Introduction

The social context surrounding relationship recognition for sexual minorities has changed rapidly over the past two decades. Prior to 1997, no state in the United States legally recognized same-sex unions (Human Rights Campaign 2018). In 2000, Vermont became the first state to recognize same-sex civil unions and in 2003 Massachusetts became the first state to legalize same-sex marriage (Human Rights Campaign 2018). Ten years after legalization of same-sex marriage in Massachusetts, roughly 40% of the United States population lived in a state that recognized same-sex marriage (Human Rights Campaign 2013). Two years later, on June 26, 2015, the United States Supreme Court ruled on Obergefell v. Hodges, effectively allowing same-sex marriage in all states. In spite of these dramatic changes in the legal recognition of same-sex unions, we know little about their implications for the formation and dissolution of these relationships.

The rapid social change regarding relationship recognition likely signals a reduction in the discrimination of sexual minorities that could shape multiple outcomes for this population, especially their physical and mental health (King and Bartlett 2006). Consistent with the minority stress perspective, and suggestive of this, Hatzenbuehler et al. (2009) found that LGB (lesbian, gay, bisexual) adults living in states with hate crime or employment discrimination protection had lower rates of psychiatric disorders than LGB adults living in states with no protections. Similarly, using longitudinal data, Hatzenbuehler et al. (2010) found that increases in psychiatric morbidity between interviews were greater for sexual minorities who resided in states that passed bans on same-sex marriage than for their counterparts whose states did not pass bans. Presumably, the legal recognition of same-sex relationships could also have implications for the formation and dynamics of same-sex coresidential unions (i.e., cohabiting unions and marriages) among sexual minorities. The minority stress perspective also suggests that the social context, as well as the legal context, is important for same-sex union outcomes among sexual minorities.

Our understanding of same-sex union transitions is largely based on research examining relationship dissolution. For example, about twenty studies had examined the stability of same-sex couples as of 2017 (Joyner et al. 2017). To our knowledge, researchers have not examined how sexual orientation identity is associated with the timing of same-sex coresidential unions and no previous research has examined how indicators of social context are associated with same-sex union formation in the United States. Given the importance of coresidential unions for supporting the health and well-being of sexual minorities (Umberson and Kroeger 2016), it is critical to assess their entry into these unions. Data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) provide a unique opportunity to address this issue. Focusing on sexual minorities (bisexuals and homosexuals), we examine how social context indicators at the family, tract, and county levels are associated with the rate of forming a first same-sex union. For comparative purposes, we also examine how social context indicators are associated with the rate of forming a first different-sex union among sexual majorities (heterosexuals). Our analyses potentially offer clues about the implications of changes in the legal recognition of same-sex couples for patterns of union formation among sexual minorities.

Background

Union Formation in Young Adulthood

Patterns of union formation have shifted dramatically over the past few decades. The median age at first marriage has reached an all-time high of 29.8 for men and 27.8 for women in 2018 (U.S. Census Bureau 2018). The delay in marriage has been accompanied by an increase in levels of cohabitation (Bumpass and Lu 2000; Manning et al. 2014; Lamidi and Manning 2016). While the age at first marriage has increased, the age at first coresidential different-sex union has remained the same at around 23 for men and 21 for women (Manning et al. 2014).

The factors associated with the timing and likelihood of marriage are well known, but those associated with different-sex coresidential union formation in general are less frequently examined. This reflects the fact that studies typically treat cohabitation and marriage as competing risks and acknowledge the factors predicting marriage are not the same as those predicting cohabitation. A recently published exception is Kuhl et al. (2012) who found using data from Add Health that race/ethnicity, family SES, and living with two biological parents in adolescence were all associated with rates of forming a first coresidential union among respondents ages 25–32. Other work based on samples with a wider age range of respondents found similar patterns of association. Specifically, African Americans were less likely to cohabit or marry than whites, respondents who had parents with higher educational attainment were less likely to form a coresidential union, and individuals who lived with two biological married parents in childhood were less likely than their counterparts who did not to form a union (Guzzo 2006; Teachman 2003).

While previous research has found that sexual majority women enter coresidential unions earlier than men (Manning et al. 2014), it is unclear how sexual minority status might complicate these patterns. Individuals are socialized to search for different-sex partners starting in adolescence (Strohm 2010), but presumably, sexual minorities begin to search for same-sex partners after developing same-sex attraction and identities (Heatherington and Lavner 2008).

Delays in marriage are also thought to indicate difficulties in finding a match (Lewis and Oppenheimer 2000). Given the fact that the sexual minority population is small, sexual minorities likely face greater constraints finding partners (Strohm 2010). Consistent with this notion, Rosenfeld and Thomas (2012) found evidence that the Internet is the main way sexual minorities meet their partners. Additionally, respondents in same-sex relationships were considerably more likely than their counterparts in different-sex relationships to report they met their partner through the Internet. Thus, we expect sexual minorities to form coresidential unions later in life than their sexual majority counterparts.

Gender differences in the median age at first different-sex coresidential unions suggest there may be differences in when sexual minority men and women form coresidential unions. As stated earlier, different-sex union formation is later for men than women. Therefore, we would expect men to form same-sex unions later in young adulthood than women. According to weighted data from Wave IV of Add Health (author’s calculations), the median age at first same-sex union formation is 23.8 for men and 22.1 for women, while the median age for first different-sex union formation is 22.4 for men and 20.9 for women. These estimates suggest that individuals form same-sex unions at least 1 year later than they form different-sex unions and this timing differs by gender.

Same-Sex Union Formation

Limited previous research has used large-scale, quantitative data to examine the correlates of same-sex coresidential union formation (Frisch and Hviid 2006; Mernitz and Pollitt 2018; Rosenfeld and Kim 2005; Strohm 2010). Using British data from the National Child Development Study (NCDS) and the 1920 British Cohort Study (BCS), Strohm (2010) found, based on life table methods, that the rates of entry into same-sex cohabitation increased steadily from age 16 to age 34. In comparison, entry into different-sex cohabitation increased through the early and mid-20s and then leveled off before declining during the early 30s. Men and women were equally likely to enter a same-sex union by age 34 (Strohm 2010). Strohm (2010) also found, based on survival models including a rich set of variables, that individuals with higher levels of education and occupational prestige were more likely than their less advantaged counterparts to enter a same-sex cohabiting union. Furthermore, individuals who were born in a later cohort and from higher SES areas had higher likelihood of entering a same-sex cohabiting union (Strohm 2010). Strohm (2010) points out that these patterns suggest that individuals who grew up in social contexts more favorable to same-sex relationships, or with enough resources to move away from unfavorable social contexts, were more likely to form same-sex unions.

Although we have limited knowledge of the dynamics of same-sex union formation, as Umberson et al. (2015) highlighted, several data sets have been used to study the demographic, health, and economic well-being of individuals in same-sex coresidential unions (see Black et al. 2000; Carpenter and Gates 2008; Badgett and Gates 2006; Institute of Medicine 2011). For example, Carpenter and Gates (2008) found that partnered gay men and lesbians were older, more likely to be white, and were more highly educated than non-partnered gays and lesbians. Rosenfeld (2014) found similar patterns for same-sex cohabiting couples using data from How Couples Meet and Stay Together data. Sexual minorities with greater levels of education appear more likely to be partnered, but it is unclear whether this is due to higher levels of resources or social climates that are more supportive of same-sex couples.

An important issue is the measurement of sexual minority couples. Studies on the outcomes of sexual minorities often rely on the gender composition of coresidential unions to determine sexual minority status. Often these studies label unions with two men “gay couples” and unions with two women “lesbian couples.” Yet, these unions may include bisexual individuals. The current study expands our understanding of same-sex union formation by defining our population of interest by sexual orientation identity, rather than eventual couple composition. Using sexual orientation identity as our marker of sexual minority status also allows us to assess whether bisexuals are as likely as homosexuals to form same-sex unions.

Stigma and Context

While prior work has focused on individual characteristics, the broader social context may have implications for the formation of same-sex coresidential unions among sexual minorities. Bronfenbrenner’s ecological systems theory posits that the context in which social phenomenon occurs can be seen as “a nested arrangement of structures, each contained within the next” (Bronfenbrenner 1977, p. 514). An ecological system is conceptualized as a multilevel system where the structural or outermost level focuses on social forces and institutions such as laws, government, and media (Cook et al. 2014). The interactions across and within these different levels influence outcomes (Eamon 2001). In other words, outcomes for individuals are influenced in unique ways by the different levels of social context, such as family and neighborhoods.

Given this framework, we assume that individuals make decisions about forming same-sex unions based on the current sociopolitical climate in which they reside. As elaborated below, we expect that sexual minorities will be more likely to form same-sex unions if they live in areas with more same-sex couples and fewer Republican voters. Our assumption is that these areas are more supportive of sexual minorities and that this support factors into decisions to move in with same-sex partner. On the basis of this assumption, we expect that supportive climate will be less consequential for decisions to form different-sex unions. We capture sociopolitical support at two different levels: county and neighborhood.

In their consideration of social context, Oswald et al. (2010) defined community context as the amount of community support for homosexuality including the existence of LGBT (Lesbian, Gay, Bisexual, Transgender) community members and services. They found that community climate affects the well-being of LGBT individuals by conveying messages of support or rejection that are then internalized (Oswald et al. 2010). Two commonly used indicators of climate include the patterns of voting for Republican or Democratic candidates and the concentration of same-sex couples.

The overall political climate of an area can be captured through aggregate voting patterns (e.g., Everett 2014; Oswald et al. 2010) which taps the attitudes of a community. Furthermore, as highlighted by Oswald and colleagues, the Republican National Committee explicitly opposes same-sex marriage. The official Republican Party platform states that “Traditional marriage and family, based on marriage between one man and one woman, is the foundation for a free society and has for millennia been entrusted with rearing children and instilling cultural values” (Republican National Committee 2016, p. 11). Given this explicit and clear opposition to same-sex marriage, it is reasonable to expect that sexual minorities living in areas with higher concentrations of Republican voters will be less likely than their counterparts residing in areas with lower concentrations of Republican voters to form same-sex coresidential unions. Like the concentration of same-sex couples, the proportion of Republican voting signifies the level of support in a given context for sexual minorities.

Concentration of same-sex couple households at the tract level (e.g., Wienke and Whaley 2017) operates as a demographic marker of a potentially supportive community climate or possibly as an indicator of access to partners sharing the same sexual identity (Schwartz and Graf 2010). Same-sex coresidential couples are not distributed evenly across regions, states, cities, and counties but concentrated in some areas of the country (Gates 2013). The first study to document the geography of these couples estimated on the basis of 1990 Census data that 59.3% of male same-sex couples were concentrated in twenty cities, while 45.5% of female same-sex couples were concentrated in twenty cities; yet, the twenty cities with the largest concentrations of gay couples included only 25.9% of the U.S. population and those with the largest concentrations of lesbian couples included only 25.4% of the population (Black et al. 2000). Between 1990 and 2010, increases in the percent of same-sex couples were large for counties located in the more socially conservative regions of the country (e.g., the South and Midwest) and modest for counties in less conservative regions (e.g., the West and East Coasts). As of the 2010 Census, all regions of the country included counties with high concentrations of same-sex couples (Gates 2013). Same-sex couples are also unevenly distributed across census tracts (Anacker and Morrow-Jones 2005; Spring 2013). Same-sex male couples, in particular, tend to reside in cities and tracts with higher housing costs (Anacker and Morrow-Jones 2005; Black et al. 2000).

Our understanding of the concentration of same-sex couples in the United States is based almost exclusively on data from the U.S. Census Bureau. Estimates of same-sex couple concentration are missing relationships between same-sex partners who do not describe their relationship as an unmarried partnership or do not live together (Carpenter and Gates 2008). The advantage of the Census Bureau estimates are that they offer an estimate across the entire United States at the local level, but admittedly do not offer an indicator of sexual minority concentration. Living with a same-sex romantic/sexual partner requires some level of willingness to publicly identify as a sexual minority. Thus, the uneven distribution of same-sex couples signals the barriers that sexual minorities face to coming out in places that are more socially conservative (Gates 2013). Despite this potential limitation, the concentration of same-sex couples is one of the most widely used proxies for social climate in studies concerning the effects of structural stigma on the outcomes of sexual minorities.

Coming out to family and friends is a process that is unique to sexual minorities (van Eeden-Moorefield et al. 2011) and found to be stressful (Rotheram-Borus and Fernandez 1995; Scherrer et al. 2015). Data from 2013 Pew Research survey reveal that while a large majority of the adult LGBT population has disclosed their identity to a close friend (86%), a smaller share of this population has told their mother or father (56% vs. 39%). Bisexual individuals are significantly less likely to be out than their gay and lesbian counterparts (Parker 2015). Open-ended responses among those not disclosing their identity to parents cited their anticipated reactions as a common deterrent (Pew Research Center 2013). The event of forming a same-sex union for the first time may prompt young adults to disclose their sexual orientation identity to parents. Young adults may feel more comfortable disclosing their identity to friends and family if they reside in less socially conservative contexts. Rosenfeld and Kim (2005) argue that living away from the parental home (e.g., in a different state) allows sexual minorities to form a same-sex coresidential union without having to disclose their sexual orientation to parents.

Solomon et al. (2004) explicitly addressed the association between being out and coresiding with a same-sex partner using survey data from the CUPPLES project. Importantly, the sampling frame for this project was same-sex couples who formed civil unions in Vermont the year after that state became the first to legally recognize same-sex unions (July 2000). The project not only recruited these couples, but also couples identified from their friendship circle who were coresiding outside of a civil union. Solomon et al. (2004) hypothesized that, given the fact that civil union records are available to the public, same-sex couples in civil unions would be more open about their sexual orientation than their counterparts who were not. They found that women in civil unions, but not men, reported higher levels of outness than their same-sex counterparts who had not formalized their union. They speculated that forming a civil union was not contingent on levels of outness for men in same-sex coresidential unions because they were a more select group of sexual minorities than the women.

This study based on CUPPLES data suggests that coming out as lesbian, gay, or bisexual could be a precursor to forming a same-sex coresidential union. The event of forming a same-sex union for the first time could also be conceptualized as a critical event in the process of coming out. We predict that sexual minorities who are out to their parents will be more likely than their counterparts who are not to form a same-sex coresidential union in young adulthood. Being out to parents does not occur in a social vacuum but is likely dependent on the broader social climate. We expect that residing in a supportive neighborhood or county will increase same-sex union formation by facilitating the process of coming out to friends and family. Importantly, we contribute to the growing body of research concerning the effects of social climate on the outcomes of sexual minorities by considering an unexplored outcome and taking into account multiple levels of context.

Current Study

The current study fills a gap in the literature by examining the association between social context and coresidential union formation, with a particular focus on sexual minorities. Drawing on data from the National Longitudinal Study of Adolescent to Adult Health, we address the following research question: How are key indicators of social context associated with same-sex union formation? To answer this question, we estimate proportional hazards models that predict the timing of forming a same-sex union. We expect that sexual minorities living in more supportive contexts (greater concentration of same-sex couples, fewer Republican voters, and being “out” to parents) will be more likely to form a same-sex union than their counterparts in less supportive contexts. To provide a comparison, we also present the results of models that predict the timing of forming different-sex unions for sexual majorities. These results are discussed in the “Supplementary Analysis” section and shown in Appendix Table 3. Importantly, our measures of sexual orientation identity and social context correspond to a point in time that precedes the period during which the vast majority of respondents were “at risk” of forming their first same-sex union, as elaborated below. This study will provide a portrait of same-sex union formation for a contemporary cohort of young adults who are sexual minorities.

Data and Methods

Data for this research were obtained from the National Longitudinal Study of Adolescent to Adult Health (Add Health). Add Health is a nationally representative, school-based, longitudinal study of a 1994–1995 cohort of 7th–12th graders (Harris et al. 2009). Add Health used audio computer-assisted self-interviewing (ACASI) and partner rosters to identify all of the partners with whom respondents had ever experienced a “romantic or sexual relationship” that eventuated in pregnancy, cohabitation, or marriage, in addition to any other partners with whom they had a romantic or sexual relationships since 2001. Add Health subsequently asked respondents to provide the gender, age, and race/ethnicity of each partner. ACASI not only maximizes privacy, but also allows for more complicated skip patterns (Paik 2015). Furthermore, Add Health contains multiple contextual databases, allowing for comprehensive examination of the influence of context. In-home interviews with the respondent were conducted in 1994–1995, 1996, 2001–2002, and 2007–2008. Overall, Add Health interviewed 20,745 adolescents at Wave I. At Wave III most respondents were between 18 and 26 years old. Wave IV, the most recent wave of Add Health, was conducted in 2007–2008 when most participants were between 24 and 32 years old.

Analytic Sample

Given that sexual orientation identity and two of the contextual measures (i.e., out to parents and same-sex concentration) were not included prior to Wave III, the analytic sample for this research consists of individuals who had not formed a same-sex coresidential union before Wave III. The sample of respondents who completed the first in-home interview (N = 20,745) was restricted in several ways. First, we limited the sample to respondents who participated in Waves III and IV and had information on the survey design variables (N = 12,286). Next, we restricted the sample to respondents who had valid information on our Wave III contextual variables (N = 12,053). Then, we limited the sample to respondents who provided valid responses to the Wave III sexual orientation identity question, as elaborated below (N = 11,919). Finally, we excluded respondents who had already formed a same-sex union prior to Wave III (N = 111), as our measures of context correspond to the period when the Wave III data were collected. This restriction means we are estimating first union formation among a sample who have had not yet coresided with a partner. Prior to Wave I, fewer than 10 respondents formed a same-sex union, and between Waves I and III, 34 men and 77 women formed a same-sex union. Respondents who formed a same-sex union prior to Wave III did not differ significantly from those who did not form a same-sex union prior to this wave for any of the variables of interest (results not shown). Individuals who did not form a same-sex coresidential union between Waves III and IV were censored at the time of the Wave IV interview.

As for non-response on control variables, respondents who were missing on race fall into the other category and those missing household roster information at Wave III are classified as not living with both biological parents. Only 11,181 of the 11,808 respondents provided information at the Wave 1 in-home interview necessary to construct our measure of family SES. To address non-response for this measure, we followed the imputation procedures of Bearman and Moody (2004) (e.g., either reliance on data from the in-school questionnaire or use of multiple imputation). The final sample includes 11,808 respondents, 304 of whom identified as bisexual or homosexual.

Studies on the formation of different-sex unions, in particular, likely include individuals who are not at risk of forming such a union. As the proportion of sexual minorities in the population is small, the inclusion of these individuals is likely inconsequential for model estimates. However, when examining same-sex union formation it is important to limit the sample to the population at risk (i.e., sexual minorities), as inclusion of respondents who do not identify as a sexual minority would attenuate the effects. We restrict our analyses of same-sex coresidential union formation to individuals who identified as bisexual, mostly homosexual, or 100% homosexual. Strohm (2010) acknowledged the importance of including information on sexual orientation identity when examining entry into same-sex unions, but could not do so because such a measure was not available in the NCDS nor the BCS.

Variables

Dependent Variable

First Same-Sex Coresidential Union Formation

At Wave IV, respondents were asked detailed questions about their cohabitation and marriage histories, including how many individuals they had been married to or living with, as well as start dates (month and year) of cohabitation and marriage for each partner identified.Footnote 1 As noted earlier, respondents were also asked demographic information about their partner, including their biological sex. Add Health is one of the only U.S. data sets that enables identification of same-sex relationships based on coresidence and also includes start dates of all coresidential unions (National Center for Family and Marriage Research 2013). For example, while the NLSY contains information about union formation, prior to 2005 it only asked about different-sex unions (Mernitz and Pollitt 2018). Based on the sex of the respondent (marked by the interviewer) and the sex of their partner (marked by the respondent), we were able to identify both same-sex and different-sex unions. For respondents with multiple same-sex coresidential partners, we included only the partner with the earliest union start date. Based on this date and the interview dates, we constructed a variable for the number of months between the Wave III interview and the time when the respondent first entered a same-sex coresidential union or reached the Wave IV interview (for those who did not form a same-sex union). We create a similar variable for the timing of forming a first different-sex coresidential union between interviews. An alternative modeling strategy would be to focus on the formation of the first coresidential union and treat same-sex and different-sex unions as competing risks (e.g., Strohm 2010); however, several respondents formed a different-sex union prior to forming a same-sex union.

Independent Variables

Sexual Minority Status

Sexual minority status was operationalized using the following question from Wave III: “Please choose the description that best fits how you think about yourself: (1) 100% heterosexual (straight), (2) mostly heterosexual (straight), but somewhat attracted to people of your own sex, (3) bisexual that is, attracted to men and women equally, (4) mostly homosexual (gay), but somewhat attracted to people of the opposite sex, (5) 100% homosexual (gay), (6) not sexually attracted to either males or females.” Following previous research (e.g., Hatzenbuehler et al. 2013), we excluded asexual respondents from our analyses. We recoded sexual minority status into four categories: (1) 100% heterosexual (straight), (2) mostly heterosexual, but somewhat attracted to people of your own sex, (3) bisexual, that is, attracted to men and women equally, and (4) mostly homosexual (gay), but somewhat attracted to people of the opposite sex or 100% homosexual (gay). For models of same-sex coresidential union formation, the sample is restricted to individuals who identified as bisexual or mostly/100% homosexual. We acknowledge that this measure conflates both attraction and identity, as the question refers to both of these features of sexual orientation. Despite this limitation, studies based on Add Health routinely use this question to operationalize sexual orientation identity (e.g., Everett and Mollborn 2014; Krueger et al. 2018).

Contextual Variables

County-Level Voting

As an attitudinal indicator of support for sexual minorities, we measure the percent of votes cast in respondent’s county for the Republican presidential candidate during the 2000 election (McVeigh and Maria-Elena 2009). This measure was obtained from the political context database appended to the Add Health by Fowler et al. (2010). Following McVeigh and Maria-Elena (2009), we used a continuous variable which ranged from 0.090 to 0.885.

Same-Sex Couple Concentration

We use the percent of households headed by same-sex unmarried partners in respondent’s tract as a demographic indicator of social support for sexual minorities. This measure was obtained from the contextual data appended to the Add Health by Swisher (2008). In supplemental analyses combining data from the U.S. Census and the 1988–2008 General Social Survey, Schwartz and Graf (2010) demonstrated that the percent of same-sex cohabiting couples across different locales was highly correlated with the percent of individuals identifying as gay or lesbian. Following prior work using the Add Health (Everett 2014), we use dummy variables in the models to distinguish respondents according to the concentration of same-sex couples in their neighborhood. Preliminary analyses indicated that the same-sex couple concentration variable had large right skew and a modal value of zero. Thus, we divided sexual minorities into three equally sized categories (or tertiles) on the basis of the concentration of same-sex cohabiting couples in their census tract. A low concentration tract was less than 0.003, a medium concentration tract was 0.003 to 0.008, and a high concentration tract was greater than 0.008. In supplemental analyses discussed later, we alternatively included a logged measure of same-sex couple concentration at the tract level. We also examined the effects of state- and county-level same-sex couple concentration in the supplemental analyses.

Out to Either Parent

We use a measure of whether or not the respondent had disclosed their sexual orientation identity to either of their parents as an indicator of family context. This measure was operationalized using the following question that immediately followed the Wave III measure of sexual orientation identity: “Which of your parents knows that you are bisexual/about your homosexuality? Neither parent knows, only mother knows, only father knows, both parents know.” Responses were recoded as (0) neither parent knows or (1) only mother knows, only father knows, or both parents know. This question was only asked of respondents who identified as bisexual, mostly homosexual, or 100 percent homosexual.

Control Variables

Consistent with prior work on union formation using Add Health (e.g., Raley et al. 2007), we include controls for gender, age, race, and family background.

Woman

We include a variable indicating whether the respondent was a woman or not.

Age at Wave III

We include a variable for respondent’s exact age at Wave III (i.e., 18.5).

Race/Ethnicity

Race/ethnicity of respondent was collected from the first wave of the study and recoded to a series of dummy variables (non-Hispanic black, Hispanic, and non-Hispanic other) with non-Hispanic white acting as the reference group.

Living with Two Biological Parents

Based on the household roster at Wave I, we established if the respondent was living with both biological parents at Wave I versus not.

Family SES

Family SES is based on a measure developed by Bearman and Moody (2004) that incorporates information on parental education and occupation from Wave I; this measure is widely used in studies based on Add Health.

Moved 50 + Miles

Following Ueno et al. (2014) we include an indicator of whether or not the respondent moved more than 50 miles between Waves I and III. This measure was obtained from the contextual data appended to the Add Health by Swisher (2008) that included geographical distances between waves. Respondents were coded as (1) migrated if they moved 50 miles or more between Waves I and III or (0) if they did not. Respondents who did not have valid geocode data were excluded from the analysis.

Analytic Strategy

We begin with a descriptive profile of individuals who identified as a sexual minority (bisexual, mostly/100% homosexual) or a sexual majority (mostly/100% heterosexual) at Wave III, contrasting these two groups of respondents with respect to same-sex union history, contextual variables, and control variables. We then turn to survival analysis to examine the timing of same-sex union formation. Specifically, we display weighted estimates of the proportion of bisexuals and homosexuals who formed a same-sex coresidential union between Waves III and IV. Next, we present the hazard ratios from Cox models of same-sex coresidential union formation that adjust for survey design effects. The risk period for the models begins with the Wave III interview and ends in the month that the respondent formed their first same-sex union (if the respondent formed a same-sex union) or the month of the Wave IV interview.

Results

Descriptive Results

Table 1 displays sample means for heterosexuals and sexual minorities. Results of the significance tests indicate that heterosexual and sexual minority groups do not differ significantly (p < 0.05) from each other on any of the variables except for same-sex union formation, different-sex union formation, and sex. Considering our key outcome, we see dramatic differences in union formation by sexual orientation identity. As expected, less than 1% (0.56%) of heterosexuals formed a same-sex coresidential union prior to Wave IV versus 20.84% of sexual minorities. In addition, greater percentages of heterosexuals were women than sexual minorities (51.16% versus 42.10%). Finally, as expected, the majority of heterosexuals (63.51%) formed a different-sex union between Waves III and IV, compared to 25.15% of sexual minorities.

Table 1 Descriptive statistics sexual identity orientation: respondents from Wave IV (N = 11,808) without any same-sex coresidential unions prior to Wave III
Full size table

Figure 1 shows results from weighted life table analyses of same-sex coresidential union formation for bisexuals and homosexuals. Following previous studies (e.g., Everett and Mollborn 2014), we collapsed mostly homosexual and 100% homosexual. The x-axis shows the number of years since the Wave III interview and the y-axis shows the cumulative weighted proportion of individuals who had entered a same-sex coresidential union by a given year. Mostly/100% homosexual respondents were considerably more likely than bisexuals to form a same-sex union. By 6 years after the Wave III interview, 35% of mostly/100% homosexual respondents had formed a same-sex union compared to roughly 11% of bisexual respondents.

Fig. 1
figure 1

Life table estimates of the proportion forming a first same-sex coresidential union between Waves III and IV: sexual minorities

Full size image

Multivariate Results

Table 2 displays the ratios from Cox models of first same-sex union formation. Model 1 includes Republican voting at the county level and the control variables. Model 2 includes the tract same-sex concentration dummies and the control variables. Model 3 displays the out to either parent variable and the control variables. Model 4 represents the full model and includes all three sets of contextual variables.

Table 2 Hazard ratios from cox models of first same-sex union: sexual minority respondents (N = 304)
Full size table

The results from Model 1 show that Republican voting is not significantly associated with same-sex union formation for sexual minorities. In Model 2, our two indicators of same-sex couple concentration are significant for sexual minorities (p < 0.05). Sexual minorities who live in tracts with medium and high concentrations of same-sex couples have significantly higher rates of forming a first same-sex union than those in tracts with a low concentration of same-sex couples. Specifically, sexual minorities living in tracts with a high concentration of same-sex couples have rates of forming a first same-sex union that are almost three times (i.e., 2.8) higher than those of their counterparts residing in tracts with a low concentration of same-sex couples; sexual minorities living in tracts with a medium concentration have rates that are more than twice as large (i.e., 2.4) as those of their counterparts in tracts with a low concentration.

The results from Model 3 reveal that sexual minorities who are out to either parent have significantly higher rates of forming a first same-sex union compared to those who are not out to either parent. Specifically, sexual minorities who are out to either parent have a rate of forming a same-sex coresidential union that is roughly 4.5 times higher than that of those who are not out. An advantage of this analysis is that the indicator of coming out precedes the transition to coresidence; however, respondents who plan to move in with a partner may feel pressure to come out to their parents.

Model 4 presents results from the full model that combines all three sets of contextual variables. Sexual minorities living in a tract with a high concentration of same-sex couples continue to have significantly higher rates of forming a first same-sex union than sexual minorities living in a tract with a low concentration of same-sex couples. Furthermore, being out to either parent continues to be significantly related to higher rates of forming a first same-sex union. In the full model, sexual minorities who are out to their parents have 4.2 times higher rates of forming a first same-sex union than those who are not. As expected, the inclusion of information on being out to parents reduces the hazard ratios for the same-sex couple concentration indicators.

Sexual minority women have roughly one-third the rate of forming a same-sex union as sexual minority men. We suspect this reflects the fact that women are more likely than men to identify as bisexual (e.g., roughly four-fifths of the women versus one-third of the men). Recall from Fig. 1 that bisexuals were considerably less likely than homosexuals to form same-sex unions between interviews. Savin-Williams et al. (2012) found that bisexual men and women alike were dramatically more likely than homosexual men and women to change their identity between Waves III and IV. Among bisexuals who changed their identity, women were more likely than men to shift towards heterosexual.

Supplemental Analyses

We also conducted several sensitivity analyses that are not displayed. We estimated models that included interactions between gender and the contextual variables and did not detect any significant differences; this reflects the fact that the effects were similar in magnitude for sexual minority men and women. We also interacted the same-sex neighborhood concentration indicators with a linear specification of time and, alternatively, a logged specification. The results from these models suggest that the gaps between sexual minorities in neighborhoods with medium or high concentrations of same-sex couples and their counterparts in neighborhoods with a low concentration did not differ significantly over the risk period. Similar tests suggested that rates for those who were out to parents versus not were also proportional over time. We additionally ran sets of left-truncated models that began the risk period with the exact age at Wave III and using as a timing variable the age at first same-sex union formation (or age at Wave IV interview if censored). The ratios and significance levels for these models were virtually identical to those displayed in Table 1.

In addition, we estimated models that included measures of same-sex concentration at the state and county levels. These variables (recoded into high, medium, and low concentration) were not significant. As a falsification test, we substituted in our models a tract-level variable for concentration of different-sex couples and failed to detect any significant effects for sexual minorities. Drawing on Rosenfeld’s (2007) premise that independence from family and geographic mobility has led to an increase in same-sex couples, we ran models that additionally included an interaction between the logged same-sex concentration variable and geographic mobility. This interaction term was not significant, suggesting the effect of same-sex neighborhood concentration did not differ for sexual minorities based on their mobility.

To take sexual fluidity into account, we ran models that included an indicator for stable sexual minority identity between Waves III and IV. Stable identity was associated with a higher likelihood of forming a same-sex union, but did not change the contextual effects. This variable did, however, explain why sexual minority women were less likely than sexual minority men to form a first same-sex union.

To address possible confounding variables, we investigated whether additional variables were associated with same-sex union formation. Since Add Health did not ask about the dates of degree completion at Wave IV, we were unable to include a time-varying covariate for education. Instead we included a Wave IV indicator of whether the respondent had a bachelor’s degree and failed to find a significant effect. We also examined how neighborhood socioeconomic disadvantage was associated with same-sex union formation using a scale that included variables for the proportion of men sixteen and older who were unemployed, the proportion of households receiving public assistance, the proportion of the population below the poverty level, the proportion household with incomes less than $75K, the proportion of those 25 years or older without a bachelor’s degree, and the proportion of those employed 16 years and over who were not in management, professional, or related occupations (e.g., South and Crowder 2000). This scale was not associated with same-sex union formation among sexual minorities using either a linear or quadratic specification in the models. We also added logged specifications of variables (one at a time) that captured median household income and housing prices, as these variables failed significance tests for normality. These indicators of neighborhood affluence were not significantly associated with same-sex union formation. We also added a set of indicators for residing in the South, Northeast, or West (vs. the Midwest) at Wave III and did not detect any significant regional differences.

Finally, we examined the effects of contextual factors in analogous models of different-sex union formation among heterosexuals who had not formed such a union at the time of the Wave III interview (presented in “Appendix”). With respect to the contextual variables, sexual majority individuals living in areas with higher percentage of Republican voting had significantly higher rates of forming a different-sex union. The effects of other variables were consistent with previous research (Guzzo 2006; Kuhl et al. 2012; Teachman 2003).

Discussion

The social landscape surrounding sexual minorities and, more broadly, union formation has been rapidly changing in the past two decades. During this time, the average age at marriage has risen (U.S. Census Bureau 2018) and sexual minorities have gained the right to legally marry in all states (Human Rights Campaign 2018). Even with these changes, and despite the documented importance of context for sexual minorities, studies have yet to directly examine the effects of contextual factors on same-sex union formation in the United States. Using data from the National Longitudinal Study of Adolescent to Adult Health, and guided by Bronfenbrenner’s ecological systems theory framework, this research sought to fill a major gap in our understanding of union formation by examining the association between social context and same-sex coresidential union formation, with a particular emphasis on sexual minorities.

Consistent with our expectations, we found that family context mattered for same-sex union formation. About half of sexual minority young adults were out to their parents and being out had clear implications for their union formation. Almost half (48.4%) of sexual minority men who were out to either parent formed a same-sex union compared to fewer than one-sixth (14%) of sexual minority men who were not out (results not shown). Among sexual minority women, over one-fifth (20.7%) who were out to either parent formed a same-sex union compared to less than one-tenth (10.9%) of sexual minority women who were not out to either parent (results not shown). As shown in the analyses, respondents who were out to either parent had significantly higher rates of forming a same-sex union than those who were not. This finding is consistent with previous research by Rosenfeld and Kim (2005) and Strohm (2010) which suggests that the family of origin has a critical influence on same-sex union formation.

Our demographic indicator of supportive context, same-sex couple concentration, also mattered for same-sex union formation. Sexual minorities living in tracts with a high concentration of same-sex couples had significantly higher rates of forming a first same-sex union than those living in tracts with a low concentration of same-sex couples. Our political measure of supportive context, county-level Republican voting, was not significantly associated with same-sex union formation for sexual minorities. We had expected that respondents residing in counties with lower percentage of voters who cast a vote for the Republican candidate (Bush) would be more likely to form coresidential unions. This political indicator of context has been important in other work on depression (Everett 2014), but does not appear to be influential for union formation.

While this paper provides new insights into union formation for sexual minorities, several limitations exist. First, because contextual data and information on sexual orientation identity is only available starting at Wave III, the analysis of union formation was limited to individuals who had not formed a same-sex union prior to Wave III. Although the sample of individuals who formed a same-sex union prior to Wave III did not differ significantly from the sexual minority sample used in this analysis, it is possible that context could function differently for individuals who form unions earlier. Second, the question used to operationalize sexual minority status combines both sexual orientation identity and attraction, thus conflating two very different concepts. Third, the question regarding disclosure of sexual orientation identity was only asked about parents and did not address parental reaction to coming out. Some respondents who were out to either parent may have experienced negative reactions that impeded their union formation. Relatedly, Add Health did not ask if the respondent was out to anyone else. It may be the case that disclosure to other groups (such as friends or co-workers) influences union formation in unique ways.

Fourth, our measures of social context came from the 2000 Census and thus fail to capture change during the period of risk that resulted from either respondents changing contexts or their contexts changing over time. Ideally, we would have examined the effects of sex-specific measures of same-sex couple concentration. Also, men in same-sex couples cluster geographically more substantially than women in same-sex couples (Madden and Ruther 2015). Therefore, the concentration variable is largely driven by male couples. Separate measures of the concentration of female same-sex households and male same-sex households are not available in the current Add Health contextual data base. It is also important to note that concerns about the accuracy of estimates of the number of same-sex couple households have been noted by the U.S. Census Bureau (O’Connell and Feliz 2011). More specifically, approximately 28% of same-sex couple households in the 2010 Census are likely different-sex couples (O’Connell and Feliz 2011). The estimates of same-sex couple households were less accurate in the South and upper Midwest than on the West coast and in the Northeast (O’Connell and Feliz 2011). Although we are unable to correct for this in the Add Health data, researchers should be aware of this potential limitation when interpreting results.

Finally, due to the period of observation (roughly 2000 to 2008), we could not distinguish formation of same-sex marital and cohabiting unions. At the time of the Wave IV interview, only two states had legalized marriage for same-sex couples, while 45 states had a constitutional or statutory ban on same-sex marriage. It may take time to see the effects of changing legal and social landscape to take place. As Frost et al. (2017) point out, “the social, political, and legal controversies surrounding same-sex marriage in the United States are deeply rooted. Their effects endure over time, and they vary across social settings and geographic locations” (p. 456). Wave V of Add Health will present a unique opportunity to examine sexual minority outcomes after the legalization of marriage to same-sex couples.

Despite these limitations, this research possesses many strengths. First, Add Health contains a larger number of sexual minorities and individuals in same-sex relationships than most other large data sets. For example, the most recent SIPP data only include approximately 200 same-sex couples across a wide age span (18–64). In addition, respondents were asked detailed questions about their cohabitation and marriage histories, allowing for a comprehensive examination of same-sex union formation. Importantly, the contextual data available with Add Health at Wave III allowed us to examine the effects of social context on union formation prior to the period of risk. Furthermore, this research is one of the first to examine the relationship between coming out and union formation. As coming out represents one of the biggest and most salient differences between sexual minorities and majorities, this research adds an important dimension missing in previous work.

In summary, we find that context matters for same-sex coresidential union formation. Sexual minorities have a higher likelihood of forming a first same-sex union the more supportive their context. Our findings illustrate the importance of considering context when examining the formation of same-sex coresidential unions among sexual minorities. Our analyses offer clues about the implications of changes in the legal recognition of same-sex couples for patterns of union formation among sexual minorities. As the social and cultural landscape of the United States continues to transform, understanding contextual factors is an important focus for future research on other outcomes for sexual minorities, including their health and well-being.