”, Contemporary Economic Policy
     . 22(1), 78-94.
Chun, Hyunbae, and Injae Lee (2001). “Why Do Married Men Earn More: Productivity or Marriage Selection?” Economic
     Inquiry, 39(2), 307-319.
Cohen, Philip N. (2002). “Cohabitation and the Declining Marriage Premium for Men,” Work and Occupations. 29(3),
     346-363.
Corwell, Christopher, and Peter Rupert (1995). “Marriage and Earnings,” Federal Reserve Bank of Cleveland Economic
     Review 31(4), 10-20.
Dougherty, Christopher (2006). “The Marriage Earnings Premium as a Distributed Fixed Effect,” The Journal of Human Resources.
     51(2), 433-443.
Frank, Jeff (2007). “Is the Male Marriage Premium Evidence of Discrimination Against Gay Men?” Lee Badgett and Jeff Frank
     (Eds) Sexual Orientation Discrimination. London: Routledge, 93-103.
Ginther, Donna K. and Madeline Zavodny (2001). “Is the Male Marriage Premium Due to Selection?  The Effect of Shotgun
     Weddings on the Return to Marriage,” Journal of Population Economics, 14, 313-328.
Gray, Jeffrey S. (1997). “The Fall in Men’s Return to Marriage: Declining Productivity Effects or Changing Selection?” Journal
     of Human Resources. 32(2), 481-504.
Hersch, Joni, and Leslie S. Stratton (2000). “Household Specialization and the Male Marriage Wage Premium,” Industrial and
     Labor Relations Review. 54(1), 78-94.
Human Rights Campaign (2001), State of the Workplace 2001. Washington, DC.
Hill, Martha (1979). “The Wage Effects of Marital Status and Children,” The Journal of Human Resources. 14(4), 579-93.
Kenny, Lawrence (1983). “The Accumulation of Human Capital During Marriage by Males,” Economic Inquiry. 21(2), 223-31.
Klawitter, Marieka M. and Victor Flatt (1998). “The Effects of State and Local Antidiscrimination Policies on Earnings for Gays
     and Lesbians,” Journal of Policy Analysis and Management. 17(4), 658-86.
Korenman, Sanders, and David Neumark (1991). “Does Marriage Really Make Men More Productive?” Journal of Human
     Resources. 26(2), 282-307.
Krashinsky, Harry A. (2004).  “Do Marital Status and Computer Usage Really Change the Age Structure?” Journal of Human
     Resources. 39(3), 774-91.
Loh, Eng Seng (1996). “Productivity Differences and the Marriage Wage Premium for White Males,” Journal of Human
     Resources. 31(3), 566-88.
Mehay, Stephen L. and William R. Bowman (2005). “Marital Status and Productivity: Evidence from Personnel Data,” Southern
     Economic Journal. 72(1), 63-77.
Nakosteen, Robert A., Michael A. Zimmer (1997). “Men, Money, and Marriage: Are High Earners More Prone than Low
     Earners to Marry?” Social Science Quarterly 78(1), 66-82.
National Gay and Lesbian Taskforce (2002), 2000 Census and Same Sex Households: A Users Guide. Washington, DC.
Zimmer, Michael (2006). “Employer Discrimination and the Earnings Premium of Married Men:  Evidence from Quantile
     Regression,” The Journal of Economics 32(1), 1-24.