The Direct Effects of Legal Same-Sex Marriage in the United States: Evidence From Massachusetts

Abstract

Introduction

In June 2015, a 5–4 decision of the United States Supreme Court in Obergefell v. Hodges (2015) found that marriage is a fundamental right that cannot be deprived of citizens in same-sex relationships under the Due Process and Equal Protection clauses of the Fourteenth Amendment to the U.S. Constitution, effectively making same-sex marriage legal throughout the country. Obergefell resolved a preexisting patchwork of laws and policies in which the legal status of a same-sex marriage varied across states, even when same-sex couples were married in a jurisdiction allowing same-sex marriage. Two years prior to Obergefell, the Supreme Court in another 5–4 decision in United States v. Windsor (2013) found the federal government’s refusal to recognize state-sanctioned legal same-sex marriages under the 1996 Defense of Marriage Act an unconstitutional violation of the Due Process clause of the Fifth Amendment. Windsor effectively legalized same-sex marriage in the eyes of the federal government for the thousands of same-sex couples who had been legally married in the handful of states that had adopted same-sex marriage through state supreme court cases, state legislatures, or direct referenda prior to Windsor. The rapidly changing legal landscape has coincided with what many observers have noted is one of the fastest shifts in public sentiment on any social issue: polls suggest that a solid majority of adults in the United States now support the right of same-sex couples to marry (Dutton et al. 2013; McCarthy 2019; Pew Research Center 2012, 2013), which was not true as recently as 2010.

Opponents of legal same-sex marriage argued that it is unlikely to be desired by sexual minorities,¹ would harm the institution of traditional heterosexual marriage, and is bad for children. Proponents argue that there is no support for these claims based on the available evidence, including the experiences of other countries that legalized same-sex unions in the 2000s (Badgett 2009; Dillender 2014; Trandafir 2014, 2015). There is, however, alarmingly little credible social science evidence using population-based data sets on the effects of legal same-sex marriage in the United States on even the most direct outcomes: marriage take-up, partnership, and union formation.

In this article, I provide some of the first direct evidence on these questions by studying the experience of Massachusetts, which was the first U.S. state to legalize same-sex marriage in a decision of the state’s Supreme Judicial Court in Goodridge v. Massachusetts Department of Public Health (2003). Specifically, I address three research questions. First, does access to legal same-sex marriage increase marriage take-up among existing same-sex couples? Second, does access to legal same-sex marriage increase the likelihood of being in a same-sex couple (i.e., does it affect union formation)? And third, how do these effects vary with sociodemographic characteristics such as age, race, education, and the presence of children in the household?

The analysis uses confidential public health survey data from the Massachusetts Behavioral Risk Factor Surveillance System (MA-BRFSS), which has asked individuals a direct question about sexual orientation every year since 2001. When pooled, these data identify more than 4,300 self-identified lesbians, gay men, and bisexual individuals and more than 117,000 heterosexuals. I combine the sexual orientation responses with information on marital status to document the dynamics of marriage and partnership for sexual minorities compared with heterosexuals around the timing of court-ordered same-sex marriage in the state.

To preview, I find that the availability of legal same-sex marriage in Massachusetts led to large, immediate, sustained, and statistically significant increases in marriage among sexual minorities relative to heterosexuals. I estimate increases in the probability that self-identified lesbians, bisexual women, and gay men report being married on the order of 27.1, 8.4, and 11.4 percentage points, respectively, relative to the associated probability for heterosexuals. I show that these changes are entirely driven by increases in marriage among sexual minorities as opposed to decreases in marriage among heterosexuals and that the results are not due to compositional changes in who identifies as gay, lesbian, or bisexual. These increases in marriage were mostly offset by statistically significant decreases in the probability of being “a member of an unmarried couple” (especially among lesbians); I find little effect of the availability of same-sex marriage on union formation.

The results in this article are important for several reasons. First, same-sex marriage remains one of the most pressing and controversial social issues, and there has been troublingly little population-based empirical social science research for informing debates about the likely effects of Obergefell and related decisions on marital status, family formation, and social outcomes. My research confirms that there is substantial and robust demand for same-sex marriage among sexual minorities. My results also speak directly to another argument commonly voiced against legalizing same-sex marriage: namely, that it will harm different-sex marriage. As measured by heterosexual marriage probabilities in the medium term (i.e., up to nine years after legalization), I do not find support for this argument, consistent with Dillender (2014) and Trandafir (2015).

Second, this article adds to a large literature on the economic and policy determinants of marriage more broadly. Scholars have demonstrated the empirical significance of the tax-cost of marriage (Alm and Whittington 1999), unilateral divorce laws (e.g., Friedberg 1998; Peters 1986; Wolfers 2006), blood test requirements (Buckles et al. 2011), minimum age requirements (Blank et al. 2009), and other public policies on marriage rates among heterosexuals. The work presented here adds to prior research by studying a much more direct change to the overall cost of getting married—access to the legal institution for same-sex couples—than has been typically examined. In this sense, this work is related to prior studies of anti-miscegenation statutes and the effects of Loving v. Virginia on interracial marriage (Fryer 2007; Gevrek 2014).

Background, Conceptual Framework, and Prior Literature

The Goodridge Decision and Related Institutional Background

This study evaluates the effects of a 2003 ruling by the Supreme Judicial Court of the State of Massachusetts. In that case, the Gay & Lesbian Advocates and Defenders (GLAD, now called GLBTQ Legal Advocates & Defenders) brought a case on behalf of seven same-sex couples who were denied marriage licenses in the state in 2001. Hillary and Julie Goodridge were the lead plaintiffs in the case; they had been together for nearly two decades and were raising a daughter together. In 2002, a Superior Court judge ruled in favor of the Massachusetts Department of Health, writing that the appropriate place to decide this issue was the state legislature (which had recently defeated legislation that would have legalized same-sex marriage) as opposed to the court. The case was appealed directly to the state’s Supreme Judicial Court and was argued in March 2003. On November 18, 2003, the Massachusetts Supreme Judicial Court ruled in a 4–3 decision that denying same-sex couples the right to marry violated the state constitution’s equal protection clause. The court issued a 180-day stay of the ruling to allow the state legislature time to determine how it would accommodate the ruling, but no agreement was reached. On May 17, 2004 the first same-sex marriage licenses were issued to couples in Cambridge, Massachusetts.

The Massachusetts ruling studied here is one in a series of interventions by state and federal courts and legislatures in the United States on this issue. In 1996, the U.S. Congress adopted the Defense of Marriage Act (DOMA), which defined marriage as between one man and one woman, effectively making same-sex marriage illegal in the eyes of the federal government. After DOMA was enacted, a handful of states took steps to recognize same-sex civil unions and domestic partnerships which usually contain some subset of the rights and responsibilities of different-sex (or “traditional”) marriage within the state (e.g., California in 1999; Vermont in 2000, Connecticut in 2005; New Jersey in 2006; and Washington, Oregon, and New Hampshire in 2007). Other states went further in formally legalizing same-sex marriage. Most early movers followed Massachusetts’ lead and did so through state supreme court decisions (e.g., California and Connecticut in 2008; and Iowa in 2009), usually finding that depriving gay and lesbian couples of marriage rights violates due process and equal protection. Other states also legalized same-sex marriage through direct legislative action (e.g., Vermont, New Hampshire, and Washington, DC, all in 2009) and through the ballot box (e.g., Maine, Maryland, and Washington in 2012). Notably, a majority of U.S. states also enacted statutes and approved ballot propositions against same-sex marriage. Much of this activity occurred in the November 2002, 2004, and 2006 elections in which many Southern and Midwestern states adopted statewide constitutional amendments defining marriage within the state as between one man and one woman, despite that most of these states already had statutes against same-sex marriage on the books.

Theoretical Considerations

How and why might access to legal same-sex marriage affect marriage and cohabitation? Most prior work has used conceptual frameworks from sociology and economics to understand possible effects on family formation outcomes for heterosexuals in response to different forms of legal recognition for same-sex couples. The economic models following Becker (1973, 1974) turn on changes in the economic value of marriage relative to alternative statuses such as unmarried cohabitation. As Trandafir (2014) noted, granting same-sex couples the right to full marriage equality could increase the value of marriage for some heterosexuals (and thus increase the traditional marriage rate) through several different mechanisms, including: reducing the likelihood that governments and employers provide benefits to unmarried partners (Rauch 2004), triggering a threat of identity to heterosexual couples (Akerlof and Kranton 2000), or increasing general interest in marriage. Granting legal same-sex marriage could also decrease the value of marriage for some heterosexuals. Cherlin (2004) has discussed the “deinstitutionalization” of marriage as characterized by the growth of alternative family formations (including same-sex relationships) and the long-term decline in traditional marriage. Access to legal same-sex marriage could further signal this deinstitutionalization and induce some different-sex couples to forego marriage. Finally, the exclusive nature of marriage to only different-sex couples likely appeals to some people in different-sex relationships and not to others; granting legal same-sex marriage should be expected to increase the relative value of marriage for those who saw exclusion of same-sex couples as discriminatory and should decrease the relative value of marriage for those who preferred its exclusive aspect. These competing factors make the net effect on different-sex marriage theoretically ambiguous.

Understanding how access to legal same-sex marriage should affect these same outcomes for sexual minorities has been less explored. On the one hand, it is inevitable and not particularly informative that moving from a policy environment where same-sex marriage is prohibited to one in which it is allowed (i.e., moving from an effectively infinite cost to a reasonably low cost) will be expected to increase marriage take-up among sexual minorities. Arguably, the more interesting theoretical questions include whether we should expect same-sex marriage rates to approach different-sex marriage rates and how we might expect same-sex marriages to compare with different-sex marriages. The latter question is beyond the scope of this article because of data limitations, although some prior related work has examined the division of labor in same-sex couples compared with different-sex couples (but did not address access to marriage per se; see, e.g., Jepsen and Jepsen 2015).

There are several theoretical reasons that one might expect take-up of same-sex marriage to be similar to take-up of traditional different-sex marriage. In addition to the direct, tangible benefits to legal marriage for same-sex couples (e.g., health insurance, parenting rights, etc.), other intangible benefits to marriage would accrue to same-sex couples in much the same way as they do to different-sex couples. Badgett (2009–2010) outlined these frameworks, which largely deal with reducing uncertainty about the division of property and custody of children if the marriage ends. For example, marriage promotes household specialization to make optimal investments in human capital and labor market skills that exploit comparative advantages in market and home production. Marriage also provides social insurance, protecting each spouse against ill health or job loss of the other. Marriage reduces transaction costs and the need to renegotiate contracts as household circumstances change, allows couples to take advantage of economies of scale, signals commitment, and promotes reciprocity and altruism. All these features are as meaningful to same-sex couples as they are to different-sex couples.

At the same time, there are also reasons that one might expect same-sex marriage take-up to be different from (lower than) the associated rates for heterosexuals, many of which Manning et al. (2016) discussed in the context of a closely related family outcome: relationship stability. First, same-sex couples—particularly gay male couples—are much less likely than different-sex couples to have (or expect to have) children. Given that a key role of marriage is to serve as legal and social protection for children, the demand for marriage by same-sex couples will likely be lower than for different-sex couples. Second, getting married is inherently a public act (at least partially), which may expose sexual minorities to a higher degree of discrimination and minority stress (Meyer 1995), potentially depressing their demand for marriage. Indeed, the granting of new legal rights is often followed by intense periods of backlash against the targeted group. Also, the population at risk of getting married (i.e., individuals in relationships) is much smaller for sexual minority men than for heterosexual men. Multiple prior studies have documented that gay men partner at substantially lower rates than heterosexual men, heterosexual women, and lesbians (see, e.g., Carpenter and Gates 2008). Although marriage may increase incentives to form partnerships, the historically lower partnering rate of gay men would predict lower marriage take-up for men in same-sex couples compared with men in different-sex couples.

Literature Review

Despite substantial legal scholarship on the topic of same-sex marriage and the role of the Goodridge decision (see, e.g., an excellent discussion in Bonauto 2005), there has been far less research in quantitative social science evaluating the effects of Goodridge specifically or of legal same-sex marriage policies in general.² This gap exists partly because of a historical lack of large population-based surveys with information on sexual orientation at the individual level, which is required to understand partnership and marriage.³ The National Health Interview Survey, for example, began asking about sexual orientation only in 2013. Some states now include sexual orientation questions on state health surveys, but most of this activity occurred after state experiences with legal same-sex marriage, thus precluding analyses of changes over time. Some published research has explored effects of same-sex marriage and related relationship recognition policies in the United States and abroad on outcomes such as health insurance coverage, health care use, and health and social outcomes (see, e.g., Boertien and Vignoli 2019; Buchmueller and Carpenter 2012; Carpenter et al. 2019; Dee 2008; Francis et al. 2012; Gonzales 2015; Hatzenbuehler et al. 2012; Raifman et al. 2017), but I am not aware of any published work in the United States on the more fundamental question of how availability of legal same-sex marriage affects union formation and marital status using population-based representative data with individual-level information on sexual orientation.

Three closely related studies, however, have examined how different forms of legal recognition of same-sex couples have affected marriage and family outcomes for different-sex couples, including one study in the United States. Dillender (2014) studied the experience of U.S. states from 1995 to 2010 in difference-in-differences models and found no effects of legal access to same-sex marriage or registered domestic partnership on marriage rates of different-sex couples. Two related studies examined the experiences of other countries. Trandafir (2014) studied the experience of the Netherlands, which granted marriage-like registered domestic partnership to same-sex couples in 1998 and full marriage equality in 2001. Using multiple empirical approaches including a synthetic control design, he found no effects of the Netherlands’ policies on overall or different-sex marriage rates in the aggregate.⁴ In a related analysis, Trandafir (2015) expanded his sample to include the experiences of all OECD countries from 1980 to 2008 and confirmed no systematic effects of access to same-sex marriage or registered domestic partnerships on family formation outcomes (e.g., marriage, divorce, or extramarital births) in those countries. None of these studies, however, examined marriage take-up or union formation by sexual minorities, mainly because of a lack of data.

Finally, although this study is the first, to my knowledge, to examine the Massachusetts ruling using population-representative individual-level data on sexual orientation, other scholars have used state administrative data on marriage licenses and registered domestic partnerships combined with census data to study the Massachusetts reform.⁵ For example, Badget and Herman (2011) estimated that 75% of lesbian couples and 59% of gay male couples in the state legally married, and marriage license data from other states also consistently show that women in same-sex relationships demand marriage and other legal relationship statuses more than men in same-sex relationships, consistent with findings of Carpenter and Gates (2008). Gates et al. (2008) also examined the age distribution of individuals in Massachusetts who married and found that same-sex couples who married in the state were systematically younger than the average age in the stock of married different-sex couples in the state. Measured in terms of flows, however, same-sex couples who married in Massachusetts were systematically older than the average different-sex couple who also obtained marriage licenses over the same period, likely reflecting that the same-sex couples were not legally allowed to marry earlier in their relationships.

To summarize, a substantial and growing body of work has examined legal relationship recognition for same-sex couples in the United States and internationally. This literature has produced a good deal of evidence on marriage take-up from administrative data on marriage licenses but lacks information on the population “at risk” of potentially marrying (that is, single sexual minorities who may want to enter into a same-sex partnership or marriage). I fill that gap in the literature with the first study using population-representative data to directly evaluate how access to legal same-sex marriage affected marriage take-up and union formation among sexual minority individuals.

Data Description and Empirical Approach

Data

Data come from confidential versions of the Massachusetts Behavioral Risk Factor Surveillance System (MA-BRFSS), a representative sample of Massachusetts adults. These data represent the state’s participation in the national BRFSS, which is coordinated by the Centers for Disease Control and administered by each individual state. The national BRFSS includes a core questionnaire that includes questions all states must ask, and states can choose to include additional questions on topics of public health interest to the state. This study makes use of the fact that Massachusetts has asked direct questions about sexual orientation to all respondents aged 18–64 since 2001; my analysis sample goes through 2013. I use information on each respondent’s sexual orientation in conjunction with information on marital status to identify the effects of Goodridge.⁶

Specifically, each individual aged 18–64 in the MA-BRFSS is asked, “Do you consider yourself to be: heterosexual or straight; homosexual or gay; bisexual; or other?”⁷ Regarding marital status, respondents are asked, “Are you: married, divorced, widowed, separated, never married, or a member of an unmarried couple?” I define a variable MARRIED that equals 1 for respondents who state they are married, and I define MEMBER OF AN UNMARRIED COUPLE similarly. I also define a variable called PARTNERED that equals 1 if the respondent reports she is either married or a member of an unmarried couple; this is a proxy for union formation.

The BRFSS marital status question is less than ideal for my purposes, and the challenges associated with using marital status for sexual minorities have been addressed elsewhere (Carpenter and Gates 2008). One important limitation of the BRFSS marital status question is that although it was intended to elicit legal marital status, the survey does not explicitly instruct respondents in this way. Thus, I cannot know whether people who report being married are, in fact, legally married or instead simply think of themselves as married. One indication of this problem is that in 2001–2003, the period prior to same-sex marriage licenses becoming available in Massachusetts, some small share of gay and lesbian respondents reported being married. Some of these individuals could have been married in Ontario, Canada, which began performing same-sex weddings in 2001 and fully legalized them in 2003 without residency or citizenship requirements through a provincial court ruling. Some of these individuals might also be gay men and lesbians who are legally married to different-sex spouses.

Empirical Approach

Notably, I also control for (but do not report in the equation) an indicator variable for observations in the DURING GOODRIDGE period (i.e., November 18, 2003–May 17, 2004) and the associated interactions between this variable and the GAY/LESBIAN and BISEXUAL indicator variables. In this interim period, there was presumably uncertainty about whether and how the legislature was going to handle the directive from the state’s Supreme Judicial Court to accommodate its ruling. For example, less than a month after the ruling the legislature asked the court if creating an institution that was similar to marriage but with a different name (e.g., civil unions) would suffice; two months later, in February 2004, the court responded that it would not. Thus, it is important to consider potential effects during this period separately from the period after licenses became available. In practice, there are very few observations during this period (particularly for sexual minority individuals). How I treat this “during” period does not change the main findings.

One concern with the estimating equation is the possibility that choice of sexual orientation label may have changed in response to the decision as a result of changes in social attitudes toward sexual minorities. The direction of this effect is not obvious because there may have been some antigay backlash in response to the decision, as there was in California during the summer of 2008 after courts in that state legalized same-sex marriage. Figs. A1 and A2 in the online appendix show the proportion of women and men, respectively, choosing the various sexual orientation categories. The most notable pattern in each figure is that refusal to answer the sexual orientation question declined over time. This trend does not appear to change around the timing of Goodridge. Importantly, the proportion of women responding lesbian or bisexual was remarkably stable over time, suggesting that my findings for women are unlikely to be biased by such composition concerns. For men, some visual evidence suggests an increasing share self-identifying as gay, but the magnitude is very modest compared with the relative increase in same-sex marriage that I estimate later.

In Eq. (1), I also control for a vector X_it, which includes demographic characteristics that are included in the MA-BRFSS data: age group (25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, and 60–64, with 18–24 as the excluded category); race/ethnicity (Black, Hispanic/Latino, and other race, with White as the excluded category); and education (high school graduate or less, college degree or more, and education missing, with some college as the excluded category). T_t is a vector of dummy variables for each survey wave. I estimate heteroskedasticity-robust White standard errors, and the models are unweighted.⁹ ε_it is assumed to be a well-behaved error term.

Results

Table 1 shows sample characteristics for the 2001–2013 MA-BRFSS data analyzed here. These patterns have been discussed elsewhere and match those found from other similar population-based representative data sets that have asked direct questions about sexual orientation and from data that permit identification of same-sex couples, such as the decennial census (e.g., Black et al. 2000; Carpenter 2005). For example, I find that very few men identified as bisexual—a finding that is in stark contrast to patterns for women, sizable numbers of whom self-identified as bisexual women. Both gay men and lesbians reported much higher levels of education than heterosexual individuals, are more likely to be White, and are much less likely to have had children present in the household.

Figure 1 plots the raw probabilities of reporting being married among heterosexual women, heterosexual men, lesbians, and gay men in the MA-BRFSS data in each year from 2001 to 2013 (Fig. A3 in the online appendix presents the same information for heterosexual women, heterosexual men, bisexual women, and bisexual men). Several patterns merit mention. First, heterosexual women and men reported being married at much higher rates than either lesbians or gay men. Second, the probability of reporting being married among heterosexual women and men did not change much across the introduction of same-sex marriage in Massachusetts, and some visual evidence suggests that marriage increased among heterosexual men.¹⁰ Third, lesbians were much more likely to report being married than were gay men, a finding consistent with prior work showing that lesbians are more likely than gay men both to be in partnerships and to formally recognize/register them with the government (Badgett and Herman 2011; Carpenter and Gates 2008). Fourth, the probability of reporting being married among lesbians and gay men was very low (below 10%) prior to 2004 but increased sharply and continuously for lesbians (and more gradually for gay men) following the legalization of same-sex marriage in the state.¹¹ Overall, the results in Fig. 1 indicate that the Goodridge ruling led to immediate and large increases in marriage probability for lesbians and more modest increases in marriage probability for gay men.

Table 2 presents the associated regression results corresponding to Fig. 1 (and Fig. A3, online appendix), along with other outcomes pertaining to partnership in columns 2 and 3. The top panel of the table presents results for women, and the bottom panel presents results for men. Each column in each panel represents the results from a separate estimation of Model 1, with coefficients on the relevant interaction terms (LESBIAN/GAY × AFTER GOODRIDGE and BISEXUAL × AFTER GOODRIDGE) in a model that includes all the other control variables described in Eq. (1). Results for the probability that respondents reported being married are presented in column 1; results for models that consider the probability of reporting being a member of an unmarried couple and the probability of reporting being either married or a member of an unmarried couple are shown in columns 2 and 3, respectively.

The results in column 1 of Table 2 confirm the patterns observed in Fig. 1. Specifically, I estimate that Goodridge led to 27.1 and 8.4 percentage point increases in the probability of reporting being married among lesbians and bisexual women, respectively, relative to the associated change for heterosexual women. Both estimates are statistically significant at conventional levels. Among gay men, as shown in the bottom panel, the ruling is estimated to have led to a significant 11.4 percentage point increase in the probability of reporting being married; the estimate for bisexual men is negative and statistically insignificant. Moving to the probability of reporting being a member of an unmarried couple, shown in column 2, I estimate statistically significant decreases in this outcome for lesbians and bisexual women following Goodridge of about 24.5 and 6.3 percentage points, respectively, suggesting that the large majority of women who were previously in unmarried partnerships formalized their relationship by getting married (although the estimate for bisexual women shown in column 2 is not statistically significant). The same pattern can be seen for gay men in the bottom panel of column 2: Goodridge significantly reduced the probability of being a member of an unmarried couple by 6.2 percentage points for gay men compared with the associated change for similarly situated heterosexual men. Finally, column 3 shows that the combination of very large increases in marriage and large decreases in being a member of an unmarried couple correspond to modest estimated increases in the likelihood of being partnered for lesbians and gay men, though neither estimate is statistically significant.¹² A visual presentation of these relationships for lesbians and gay men in Figs. 2 and 3, respectively, shows that although union formation was fairly stable over this period, increases in marriage were accompanied by reductions in the likelihood of being a member of an unmarried couple.

In results not reported but available upon request, I found similar patterns when incorporating information in the survey on household sex composition. Specifically, individuals were asked how many adult men and how many adult women were in the household. Because the vast majority of heterosexuals, gay men, and lesbians in marriage-like relationships do, in fact, cohabit with their partners (Carpenter and Gates 2008), and because Goodridge should have changed family formation differentially only for same-sex couples (given that different-sex marriage was legal throughout the period), it is natural to consider outcomes that cross the self-reported sexual orientation information with minimum household sex composition requirements. I therefore considered an outcome in which lesbians and bisexual women were defined as MARRIED only if they reported both being married and having at least two adult women in the household (and logically similarly for gay and bisexual men). Although this approach is not perfect (e.g., it incorrectly counts same-sex adult children or parents of sexual minorities as the likely spouses of those sexual minorities), it isolates the group of sexual minorities who are most “at risk” of same-sex marriage, which is the only institution that changed differentially with the Goodridge decision in Massachusetts. Importantly, the results on marriage and partnership were largely unaffected when the additional information on household sex composition was incorporated. That is, I still found similarly sized and significant increases in marriage using this alternative marriage definition. I choose to use models without the household composition requirements as the baseline because they are more conservative.

Table 3 shows the two key coefficients of interest on marriage for separate demographic groups. As in Table 2, results for women are shown in the top panel, and results for men are shown in the bottom panel. Columns 1–2 present results by age group (18–44 vs. 45–64); columns 3–4 present results by educational attainment (some college or less vs. college degree or more); columns 5–6 present results by race (White vs. non-White); and columns 7–8 present results by presence of children in the household (which was asked only in 2001–2008). In addition to being descriptively interesting, some of these differences are also potentially revealing about the theoretical considerations for same-sex marriage. Sexual minorities with higher socioeconomic status may be able to realize the tangible benefits of marriage more directly (e.g., because they may have more assets to protect), although they would also be more likely than sexual minorities with lower socioeconomic status to have had alternative legal arrangements prior to Goodridge. Moreover, marriage probabilities among the comparison group (heterosexuals) may have also varied by demographic characteristics. It is well documented that younger and more highly educated individuals hold more liberal and accepting attitudes toward homosexuality than older and less educated individuals (Herek 2000), and so one might expect that the previously discussed theoretical considerations vary systematically with, say, age and education. For example, younger heterosexuals may have been more likely to forgo marriage because of its discriminatory nature prior to Goodridge (which would predict a smaller increase in relative marriage take-up for younger sexual minorities compared with younger heterosexuals), and more highly educated heterosexuals may have also felt less of an identity threat in response to the court ruling (which would predict a larger increase in relative marriage take-up for highly educated sexual minorities compared with highly educated heterosexuals).

Perhaps the most salient and theoretically unambiguous demographic difference of interest in Table 3 is the presence of children in the household. Although the data do not permit me to identify whether the children came before or after the marriage, other research has shown that many people in same-sex couples are raising children from prior heterosexual relationships and different-sex marriages, so some share of children predated the marriage. Households with children from prior relationships likely have the strongest benefits to legal marriage because many of the tangible rights and responsibilities that come with marriage bear directly on the presence of children, including legal parental rights. For households where the children came after the marriage, it is plausible or even likely that the couple had expectations about having children prior to marriage, particularly because special effort is required to add children to a household headed by same-sex couples. All else equal, couples who expect to have children will have stronger incentives to marry than other couples because marriage is likely to save them substantial future costs in the form of complicated legal arrangements with respect to childrearing.¹³ There are other possible differential benefits by parental status as well: tangible benefits, such as access to health insurance, which are likely to be more valuable for households with children than for households without children; and intangible benefits, such as the value of the increased legitimacy of the relationship in the eyes of other community members (e.g., teachers and schools).

As shown in the top panel of Table 3, increases in marriage among women who are sexual minorities compared with heterosexuals were very broad-based across demographic groups. Lesbians in all groups experienced large estimated increases in marriage, although larger effects are estimated for older lesbians, more educated lesbians, White lesbians, and lesbians in households with children present (compared with the associated changes for heterosexual women). Extremely broad-based increases in marriage with respect to age and education are also estimated for men, as shown in the bottom panel. As with lesbians, marriage increased significantly for White gay men, although the samples of non-White sexual minorities are very small. Finally, although very few gay men reported children present in the household, I also estimate a much larger and significant Goodridge effect on marriage for gay men with children relative to the associated marriage effect for gay men without children. The larger increases in marriage for gay men and lesbians with children relative to those without children in the household are consistent with differential returns to marriage for same-sex couples with children present or who expect to have children in the future.

Discussion, Limitations, and Conclusion

I provide evidence from the United States on the direct effects of court-ordered legal same-sex marriage on marriage and partnership among sexual minorities by studying the case of Massachusetts before and after its state Supreme Court legalized same-sex marriage. The results indicate that the Goodridge ruling led to large, immediate, and statistically significant increases in the likelihood of marriage for lesbians, with more modest (but statistically significant) increases in marriage for bisexual women and gay men, controlling for the associated changes for heterosexuals. Notably, I found coincident and large reductions in the likelihood of being a member of an unmarried couple among lesbians and gay men in Massachusetts following the Goodridge ruling. I also show that heterosexual unions were not negatively affected up to nine years after the ruling as measured by marriage probabilities.

Although the finding that lesbians had much larger marriage take-up effects than gay men matches much prior work (e.g., Badgett and Herman 2011; Carpenter and Gates 2008), the findings that bisexual women also had significant increases in the probability of marriage is surprising given prior work indicating that bisexual women are not likely to have same-sex partners. Parker (2015) reported that among self-identified bisexual individuals in partnerships in a Pew Research Center survey, 84% had different-sex partners while only 9% had same-sex partners.¹⁴ In results not reported, I found that the relative increase in the likelihood of reporting being married among bisexual women in Massachusetts after Goodridge is driven by bisexual women in households with exactly two adult women and no adult men (i.e., households where the bisexual female respondent is likely to have a same-sex partner). More research and data on bisexual women is needed using other sources of variation to determine whether these patterns are unique to Massachusetts.

This study is subject to several limitations. For example, I do not have good information on dissolutions or relationship stability. Now that Massachusetts has had several years of experience with same-sex marriage, it is likely that same-sex divorce is also relevant, and prior research has suggested that the policy environment may be related to relationship stability for same-sex couples (Manning et al. 2016).¹⁵ Also, I do not examine other health, labor market, and social outcomes that could be affected by same-sex marriage, primarily because of data limitations: the sample sizes of sexual minorities are large enough to clearly document effects on marriage, but beyond this, statistical power is lacking to credibly identify effects on other outcomes.¹⁶

Another important limitation of this analysis is that I cannot account for the possible role of cross-state migration. That is, same-sex couples with a particularly strong demand for marriage may have moved to Massachusetts to take advantage of the new legal and social benefits afforded through Goodridge. Given the close geographic proximity of major population centers in the state to other nearby states in the Northeast, this is a legitimate concern. Migration is unlikely to play a large role at explaining the marriage effects observed here for several reasons. First, most same-sex couples who wanted to migrate to Massachusetts to obtain a marriage license would have had to engage in a costly permanent move. Massachusetts’ governor at the time, Mitt Romney, invoked a 1913 state law stating that the state could not issue marriage licenses to individuals if the marriage would not be recognized in their home state; this effectively prohibited out-of-state applicants from obtaining marriage licenses because Massachusetts was the first and only state to grant legal same sex marriage from 2004 through 2008. Moreover, Romney’s position was highly publicized and covered in major newspaper outlets; a prominent article in The New York Times quoted Romney as stating “Massachusetts should not become the Las Vegas of same-sex marriage. We do not intend to export our marriage confusion to the entire nation” (Belluck 2004a).¹⁷

The next governor of Massachusetts, Deval Patrick, repealed the 1913 law in July 2008 (Associated Press 2008), by which time gay or lesbian couples who wanted to travel out of state to get married could have also gone to California (summer 2008) or Connecticut (October 2008) to do so—and many did.¹⁸ Of course, I cannot rule out that some same-sex couples with a strong demand for marriage did establish residence in Massachusetts to take advantage of legal marriage; the Massachusetts BRFSS data contain no information on length of time in the state, prior state of residence, or state of birth. Moreover, other data sets, such as the American Community Survey, that do identify cross-state migration are problematic because they do not consistently identify a credible sample of same-sex couples over this period and because year-over-year cross-state migration rates are extremely low (Ihrke and Farber 2012).¹⁹ Even if migration is nontrivial, however, the results still indicate a very strong demand for marriage among sexual minorities.

An issue related to the migration concern is that I cannot be certain that all the people who reported being married obtained their marriage license in Massachusetts (and this is true for both heterosexuals and sexual minorities). As noted earlier, same-sex wedding ceremonies were being performed in Ontario, Canada, as early as 2001, and that province legalized same-sex marriage without citizenship or residency requirements in 2003. Civil unions for same-sex couples were also available in nearby Vermont without residency requirements as early as 2000, and same-sex couples from Massachusetts who obtained a civil union in Vermont may have also described themselves as married to the BRFSS interviewer because Vermont’s civil unions gave many of the same rights and responsibilities as full marriage.²⁰ These factors may partly explain why the pre-Goodridge marriage probability among sexual minorities was not exactly zero.

Despite these limitations, this study offers the first direct evidence using a credible, representative, population-based data set that court-ordered same-sex marriage in the United States—one of the most salient contemporary social issues of the day—significantly affected marriage and partnership for sexual minorities relative to heterosexuals. Contrary to two of the most commonly voiced arguments against same-sex marriage, I do not find consistent evidence that heterosexual marriage suffered when same-sex couples were allowed to get married, and moreover, I find that there was a robust demand for marriage among sexual minorities in the United States, confirming results presented by Gates et al. (2008) and Badgett and Herman (2011). The U.S. Supreme Court ruling extending same-sex marriage to the entire United States in Obergefell v. Hodges in 2015 may be expected to further increase marriage among sexual minorities by reducing legal uncertainty and increasing the legitimacy of same-sex relationships nationwide. This should offer social scientists fertile new opportunities to test important theories about marriage, family formation, and household specialization.

Acknowledgments

I thank Lee Badgett, Marianne Bitler, Patrick Button, Jeff Frank, Gilbert Gonzales, Helen Hawk, Bill Jesdale, Marieka Klawitter, Stewart Landers, Trevon Logan, Hai Nguyen, Liz Peters, Manisha Shah, Mark Stehr, and seminar participants at the U.S. Census Bureau, Johns Hopkins University, The Ohio State University, UC Santa Barbara, University of California Irvine, University of Illinois-Chicago, University of New Hampshire, University of Southern California, Vanderbilt, and the 2012 APPAM meetings for very useful comments and discussions. The data used in this paper are protected by a confidentiality agreement; researchers interested in the data can contact the author for information on how to obtain access. I thank Helen Hawk, Maria McKenna, and Liane Tinsley for help with the Massachusetts BRFSS data. The contents of this paper do not reflect the views of the Massachusetts Department of Health or any other organization. All errors and omissions are my own.

Data Availability

The data sets analyzed for the current study are not publicly available, but researchers can contact the author for information on how to apply for access. Permission must be obtained by the Massachusetts Department of Health.

Compliance With Ethical Standards

Ethics and Consent

The author confirms that guidelines from the Committee on Publication Ethics (COPE) were followed for this study.

Conflict of Interest

The author declares he has no conflict of interest.

Notes

References

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.