week3

ORIGINAL ARTICLE

Sexual Identification in the United States at the Intersections of Gender, Race/Ethnicity, Immigration, and Education

Tony Silva1 & Clare R. Evans2

# Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract Sexual identification is shaped by social processes that vary across multiple axes of marginalization and social position— including gender, race/ethnicity, immigration status, and education. However, to date quantitative findings on sexual identity formation have been inconsistent and most existing studies do not use intersectional frameworks. Drawing on intersectional theory and using an innovative multilevel method for measuring intersectional effects, we address this gap in our understanding of sexual identification by examining how the likelihood to adopt an exclusively heterosexual sexual identity varies along the intersecting axes of gender, race/ethnicity, immigration status, and education. We analyze data from 15,340 U.S. young adults, 24–32 years-old, who answered Wave IVof the nationally representative Add Health survey. Among strata of women, there was considerable variability in propensity to exclusively heterosexual identify across racial/ethnic and immigrant status categoriza- tions: White, Black, Native American, immigrant Asian/Pacific Islander, non-immigrant Asian/Pacific Islander, immigrant Latinx, and non-immigrant Latinx. Among strata of men, the propensity to heterosexual identify was considerably higher overall and there was less variability across racial/ethnic and immigrant status categorizations. Results suggest that across races/ ethnicities and immigration statuses men seem to be similarly affected by heteronormative expectations, whereas more compli- cated processes involving race/ethnicity and immigration status shape women’s propensity to exclusively identify as heterosex- ual. For most intersectional strata, the propensity to exclusively heterosexual identify did not differ by education level. Practitioners and researchers should be aware of how race/ethnicity/immigrant status may shape sexual identification, but in gendered ways.

Keywords Sexuality . Gender . Sexual identity . Heterosexuality . LGBT . Multilevel modeling . MAIHDA . Masculinity .

Femininity

Unlike sexual attraction which may have biological influences (Bailey et al. 2016), sexual identity is socially determined and differs across social, cultural, and historical contexts. Thus identity adoption can be analyzed sociologically. Although

today there exist general understandings of what it means to be heterosexual, lesbian/gay, or bisexual, there are nonetheless multiple factors that affect how individuals identify and ex- press their sexuality, whether they do so openly or privately (as with LGBQ people who are “closeted”). In both interview studies (Budnick 2016; Carrillo and Hoffman 2016, 2018; Silva 2017; Walker 2014) and surveys (Copen et al. 2016; Kuperberg and Walker 2018; Silva 2019; Silva and Whaley 2018), individuals report identifying as heterosexual despite also reporting same-sex attractions or behaviors. Sexual iden- tification does not simply reflect attractions and sexual prac- tices; instead, although identification may be related to both, it is best theorized as a distinct process.

In particular, heteronormativity is a dominant, institution- alized system of oppression that is more or less ubiquitous in its reach. The degree to which individuals are held to heteronormative standards, however, and the extent to which

Electronic supplementary material The online version of this article (https://doi.org/10.1007/s11199-020-01145-x) contains supplementary material, which is available to authorized users.

* Tony Silva tony.silva@northwestern.edu

Clare R. Evans cevans@uoregon.edu

1 Department of Sociology, Northwestern University, 1808 Chicago Avenue, Evanston, IL 60208-1330, USA

2 Department of Sociology, University of Oregon, Eugene, OR, USA

https://doi.org/10.1007/s11199-020-01145-x

Published online: 27 March 2020

Sex Roles (2020) 83:722–738

they are able to resist these pressures if they seek to identify as LGBQ (lesbian, gay, bisexual, or queer), is unequal across society. This is likely for three interrelated reasons: people in different intersections of social positions (a) are held to differ- ing heteronormative pressures and expectations; (b) have varying levels of awareness of LGBQ identities and different levels of internalized bias against them, thus affecting how they see themselves; and (c) have differential resources and social support for resisting heteronormativity if they want to do so.

Importantly, there is some evidence that likelihood to em- brace heterosexual or sexual minority identities varies by gen- der, race/ethnicity, and educational attainment, three key so- cial positions in American society which suggest that social inequalities and other aspects of social identity affect heteronormative expectations for sexual identification in com- plex ways. However, there are inconsistent findings in this literature, and most existing studies only analyze one or two identities as they relate to sexual identification, such as gender or race/ethnicity. Little is known about the intersectional pat- terning of sexual identification in a nationally representative sample. Processes such as social norms and experiences that impact sexual identification may operate differently in, or be entirely unique to, particular intersectional social locations. Examining sexual identity at the intersections of multiple so- cial locations enables us to capture the totality of the effects of these social processes in shaping sexual identification.

Using an innovative multilevel intersectional methodology (Evans 2015; Evans et al. 2018; Jones et al. 2016) and a large nationally representative sample from the National Longitudinal Study of Adolescent to Adult Health (Add Health), we address this substantive gap in our understanding of sexual identification by examining how the likelihood to adopt an exclusively heterosexual sexual identity varies along the intersecting axes of gender, race/ethnicity, immigration status, and education in the United States. We examine the intersections of these four axes specifically because they are among the most salient and consequential dimensions of in- equality and marginalization in the United States. Race, class, and gender have long been central to sociological analyses of social experience and inequality (Collins 1990) and to inter- sectional analysis (Choo and Ferree 2010; Cho et al. 2013; Collins 2015; Crenshaw 1991; Grzanka 2014; McCall 2005). We draw on these rich frameworks to analyze how they can shape and reflect processes related to sexual identification. Relatedly, although immigration status has long been an im- portant dimension of social experience in the United States, most analyses of sexual identification do not consider it.

We build upon the theoretical work of intersectionality studies, which Black women feminists developed to address how Black women experience multiple interlocking forms of disadvantage (e.g., Crenshaw 1991). We use that theoretical framework here to reflect how individuals’ experiences at the

micro level are deeply shaped by interlocking systems of in- equality at the macro level. These forces include racism, sex- ism, heterosexism, and xenophobia. Structural inequalities re- lated to race, gender, immigration status, and socioeconomic class, and in particular their intersections, we posit, shape the sexual identities individuals adopt.

Our goals are to fill a gap in the literature about sexual identification at the intersections of gender, race/ethnicity, im- migration status, and education, as well as to help clarify in- consistent findings from prior studies, particularly those relat- ed to race and sexual identification. We also hope to add additional nuance by analyzing immigrant status as a distinct social axis and by examining Native American and Asian/ Pacific Islander respondents, who due to small sample sizes have been mostly unexamined in prior research. By doing so, we contribute to a sociological understanding of how intersecting social locations shape how individuals understand and label themselves.

Unlike most studies about sexual identification, we analyze heterosexuality rather than sexual minority identities. The rea- sons for this focus are twofold. First, little prior research has analyzed heterosexual identification. Heterosexuality is gen- erally treated as a “default” or “normal” identity, when in fact individuals consciously adopt it just like any other identity. By focusing on heterosexuality, we help to correct the mispercep- tion that it is only LGBQ people who consciously adopt iden- tity labels. Second, our focus on heterosexual identification mirrors our interest in how heteronormative expectations op- erate differently across strata and how individuals at different intersecting social locations defined by race, class, and gender are able to resist the pressures of heteronormativity more ef- fectively than others.

Social Forces Shaping Identities

Identity adoption occurs within contexts and social processes that shape how individuals understand themselves. All types of social identities are affected by social processes to at least some extent, including gender and race, even as physical char- acteristics shape how people are perceived and the identities others interpret as appropriate for them. Racial identification is partially dependent upon phenotypic characteristics and fam- ily history, but which characteristics are given social impor- tance and how these give rise to particular identities differ based on social context, time period, and even individuals’ social status (Alba et al. 2016; Kramer et al. 2016; Omi and Winant 2015; Penner and Saperstein 2013; Saperstein and Penner 2012).

Sexual identification, like race, is sometimes assumed to stem from biological origins, yet this is only true if identifica- tion exactly mirrors sexual attractions, which it often does not. Social processes also shape sexual identification. In the 2013–

723Sex Roles (2020) 83:722–738

2015 National Survey of Family Growth (NSFG), for in- stance, conservative beliefs about gender and sexuality are associated with higher odds of heterosexual identification among people with two or more lifetime same-sex partners and/or substantial same-sex attraction (Silva 2019). Similarly, increased conservatism and religiosity over survey waves in Add Health are associated with changing to a het- erosexual identity from a sexual minority identity (Silva 2019). Both findings suggest that conservative and religious beliefs shape sexual identification. Patterns associated with heterosexual identification reflect social forces that affect how individuals understand their sexuality. Yet how other so- cial identities shapes sexual identification is less clear.

Identities are shaped by social processes and structures of inequality, and these identities in turn shape how individuals adopt and express other identities. A key question, then, asks in what ways identification differs by some of the most con- sequential social axes in the United States: race/ethnicity, im- migration status, gender, and social class. To the extent that race/ethnicity, immigration, gender, and education shape lived experiences, it is reasonable to assume that they may also shape the propensity to identify sexually.

We theorize that heteronormativity affects people different- ly depending on their social position along intersections of race/ethnicity, immigration status, gender, and education. Individuals at the intersections of these social locations will experience heteronormativity differently, be exposed to differ- ent beliefs about LGBQ people, and have different opportuni- ties to resist heteronormativity and embrace an LGBQ identi- ty. How exactly that happens reflects a series of complex so- cial processes. Whereas differences related to gender may op- erate according to a status hierarchy (Mize and Manago 2018), there is little research on whether similar processes operate with regard to race/ethnicity, immigration status, or education, which our paper helps to address. Relatedly, how social pro- cesses related to each of these axes combine to affect group probabilities of sexual identification has not yet been exam- ined, and which we explore in the present paper.

In the following sections we review literature on gender, race, immigrant status, and education separately, not because their effects are singular or additive, but because much of the literature on sexual identification has examined these axes separately. For instance, most quantitative sociological work uses conventional regression models that analyze race, gender, and education as separate variables without examining their interactions. After discussing these axes separately, we detail the intersectional framework that can synthesize these findings.

Gendered Sexual Identification

Gendered social forces affect men and women such that they experience sexuality quite differently, including as it relates to

sexual identification. Nationally representative surveys con- sistently show that a higher proportion of women than men identify as something other than exclusively heterosexual (England et al. 2016; Gates 2014; Savin-Williams and Vrangalova 2013). One reason for this difference is that wom- en have a lower social status than men and are seen as less socially valuable, so they have more leeway to engage in practices that challenge gender norms, including engaging in same-sex behavior and identifying as something other than exclusively heterosexual (England 2016). Further, because women are already lower in social hierarchies than men, they have less social status to lose for LGBQ identification than men, thus encouraging a greater proportion of men than wom- en to identify as heterosexual (England 2016). This also helps explain why experiments show that people are much likelier to believe that men are bisexual or gay because of a single same-sex encounter than women (Mize and Manago 2018).

Another related reason a higher proportion of women than men identify as something other than exclusively heterosexual is because heterosexuality and homophobia are more central to masculinity than to femininity. In other words, more men than women perceive same-sex sexuality as a threat to how they perceive themselves in terms of gender. Nationally, men report greater homophobic attitudes on average than women (England 2016; Greenberg et al. 2019; Silva 2019). Experiments also show that men are likelier to report homo- phobic attitudes when they feel their masculinity is threatened (O’Connor et al. 2017; Weaver and Vescio 2015), a dynamic that rarely occurs with women when they feel their femininity is threatened (Falomir-Pichastor and Mugny 2009; Nagoshi et al. 2019; Willer et al. 2013). Further, many men use hetero- sexual identification and sexualized power over women to bolster their sense of themselves as masculine (Pascoe 2011). In part because heterosexuality and homophobia are more central to masculinity than femininity, men are likelier to identify as exclusively heterosexual than women. This could be in part because heterosexual identification, for men, is a “manhood act” (Schrock and Schwalbe 2009, p. 279) that men use to construct normative masculinity and secure social advantages that men have relative to women (and heterosex- ual people relative to LGBQ people).

Race/Ethnicity, Immigration, and Sexual Identification

Research suggests that there are racial differences in LGBQ identification, although this is not uniform across surveys. Gates (2014) examined four nationally representative surveys—Gallup (2014), the NSFG (2006–2010), the General Social Survey (GSS, 2008–2012), and the National Health Interview Survey (NHIS, 2013)—and found that al- though most did show some differences in LGBT

724 Sex Roles (2020) 83:722–738

identification across racial/ethnic identities, most differences were not significant. Only in Gallup did significant differences emerge, with Black, Latinx, Asian, and multiracial respon- dents likelier to identify as LGBT than White people. These racial/ethnic differences in LGBT identification appear to have grown in Gallop’s 2016 survey (Gates 2017). Similarly, Mustanski et al. (2014) analyzed the 2005 and 2007 Youth Risk Behavior Surveillance System and found that Black and Latinx adolescents were likelier to identify as lesbian/gay than White people and that Asian respondents and those of “other” races were likelier to identify as unsure than White people. Among Millennials, Latinx people appear to have a greater likelihood of LGBT identification than other racial/ethnic groups, according to the representative 2018 GenForward sur- vey (Cohen et al. 2018). The 2018 GSS survey shows that Black women are likelier to identify as bisexual than women of all other races (Compton and Bridges 2019). No known research using a nationally representative, generalizable sam- ple has examined Native sexual identification. Overall, some (but not all) research suggests that People of Color are likelier to identify as a sexual minority than White people, although which racial/ethnic identity has significant differences is not consistent across surveys. Why People of Color may be like- lier than White people to identify as LGBQ is unclear, al- though it could echo why women are likelier to identify as LGBQ than men: People of Color are lower in racial/ethnic hierarchies than White people, so they have less social status to lose when identifying as LGBQ.

Of course, racial/ethnic categories like “White,” “Black,” “Asian,” “Latinx,” and “Native American” are broad umbrella terms with extensive diversity within them, but most nation- ally representative surveys—including Add Health—either do not gather further information about respondents’ racial/ethnic identities or do not have a large enough sample to provide more nuanced analyses. Regardless, analyzing the five main racial/ethnic categories that do exist on these surveys is bene- ficial because these are among the key racial/ethnic categories that Americans use to understand race. They also capture in- equality on measures of income, education, and self-rated health better than many alternatives (Howell and Emerson 2017). Thus, despite their imperfections, these categories are reasonable measures for racialized social processes. They do not necessarily reify racial categories so long as scholars use them provisionally to better understand how systems of power reproduce inequality (Cho et al. 2013), which is one of our goals.

Almost no research exists on sexual identification among immigrants in the United States, so this literature is highly underdeveloped. The 2002–2013 NSFG shows that whereas Latina immigrants are less likely to identify as bisexual than U.S.-born Latina, Black, and White women, the rates of bi- sexual identification are about the same among all categories of men (England et al. 2016). Being an immigrant seems to

make LGBQ identification less likely among women, al- though little research has explored why. Research using na- tionally representative samples from Europe shows that immi- grants are more homophobic on average than their native counterparts and that this difference declines over time as immigrants integrate in their new country, over new genera- tions as immigrants’ children are socialized within the culture of the host country, and with each new cohort (Van der Bracht and Van de Putte 2014). This change is slower among immi- grants and their children who are poor, very religious, and/or use their native language (Soehl 2017) or who are Muslim (Röder 2015). Homophobic attitudes and changes to them also differ depending on the distribution of homophobic attitudes in immigrants’ country of origin (Soehl 2017). It is likely that homophobic beliefs shape sexual identification: Higher levels of homophobia and greater emphasis on conventional family formation among immigrants likely encourages heterosexual identification, but less so with successive generations and over time. How immigration status intersects with other social lo- cations, however, is unclear, because little research on the topic exists.

One of the best (and only) examples of research that ana- lyzes how intersections of race/ethnicity, gender, and immi- gration status predict sexual identification is England et al.’s (2016) study using the 2002–2013 NSFG to examine bisexual identification and behavior over different cohorts of White, Black, and immigrant and U.S.-born Latinx men and women. They found that whereas bisexual identification has risen over successive cohorts for White, Black, and U.S.-born Latina women, it for the most part has not for any cohort of men. They also find that although men have similarly low proba- bilities of bisexual identification, immigrant Latina women have lower probabilities than all other U.S.-born women. We build on this study by (a) analyzing respondents’ education as a distinct social axis; (b) extending analysis to Native American and immigrant and non-immigrant Asian and Pacific Islander people, three groups underrepresented in sex- uality research; (c) using a survey that provided more nuanced identity options than the NSFG (including “mostly heterosex- ual”); (d) analyzing heterosexual identification, which is rare- ly analyzed in sexuality studies; and (e) using innovative in- tersectional multilevel models that provide more reliable and interpretable results than single-level models (Bell et al. 2019; Evans 2019; Evans et al. 2018).

Education and Sexual Identification

Like the relationship between race/ethnicity and sexual iden- tification, the relationship between educational attainment and sexual identification differs across surveys. Differences in identification across education categories may emerge through a combination of selection effects (i.e., who attends

725Sex Roles (2020) 83:722–738

college in the first place) and causal effects of education on self-perceptions. The 2008–2012 GSS and 2013 NHIS show that among people age 25 years or older LGB individuals are likelier than heterosexual people to have a bachelor’s degree or more, but the 2006–2010 NSFG and the 2014 Gallop did not find this pattern—and even the differences in the GSS and NHIS were not statistically significant (Gates 2014). In con- trast, the 2012 American National Election Survey Time Series Study shows that LGBQ respondents have significantly higher levels of education (Grollman 2017). Using Add Health, Mollborn and Everett (2015) also found significant differences: Sexual minority women have lower rates of edu- cational attainment than exclusively heterosexual women. In contrast, they found that some sexual minority men—those who identify as mostly heterosexual, mostly gay, or exclusive- ly gay—have higher rates than exclusively heterosexual men. Disadvantaged family backgrounds and schooling contexts helped explain some but not all of the disparities for women, suggesting that disadvantage on the basis of gender, sexual identity, and other forms of marginalization compound to af- fect educational attainment. For men, on the other hand, prob- lematic aspects of heterosexual masculinity may be partially responsible: Many male adolescents use academic under- achievement as a way to build masculinity (Hsin 2018; Morris 2008). Additionally, persistent wage gaps mean that men without a college degree will earn much more on average than women without a college degree (Gould and Kroeger 2017; Carnevale et al. 2018). This means that heterosexual men have fewer economic incentives than women (of any identity) to pursue higher education due to earning advantages related to their gender and sexuality.

There are reasons other than selection effects that could explain why educational attainment may be related to sexual identification. Higher education may help facilitate critical thinking skills that allow some individuals to reject the asso- ciations between attractions or sexual behaviors and sexual identification. Indeed, Silva and Whaley (2018) found that among men with same-sex sexuality in the 2011–2013 NSFG, educational attainment of a bachelor’s degree or above was associated with heightened odds of heterosexual identifi- cation, after controlling for other factors. Although their study was conducted in a very specific population—men with same- sex behaviors and/or attractions—the mechanisms at play there may occur in other populations as well. Further, some individuals with class advantage may not wish to threaten their status by adopting a marginalized identity. Indeed, sexual minority women and men usually face more economic disad- vantages than their heterosexual counterparts (Conron et al. 2018). On the other hand, educational institutions expose in- dividuals to new groups, networks, and ideas, each of which may influence some individuals to adopt a non-exclusively heterosexual identity. This could be because some individuals reinterpret their identity or because some individuals feel

newly confident to claim an identity they have always wanted (e.g., those who were previously “closeted”). Regardless of the causal link between education and sexual identity, and whether this facilitates heterosexual or sexual minority iden- tification, little research has explored how education—in com- bination with race/ethnicity, immigration, and gender— combine to shape the likelihood of heterosexual identification. Further, the inconsistent research findings indicate that more research on education and sexual identification is necessary.

Intersectionality and Heterosexual Identification

Intersectionality is a theoretical framework originating in Black feminist scholarship (e.g., Crenshaw 1991, and Collins 1990) that recognizes how social locations such as race/ethnicity, class, gender, and sexuality affect lived experi- ences by interacting with one another within systems of in- equality that advantage some social locations over others. The effects of these social locations are not additive but rather are dynamic and complex. Thus, although gender differences in sexual identification may be in part due to status differentials, differences in identification along the lines of race/ethnicity, immigration status, and education may be due to different mechanisms—and in any case, intersectional frameworks re- quire that these mechanisms be analyzed at the intersections of social positions rather than at just one (e.g., gender). Intersectional frameworks analyze and critique power rela- tions and structural inequalities (Cho et al. 2013; Choo and Ferree 2010; Collins 2015; Grzanka 2014; McCall 2005). Macro systems of inequality affect lived experiences, includ- ing micro meaning-making processes and interpersonal inter- actions (Choo and Ferree 2010). Consequently, intersections of social locations will affect both how individuals experience identities and their propensity to adopt certain identities over others.

Applied to identification, the framework of intersectionality allows us to analyze how, for instance, het- erosexual men of different races may differently experience their heterosexual identity and how different racial/ethnic groups may have different probabilities of identifying as het- erosexual. Qualitative research is particularly valuable for the first aim, and quantitative research can help to address the second. In line with McCall’s (2005, p. 1773) “intercategorical complexity” approach, we use existing cate- gories of social locations—while socially constructed—to un- derstand how systems of inequality affect individuals’ likeli- hood to identify as exclusively heterosexual rather than some- thing else. We cannot quantitatively examine how individuals experience their identity, but we can quantitatively analyze how social forces shape group probabilities of adopting a par- ticular identity.

726 Sex Roles (2020) 83:722–738

The most common approach to modeling intercategorical (McCall 2005) intersectional inequalities relies on fitting fixed effects regression models that include all “additive” effects (e.g., female gender, Latinx ethnicity) and all possible interac- tions (e.g., female and Latinx). This approach has a number of methodological limitations as more axes of inequality and social process are added to the model, namely: (a) the number of parameters required increases geometrically, creating prob- lems related to model parsimony; (b) the small sample size in many social strata means that stratum-level estimates may become unreliable; and (c) interpreting results becomes un- wieldy and complex (Evans 2019). Recent advances in model- ing intersectional inequalities using multilevel models have helped to address these issues (Evans 2015; Evans 2019; Evans et al. 2018; Jones et al. 2016).

Although a variety of projects has explored the relation- ships between one or more social locations and sexual identi- fication, none is known to have examined all of these social locations and processes (race, immigration status, gender, ed- ucation) as they intersect with one another and none have applied the novel multilevel approach to intersectional analy- ses of sexual identification. In the present study we apply cutting-edge multilevel analytic techniques in order to deter- mine how predicted probabilities of heterosexual identifica- tion differ across social strata defined by the intersections of race/ethnicity, immigration status, gender, and educational at- tainment. By doing so, we hope to address a key sociological question: How do intersecting social locations and processes, reflecting systemic inequalities, shape how individuals under- stand and label themselves?

Method

Data

We drew all data from Add Health, a longitudinal study that follows a representative sample of individuals who were in grades 7–12 in the 1994–1995 school year. Researchers first selected a stratified random sample of high schools in the United States that had an 11th grade and 30 or more students enrolled. Schools were selected with a probability proportion- al to their enrollment size and were stratified based on school size, racial/ethnic composition, school type (private, religious, public), metropolitan context (rural, suburban, urban), and region. Researchers also sampled middle schools with a 7th grade that sent students to one of the sampled high schools after graduation, with probability proportional to how many students those middle schools sent to the sampled high schools. Over 70% of sampled schools participated, and re- fusals were replaced with a similar school based on strata characteristics. Overall, 132 schools in 80 communities were sampled.

Of the 90,118 students sampled for the in-school portion of the survey in Wave I, a subsample known as the “core sample” were selected to be surveyed in subsequent waves (n = 20,745). Sexual identification—the dependent variable in our analyses—was measured only in Waves III and IV, in 2001–2002 and 2008, respectively. In total 15,701 respon- dents participated in Wave IV. Of these, we dropped 361 for missing data, including 69 for missing the dependent variable and 292 for missing data on gender, race/ethnicity, immigra- tion status, educational attainment, or age. We only examine Wave IV responses because the number of respondents who changed identities between waves was too small to analyze intersectionally. In Wave IV respondents were between the ages of 24 and 32 years-old.

To ensure respondent confidentiality, we followed all pro- cedures mandated by the Carolina Population Center, which oversees the distribution and protection of Add Health data. Our study and all data protection procedures were approved by the University of Oregon Institutional Review Board.

Exclusively Heterosexual Identification

The dependent variable is whether the respondent identified as exclusively heterosexual in Wave IV. This was measured by the item: “Please choose the description that best fits how you think about yourself.” Unlike most nationally representative surveys, identity options were presented on a continuum. Options included “100% heterosexual (straight)”; “mostly heterosexual (straight), but somewhat attracted to people of your own sex”; “bisexual, that is, attracted to men and women equally”; “mostly homosexual (gay), but somewhat attracted to people of the opposite sex”; “100% homosexual (gay)”; “not sexually attracted to either males or females”; and “don’t know.” We dichotomized responses into 100% heterosexual (hereafter: heterosexual) versus anything else, but coded “don’t know” as missing. We examined sexual identity dichot- omously to reflect how identification as anything other than exclusively heterosexual is considered non-normative in the United States.

Although our measure of sexual identity combines attrac- tion and identity, many respondents in Add Health identified in ways not simply based on attractions or sexual practices. For instance, many respondents who reported same-sex attrac- tions or practices in Wave IV nonetheless chose an exclusively heterosexual identity, just as many others chose a “mostly heterosexual” identity despite not reporting same-sex attrac- tions or practices in Wave IV. “Discordance” between identity and attractions and/or behavior is common in Add Health (Caplan 2017; Lourie and Needham 2017), even despite ques- tion wording, much like it is in other surveys. In the present paper, we do not examine discordance, which is an entirely separate topic, largely because the number of respondents who report identity-behavior discordance is too small to analyze

727Sex Roles (2020) 83:722–738

intersectionally. Further, although we examine identity specif- ically, the social processes that shape sexual identification also likely similarly affect engaging in same-sex behavior and reporting same-sex attraction. Any differences in rates of same-sex behavior and attraction are not indicators of biolog- ical sexual differences between intersections of social loca- tions, but rather are themselves the result of racialized, gen- dered, classed, and immigration-related social processes that are also related to sexual identification, which is the outcome we examine here.

Although the sexual identity question appears continuous, the concept of sexual identification is not. There are three main reasons for this. First, a continuous measure of sexual identity does not capture different levels of normativity con- nected to different identities. For instance, bisexuality is deep- ly stigmatized (Dodge et al. 2016). Due to stigmatization, bisexual people face higher rates of many physical and mental health issues than gay, lesbian, and heterosexual people (Dyar et al. 2019; Newlin Lew et al. 2018; Paschen-Wolff et al. 2019; Ross et al. 2018; Taggart et al. 2019). Whereas being open about one’s sexual identity leads to better mental health and less substance abuse among gays and lesbians, for bisex- uals being more open leads to worsened health outcomes (Feinstein et al. 2014). What these findings show is that bi- sexuality is not necessarily more normative or less stigmatized compared to gayness/lesbianism and is thus not a midpoint in a continuum of normativity.

Second, as research with nationally representative surveys has shown, there are multiple dimensions to how individuals experi- ence their sexuality, including romantic attraction, physical at- traction, and sexual behavior (Mishel 2019; Priebe and Svedin 2013). Although often overlapping, each is analytically distinct and for many people do not perfectly align. Because of numerous distinct aspects of sexual orientation that do not necessarily align, a continuous measure of sexual identification is not equivalent to a continuous measure of sexual orientation.

Third, there are also many criteria people use to identify themselves that do not support the treatment of sexual identi- fication as continuous. For instance, some people identify themselves based on the gender of their current partners rather than on their entire sexual history or their sexual attractions (Rust 1992). There are many subjective elements of self- identification that are not well represented by a continuous construct.

For all those reasons, a seemingly continuous measure of sexual identity does not reflect a continuous measure of sexual identity in terms of normativity, sexual orientation, or reasons for sexual identification. Categories of sexual identification represent meaningful differences that are not equivalent to a continuous underlying measure, and as a result we dichoto- mize this variable in models according to exclusively hetero- sexual identification, which is not stigmatized, and all other identities which face varying degrees of stigmatization.

Independent Variables and Social Strata Categorizations

Our primary analysis is a “three-axes” model (Model 1) ad- dressing the intersections of gender, race/ethnicity and immi- gration status. Gender was coded dichotomously: (1) man and (2) woman. Although not comprehensive or reflective of gen- der diversity (see Magliozzi et al. 2016 for better survey options) or biologically intersex individuals (Davis 2015), this was the only coding possible in Add Health data.

Immigration status was measured in Add Health in Wave IV using the measure “Were you born a U.S. citizen?” Those who responded “yes” were coded as non-immigrant and those who responded “no” were coded as immigrant. We did not examine different countries of origin because there were too few respondents from any given country to make that feasible in analyses. Race/ethnicity was measured in Wave I, and when respondents indicated more than one racial/ethnic identifica- tion, a Census Bureau algorithm was used to code a single categorization. Finally, due to data limitations it was not pos- sible to differentiate by immigration status for White, Black, and Native American respondents. Therefore, we combined race/ethnicity and immigration status into a single variable, and we did not differentiate between immigrant and non- immigrant Black, White, and Native American respondents. Options included seven racial/ethnic/immigration categories: immigrant Latinx, all races; non-immigrant Latinx, all races; Black, any immigration status; immigrant Asian/Pacific Islander (PI); non-immigrant Asian/PI; Native American, any immigration status; and White, any immigration status. In the primary three-axes model of gender by race/ethnicity and immigration status there are therefore 14 possible inter- sectional social strata. Each stratum had over 100 respondents.

We then progress to a “four-axes” model (Model 2) which intersects educational levels with gender, race/ethnicity, and immigration status. Education was coded as: high school or less (coded 1), some college or vocational training (2), and college degree or more (3). We use education as a proxy for class. These three education levels were intersected with the 14 gender by race/ethnicity and immigration status strata for a total of 42 social strata. Of these 42 strata, 40 had 20 or more respondents and the remaining two (i.e., immigrant Asian/PI men and women with a high school degree or less) had be- tween 15 and 19, indicating a reasonable sample size per stra- tum for estimation in a multilevel model.

For the purposes of multilevel modeling, social strata are treated as level 2 units, necessitating the creation of unique ID numbers for each possible intersectional stratum. Three-digit ID codes were created corresponding to the three dimensions evaluated: gender (digit 1), race/ethnicity and immigration status (digit 2), and education (digit 3). Numerical values in these digits correspond to the coding for these variables (see previous). Where education was not included in the analysis,

728 Sex Roles (2020) 83:722–738

as in the three-axes models, a “0” was used in the third digit. For instance, in the three-axes model the social stratum code 260 refers to women (2) who are Native American (6) with education level not evaluated (0). In the four-axes model, 171 refers to men (1) who are White (7) and have a high school degree or less (1).

Control Variable: Age

Age is related to sexual identification, but the age range of respondents in the sample is insufficient to examine age sep- arately as an axis intersecting with the others. Respondents’ age in Wave IV centered to the median (29 years) is therefore included as a control. We did not include other controls be- cause the purpose of our paper is to obtain reliable stratum- level estimates of exclusively heterosexual identification, which quantify the totality of the effects of social processes that shape sexual identification. Pinpointing more precisely why these differences exist is a direction for future research.

Analysis Plan

We apply the innovative multilevel approach to modeling inter-categorical intersectional inequalities now known as multilevel analysis of individual heterogeneity and discrimi- natory accuracy (MAIHDA) (Evans 2015; Evans et al. 2018; Jones et al. 2016; Merlo 2018). Multilevel models are typical- ly used to address statistical dependency in respondents’ out- comes due to them sharing physical contexts, such as neigh- borhoods. Nesting individuals within intersectional strata is similar, except that the shared context is a social location within intersecting social processes and structures rather than a literal physical context. In our models, individuals (level 1) are clustered hierarchically within intersectional social strata (level 2).

In a linear version of a MAIHDA model where all additive main effects (e.g., woman, Native American) are included but no interaction terms between these axes of marginalization are included, the stratum-level residual estimates are interpretable as the remaining “interaction” effect that is unexplained by additive effects (assuming an absence of omitted variable bi- as). Logistic intersectional multilevel models can also be fit and used to estimate the total predicted probability of the outcome in each stratum after adjustment for any covariate controls, such as age. Our analytical interest is in the total predicted values of propensity to exclusively heterosexual identify in each stratum. We therefore fit MAIHDA models and recombine the additive fixed effects with stratum-specific residual interaction effects in order to arrive at the total pre- dicted value.

We use MAIHDA to estimate predicted probabilities rather than simply calculating “raw” probabilities in each stratum or using conventional regression approaches because MAIHDA

provides estimates that are more reliable when the size of some strata is small (Bell et al. 2019; Evans 2015; Evans et al. 2018), which frequently occurs under conditions of high-dimensional interactions such as the present model. Using conventional approaches, estimates for small strata would be highly unreliable. Although this problem is not en- tirely solved in MAIHDA (Bell et al. 2019), it is substantially reduced because multilevel models provide precision- weighted estimates, thus automatically adjusting estimates based on sample size.

We begin with the three-axes version of the intersectional analysis (intersecting gender by race/ethnicity and immigra- tion status) and proceed to the four-axes version which incor- porates education as well. We present two versions of each model: a null model which included no covariates and a model with all additive effects that controlled for age. The equation for the four-axes model is:

straightij∼binary πij � �

log πij

1−πij � � ¼ β0 þ β1 Womanj

� � þ β2 Imm Latinxj

� �

þ β3 NonImm Latinxj � �

þ β4 Black j � �

þ β5 Imm Asian=PI j � �

þ β6 NonImm Asian=PI j � �

þ β7 Native Americanj � �

þ β8 HighSchool or Lessj � �

þ β9 Some Collegej � �

þ β10 Ageij � �

þ μ0 j

μ0 j∼ 0; σ 2 μ0

h i

Var straightijjπij � �

¼ πij 1−πij � �

The equation for the three-axes model was identical, minus the education variables and strata specifications.

For each model we calculated the variance partition coef- ficient (VPC), which is the percent of the total variation in the dependent variable that is attributable to the stratum-level. In the null model, the VPC represents the total variability be- tween-strata, whereas in the additive model it represents the between-stratum variance that remains unexplained after ad- justment for the additive effects and age. The VPC is calcu- lated by dividing the variance at the stratum-level by the total variance, which is the sum of variances at the stratum-level (σ2μ0 ) and individual level (σ

2 e0 ). In logistic models, the

individual-level variance can be estimated using the latent variable approach (Goldstein et al. 2002) in which σ2e0 ¼ π2/ 3 = 3.29. The full equation is:

VPC ¼ Between Stratum Variance Total Variance

� 100% ¼ σ2μ0

σ2μ0 þ σ2e0 � 100%

Because Add Health used school-based cluster-sampling, we evaluated the robustness of our results by adjusting for

729Sex Roles (2020) 83:722–738

residual school-based clustering. This was accomplished by using a cross-classified multilevel model (CCMM) of respon- dents (level 1) simultaneously nested in intersectional social strata (level 2) and schools (level 2) as outlined by Evans (2019). Our results were largely robust to adjustment for cross-classification by school context, so we focus our discus- sion on the more interpretable hierarchical model. (CCMM results are available in the online supplement to this article.) The only substantive difference is that the level 1 coefficient for immigrant Latinx, compared to White, lost significance at the .05 level in the four-axes CCMM analyses.

Data cleaning was completed in SAS 9.4 and models were run in MLwiN 2.32 (Rasbash et al. 2015) using Bayesian Markov Chain Monte Carlo (MCMC) techniques (Browne 2015) from Stata 14.2 (using the “runmlwin” command). The regression models were first fit using restrictive iterative generalized least squares (RIGLS) estimation for second- order predictive (or penalized) quasi-likelihood estimates (PQL2) to provide the Bayesian MCMC procedure with initial values, much as Axelsson Fisk et al. (2018) outline. Burn-in of 5000 iterations were used in all analyses.

Results

We first detail results for Model 1 (the three-axes model), which interacts gender and race/ethnicity/immigration status, and then we address Model 2 (the four-axes model), which adds interaction with education (model results presented in Table 1). Thus, we first address intersections of race, immi- gration status, and gender before adding nuance with education.

In Model 1 there is considerable between-stratum variability in the propensity to heterosexual identify. This is reflected both in the stratum-level VPC of 14.72% in the null model (see Table 1a) and in the predicted probabilities of exclusively heterosexual identification by gender, race/ethnicity, and immigration status that were derived from Model 1 (see Fig. 1 and Table 2). Predicted probabilities (PP) of exclusively heterosexual identifi- cation range from 70.36% for Native American women to 95.94% for immigrant Asian/PI men—a substantial range. Table 2 includes a ranking of strata according to predicted prob- ability, with 1 representing the stratum with the lowest and 14 representing the stratum with the highest PP.

Importantly, the between-stratum differences in propensity to heterosexual identify are socially patterned. In purely additive terms (see parameters in Table 1a), strata composed of women are substantially less likely to heterosexual identify than strata composed of men (OR = .30, p < .001). Put differently, men’s strata (1/.3 = 3.33) have substantially higher odds of heterosexual identification than women’s. Meanwhile the racial/ethnic/immi- gration status axes appear less relevant, with only one statistically significant difference: Immigrant Asian/PI strata are more likely

than White strata to heterosexual identify (OR = 1.98, p = .039). However, these additive dimensions of the model obscure a key finding: Between-stratum differences are greater among strata of women than among strata of men (see Fig. 1), indicating an important interaction between these axes. As is clear from this visualization, racial/ethnic/immigration differences in heterosex- ual identification are most evident among women. Findings from prior studies on the relationship between race/ethnicity and sex- ual identification have been inconsistent. These analyses reveal that this is likely because gender moderates the relationship be- tween race/ethnicity/immigration and sexual identification.

Among women, the highest and lowest predicted probabil- ities of exclusively heterosexual identification include Native American women (70.36%) and immigrant Asian/PI women (88.02%), a difference of 17.66% (see Table 2). Among men, the highest and lowest predicted probabilities include non- immigrant Latino men (91.67%) and immigrant Asian/PI men (95.94%), a difference of only 4.27%. This was the only significant difference among men’s strata. Thus, racial/ethnic/ immigration differences are substantial, but mostly among women. These results suggest that men of all racial/ethnic/ immigration statuses seem to be similarly affected by expec- tations of masculinity, whereas more complicated processes involving racial/ethnic/immigration status shape women’s propensity to heterosexual identify. Thus, the centrality of heterosexuality to femininity differs somewhat by intersec- tions of race/ethnicity and immigration status.

Interestingly, the role of immigration status also appeared to be meaningful, although the magnitude of this effect dif- fered again by gender and race/ethnicity. Immigration status increased the propensity to heterosexual identify, with immi- grant Latinas and immigrant Asian/PI women being more likely to heterosexual identify than their non-immigrant coun- terparts. The 95% CI of these PP estimates overlapped among men whereas they did not overlap among women, indicating that the impact of immigration status was more important among women than men. Among Asian/PI women immigra- tion status appeared to have a particularly large effect, with 78.9% of non-immigrants and 88.0% of immigrants reporting exclusively heterosexual identification. Thus, immigrant sta- tus substantially shapes the likelihood of heterosexual identi- fication among Asian and Latina women, a finding that rein- forces the importance of examining immigrant status in socio- logical research.

When education is intersected with the other axes, the gen- eral patterns described previously hold. Model 2 results are presented in Table 1b (regression results) and Table 3 (PP by stratum). The results from this analysis are also intriguing and indicative of an intersectional story. For instance, the additive education parameters (Model 2, Table 1b) are mostly not sta- tistically significant, indicating that for most gender by race/ethnicity/immigration intersections the propensity to het- erosexual identify does not differ by education level.

730 Sex Roles (2020) 83:722–738

However, education does appear to be important among White men and White women, although the associations are in opposite directions. Among White men, education is related to heterosexual identification in a graded fashion, with each additional step toward higher education being associated with lower propensity to heterosexual identify (high school or less: PP = 94.8%, some college: PP = 93.3%, college plus: PP = 91.0%). Notably the 95% credible interval (CI: these are sim- ilar to confidence intervals, but are distinct and reflect differ- ences between Bayesian and frequentist modeling) of the PP estimate for “some college” overlap minutely with the CI for the higher and lower education levels; however, estimates for the highest and lowest education levels do differ from each other (with non-overlapping CIs). Among White women the opposite story emerges. White women with higher education (college degrees or more) are more likely to exclusively het- erosexual identify (high school or less: PP = 77.4%, some

college: PP =77.6%, college plus: PP 82.1%), and estimates for the highest and lowest education levels significantly differ from one another.

Although the findings for education may partially reflect se- lection effects (Mollborn and Everett 2015), it is notable that the strongest and statistically significant patterns are among White people. It appears that White masculinity and White femininity are associated with the likelihood to adopt an exclusively hetero- sexual identity at different levels of education, although in oppo- site directions.

Discussion

The present study is the first known to apply an innovative multilevel intersectional framework in order to estimate the probability of exclusive heterosexual identification across

Table 1 Nested models showing the association between social locations and exclusively heterosexual identification using multilevel modeling

(a) Model 1 (3-Axes) (b) Model 2 (4-Axes)

Version A

OR (SD)

p

Version B

OR (SD)

p

Version A

OR (SD)

p

Version B

OR (SD)

p

Fixed Effects

Age 1.073 (.015) p = .001 1.072 (.014) p = .001

Gender (Woman) .302 (.058) p = .001 .315 (.033) p = .001

Race/Ethnicity/Immigration

Immigrant Latinx 1.461 (.479) p = .142 1.471 (.331) p = .046

Non-immigrant Latinx .971 (.300) p = .373 .936 (.151) p = .318

Black 1.338 (.455) p = .152 1.288 (.194) p = .057

Immigrant Asian/PI 1.984 (.719) p = .039 1.886 (.473) p = .008

Non-immigrant Asian/PI .981 (.304) p = .408 .927 (.175) p = .305

Native American .821 (.362) p = .183 .746 (.161) p = .067

White – –

Education

High school or less .978 (.138) p = .392

Some college .974 (.134) p = .375

Bachelor’s or above –

Random Effects Est [CI]

(VPC %)

Est [CI]

(VPC %)

Est [CI]

(VPC %)

Est [CI] (VPC %)

Strata-Level .591

[.237, 1.421] (14.72%)

.098

[.002, .441]

(2.73%)

.495

[.268, .855]

(12.96%)

.055

[.015, .128] (1.61%)

Fit Statistics

DIC 11,863.02 11,841.75 11,843.15 11,813.26

OR = odds ratio. Estimates are not weighted because there is no established method for weighting data using these modeling strategies. CI = 95% Credible Interval. Credible intervals are similar in interpretation to confidence intervals, but are distinct and reflect differences between Bayesian and frequentist modeling. VPC = Variance Partition Coefficient. We used the latent variable approach Goldstein et al. (2002) described for logit multi-level models. The VPC is calculated by dividing the variance at a particular level-2 context by the total variance, which includes all level-2 variance and the individual-level variance (which is π2 /3 = 3.29). BIC stands for Bayesian Deviance Information Criterion. This is an indicator of model fit for multilevel models, with lower values demonstrating better fit

731Sex Roles (2020) 83:722–738

intersections of gender, race/ethnicity, immigration status, and education level in a nationally representative U.S. sample. Typically, intersectional studies of sexual identity focus on the lived experiences of individuals at particular intersectional social locations (most commonly in qualitative studies: e.g., Hunter 2010; Moore 2011), or else they examine the effects of these experiences, such as on respondent well-being (e.g., Gonzales and Ortiz 2015; Liu et al. 2017). Studies treating sexual identification itself as the dependent variable are rare (England et al. 2016 is an exception). Our study is also the first known to include underrepresented racial/ethnic/immigrant groups in sexuality research including Native Americans, non-immigrant Asian/PIs, and immigrant Asian/PIs.

Heterosexual identification, despite being subject to patterned social forces like sexual minority identities, has also been mostly unexamined in prior quantitative research.

Using a large, nationally representative U.S. sample and novel multilevel intersectional methods (Evans 2015; Evans et al. 2018; Jones et al. 2016) we find evidence of considerable variability between strata in the propensity to exclusively het- erosexual identify. This evidence suggests that sexual identity formation and adoption is a social process influenced by mul- tiple, intersecting axes of social experience within interlocking systems of power and privilege. How people identify is in part a reflection of their embeddedness in heteronormative social processes and contexts, which are evidently more

Table 2 Predicted probability of heterosexual identification by strata, 3-axes: Gender and race/ethnicity/immigration status

Stratum Race/Ethnicity/Immigration Category PP 95% CI Stratum Rank Number in Stratum

(a) Men

110 Immigrant-Latinx 93.600 [90.003, 96.043] 12 240

120 Nonimmigrant-Latinx 91.672 [89.917, 93.221] 8 939

130 Black 94.250 [93.117, 95.288] 13 1499

140 Immigrant-Asian 95.939 [93.404, 97.684] 14 175

150 Nonimmigrant-Asian 92.248 [89.673, 94.590] 9 321

160 Native American 92.434 [87.861, 96.719] 10 139

170 White 93.033 [92.281, 93.761] 11 3911

(b) Women

210 Immigrant-Latinx 86.983 [82.568, 90.627] 6 251

220 Nonimmigrant-Latinx 80.147 [77.824, 82.473] 4 1043

230 Black 83.783 [82.186, 85.341] 5 1962

240 Immigrant-Asian 88.020 [83.647, 91.752] 7 196

250 Nonimmigrant-Asian 78.933 [74.187, 83.175] 2 278

260 Native American 70.362 [62.922, 77.496] 1 137

270 White 79.243 [77.907, 80.528] 3 4339

PP is Predicted Probability based on main and interaction effects. 95% CI is the Credible Interval for the Predicted Probability estimate. Stratum Rank: lowest rank number indicates stratum with lowest predicted probability of exclusively heterosexual identification

Men Women

Fig. 1 Predicted probability of exclusively heterosexual identification by race/ethnicity, immigration status, and gender. Markers indicate estimates; spikes indicate 95% credible intervals. Solid horizontal lines represent highest and lowest predicted probabilities of heterosexual

identification among strata of men. Dashed horizontal lines represent highest and lowest predicted probabilities of heterosexual identification among strata of women

732 Sex Roles (2020) 83:722–738

Table 3 Predicted probability of heterosexual identification by stratum, 4-axes: Gender, race/ethnicity/immigration status, and education

Stratum Race/Ethnicity/Immigration Category Education PP 95% CI Stratum Rank Number in stratum

(a) Men

111 Immigrant-Latinx HS or less 93.807 [90.030, 96.531] 34 95

112 Immigrant-Latinx Some college 94.838 [91.941, 96.965] 39 108

113 Immigrant-Latinx College+ 94.048 [89.582, 96.586] 35 37

121 Nonimmigrant-Latinx HS or less 91.980 [89.529, 94.024] 28 325

122 Nonimmigrant-Latinx Some college 92.126 [89.803, 94.035] 29 453

123 Nonimmigrant-Latinx College+ 90.843 [86.828, 93.781] 22 161

131 Black HS or less 94.533 [92.810, 96.081] 37 502

132 Black Some college 94.424 [92.936, 95.795] 36 672

133 Black College+ 93.590 [91.335, 95.450] 33 325

141 Immigant-Asian HS or less 95.228 [91.175, 97.570] 40 17

142 Immigrant-Asian Some college 96.091 [93.464, 97.890] 42 70

143 Immigrant-Asian College+ 96.006 [93.322, 97.879] 41 88

151 Nonimmigrant-Asian HS or less 92.340 [88.489, 95.324] 31 65

152 Nonimmigrant-Asian Some college 92.255 [88.495, 95.045] 30 100

153 Nonimmigrant-Asian College+ 91.723 [88.282, 94.379] 27 156

161 Native American HS or less 91.284 [86.410, 95.090] 26 51

162 Native American Some college 91.226 [86.886, 94.886] 25 63

163 Native American College+ 91.036 [85.791, 95.263] 24 25

171 White HS or less 94.824 [93.449, 96.093] 38 1034

172 White Some college 93.313 [92.177, 94.350] 32 1677

173 White College+ 90.954 [89.213, 92.498] 23 1200

(b) Women

211 Immigrant-Latinx HS or less 86.301 [80.254, 91.298] 16 66

212 Immigrant-Latinx Some college 86.132 [80.606, 90.656] 15 117

213 Immigrant-Latinx College+ 86.688 [80.520, 91.491] 17 68

221 Nonimmigrant-Latinx HS or less 79.851 [75.449, 83.730] 10 253

222 Nonimmigrant-Latinx Some college 79.184 [75.919, 82.152] 8 513

223 Nonimmigrant-Latinx College+ 81.422 [77.356, 85.348] 11 277

231 Black HS or less 82.351 [78.659, 85.509] 14 397

232 Black Some college 82.305 [79.984, 84.625] 13 942

233 Black College+ 86.731 [84.300, 89.121] 18 623

241 Immigrant-Asian HS or less 87.521 [80.076, 93.199] 20 15

242 Immigrant-Asian Some college 86.904 [81.257, 91.712] 19 86

243 Immigrant-Asian College+ 89.254 [84.436, 93.351] 21 95

251 Nonimmigrant-Asian HS or less 76.390 [66.852, 84.025] 4 36

252 Nonimmigrant-Asian Some college 79.177 [72.636, 84.637] 7 104

253 Nonimmigrant-Asian College+ 79.643 [74.235, 84.651] 9 138

261 Native American HS or less 71.808 [61.813, 80.191] 2 46

262 Native American Some college 70.396 [61.575, 78.866] 1 63

263 Native American College+ 74.135 [63.659, 82.613] 3 28

271 White HS or less 77.438 [74.551, 80.224] 5 816

272 White Some college 77.638 [75.705, 79.576] 6 1863

273 White College+ 82.083 [80.235, 83.781] 12 1660

For Education: HS or less = high school or less; College+ = college/Bachelor’s degree or above. PP is Predicted Probability based on main and interaction effects. 95% CI is the Credible Interval for the Predicted Probability estimate. Stratum Rank: lowest rank number indicates stratum with lowest predicted probability of exclusively heterosexual identification

733Sex Roles (2020) 83:722–738

consequential in terms of identification for some groups than others. Additionally, our findings about immigrant status sug- gest that it is a key axis of social life that should often be included when conducting research with large surveys.

In particular, we find that gender modifies the relationship between race/ethnicity/immigration status and sexual identifi- cation. Whereas there were substantial differences in hetero- sexual identification by race/ethnicity/immigrant status among women, racial/ethnic/immigrant status differences among men were smaller and mostly not statistically signifi- cant. This could be because men of all racial/ethnic/immigra- tion statuses are expected to embody hegemonic masculine ideals in part through heterosexual identification (Connell 1987). Previous survey research examining race/ethnicity and sexual identification had inconsistent findings likely in part because researchers did not more fully examine how this relationship differed by gender and immigration status. Women of different races/ethnicities and immigration statuses seem to experience different heteronormative pressures and opportunities to resist this heteronormativity.

Exactly why immigration status affects women in such varied ways requires further research. Immigrant Asian/PI women had the highest likelihood to identify as exclusively heterosexual among women. Similarly, immigrant Latinas had the second-highest predicted probability among women, which was significantly higher than non-immigrant Latinas. Among many immigrant Asian/PI and Latina women, hetero- sexuality appears to be tightly connected to normative femi- ninity, and more so than among their U.S.-born counterparts. Women who immigrated from developing nations may desire to remain connected to heteronormative immigrant communi- ties and families and may (a) feel that identification as any- thing other than exclusively heterosexual would comprise their familial and community relationships and (b) have inter- nalized different beliefs about LGBQ identification than their U.S.-born counterparts. These results may also indicate that immigrant women identify as heterosexual to appear respect- able to those in their immigrant and non-immigrant commu- nities, using heterosexuality as a way to bolster their status despite disadvantages based on race, national origin, and gender.

With regard to race, Native American women had the low- est predicted probability of heterosexual identification. This finding may reflect historical non-binary understandings of gender and sexuality in some tribes/nations. Some versions of these understandings may continue today, but in gendered ways. Little other survey research exists on this topic because Add Health is the only large-scale, nationally representative survey that offers a large enough sample to document the experiences of Native people today. The gap between Native men and women in their predicted probabilities of heterosex- ual identification—92.43% and 70.36%, respectively—is much larger than any other racial/ethnic/immigrant group,

again indicating uniquely gendered social processes that dy- namically interact with race/ethnicity and that little existing research has explored.

Adding education level into the intersectional model was also instructive. For most intersections of gender by race/ethnicity/immigration status there were no substantial differences in propensity to heterosexual identify by education level. The exceptions to this were among White men and White women, and the direction of the association appeared to be opposites in these cases. White men with a college de- gree were significantly less likely to heterosexual identify than White men with high school degrees or less. In contrast, among White women higher education (college degree or more) was associated with higher likelihood to heterosexual identify. Although causality is impossible to discern using the present study design, these results suggest that education/ social class may uniquely affect White people’s propensity to heterosexual identify in ways it does not for other racial/ ethnic and immigrant groups. Selection effects may explain part of this pattern (Hsin 2018; Mollborn and Everett 2015; Morris 2008), but it does not explain why the pattern is only significantly different for White women and men.

Overall, our results suggest that sexual identification in part reflects placements in gendered status hierarchies, but that more complicated processes related to race/ethnicity and im- migration status shape women’s probability of heterosexual identification. Men of all racial/ethnic/immigrant statuses are likelier to identify as heterosexual than their women counter- parts, excepting Latinx immigrants, the CIs of which slightly overlap for men and women. Women have greater flexibility to resist heteronormativity both because of their lower place- ment in status hierarchies (England 2016; Mize and Manago 2018) and because heterosexuality is less central to femininity than it is to masculinity. Lower status leading to greater prob- abilities of sexual minority identification does not hold for immigrant status and race/ethnicity, however. Immigrant women are likelier to identify as heterosexual than their non- immigrant counterparts, and whereas White women have higher status than Women of Color, only Native American women had lower probabilities of heterosexual identification. Clearly, status hierarchies for immigration status and race/ ethnicity do not shape sexual identification in ways that mirror gender.

Different probabilities of heterosexual identification among women exist likely because women at different inter- sections of race/ethnicity and immigration status experience different pressures of heteronormativity and opportunities to resist heteronormativity. Results for race/ethnicity among women likely reflect racial processes and marginalization spe- cific to particular racial groups, especially Black and Native women, that themselves shape different experiences with heteronormativity and different ways in which heterosexuality is related to the construction of femininity among Black and

734 Sex Roles (2020) 83:722–738

Native women. Although “messy,” our results show exactly what intersectionality theory suggests: Adding identities does not lead to predetermined outcomes, like lower status auto- matically resulting in lower probability of heterosexual iden- tification. Instead, the results are much more complex, and especially among women reflect the workings of numerous status hierarchies and social processes related to particular racial/ethnic/immigration intersections which interact to lower the probability of heterosexual identification (gender), in- crease it (immigrant experience), or have different effects based on specific identities (race/ethnicity).

Limitations and Future Research Directions

There are several directions for future research. Add Health data are becoming dated, so more recent survey data should be col- lected and examined in order to confirm these findings and to determine if patterns differ across age cohorts. Additionally, fu- ture studies should consider how other axes, such as religious identity or national country of origin for immigrants, shape sex- ual identity. It was not possible to do that in the current paper because the strata would have been too small to provide reliable results, but future studies could emphasize these axes more strongly. Further, quantitative methods cannot explore all aspects of intersectionality. We can analyze the likelihood of a particular outcome, such as heterosexual identification, but not the experi- ence of that outcome. Qualitative research thus remains critical to understanding how intersectionality affects lived experiences.

Practice Implications

The present findings can benefit mental health professionals, activists, and educators. Therapists and educators can tailor their messages to reflect that race/ethnicity and immigration status may shape sexual identification in gendered ways. For instance, men of a variety of backgrounds could be educated about how expectations of masculinity shape men to identify as heterosexual and could be given information on how to affirmatively claim an LGBQ identity if they choose to do so. Women could be educat- ed about how their immigrant and/or racial/ethnic communities shape sexual identification and can be connected to resources to help those who wish to claim an LGBQ identity be able to do so while retaining ties to their communities. Activists for LGBQ rights can use these findings to tailor outreach to particular com- munities, being mindful of how intersections of race/ethnicity, immigration status, and gender shape sexual identification due to structural inequalities—not simply individual prejudices—and that dismantling systems of inequality for LGBQ people requires dismantling all systems of inequality. Thus, activism for sexual minority equality goes hand-in-hand with activism for equality of opportunity along the lines of race/ethnicity, gender, and immi- gration status.

Conclusion

Our results show that quantitative intersectional methods are equipped to capture some of the effects of complicated social processes that shape sexual identification. Previous research has shown that sexual identity is a primary axis of social difference (Grollman 2017; Schnabel 2018) and is itself shaped by complex social factors rather than just attractions and behaviors. Sexual identity is, consequently, an ideal case for applying intersectional methods. Our analyses add to the literature on sexual identifica- tion by showing that key axes of social differentiation, including race/ethnicity, immigration status, gender, and education, may shape sexual identification in line with what intersectionality theory suggests. Our results highlight the utility of quantitative intersectional methods for sociologists of all interest areas. They also suggest how intersecting social locations substantially shape the predicted probability to identify as heterosexual, due to both additive main effects and their interactions. Identities, including sexual identities, are mutually constituted and shaped by multiple intersecting social locations that reflect social inequalities along the lines of gender, race/ethnicity, immigration status, and class.

Acknowledgements The first author would like to thank the Sexualities Project at Northwestern (SPAN) for providing him the postdoctoral fel- lowship that gave him the ability to work on this paper.

Compliance with Ethical Standards

Conflict of Interest The authors report no conflicts of interest. The au- thors followed all requirements set forth by the Carolina Population Center to guarantee confidentiality for participants of Add Health. The authors also gained approval from the University of Oregon Institutional Review Board to work with the Add Health dataset.

References

Alba, R. D., Insolera, N. E., & Lindeman, S. (2016). Is race really so fluid? Revisiting Saperstein and Penner’s empirical claims. American Journal of Sociology, 122(1), 247–262. https://doi.org/ 10.1086/687375.

Axelsson Fisk, S., Mulinari, S., Wemrell, M., Leckie, G., Perez Vicente, R., & Merlo, J. (2018). Chronic obstructive pulmonary disease in Sweden: An intersectional multilevel analysis of individual hetero- geneity and discriminatory accuracy. SSM - Population Health, 4, 334–346. https://doi.org/10.1016/j.ssmph.2018.03.005.

Bailey, J. M., Vasey, P. L., Diamond, L. M., Breedlove, S. M., Vilain, E., & Epprecht, M. (2016). Sexual orientation, controversy, and sci- ence. Psychological Science in the Public Interest, 17(2), 45–101. https://doi.org/10.1177/1529100616637616.

Bell, A., Holman, D., & Jones, K. (2019). Using shrinkage in multilevel models to understand intersectionality: A simulation study and a guide for best practice. Methodology European Journal of Research Methods for the Behavioral and Social Sciences, 15(2), 88–96. https://doi.org/10.1027/1614-2241/a000167.

Browne, W. J. (2015). MCMC estimation in MLwiN Version 2.32. Centre for Multilevel Modelling, University of Bristol.

735Sex Roles (2020) 83:722–738

Budnick, J. (2016). “Straight girls kissing”? Understanding same-gender sexuality beyond the elite college campus. Gender & Society, 30(5), 745–768. https://doi.org/10.1177/0891243216657511.

Caplan, Z. (2017). The problem with square pegs: Sexual orientation concordance as a predictor of depressive symptoms. Society and Mental Health, 7(2), 105–120. https://doi.org/10.1177/ 2156869317701266.

Carnevale, A. P., Smith, N., & Gulish, A. (2018). Women can’t win: Despite making educational gains and pursuing high-wage majors, women still earn less than men. Retrieved from Georgetown University, Center on Education and the Workforce website: https://cew.georgetown.edu/cew-reports/genderwagegap/#full- report.

Carrillo, H., & Hoffman, A. (2016). From MSM to heteroflexibilities: Non-exclusive straight male identities and their implications for HIV prevention and health promotion. Global Public Health, 11(7–8), 923–936. https://doi.org/10.1080/17441692.2015. 1134272.

Carrillo, H., & Hoffman, A. (2018). ‘Straight with a pinch of bi’: The construction of heterosexuality as an elastic category among adult US men. Sexualities, 21(1–2), 90–108. https://doi.org/10.1177/ 1363460716678561.

Cho, S., Crenshaw, K. W., & Mccall, L. (2013). Toward a field of intersectionality studies: Theory, applications, and praxis. Signs, 38(4), 785–810. https://doi.org/10.1086/669608.

Choo, H. Y., & Ferree, M. M. (2010). Practicing intersectionality in so- ciological research: A critical analysis of inclusions, interactions, and institutions in the study of inequalities. Sociological Theory, 28(2), 129–149. https://doi.org/10.1111/j.1467-9558.2010.01370.x.

Cohen, C. J., Fowler, M., Medenica, V. E., & Rogowski, J. C. (2018). Millennial attitudes on LGBT issues: Race, identity, and experience. Chicago: University of Chicago.

Collins, P. H. (1990). Black feminist thought: Knowledge, consciousness, and the politics of empowerment. Boston: Unwin Hyman.

Collins, P. H. (2015). Intersectionality’s definitional dilemmas. Annual Review of Sociology, 41, 1–20. https://doi.org/10.1146/annurev- soc-073014-112142.

Compton, D., & Bridges, T. (2019, April 12). 2018 GSS update on the U.S. LGB population [Wordpress]. Retrieved from inequality by (interior) design website: https://inequalitybyinteriordesign. wordpress.com/2019/04/12/2018-gss-update-on-the-u-s-lgb- population/.

Connell, R. (1987). Gender and power: Society, the person, and sexual politics. Stanford: Stanford University Press.

Conron, K. J., Goldberg, S. K., & Halpern, C. T. (2018). Sexual orienta- tion and sex differences in socioeconomic status: A population- based investigation in the National Longitudinal Study of adolescent to adult health. Journal of Epidemiology and Community Health, 72(11), 1016–1026. https://doi.org/10.1136/jech-2017-209860.

Copen, C. E., Chandra, A., & Febo-Vazquez, I. (2016). Sexual behavior, sexual attraction, and sexual orientation among adults aged 18-44 in the United States: Data from the 2011-2013 National Survey of family growth. National Health Statistics Reports, 88, 1–14 https://www.cdc.gov/nchs/data/nhsr/nhsr088.pdf.

Crenshaw, K. (1991). Mapping the margins: Intersectionality, identity politics, and violence against women of color. Stanford Law Review, 43(6), 1241–1299. https://doi.org/10.2307/1229039.

Davis, G. (2015). Contesting intersex: The dubious diagnosis. New York: NYU Press.

Dodge, B., Herbenick, D., Friedman, M. R., Schick, V., Fu, T.-C. (Jane), Bostwick, W., … Sandfort, T. G. M. (2016). Attitudes toward bisex- ual men and women among a nationally representative probability sample of adults in the United States. PLoS One, 11(10).

Dyar, C., Taggart, T. C., Rodriguez-Seijas, C., Thompson, R. G., Elliott, J. C., Hasin, D. S., … Eaton, N. R. (2019). Physical health dispar- ities across dimensions of sexual orientation, race/ethnicity, and sex:

Evidence for increased risk among bisexual adults. Archives of Sexual Behavior, 48(1), 225–242. https://doi.org/10.1007/s10508- 018-1169-8

England, P. (2016). Sometimes the social becomes personal: Gender, class, and sexualities. American Sociological Review, 81(1), 4–28. https://doi.org/10.1177/0003122415621900.

England, P., Mishel, E., & Caudillo, M. (2016). Increases in sex with same-sex partners and bisexual identity across cohorts of women (but not men). Sociological Science, 3, 951–970. https://doi.org/ 10.15195/v3.a42.

Evans, C. R. (2015). Innovative approaches to investigating social deter- minants of health—Social networks, environmental effects and intersectionality (Harvard University). Retrieved from http://nrs. harvard.edu/urn-3:HUL.InstRepos:23205168

Evans, C. R. (2019). Adding interactions to models of intersectional health inequalities: Comparing multilevel and conventional methods. Social Science & Medicine, 221, 95–105. https://doi.org/ 10.1016/j.socscimed.2018.11.036.

Evans, C. R., Williams, D. R., Onnela, J.-P., & Subramanian, S. V. (2018). A multilevel approach to modeling health inequalities at the inter- section of multiple social identities. Social Science & Medicine, 203, 64–73. https://doi.org/10.1016/j.socscimed.2017.11.011.

Falomir-Pichastor, J. M., & Mugny, G. (2009). “I’m not gay. . . . I’m a real man!”: Heterosexual men’s gender self-esteem and sexual prejudice. Personality and Social Psychology Bulletin, 35(9), 1233–1243. https://doi.org/10.1177/0146167209338072.

Feinstein, B. A., Wadsworth, L. P., Davila, J., & Goldfried, M. R. (2014). Do parental acceptance and family support moderate associations between dimensions of minority stress and depressive symptoms among lesbians and gay men? Professional Psychology: Research and Practice, 45(4), 239–246. https://doi.org/10.1037/a0035393.

Gates, G. J. (2014). LGBT demographics: Comparisons among population-based surveys. Retrieved from The Williams Institute on Sexual Orientation and Gender Identity Law and Public Policy website: https://williamsinstitute.law.ucla.edu/wp-content/uploads/ lgbt-demogs-sep-2014.pdf

Gates, G. J. (2017, January 11). In U.S., more adults identifying as LGBT. Pew. Retrieved from https://news.gallup.com/poll/201731/lgbt- identification-rises.aspx

Goldstein, H., Browne, W., & Rasbash, J. (2002). Partitioning variation in multilevel models. Understanding Statistics, 1(4), 223–231. https:// doi.org/10.1207/S15328031US0104_02.

Gonzales, G., & Ortiz, K. (2015). Health insurance disparities among racial/ethnic minorities in same-sex relationships: An intersectional approach. American Journal of Public Health, 105(6), 1106–1113.

Gould, E., & Kroeger, T. (2017). Women can’t educate their way out of the gender wage gap. Retrieved from Economic Policy Institute website: https://www.epi.org/publication/women-cant-educate- their-way-out-of-the-gender-wage-gap/.

Greenberg, D., Najle, M., Bola, O., & Jones, R. P. (2019). Fifty years after stonewall: Widespread support for LGBT issues – Findings from American values atlas 2018. Retrieved from Public Religion Research Institute website: https://www.prri.org/research/fifty- years-after-stonewall-widespread-support-for-lgbt-issues-findings- from-american-values-atlas-2018/.

Grollman, E. A. (2017). Sexual orientation differences in attitudes about sexuality, race, and gender. Social Science Research, 61, 126–141. https://doi.org/10.1016/j.ssresearch.2016.05.002.

Grzanka, P. R. (Ed.). (2014). Intersectionality: A foundations and fron- tiers reader. New York: Routledge.

Howell, J., & Emerson, M. O. (2017). So what “should” we use? Evaluating the impact of five racial measures on markers of social inequality. Sociology of Race and Ethnicity, 3(1), 14–30.

Hsin, A. (2018). Hegemonic gender norms and the gender gap in achieve- ment: The case of Asian Americans. Sociological Science, 5, 752– 774. https://doi.org/10.15195/v5.a32.

736 Sex Roles (2020) 83:722–738

Hunter, M. A. (2010). All the gays are white and all the blacks are straight: Black gay men, identity, and community. Sexuality Research & Social Policy, 7(2), 81–92. https://doi.org/10.1007/ s13178-010-0011-4.

Jones, K., Johnston, R., & Manley, D. (2016). Uncovering interactions in multivariate contingency tables: A multi-level modelling explorato- ry approach. Methodological Innovations, 9, 1–17. https://doi.org/ 10.1177/2059799116672874.

Kramer, R., DeFina, R., & Hannon, L. (2016). Racial rigidity in the United States: Comment on Saperstein and Penner. American Journal of Sociology, 122(1), 233–246. https://doi.org/10.1086/ 687374.

Kuperberg, A., & Walker, A. M. (2018). Heterosexual college students who hookup with same-sex partners. Archives of Sexual Behavior, 47(5), 1387–1403. https://doi.org/10.1007/s10508-018-1194-7.

Liu, H., Reczek, C., Mindes, S. C. H., & Shen, S. (2017). The health disparities of same-sex cohabitors at the intersection of race- ethnicity and gender. Sociological Perspectives, 60(3), 620–639. https://doi.org/10.1177/0731121416663685.

Lourie, M. A., & Needham, B. L. (2017). Sexual orientation discordance and young adult mental health. Journal of Youth and Adolescence, 46(5), 943–954. https://doi.org/10.1007/s10964-016-0553-8.

Magliozzi, D., Saperstein, A., & Westbrook, L. (2016). Scaling up: Representing gender diversity in survey research. Socius: Sociological Research for a Dynamic World, 2, 1–11. https://doi. org/10.1177/2378023116664352.

McCall, L. (2005). The complexity of intersectionality. Signs, 30(3), 1771–1800. https://doi.org/10.1086/426800.

Merlo, J. (2018). Multilevel analysis of individual heterogeneity and dis- criminatory accuracy (MAIHDA) within an intersectional frame- work. Social Science & Medicine, 203, 74–80. https://doi.org/10. 1016/j.socscimed.2017.12.026.

Mishel, E. (2019). Intersections between sexual identity, sexual attraction, and sexual behavior among a nationally representative sample of American men and women. Journal of Official Statistics, 35(4), 859–884. https://doi.org/10.2478/jos-2019-0036.

Mize, T. D., & Manago, B. (2018). Precarious sexuality: How men and women are differentially categorized for similar sexual behavior. American Sociological Review, 83(2), 305–330. https://doi.org/10. 1177/0003122418759544.

Mollborn, S., & Everett, B. (2015). Understanding the educational attain- ment of sexual minority women and men. Research in Social Stratification and Mobility, 41, 40–55. https://doi.org/10.1016/j. rssm.2015.04.004.

Moore, M. (2011). Invisible families: Gay identities, relationships, and motherhood among black women. Berkeley: University of California Press.

Morris, E. W. (2008). “Rednecks,” “rutters,” and rithmetic: Social class, masculinity, and schooling in a rural context. Gender & Society, 22(6), 728–751. https://doi.org/10.1177/0891243208325163.

Mustanski, B., Birkett, M., Greene, G. J., Rosario, M., Bostwick, W., & Everett, B. G. (2014). The association between sexual orientation identity and behavior across race/ethnicity, sex, and age in a proba- bility sample of high school students. American Journal of Public Health, 104(2), 237–244.

Nagoshi, C. T., Raven Cloud, J., Lindley, L. M., Nagoshi, J. L., & Lothamer, L. J. (2019). A test of the three-component model of gender-based prejudices: Homophobia and transphobia are affected by raters’ and targets’ assigned sex at birth. Sex Roles, 80(3), 137– 146. https://doi.org/10.1007/s11199-018-0919-3.

Newlin Lew, K., Dorsen, C., & Long, T. (2018). Prevalence of obesity, prediabetes, and diabetes in sexual minority men: Results from the 2014 behavioral risk factor surveillance system. The Diabetes Educator, 44(1), 83–93. https:/ /doi.org/10.1177/ 0145721717749943.

O’Connor, E. C., Ford, T. E., & Banos, N. C. (2017). Restoring threatened masculinity: The appeal of sexist and anti-gay humor. Sex Roles, 77(9–10), 567–580. https://doi.org/10.1007/s11199-017-0761-z.

Omi, M., & Winant, H. (2015). Racial formation in the United States. New York: Routledge/Taylor & Francis Group.

Paschen-Wolff, M. M., Kelvin, E., Wells, B. E., Campbell, A. N. C., Grosskopf, N. A., & Grov, C. (2019). Changing trends in substance use and sexual risk disparities among sexual minority women as a function of sexual identity, behavior, and attraction: Findings from the National Survey of family growth, 2002–2015. Archives of Sexual Behavior, 48(4), 1137–1158. https://doi.org/10.1007/ s10508-018-1333-1.

Pascoe, C. J. (2011). Dude, you’re a fag: Masculinity and sexuality in high school (New ed.). Berkeley: University of California Press.

Penner, A. M., & Saperstein, A. (2013). Engendering racial perceptions: An intersectional analysis of how social status shapes race. Gender & Society, 27(3), 319–344. https://doi.org/10.1177/ 0891243213480262.

Priebe, G., & Svedin, C. G. (2013). Operationalization of three dimen- sions of sexual orientation in a national survey of late adolescents. Journal of Sex Research, 50(8), 727–738. https://doi.org/10.1080/ 00224499.2012.713147.

Rasbash, J., Steele, F., Browne, W. J., & Goldstein, H. (2015). A user’s guide to MLwiN: version 2.32. Centre for Multilevel Modelling, University of Bristol.

Röder, A. (2015). Immigrants’ attitudes toward homosexuality: Socialization, religion, and acculturation in European host societies. International Migration Review, 49(4), 1042–1070. https://doi.org/ 10.1111/imre.12113.

Ross, L. E., Salway, T., Tarasoff, L. A., MacKay, J. M., Hawkins, B. W., & Fehr, C. P. (2018). Prevalence of depression and anxiety among bisexual people compared to gay, lesbian, and heterosexual individ- uals: A systematic review and meta-analysis. The Journal of Sex Research, 55(4–5), 435–456. https://doi.org/10.1080/00224499. 2017.1387755.

Rust, P. (1992). The politics of sexual identity: Sexual attraction and behavior among lesbian and bisexual women. Social Problems, 39(4), 366–386. https://doi.org/10.2307/3097016.

Saperstein, A., & Penner, A. M. (2012). Racial fluidity and inequality in the United States. American Journal of Sociology, 118(3), 676–727. https://doi.org/10.1086/667722.

Savin-Williams, R. C., & Vrangalova, Z. (2013). Mostly heterosexual as a distinct sexual orientation group: A systematic review of the empir- ical evidence. Developmental Review, 33(1), 58–88. https://doi.org/ 10.1016/j.dr.2013.01.001.

Schnabel, L. (2018). Sexual orientation and social attitudes. Socius, 4, 1– 18. https://doi.org/10.1177/2378023118769550.

Schrock, D., & Schwalbe, M. (2009). Men, masculinity, and manhood acts. Annual Review of Sociology, 35, 277–295.

Silva, T. (2017). Bud-sex: Constructing normative masculinity among rural straight men that have sex with men. Gender & Society, 31(1), 51–73. https://doi.org/10.1177/0891243216679934.

Silva, T. J. (2019). Straight identity and same-sex desire: Conservatism, homophobia, and straight culture. Social Forces, 97(3), 1067–1094. https://doi.org/10.1093/sf/soy064.

Silva, T. J., & Whaley, R. B. (2018). Bud-sex, dude-sex, and heteroflexible men: The relationship between straight identification and social attitudes in a nationally representative sample of men with same-sex attractions or sexual practices. Sociological Perspectives, 61(3), 426–443. https://doi.org/10.1177/0731121417745024.

Soehl, T. (2017). From origins to destinations: Acculturation trajectories in migrants’ attitudes towards homosexuality. Journal of Ethnic & Migration Studies, 43(11), 1831–1853. https://doi.org/10.1080/ 1369183X.2016.1246178.

Taggart, T. C., Rodriguez-Seijas, C., Dyar, C., Elliott, J. C., Thompson, R. G., Hasin, D. S., & Eaton, N. R. (2019). Sexual orientation and sex-

737Sex Roles (2020) 83:722–738

related substance use: The unexplored role of bisexuality. Behaviour Research and Therapy, 115, 55–63. https://doi.org/10.1016/j.brat. 2018.12.012.

Van der Bracht, K., & Van de Putte, B. (2014). Homonegativity among first and second generation migrants in Europe: The interplay of time trends, origin, destination and religion. Social Science Research, 48, 108–120. https://doi.org/10.1016/j.ssresearch.2014. 05.011.

Walker, A. (2014). “Our little secret”: How publicly heterosexual women make meaning from their “undercover” same-sex sexual experi- ences. Journal of Bisexuality, 14(2), 194–208. https://doi.org/10. 1080/15299716.2014.902347.

Weaver, K. S., & Vescio, T. K. (2015). The justification of social inequal- ity in response to masculinity threats. Sex Roles, 72(11–12), 521– 535. https://doi.org/10.1007/s11199-015-0484-y.

Willer, R., Rogalin, C. L., Conlon, B., & Wojnowicz, M. T. (2013). Overdoing gender: A test of the masculine overcompensation thesis. American Journal of Sociology, 118(4), 980–1022. https://doi.org/ 10.1086/668417.

Publisher’s Note Springer Nature remains neutral with regard to jurisdic- tional claims in published maps and institutional affiliations.

738 Sex Roles (2020) 83:722–738