The 2001, 2003, and 2005 waves of the CHIS were telephone surveys of over 40,000 households in California each year. The CHIS uses a multistage sampling design in which a random adult is selected within each household using random-digit dialing (RDD) methods. When weighted, the sample is representative of the noninstitutionalized population of California. We use confidential versions of these data that contain information on the respondent’s self-reported sexual orientation. Respondents also provide individual information on a variety of health conditions, health behaviors, and demographic characteristics. At the end of the “demographics” section (where age, race, and education information is elicited), adult respondents in the 2001 wave were asked the following: “The next question is about your sexual orientation, and I want to assure you that your answers are completely confidential. Are you gay [lesbian] or bisexual?”4
In 2003 and 2005, individuals were asked, “Do you think of yourself as straight or heterosexual, as gay[lesbian] or homosexual, or bisexual?”5
We use responses to these questions to identify gay men and lesbians.6
The CHIS also includes information on each individual’s partnership status. Specifically, respondents are asked to state their marital status, and one of the choices is “living with a partner.” We identify partnered gay men and lesbians as individuals who reported being gay or lesbian and who concurrently reported living with a partner.
Our approach for identifying partnership among the sample of gay men and lesbians in CHIS has a few drawbacks. Most importantly, we identify partnership on the basis of a question about marital status, and respondents are forced to choose among several categories that need not be mutually exclusive. Another potential problem with our measure is that we do not actually observe the overall sex composition of the household. While it is reasonable to assume that a gay man who reports he is “living with a partner” is, in fact, living with a man (and similarly for lesbians), we cannot verify this to be true. This source of error is likely trivial. In the Tobacco Survey, we find only a single observation of a self-identified lesbian or gay man who reports living with a partner of a different sex.
We note that the measure implies cohabitation
with one’s partner and therefore excludes other types of “dating” relationships in which the individuals do not live together.7
For the CHIS and all subsequent samples, we consider adults aged 18–59 (inclusive): this yields 1,306 self-identified gay men and 809 lesbians pooled across the three CHIS waves.
We complement the CHIS with the 2003 California LGBT Tobacco Survey. The Tobacco Survey is a sample of self-identified sexual minorities and individuals reporting same-sex sexual behavior. When weighted, this sample is designed to be representative of California’s lesbian and gay population. The study was commissioned by the California Department of Health and performed by the Field Research Corporation. The telephone-based Tobacco Survey used a disproportionate stratified RDD design and a weighting scheme that explicitly made use of “high-density” gay and lesbian zip codes; importantly, these high-density zip codes were determined by using information on the geographic distribution of same-sex unmarried partners from the 2000 decennial census. This component of the sampling strategy must be kept in mind in the context of our partnership estimates because it is possible that the geographic distribution of sexual minorities varies according to partnership status, and the census identifies only partnered gay men and lesbians. If this is the case, then these data may overstate the proportion of lesbians and gay men who are partnered because the sampling strategy might be biased toward geographic areas where there are more same-sex couples.
All households were first screened using a question that asked whether the respondent was gay, lesbian, or bisexual or had a same-sex sexual experience since age 14. Our analysis sample includes adults aged 18–59 who self-identified as either “gay” or “lesbian,” which includes 770 and 266 individuals, respectively.8
A key advantage of the Tobacco Survey relative to the other data sources is the high level of detail individuals were asked to give about current and previous partnership situations. Specifically, individuals were asked whether they had ever been legally married, as well as their current marital status. Individuals were then asked whether they had a current “primary” partner, which was explained as “someone you love more than anyone else and feel a unique commitment to.”
It is notable that the Tobacco Survey asks about partnership separately from marital status. That is, CHIS identifies individuals who report “living with a partner” as one of the response options to a question about current marital status. Unfortunately, these response options need not be mutually exclusive (e.g., “living with a partner” and “divorced”). The Tobacco Survey, in contrast, asks about marital status and partnership separately. First, the survey asks whether the respondent has ever been legally married. Of those individuals who report having ever been legally married, the survey then asks the respondent’s marital status, intended to elicit current legal marital status. All individuals who did not report that they were currently married were then asked the question about a “primary” partner. In the CHIS, a respondent who is both divorced and currently in a same-sex partnership might not be counted as partnered if he or she chose the “divorced” option on the marital status question. The Tobacco Survey, in contrast, allows respondents to indicate that they are both divorced and in a cohabiting partnership.
Individuals with a current primary partner were then asked whether the partner is same-sex or opposite sex, as well as whether the individual is living with that primary partner. We use combinations of these responses to define “partnered” individuals in the Tobacco Survey as respondents who report living with a same-sex primary partner. Under this definition, individuals who reported a current primary partner but did not report cohabiting with that partner are not considered “partnered” per se. We provide evidence (and prevalence estimates) on these relationships that are likely a mix of those who are seriously dating and those who are “living apart together.” But we impose the general cohabitation requirement to create measures of partnership that are most consistent across data sources. Unfortunately, we do not observe the “has a partner but is not cohabiting with that partner” group in the CHIS. The census data, which we describe later, also require the presence of a same-sex unmarried partner living in the household for identification of gay and lesbian couples.
Individuals in the Tobacco Survey who reported living with a primary partner were also asked about the length of their cohabitation, as well as the length of the overall relationship, and all respondents were asked a battery of standard demographic questions such as race, age, income, education, and the presence of children in the household. Finally, respondents who reported living with a same-sex partner and who also reported not being currently married were asked whether their partnership was registered with the local or state government. We use responses to this question to provide estimates of the prevalence of official domestic partner registrations in California, and we examine the relationship between observable demographic characteristics and domestic partner registration.
Finally, we complement our analyses of California statewide individual-level data with the more well-known Census 2000 data. The Census 2000 5% and 1% Public Use Microdata Samples (PUMS) are drawn from the approximately 20% of households in the United States who received a census “long-form” that asks detailed demographic and economic questions. The PUMS are designed as a 1 in 20 and 1 in 100 sampling of the total U.S. population. We combine the two samples because they are independent draws from the long-form responses. The census does not ask any direct questions about sexual orientation or sexual behavior.9
Rather, census forms include relationship categories that define how individuals in a household are related to the householder. These fall into two broad categories: related persons (e.g., husband/wife, son/daughter), and unrelated persons (e.g., roomer/boarder, unmarried partner). If the householder designates another adult of the same sex as his or her “husband/wife” or “unmarried partner,” Census 2000 enumerates this household as a same-sex unmarried partner couple.10
To accord with our other statewide individual-level data, we present Census 2000 results for California same-sex couples, thus providing an important check on data quality.
Gates and Ost (2004)
and Black et al. (2006)
suggested a possible serious measurement error problem with census same-sex couple data. Census Bureau coding procedures recode any same-sex “husband” or “wife” from the household roster as an “unmarried partner.” As a result of this procedure, any different-sex married couples that inadvertently miscode the sex of one of the spouses will be coded as same-sex “unmarried partner” couples. Given the 90-to-1 ratio between married and unmarried partners in the census, even rare sex miscodes could significantly contaminate the same-sex couple sample with different-sex married couples. We use the method advanced in Black et al. (2006)
and restrict attention to same-sex couples for whom marital status was not allocated for either member of the couple. Census Bureau coding procedures did not permit an “unmarried partner” to have a marital status of “currently married” and allocated any such response. A same-sex “unmarried partner” could be listed as “currently married” for two primary reasons: (1) he or she is part of a same-sex couple in which the partners consider themselves to be married, or (2) he or she is part of a different-sex married couple in which the sex of one of the spouses was miscoded (as described above). By restricting the sample to couples without any marital status allocations, we eliminate the group that is likely to be most prone to this error. Unfortunately, we potentially bias some of the demographic characteristics if same-sex couples who consider themselves to be married differ from those who consider themselves to be “unmarried partners.”