Introduction

Sexual Minority Stress and Suicide

Gay and bisexual men are two- to five-times more likely than heterosexual men to attempt suicide (Hottes, Bogaert, Rhodes, Brennan, & Gesink, 2016). The most prominent theory currently used to understand this disparity is the minority stress model. The theory was first articulated in the context of sexual minority individuals’ lives by Meyer (2003a) and has subsequently been developed and refined by others (Hatzenbuehler, 2009). Minority stress theory posits that sexual minority individuals cumulatively experience increased stress as they encounter sexual (antigay/bisexual) stigma over the course of their lives. Meyer outlines two types of minority stress processes: distal stressors, which include enacted prejudice events (such as violence or discrimination), and proximal stressors, which include anticipated rejection (or “hyper-vigilance”), concealment of sexual identities, and internalized homophobia (Meyer, 2003a). Minority stress is unique, chronic, and socially based and is hypothesized to cause multiple psychological and social problems, for example, emotional dysregulation, hopelessness, negative self-schemas, and felt social rejection (Hatzenbuehler, 2009). To cope, individuals affected by this process may choose a variety of adaptive strategies, including healthcare-seeking, drug and alcohol use, or social isolation. If these strategies are ineffective, some will turn to suicide.

Research on sexual minority suicide has disproportionately focused on youth (Hottes et al., 2016). While rates of suicide attempt among sexual minority individuals decrease with age (Russell & Toomey, 2012) (an epidemiologic trend also apparent in heterosexual populations [Nock et al., 2008]), they remain elevated, as compared with heterosexuals, throughout adulthood (Blosnich, Nasuti, Mays, & Cochran, 2016; Hottes et al., 2016)—an age when fatal suicide rates are highest (Safer, 1997). These distinct age-related patterns and the dearth of research on suicidal behavior in sexual minority adults underscore the importance of suicide research using samples of sexual minority adults.

At least six empirical studies have explicitly tested sexual minority stress theory in relation to suicide ideation or attempts in sexual minority adults (McLaren, 2016; Michaels, Parent, & Torrey, 2015; Ploderl, Faistauer, & Fartacek, 2010; Ploderl et al., 2014; Ploderl & Fartacek, 2009; Wang, Plöderl, Häusermann, & Weiss, 2015). Other studies have implicitly used the theory by measuring minority stress-related constructs (D’Augelli, Grossman, Hershberger, & O’Connell, 2001; Diaz, Ayala, Bein, Henne, & Marin, 2001; Ferlatte, Dulai, Hottes, Trussler, & Marchand, 2015; Igartua, Gill, & Montoro, 2009; Irwin, Coleman, Fisher, & Marasco, 2014; Paget et al., 2016). Each of these studies has focused on one (or both) of two constructs from the minority stress model: enacted stigma (a distal stressor) and internalized homophobia (a proximal stressor). Enacted stigma is defined as stigmatizing behavior that is explicitly expressed through interpersonal interactions, for example, antigay/bisexual slurs, discrimination, exclusion, verbal harassment, and in some cases violence (Herek, Gillis, & Cogan, 2009). In contrast, internalized homophobia is an internal and proximal form of stress that comes from devaluing the self in response to societal stigma (Herek et al., 2009; Meyer, 2003a). Evidence is most robust for an association between measures of enacted stigma and suicide-related outcomes (D’Augelli et al., 2001; Diaz et al., 2001; Ferlatte et al., 2015; Irwin et al., 2014; Paget et al., 2016; Ploderl et al., 2010, 2014; Ploderl & Fartacek, 2009). Two studies have found internalized homophobia to be associated with suicide ideation, though not with suicide attempts (McLaren, 2016; Ploderl et al., 2014). Meyer notes that despite conceptual and measurement-related distinctions between distal and proximal minority stressors, the processes are interdependent: for example, experiencing antigay violence (distal) may lead one to feel ashamed (i.e., internalized homophobia) or to anticipate future experiences of prejudice (Meyer, 2003a).

Other Perceived Causes of Sexual Minority Suicide

In contrast to the deductive studies of minority stress theory outlined above, Wang et al. (2015) found that among 116 Swiss gay men who had attempted suicide, few named enacted stigma as a cause of their attempts. Rather, the most commonly cited causes were problems with romantic relationships (19%), accepting one’s homosexuality (akin to internalized homophobia; 16%), and depression (11%). The discordance between results from the deductive, associational studies that emphasize the importance of enacted stigma, and Wang’s inductive study, which emphasizes other causes, may be explained by limitations in measurement of experiences of sexual stigma, or by attribution to mediators rather than fundamental causes. In other words, suicidal individuals who have experienced minority stress may be more likely to attribute their suicide attempts to more proximal factors, regardless of whether minority stress in fact contributed to their suicidal thoughts and attempts.

Challenges and Opportunities in Minority Stress Measurement

Meyer (2003b) noted that a critical challenge to measurement of sexual stigma is the distinction between overtly discriminatory, acute events that will be memorable (e.g., an incident of sexuality-motivated harassment or bullying) and a more chronic, daily strain of being vigilant or on guard against said events. The former category may be easier to measure because of the rarity and salience of these events, though perhaps less important than the daily hassle of being “hypervigilant” in explaining the cumulative stress load of sexual minority individuals. The rarity of the former category may also explain why it was less commonly cited in Wang et al.’s study. On the other hand, overt acts of discrimination and victimization, though rarer, may be more acute and severe than the daily experience of hypervigilance and therefore may have a larger total effect on suicidal thoughts and behavior.

A second challenge with regard to measurement of sexual stigma concerns Meyer’s (2003a) distinction between distal and proximal stressors. Stigma need not be enacted upon an individual to be felt. Measures of hypervigilance get at one aspect of felt stigma (a proximal stressor)—that is, the worry and anticipation of enacted stigma. However, felt stigma may also operate through chronically and strategically concealing one’s stigmatized identity. Concealment in response to sexual stigma may simultaneously confer benefits—by shielding sexual minority individuals from enacted stigma—and activate its own deleterious psychological process that induces shame, hostility, or psychosocial challenges described in relation to the broader minority stress model above (Meyer, 2003a; Pachankis, 2007). Binary measurements of concealment are notably inadequate because most sexual minority individuals will carefully evaluate particular social contexts and relationships and selectively disclose their sexuality only when perceived to be safe (Horowitz & Newcomb, 2001; Pachankis, 2007).

Traditional Suicide Risk Factors

In addition to the issues of minority stress measurement outlined above, gay men in Wang et al.’s study may have articulated more proximal causes of their suicide attempts (e.g., depression, relationship troubles) because they are more recent and acute when contrasted to the cumulative experiences of sexual stigma. Associational studies examining measured experiences of sexual stigma and studies of traditional risk factors (e.g., depression, isolation) in relation to suicide attempts among sexual minority individuals may be unified through mediation models that test the extent to which traditional risk factors explain associations between sexual stigma and suicide-related outcomes. Moreover, mediation analyses are necessary to identify proximal points of intervention, in order to prevent suicide among those who have already accumulated experiences of stigma across the life course, thereby failing to benefit from recently enacted measures to reduce stigma (e.g., legislative or policy reform) (Hatzenbuehler & Keyes, 2013).

Integrating Minority Stress Measurement and Mediation

In the present study, we attempted to address several gaps and limitations in this body of literature by first, developing a measurement model of sexual minority stress inclusive of the less-studied constructs of anticipated prejudice and concealment of sexuality; second, using structural equation modeling to assess the magnitude of associations between minority stress constructs and suicide attempts; and third, looking for evidence suggestive of mediation of these associations by commonly studied psychosocial causes of suicide. Our study is informed by Hatzenbuehler’s (2009) extension of Meyer’s minority stress model to an integrated psychological mediation model. Hatzenbuehler hypothesized that there are “multiple pathways through which stigma-related stress ultimately influences the pathogenesis of mental health…disorders among sexual minorities,” including pathways related to coping (e.g., alcohol and substance use), social isolation, and cognitive processes (e.g., those that confer a higher risk of depression). Following Hatzenbuehler’s approach, we tested mediators representing each of these three conceptual pathways. Our analysis was completed using a sample of approximately 8000 gay and bisexual men, enabling comparisons of associations between minority stress and suicide attempts across key subgroups of sexual minority men on the basis of sexual identity, age, and history of suicide attempts.

Method

Participants

This study was a secondary analysis of an existing periodic survey, Sex Now, and thus was cross-sectional in design. Sex Now is the largest Canadian sample of men having sex with men (http://cbrc.net/sexnow). The survey provides a platform for assessing a broad array of health and social issues currently affecting Canadian gay and bisexual men. Survey content and analysis are determined as part of a larger community-based participatory research cycle, including community dialogues, focus groups, and meetings.

Participants were recruited between October 2014 and April 2015, from across Canada through dating and sex-seeking mobile applications and websites, social media, a database of previous survey participants, and word-of-mouth. Eligible participants resided in Canada, identified as a man who has sex with other men, and were 16 years of age or older. The survey was administered entirely online and offered in English and French. Responses were anonymous, and no honorarium was provided. Informed consent was obtained from all individual participants included in the study. Participation was restricted by IP address to avoid duplication. Furthermore, given the lack of monetary incentive to participate, duplicate responses are unlikely. Nonetheless, 41 potential duplicates (0.5% of sample) were identified using a subset of variables assumed to represent unique individuals (language, sexual identity, ethnicity, income, education, age, height, weight, and postal code) and were excluded in sensitivity analyses.

Measures

The survey instrument for the 2014–2015 cycle of Sex Now was developed in 2014 using a “world café” approach (Brown & Isaacs, 2005). World cafés involve a series of round table discussions, focused around specific questions, in order to create space where multiple perspectives and forms of knowledge (personal, professional, and academic) can be expressed and heard. As with focus groups, world cafés are facilitated, with a “host” at each table. Participants rotate between tables, with approximately 10–15 min of discussion per round. Three world cafés were conducted by the research team in early 2014, including approximately 50 gay and bisexual men of diverse ages, ethnicities, and occupations. Many of the participants were study volunteers who actively participated in all stages of the research process (i.e., survey instrument design, recruitment, and analysis). Consistent with recommendations of community-based research, questionnaire items were inductively developed, arising from topics of priority identified by community members themselves (Leung, Yen, & Minkler, 2004).

For the 2014–2015 survey, one of the dedicated world café topics was “everyday experiences of stigma and prejudice.” Participants were asked to reflect on their personal and everyday reactions to experiences of stigma and were specifically asked about a sense of hyper-vigilance related to their sexuality. Comments elicited from the world cafés included not only reports of enacted stigma measured in previous editions of the survey, but also responses to a range of diverse settings (e.g., gym, healthcare facilities, border crossings) that induced worry or anticipation of prejudice. Participants also reflected on the multiple audiences to which they needed to either conceal or disclose their sexual identity. These observations were ultimately translated into novel investigator-created minority stress measures that emerged from the lives of gay and bisexual community members, with special emphasis given to different experiences of antigay stigma (settings, social interactions, and vigilance against types of anticipated prejudice), not previously studied in the minority stress literature.

For the present analysis, we restricted the Sex Now dataset to Canadian residents who identified as man, transgender, genderqueer, two-spirit, or other. Sex Now participants who opt not to answer a question exit the survey, and their responses are not retained; therefore, there are no missing data. We used five types of measured variables in this analysis, as detailed below.

Outcome

The outcome was any reported suicide attempt in the last 12 months (binary variable).

Minority Stress Indicators

We conceptually grouped manifest indicators of sexual minority stress into three constructs, based on discussion from the focus groups described above, as well as published articulations of the minority stress model (Hatzenbuehler, 2009; Meyer, 2003b): enacted stigma (eighteen binary variables); anticipated prejudice, or “hypervigilance” (nine 5-point Likert ordinal variables); and concealment of sexuality (five 3-point Likert ordinal variables, and one binary variable). All minority stress items were framed in relation to the respondent’s sexuality. All items related to anticipated prejudice and several items related to enacted stigma were asked with respect to a range of settings (e.g., family events, work, school, gym, travel). Items related to concealment were asked with respect to a range of audiences (e.g., family, friends, coworkers) and as an ordinal scale, allowing for gradations of concealment to each audience. Minority stress experience was measured cumulatively, i.e., in reference to lifetime experience. Anticipated prejudice (Hatzenbuehler, Nolen-Hoeksema, & Erickson, 2008) is previously untested in relation to suicide attempts in sexual minority adults; likewise, concealment is less-measured in empirical studies of minority stress. Accordingly, exploratory (factor analysis) methods were used. We conceived of concealment as a continuous underlying construct (as opposed to a dichotomous measure).

Psychosocial Risk Factors (Mediators)

We selected three psychological or social variables that correspond to Hatzenbuehler’s (2009) three proposed pathways (cognitive, coping, and social) and are consistently associated with suicide risk in the general literature as potential mediators (Nock et al., 2008; O’Connor & Nock, 2014; Van Orden et al., 2010): depression, drug or alcohol use, and social isolation. According to minority stress theory, these variables are conceptually proximal to suicide attempts; i.e., minority stress is hypothesized to accumulate and eventually lead to emotional pain and hopelessness (and in some cases depression), coping strategies (e.g., drug or alcohol use), and social withdrawal. In the survey, depression and drug or alcohol use were measured as discussion with a healthcare provider about these self-defined problems in the past 12 months. Current social support was measured as the number of people respondents can count on when they need help or when something goes wrong.

Covariates

We used directed acyclic graphs to identify the minimally sufficient set of sociodemographic variables to control for confounding of the effects of minority stress on the outcome, suicide attempt (Greenland, Pearl, & Robins, 1999). These included: age, geography (urban, suburban, rural), income, education, Aboriginal identity, and HIV status, all of which have been identified as risk factors for suicide and suicide attempts (Gurm et al., 2015; Navaneelan, 2012; Nock et al., 2008; Tjepkema, Wilkins, & Long, 2013) and are hypothesized to affect experiences of sexual minority stress, thereby constituting potential confounding variables.

Effect Modifiers (Stratification Variables)

We selected three variables a priori as potential effect modifiers of associations between minority stress and suicide attempt: sexual identity (gay, bisexual), age (<30, 30–49, 50+ years), and history of suicide attempt (Rhodes et al., 2013). Sexual identity is a fundamental aspect of the lives of sexual minority individuals, affecting participation in social networks and venues, as well as identity, and may therefore moderate associations between sexual minority stress and mental health. Age is used primarily to reflect differing birth cohorts, also known to be fundamental to experiences of sexual minority stress (Hammack & Cohler, 2009). Those over 50 were born before 1964 and thus aged into adulthood (and likely sexual experience) before widespread visibility or social acceptance of a gay identity and also notably before the emergence of the AIDS epidemic. Those between 30 and 49 were born between 1964 and 1984 and thus aged into adulthood and sexual experience during a period characterized by the HIV/AIDS health crisis (before widespread availability of effective treatment) and changing legal and social support for gay men. Those under 30 were born after 1984 and thus aged into adulthood after the advent of effective HIV combination treatment and after the legalization of same-sex marriage in Canada. Lastly, history of suicide attempts was selected given previous research that suggests different causes of first versus subsequent attempts among sexual minority men (Wang et al., 2015), as well as the significance of a history of attempts as a predictor of subsequent attempts in the general suicide literature (Joiner et al., 2005).

Statistical Analysis

Factor Analysis

We built and tested a measurement model using factor analysis. Prior to factor analysis we randomly split the dataset in half, the first half of observations used for exploratory factor analysis (EFA) and second half for confirmatory factor analysis (CFA). We used 33 manifest indicators of sexual minority stress: 19 binary indicators and 14 ordinal indicators, all of which were treated as categorical variables in EFA. Robust weighted least squares was used to estimate the relations between all 33 manifest variables and an undetermined number of latent factors, hypothesized to be 3, based on the minority stress model and the questionnaire development process, described above. We evaluated incremental factor counts using the following goodness-of-fit measures and corresponding cutoff criteria proposed by Schreiber, Nora, Stage, Barlow, and King (2006) standardized root-mean-square residual (RMSR ≤ 0.08), root-mean-square Error of Approximation (RMSEA ≤ 0.06), comparative fit index (CFI ≥ 0.95), Tucker–Lewis Index (TLI ≥ 0.96), and χ 2, though emphasis was given to RMSR, RMSEA, CFI, and TLI, given that χ 2 is sensitive to large sample size (Streiner, 2006). We examined factor loadings for each possible EFA solution, and only retained factors loaded with ≥3 manifest variables with loadings ≥|0.32| (Gie Yong & Pearce, 2013). We also reviewed factor loadings for theoretical meaningfulness. Variables with complex loadings (i.e., manifest variables cross-loading on more than one factor) were not used unless conceptually interpretable. We dropped manifest variables with loadings ≤0.32 (Gie Yong & Pearce, 2013). We hypothesized that the factors would be correlated; therefore, oblique rotation was used.

CFA was used to evaluate the fit of the measurement model identified in EFA, within the second half of the dataset. The same goodness-of-fit measures were used, as in EFA, and in the case of a poor fit, a conservative approach to model modification allowed for the addition of theoretically consistent item error covariances until fit measures reached a priori cut-points (Schreiber et al., 2006). We made modifications iteratively, using chi-square modification tests and starting with within-factor item error correlations.

Structural Equation Modeling (SEM)

SEM was guided by an analytical model that describes hypothesized causal relations between distinct sexual minority stress constructs (enacted stigma, anticipated prejudice, and concealment of sexuality) and suicide attempts (Fig. 1). We used probit regression with robust least squares to estimate relations between the latent variables (factors) identified in CFA and the outcome. We then evaluated the hypothesized model—including manifest variable error covariances added in CFA—for goodness-of-fit and kept further modifications to a minimum. We added error correlations between factors (now dependent variables in SEM) due to the possibility of unmeasured third variables that could affect multiple dependent factors, and we re-evaluated fit measures after model modifications.

Fig. 1
figure 1

Analytical model of hypothesized relations between three latent sexual minority stress constructs (F1, F2, F3), traditional psychosocial risk factors (M1, M2, M3), and past 12-month suicide attempts, among sexual minority men. Note Direct pathways illustrated with bold arrows; indirect pathways illustrated with black arrows; sociodemographic covariate relations illustrated with gray arrows; residual arrows annotated by “e” represent error terms (omitted variables and measurement error) for all endogenous/dependent variables

We conducted SEM analysis in four steps. First, direct, indirect, and total associations were estimated for each factor, with adjustment for all minority stress latent constructs (multivariable Model 1). Second, the model was adjusted for potential sociodemographic confounders (multivariable Model 2). Third, we calculated predicted probabilities +/− three standard deviations from the mean factor score of each of the latent variables, holding all covariates at their mean or referent category (Muthén & Asparouhov, 2002). Fourth, we ran stratified (fully adjusted) models for each hypothesized effect modifier.

We report both unstandardized and standardized coefficients. We evaluated strength of associations using standardized coefficients; emphasis is given to magnitude of associations, though statistical significance (illustrated using 95% confidence intervals [CI]) is also considered (Tomarken & Waller, 2005). We calculated the proportion of the total association mediated for each latent factor and treated these values as suggestive evidence of mediation, with the acknowledged limitation that true mediation cannot be definitively established within the cross-sectional design of the study. Given the possibility of inconsistent mediation (mediating pathways operating in opposite directions), proportion mediated was calculated using the absolute value of coefficients (MacKinnon, 2008). Finally, we assessed effect modification in two ways. First, we examined qualitative differences in the direction and magnitude of path coefficients across subgroups. Second, we applied a statistical test of difference across levels of effect modifiers, p < .05 considered statistically significant (Fairchild & MacKinnon, 2009). All analyses were completed using MPlus 7 for Mac (version 1.4).

Results

Sample

A total of 8037 respondents completed the survey; 128 respondents living outside Canada and 43 who exclusively identified as a woman were removed, resulting in an analytic sample of 7872. A total of 265 of these respondents (3.4%) reported a suicide attempt in the previous 12 months. The sample was diverse with respect to sociodemographic characteristics, though was predominantly gay-identified, over 30 years of age, and Caucasian (Table 1).

Table 1 Characteristics of sample of sexual minority men used for structural equation modeling of minority stress and suicide attempts, N = 7872

Factor Analysis

One-, two-, three-, and four-factor measurement models were examined (Table 2). All goodness-of-fit measures reached the recommended cutoff with a three-factor solution. Fit measures continued to improve with a four-factor solution; however, the fourth factor failed to load with at least 3 manifest variables (with loadings ≥0.32), and the factor loadings were not theoretically consistent. Thus, a three-factor solution was carried forward for CFA and SEM. Sixteen manifest indicators loaded to factor 1, hereafter termed “enacted stigma”; 9 manifest indicators loaded to factor 2, “anticipated prejudice”; and 5 manifest indicators loaded to factor 3, “concealment of sexuality” (Table 3). Four enacted stigma indicators (verbal violence, antigay discrimination in healthcare setting, antigay discrimination at gym, and antigay discrimination by police/law courts) cross-loaded to the anticipated prejudice and concealment factors, and two anticipated prejudice indicators (worry at family events and employment) cross-loaded to the concealment factor, but all cross-loaded indicators were dropped from the secondary factors because they were not conceptually consistent with those latent constructs. Two enacted stigma indicators (antigay discrimination at Canadian border and antigay discrimination at other country borders) and one concealment indicator (attended event with a woman to pass as straight) failed to load (i.e., loadings <0.32) to their hypothesized factors but did load to other factors; all three were likewise dropped due to conceptual inconsistency.

Table 2 Comparison of exploratory factor analysis models of sexual minority stress, N = 3936 (split dataset)
Table 3 Sexual minority stress factor loadings, N = 7872 (dataset split for exploratory and confirmatory analysis)

The initial model tested in the second half of the dataset with CFA converged but had only adequate fit (RMSEA = 0.062; CFI = 0.946; TLI = 0.942). Therefore, error correlations were added to the hypothesized model, starting with the following four within-factor indicator error correlations: rejected/dismissed from job opportunity and antigay discrimination at work (conceptually consistent as both relate to prejudice in the workplace); called out as “homo,” “faggot,” etc. and verbal violence (conceptually consistent as both refer to verbal antigay slurs); verbal violence and physical violence (conceptually consistent as both refer to overt expressions of homophobia); anticipated prejudice at Canadian border entry and anticipated prejudice at foreign border entry (conceptually consistent as both refer to prejudice during travel). The following three between-factor indicator error correlations were added given conceptual consistency: antigay discrimination at family events and anticipated prejudice at family events (same setting); antigay discrimination in healthcare setting and anticipated prejudice in healthcare setting (same setting); anticipated prejudice at family events and concealment of sexuality to family (same audience). Modifications improved all goodness-of-fit measures, such that the final model reached the recommended cutoffs for all measures (RMSEA = 0.057, CFI = 0.960, TLI = 0.951) (Table 3). Cronbach’s alpha was 0.85 for the enacted stigma factor, 0.87 for the anticipated prejudice factor, and 0.93 for the concealment factor.

Structural Equation Modeling

The hypothesized SEM model was estimated using the analytical model (Fig. 1), including the three latent constructs tested and confirmed in factor analysis. The initial model had poor fit; therefore, minor modifications were made. Specifically, error covariances were added between each pair of latent factors, and between depression and drug/alcohol use, as suggested by modification indices. Correlations between latent factors were 0.631 (95% CI 0.582, 0.680) for enacted stigma/anticipated prejudice, −0.385 (95% CI −0.422, −0.348) for enacted stigma/concealment, and −0.097 (95% CI −0.111, −0.083) for anticipated prejudice/concealment. All three mediators were significantly associated with changes in risk of suicide attempt: those who discussed depression (standardized coefficient 0.355, 95% CI 0.253, 0.457) or drug or alcohol use (0.125, 95% CI 0.009, 0.241) with a healthcare provider were more likely to have attempted suicide, while those with social support were less likely to have attempted (−0.131, 95% CI −0.205, −0.057).

Associations between minority stress constructs and suicide attempts were statistically significant for all three latent factors, though direction of associations and specific indirect paths varied across constructs (Table 4). The total association was largest for enacted stigma, and approximately half of this was a direct association, with an additional 33% mediated by depression; that is, enacted stigma was positively associated with depression, which in turn was positively associated with suicide attempts. The total association of anticipated prejudice was smaller and not statistically significant. Indirect associations between anticipated prejudice and suicide attempts were statistically significant for depression (37% of total association) and social isolation (27% of total association). Anticipated prejudice was positively associated with depression, which in turn was positively associated with suicide attempts. Anticipated prejudice was also positively associated with social isolation, which in turn was positively associated with suicide attempts. Inconsistent mediation was apparent for concealment. That is, concealment was negatively associated with depression, which in turn was positively associated with suicide attempts—ultimately resulting in a negative association between concealment and suicide attempts (14% of total association). Concealment was also positively associated with social isolation, which in turn was positively associated with suicide attempts (33 of total association).

Table 4 Structural equation model path coefficients for direct, indirect, and total associations between three latent sexual minority stress constructs (factors) and past 12-month suicide attempts, N = 7872

Associations between minority stress factors and suicide attempt were slightly attenuated after adjustment for sociodemographic covariates (Table 4). Four sociodemographic characteristics were significantly associated with suicide attempts: personal income (>$30,000) and university degree were inversely associated, while Aboriginal identity and HIV positive status were positively associated.

Predicted probabilities illustrate the magnitude of association for the range of minority stress construct scores, as well as selected sociodemographic variables (Fig. 2). The baseline probability of suicide attempt in the past 12 months was 0.036 (95% CI 0.026, 0.050). The largest range in probabilities was observed for enacted stigma: those at −3 standard deviations from the mean score had a 0.004 (95% CI 0.001, 0.010) probability of a suicide attempt in the past 12 months, while those at +3 standard deviations had a 0.178 (0.102, 0.281) probability.

Fig. 2
figure 2

Probability of suicide attempt in last 12 months (with 95% confidence interval), as predicted by a structural equation model including measures of sexual minority stress and social characteristics, N = 7872. Note Threshold = probability based on intercept (threshold) of probit model (holding all latent variables at mean values, and binary covariates at referent group); all other probabilities reflect change in variable indicated while fixing all latent variables at mean values and binary covariates at referent group; SD standard deviation

Effect Modification

Path coefficients for enacted stigma and anticipated prejudice were similar between gay and bisexual men; however, the direct association of concealment with suicide attempt was larger for bisexual men than for gay men (Table 5). The direct association of enacted stigma with suicide attempt was larger in older age cohorts than in younger age cohorts. The total association of enacted stigma with suicide attempt was reduced and lost statistical significance when the sample was restricted to those with a history of suicide attempt. Across-strata differences in path coefficients were statistically significant for two comparisons: the total association of enacted stigma and indirect association mediated by depression in those with (vs. those without) a prior suicide attempt.

Table 5 Structural equation model path coefficients for direct, indirect, and total associations between three latent sexual minority stress constructs (factors) and past 12-month suicide attempts, stratified by subgroups of interest identified a priori

Discussion

This empirical evaluation of the sexual minority stress model in a community-based sample of 7872 gay and bisexual men underscores the salience of enacted stigma in explaining suicide-related behavior, while also emphasizing the existence of multiple related pathways from varied experiences of minority stress to suicide attempts. After identifying a measurement model with three distinct constructs of minority stress—enacted stigma, anticipated prejudice, and concealment of sexuality—the largest total association with past 12-month suicide attempts was observed for enacted stigma (Fig. 2); 33% of this association was mediated by depression (Table 4). Below, we interpret the findings related to each of these three constructs, in sequence.

Enacted Stigma

Previous studies identified a consistent association between measures of enacted stigma and suicide-related behavior among sexual minority individuals (D’Augelli & Grossman, 2001; Diaz et al., 2001; Ferlatte et al., 2015; Irwin et al., 2014; Ploderl et al., 2010, 2014; Ploderl & Fartacek, 2009). The present study extends our understanding of the relations between enacted stigma and suicide attempts in three ways. First, the magnitude of the association of enacted stigma is far greater than the associations of anticipated prejudice and concealment. This is most clearly illustrated in Fig. 2, which shows a tenfold or greater increased risk in past 12-month suicide attempt between those at the highest and lowest end of the range of experiences of enacted stigma.

Second, we found that less than half of the total association of enacted stigma was explained by three traditional psychosocial risk factors: depression, drug or alcohol use, and social isolation. Notwithstanding the need to test additional potential mediators, this suggests that general suicide models [for example, the interpersonal theory of suicide (Joiner, 2007) and the integrated motivational-volitional model (O’Connor, 2011)] are likely insufficient to explain the elevated rates of suicide attempts in sexual minority populations (Ploderl et al., 2014). Additional research should test other potential mediators that are specific to the lives of sexual minority individuals. One example of a sexual minority-specific pathway is over-investment in achievement-related domains of contingent self-worth as a means to “guard against” the potential consequences of being discovered as a stigmatized minority, as recently theorized by Pachankis and Hatzenbuehler (2013). Another possible pathway of particular importance to sexual minority suicides is the unique experiences of same-gender sexual and romantic relationships. In the Swiss study cited in the introduction, 19% of all suicide attempts among gay men were attributed to problems with love/relationships (Wang et al., 2015). Although changes in relationship status (especially divorce/separation or widowing) are among the most consistent social predictors of suicide and suicide attempt in the general (heterosexual) population (Brent et al., 1993; Nock et al., 2008; Yen et al., 2005), relationship factors may require separate, group-specific study among sexual minorities, given that sexual minority men often adopt non-traditional (i.e., non-heterosexual) norms, practices, and roles (i.e., less likely to be married and/or monogamous, and more likely to endorse an “egalitarian and negotiated” approach to domestic labor division) (Green, 2010; Operario et al., 2015).

Third, we found evidence of a modified association of enacted stigma in comparing those with and without a history of suicide attempts, whereby the association was removed in those with a history of attempts. This is perhaps not surprising, given that prior suicide attempts are one of the strongest predictors of subsequent attempts (Joiner et al., 2005; Nock et al., 2008). Nonetheless, this finding suggests that minority stress may have a larger impact on the onset of suicidal behavior than on subsequent suicide attempts. We also saw a larger direct association of enacted stigma among older cohorts of men than among the youngest cohort, though this difference was not statistically significant.

Anticipated Prejudice

To our knowledge, measures of anticipated prejudice (or hypervigilance [Pachankis & Goldfried, 2006], or “rejection sensitivity” [Pachankis, Hatzenbuehler, & Starks, 2014]) have not previously been tested in relation to suicide attempts, though an association has been found with depression (Hatzenbuehler et al., 2008). We found a statistically significant association between anticipated prejudice and suicide attempts, mediated by depression. This association was much smaller in magnitude than the association between enacted stigma and suicide attempts, which may reflect the greater impact of enacted versus anticipated stigma. Alternative explanations merit discussion as well, however, particularly in light of the “chronic strain” hypothesis that posits that daily hassles of hypervigilance contribute to a large degree of the cumulative minority stress experience (Meyer, 2003b). One explanation is that enacted stigma and anticipated prejudice are too highly correlated to identify independent associations. In our model, the correlation coefficient between enacted stigma and anticipated prejudice factors was large (0.63) (Cohen, 1992). Furthermore, exploratory univariate analysis with each of the minority stress factors yielded total associations for enacted stigma and anticipated prejudice that were both greater than their adjusted estimates in multivariable SEM and estimates more comparable to one another than in multivariable models (Table 6, cf. Table 4). In this analysis we treated enacted stigma and anticipated prejudice as independent constructs; however, this distinction may not be so clear in the lives of sexual minority individuals who likely experience both sets of stressors as part of cumulative excess stress and related chronic strain (Meyer, Schwartz, & Frost, 2008).

Table 6 Path coefficients for direct, indirect, and total associations between latent sexual minority stress constructs (factors) and past 12-month suicide attempts, tested in separate (univariate) models for each factor, N = 7872

Concealment

Concealment was independently associated with suicide attempts, though the associations were smaller in magnitude, and mediation was inconsistent. Concealment was inversely associated with suicide attempts (i.e., protective association) when mediated by depression, and positively associated with suicide attempts (harmful association) when mediated by social isolation (Table 4). In the first instance, as concealment increased, depression decreased (hence, the inverse association), and in turn, the probability of suicide attempts also decreased. In the second instance, as concealment increased, social isolation increased, and as social isolation increased, suicide attempts also increased. Our large sample enabled exploration of these associations across subgroups of gay and bisexual men, which are further interpreted below.

Each of the contrasting associations has some degree of theoretical and empirical support. Concealing one’s sexual identity may confer some protection from enacted stigma perpetuated against known sexual minority individuals, thereby reducing the cumulative load of overt minority stressors among those who conceal (Fredriksen-Goldsen et al., 2013). Concealment and enacted stigma constructs were indeed inversely correlated in the present study (−0.38). In addition, the direct and total protective associations of concealment were larger among bisexual men than among gay men, though statistical tests of difference were not significant. At the same time, concealment of gay or bisexual identities reduces opportunities for social connection with other gay and bisexual men and thus limits social support through these channels (Hatzenbuehler, 2009); accordingly, the positive association between concealment and suicide attempts in our sample was mediated by lack of social support. These contrasting associations of concealment notably would have gone undetected without testing for mediators, as the total association of concealment was not statistically significant (Table 4) (MacKinnon, 2008).

One recent study of concealment of sexuality among bisexual men found a positive association between concealment and depression, though these authors notably distinguish between concealment and disclosure, defining the latter as a construct more concerned with negative attitudes or expectations related to disclosing a minority sexuality (Schrimshaw, Siegel, Downing, & Parsons, 2013). Another recent study identified a “U”-shaped distribution, whereby sexual minority men who were “not out” had the lowest rates of depression and anxiety, while those who were “recently out” had the highest rates, and those who were “distantly out” fell in between (Pachankis, Cochran, & Mays, 2015). Together these additional studies confirm the importance of investigating complex, nonlinear trends in the mental health effects of concealment, as well as the importance of subgroup analysis by sexual identity. Notably, neither of these studies investigated suicide attempt as an outcome. Future research thus may help to disentangle these bidirectional associations between concealment and suicide-related outcomes by measuring these more nuanced facets of concealment-related behavior, including attitudes about concealment, motivation for concealment, and timing of concealment.

Limitations

While SEM is a powerful and flexible analytic tool, interpretation of SEM results requires caution and qualification (Tomarken & Waller, 2005). Limitations of our application of SEM broadly relate to matters of cross-sectional study design, measurement, and confounding. In this study, we were primarily interested in the lifelong cumulative effects of sexual stigma on mental health. In the absence of large-sample longitudinal studies of sexual minority populations, we relied upon a cross-sectional design. The benefit of this design is the large sample size, which enabled subgroup analyses. The cost was reduced certainty of the temporal sequence of “exposures,” mediators, and outcomes. We attempted to address this limitation by using exposure measures related to cumulative/lifetime experiences of minority stress and by restricting mediators and outcome to the past 12 months. Furthermore, there is theoretical support for the direction of effects, since suicide attempts cannot plausibly cause enacted sexual stigma. Nonetheless, we cannot rule out the possibility that the mediators and outcome preceded the exposure variables we investigated. This failure to capture longitudinal effects with a cross-sectional design may have induced a bias whereby we have over-estimated mediating associations in the absence of true longitudinal mediation; addressing this limitation will require repeating our analyses using multiple waves of data collection (Maxwell, Cole, & Mitchell, 2011). The cross-sectional design additionally induces a form of survival bias, whereby those gay and bisexual men who died—most notably by suicide—are excluded. Assuming minority stress is a major attributable cause of suicide in our population, this selection bias may have induced underestimation of associations between minority stress constructs and suicide attempts. Our overly strict missing data policy also may have induced a bias by excluding respondents who opted not to answer a question.

With regard to measurement, our study is both strengthened and limited by the use of novel, community-designed indicators of minority stress experiences. These indicators resulted from discussions by Canadian gay and bisexual men who reflected on their own life experiences, and many are previously untested in empirical minority stress research. While the hypothesized measurement model ultimately met a priori cut-points for goodness-of-fit with minor modifications in CFA, additional approaches to improve upon measurement of under-studied constructs, such as anticipated prejudice, are required. As well, we were unable to assess the convergent and discriminant validity of the minority stress factors, as the survey did not include previously validated minority stress scales (Pachankis, Goldfried, & Ramrattan, 2008). In addition, we note that internalized homophobia has been associated with suicide-related outcomes in other studies (McLaren, 2016; Ploderl et al., 2014) but was not measured in our survey. Given that internalized homophobia is conceptually related to other proximal minority stress processes, i.e., concealment and anticipated prejudice, we might expect that some of the positive associations we observed between these two constructs and suicide attempts in fact represent internalized shame. Other research has indeed found moderate correlations between internalized homophobia and concealment and anticipated prejudice constructs (Pachankis et al., 2008; Ross & Rosser, 1996).

While the associations we observed between enacted stigma and suicide attempts were large, minority stress experiences may not be the cause of the attempts. This study was ultimately a deductive test of one particular theory; inductive, qualitative approaches are required to ascertain perceived causes or precipitants of suicide attempts (Wang et al., 2015). Adjustment for sociodemographic characteristics conceptually related to both minority stress and suicide-related behavior only slightly attenuated associations between minority stress constructs and suicide attempts; however, other unmeasured confounders may explain some or all of these associations. Most notably, traits that may predispose individuals to experience or anticipate enacted stigma, or conceal their sexual identity, and attempt suicide (e.g., impulsivity, aggression, extroversion, resilience) were not measured (O’Connor & Nock, 2014; Ploderl et al., 2014).

Two of the mediators we tested, depression and problematic drug or alcohol use, were measured as having a discussion with a healthcare provider for these specific problems. These measures therefore exclude those who did not utilize healthcare for these problems. In Canada, as few as 40% of adults with depression receive professional medical or mental health care (Cheung & Dewa, 2007; Crabb & Hunsley, 2006). On the other hand, healthcare discussion related to depression or drug or alcohol use is a more specific measure, one that likely reflects greater severity of these issues. Additionally, our study relied upon self-reported suicide attempts. In the face of persistent societal stigma of suicide, some suicide attempts will go unreported; meanwhile, other survey respondents may report behaviors that would be more accurately classified as an incomplete suicide plan or self-harm without intent to die than as suicide attempts (Silverman, Berman, Sanddal, O’Carroll, & Joiner, 2007). Finally, as with most other studies of sexual minority suicide (Haas et al., 2011), our study was conducted in North America, and thus may not be generalized to countries with fewer legal protections against enacted stigma.

Implications

This study adds weight to an increasing body of evidence that societal stigma is a fundamental cause of ill health among sexual minority individuals (Hatzenbuehler et al., 2014a; Hatzenbuehler, Phelan, & Link, 2013). While enacting policies and practices to reduce societal stigma are important strategies for suicide prevention among youth and future generations of adults (Hatzenbuehler, Birkett, Van Wagenen, & Meyer, 2014b; Saewyc, Konishi, Rose, & Homma, 2014), they are inadequate to prevent suicide among the current living cohort of sexual minority adults who have already accumulated a lifetime of sexual stigma. Our study points to specific mediating factors that should be the focus of more immediate suicide prevention efforts. These proximal points of intervention include both clinic-based (counseling or healthcare support for depression or drug or alcohol use) and community-based interventions (social support); notably, neither will alone be sufficient to avert the suicide attempts we measured. These services may be enhanced or tailored for sexual minority individuals by explicitly acknowledging the pervasive sexual stigma that often precedes experiences of mental distress. Ultimately, the distinct pathways identified in our study belie a one-size-fits-all strategy; as with general population strategies (World Health Organization, 2014), effective suicide prevention for sexual minority populations must be comprehensive, multifaceted, and inclusive of community-based approaches.