Birth Cohort and the Specialization Gap Between Same-Sex and Different-Sex Couples

Abstract

We examine differences in household specialization between same-sex and different-sex couples within and across three birth cohorts: Baby Boomers, Generation X, and Generation Y. Using three measures of household specialization, we find that same-sex couples are less likely than their different-sex counterparts to exhibit a high degree of specialization. However, the “specialization gap” between same-sex and different-sex couples narrows across birth cohorts. These findings are indicative of a cohort effect. Our results are largely robust to the inclusion of a control for the presence of children and for subsets of couples with and without children. We provide three potential explanations for why the specialization gap narrows across cohorts. First, different-sex couples from more recent birth cohorts may have become more like same-sex couples in terms of household specialization. Second, social and legal changes may have prompted a greater degree of specialization within same-sex couples relative to different-sex couples. Last, the advent of reproductive technologies, which made having children easier for same-sex couples from more recent birth cohorts, could result in more specialization in such couples relative to different-sex couples.

Appendix: Coding Issues for Identifying Same-Sex Couples

In 1990, the U.S. Census Bureau began to include an answer to the “relationship to householder” variable that allowed individuals to indicate that they live in a household with an unmarried partner. This was the first year in which researchers could combine this variable with the “sex” variable to identify same-sex households. The inclusion of this question introduced complications that forced the census bureau to make choices in terms of recoding. In the 1990 census, for example, when a householder named a person of the same sex as his or her spouse (not “unmarried partner”), the census bureau considered this a logical contradiction. At the time, two people of the same sex could not be considered legally married to each other in any state. To deal with this, the census bureau changed the sex of one person in the relationship. Ultimately, the recoding led to the formation of different-sex married couples out of couples who had originally (intentionally or because of error) named themselves same-sex married couples. As a result, the only same-sex couples who are observable in the 1990 data are those who had originally coded their relationship to each other as “unmarried partners.” In the 2000 census and each of the ACS surveys from 2000 onward, when a householder considered a person of the same sex his or her spouse, the census bureau did not change the sex of one member of this couple but instead changed the “relationship to householder” of the second person from spouse to “unmarried partner.”

The first important implication of this change has to do with the chances of the same-sex-couple sample being contaminated by different-sex couples, who were falsely identified as same-sex couples. In 1990, only unmarried different-sex couples who had miscoded the sex of one of their members would end up in the same-sex couple group, as there could be no married same-sex couples. However, in the 2000 and subsequent years, any different-sex couple who miscoded their sex, married or unmarried, could end up in the same-sex couple group. Because the contamination problem is, therefore, likely very small in 1990, we follow Black et al. (2000) and leave the sample as it is. However, the 2000–2011 data is more likely to suffer from contamination, and we correct for that problem using a second difference in the way the data was processed between 1990 and 2000: the use of the “marital status” question.

In 1990, when people who reported their relationship to each other as unmarried partners, they were permitted to say that their marital status was married (or “other than currently married,” or anything else); the census bureau generally left the marital status answer as recorded by the respondent. Beginning in 2000, the census bureau did not allow persons in an unmarried same-sex partnership to consider themselves to have “married” as a marital status, which followed the 1996 Defense of Marriage Act (DOMA). Same-sex couples who called each other “spouses” had their marital status allocated to something other than “married,” and we assigned an allocation flag on the marital status variable. Thus, same-sex couples (who might have been different-sex couples with a miscoded sex) who considered themselves “married spouses” became “unmarried partners,” with an allocated marital status. To eliminate this potential contamination of different-sex couples in the same-sex couple sample, we drop any couple in the 2000–2011 ACS data for which either or both people in the couple has an allocated marital status (Black et al. 2000).

One important difference exists between the decennial censuses and the annual ACS. As of 2005: approximately one-third of ACS respondents completed their survey with a Computer Assisted Telephone or Personal Interview (CATI/CAPI), while the remaining share of participants mailed in their responses. In the computer-assisted interviews, respondents were asked to verify the sex of a same-sex spouse, which can reduce the possibility of a sex miscode. According to Gates and Steinberger (2009:11, figure 1), the exclusion of couples in which one or both partners have a marital status allocation and who had submitted their responses by mail results in a sample of same-sex couples that is largely free of measurement error. Our main results use the sample restriction suggested by Black et al. (2000). However, we check the robustness of our estimates to this sample restriction by imposing the restrictions recommended by Gates and Steinberger (2009). Rather than present these estimates, it is sufficient to note that our results are not sensitive to which sample restriction is imposed.