Abstract
Sexual-minority women (SMW) report higher rates of substance use and disorder across the life span and greater levels of minority stress in adolescence and young adulthood. Minority stress mediation models propose that higher levels of social stressors may increase emotion dysregulation, which in turn increases the propensity toward substance misuse. Few studies, however, have prospectively examined the impact of stressors and emotion dysregulation among SMW on early and escalating substance use. In this longitudinal study, we examined whether emotion dysregulation and social stress mediated the association between sexual-minority status and developing substance use (ages 17–22) in a sample of 2,201 heterosexual and 246 SMW participants in the Pittsburgh Girls Study. Results supported serial mediation processes of marijuana-use risk: SMW reported higher levels of social stress in late adolescence, which in turn predicted greater emotion dysregulation that was associated with greater marijuana use by young adulthood.
Rates of substance use disorder (SUD) among sexual-minority women (SMW) are elevated relative to heterosexual populations (Marshal et al., 2008; S. E. McCabe et al., 2005, 2009; Meyer, 2003; Vrangalova & Savin-Williams, 2014). For instance, meta-analytic reviews (Meyer, 2003) and large, nationally representative samples suggest that SMW are 3.5 times more likely to have a lifetime occurrence of an SUD relative to heterosexual women (Cochran et al., 2004). Relative to heterosexual women, SMW are also 2 to 4 times as likely to report alcohol dependence, 2 to 5 times as likely to report cigarette use, and 11 times as likely to report marijuana dependence (Marshal et al., 2008; S. E. McCabe et al., 2005; Meyer, 2003).
Adolescence and young adulthood are especially crucial periods for the initiation and progression of substance use among SMW. Across sexual orientation populations, young adulthood (ages 18–25) is a developmental period during which problem and heavy substance use reaches a lifetime peak (Chassin et al., 2003), and substance-use patterns during this period are among the most robust predictors of substance misuse and disorder across the life span (Chassin et al., 2002; McGue & Iacono, 2005). Moreover, racial differences in the development of substance use suggest that substance-use disparities may vary as a function of intersecting racial and other marginalized identities (Bowleg, 2008; Crenshaw, 1991). For instance, whereas Black youths have historically reported relatively lower levels of substance use compared with White youths (S. E. McCabe et al., 2007), this gap may be closing for marijuana use (Lanza et al., 2015), and some evidence suggests that Black women may report greater marijuana use by young adulthood (Keyes et al., 2015).
This developmental window may also be particularly crucial in explaining SUD risk among sexual-minority populations, especially for SMW. Multiple prior studies have shown that heightened substance use in SMW first emerges in adolescence (Coker et al., 2010; Marshal et al., 2008; Needham, 2012). Longitudinal studies indicate that disparities in use either persist or increase (Dermody et al., 2020; Marshal et al., 2009) through young adulthood. For instance, prior work involving the Pittsburgh Girls Study data had shown that relative to heterosexual youths, SMW reported greater frequency of alcohol, marijuana, and cigarette use in adolescence and young adulthood. SMW also increased more quickly in use through young adulthood than their heterosexual peers, and disparities in growth rates between SMW and heterosexual peers are most pronounced in early adolescence compared with later adolescence and young adulthood (Dermody et al., 2020). Among studies addressing intersections of race and sexual-minority status, findings have been somewhat inconsistent. For instance, although studies have shown that women of color report greater substance-use problems than White SMW (Mereish & Bradford, 2014), other studies have suggested no such differences in substance-related problems (Hughes et al., 2006; Parks & Hughes, 2005). Our goal in the current study is to therefore understand the factors explaining early and escalating substance use among SMW and to explore differences in these associations by race.
Stress, Emotion Dysregulation, and Developing Substance Use
Sexual-minority-stress theories suggest that SMW may report higher rates of substance use and SUD because of greater exposure to stigma-related social stressors, such as minority-related victimization, discrimination, and peer rejection (Hatzenbuehler, 2009; Meyer, 2003; Newcomb & Mustanski, 2010). The link between stress and increased risk for SUD in both the general population (Grant et al., 2003) and among sexual-minority individuals (S. E. McCabe et al., 2010) is well established. Among SMW, for instance, approximately 46% who reported victimization experiences within the past year also reported a past-year diagnosis of SUD, a figure that is 4 times as high as SMW who report no victimization (S. E. McCabe et al., 2010). Moreover, meta-analytic findings suggest that a higher level of general social stress is the strongest predictor of subsequent substance use among sexual-minority adolescents compared with other common risk factors, including co-occurring internalizing and externalizing problems and reports of general distress (Goldbach et al., 2014). Stress experiences in adolescence have also partly explained the observed disparity in substance-use outcomes (tobacco and alcohol use) between heterosexual women and SMW in prior research (Austin et al., 2008; McLaughlin et al., 2012). Together, these findings suggest that stress-related processes that heighten risk for substance-use behaviors may be intervening factors in the association between sexual-minority status and SUD.
Current models of SUD risk among SMW indicate that emotion dysregulation may be a principal mechanism linking stress with the development of SUD (Hatzenbuehler, 2009). Emotion dysregulation has been conceptualized as difficulty in modulating emotions to respond appropriately to demands in the environment (Aldao et al., 2010). For instance, individuals with greater emotion dysregulation may have poorer emotional awareness and understanding of emotion, may have limited access to resources or adaptive strategies for coping with emotion, and may be more likely to engage in maladaptive patterns of responding to emotion, such as rumination, among other characteristics (Gratz & Roemer, 2004). Stress experiences have been directly linked with emotion dysregulation behaviors (Aldao et al., 2010; Ward et al., 2003) and are thought to heighten emotion dysregulation by both increasing emotional reactivity and disrupting the use of effective regulation strategies (Cohen et al., 1995; Pechtel & Pizzagalli, 2011; Sheridan & McLaughlin, 2014).
The effect of stress on emotion dysregulation may in turn increase risk for substance-use initiation and escalation through young adulthood. Both sexual-minority-stress theory (Hatzenbuehler, 2009) and general developmental theories on the risk of early substance use (Hussong et al., 2011) propose that experiences of marginalization tax an individual’s coping resources such that substances are used as a strategy for downregulating negative affect. Supporting evidence from the general population has shown a robust relation between coping motives (for a review, see Cooper et al., 2016) and emotion dysregulation (Aldao et al., 2010; Giombini, 2015; Nolen-Hoeksema et al., 2007) with substance-use behavior. Moreover, research has shown that individuals with histories of early stress may be particularly sensitized to the reinforcing properties of substance use within adolescence and young adulthood (Andersen & Teicher, 2009; Brady & Sinha, 2005; Enoch, 2011; Sapolsky, 2003) such that stress-related motivation toward use progresses more rapidly into heavier and disordered use over time (Andersen & Teicher, 2009). Thus, minority-stress experiences and emotion dysregulation may characterize increased risk for a more rapid progression from substance-use experimentation into heavier use through young adulthood. Although evidence has supported emotion dysregulation as a mediator of the effects of sexual-minority stress on internalizing symptoms (e.g., Hatzenbuehler et al., 2008, 2009; Szymanski & Henrichs-Beck, 2014), we know of no studies that have prospectively examined whether the effect of minority stress on emotion dysregulation, in turn, predicted substance use among SMW or whether this pathway resulted in a trajectory of increasingly severe use.
Current Study
Our goal in this study was to examine whether and how stress experiences and emotion dysregulation among SMW explain changes in substance use from late adolescence into young adulthood. Using longitudinal data on 2,447 women (5.9% SMW) followed from ages 17 to 22, we pursued these questions by first establishing whether sexual-minority status was associated with level and change in emotion dysregulation through young adulthood and whether this, in turn, was associated with escalating substance use (alcohol, cigarette, and marijuana). We then tested social stress as a mechanism in these relations by examining whether the effect of stress on emotion dysregulation mediated the relation between sexual-minority status and alcohol, cigarette, and marijuana use in the sample. We hypothesized that greater social stress among SMW in late adolescence would predict subsequently higher levels of emotion dysregulation, greater frequency of substance use, and steeper increases in substance use by young adulthood. We further posited that higher levels and faster growth in substance use among SMW will be explained by the mediating effects of adolescent social stress on level and change in emotion dysregulation. Finally, we examined exploratory hypotheses about whether these effects differed among White and Black women.
Method
Research design
In this study, we used existing data from the longitudinal Pittsburgh Girls Study comprising a large, urban, community-based sample designed to assess the development of behavioral and emotional disorders and substance use from middle childhood to young adulthood (Keenan et al., 2010). Data were collected using an accelerated longitudinal design in which four age-based cohorts were recruited, ranging from ages 5 to 8, and were followed annually. This design resulted in planned missingness patterns ranging up to 37.3% in variables involved in the current study (see Table S1 in the Supplemental Material available online), which were accommodated using a multiple imputation approach (described in the Analytic Strategy section). At the time of the current analyses, data for the sample were collected through age 22 in the oldest cohort. We conducted secondary data analysis using six waves of data (ages 17–22), examining sexual-minority status (age 18) and social stress experiences in late adolescence (age 17), levels and growth in emotion dysregulation from age 19 to age 21, 1 and alcohol, cigarette, and marijuana use from age 17 to age 22. Retention rates for the study were high across annual time points, ranging from 85.4% to 97.2%. More detailed descriptions of the Pittsburgh Girls Study sample, recruitment, and retention are provided elsewhere (Hipwell et al., 2002; Keenan et al., 2010).
For these analyses, a total of 2,447 women were in the sample at the initial time point (age 17), and 246 women reported sexual-minority status at age 18. Although sexual-minority status was collected at earlier time points, age 18 was the first age at which women were the sole informants of sexual orientation. Approximately half of the young women were Black (53%), 41% were White, and the remaining participants self-identified as another race or multiracial. Low-income neighborhoods were oversampled; approximately 40% of families received public assistance at Wave 1. Approximately 47% of caregivers in the sample at Wave 1 reported 12 or fewer years of education.
Study measures
Annual interviews were conducted in the home and were completed by trained interviewers using a laptop computer. The current study included measures of sexual orientation, childhood and adolescent social stress, emotion dysregulation, and alcohol, cigarette, and marijuana use. Nearly all measures were adapted from previously existing measures used in prior studies.
Sexual-minority status
Self-report of sexual orientation was assessed using the following question asked at age 18: “Do you consider yourself to be (a) Heterosexual or straight; (b) Gay or lesbian; or (c) bisexual?” Given the relatively small number of lesbian (n = 64) and bisexual (n = 182) respondents and given that both groups are at similar risk for alcohol and other drug-use disorders (e.g., Green & Feinstein, 2012), these groups were combined to form a single SMW group.
Social stress
Stressors were measured as a composite index derived from three social domains measured at age 17: peer victimization, discriminatory experiences, and harsh parenting.
Reports of victimization by peers were measured using the Peer Experience Questionnaire (Vernberg et al., 1999) at age 17. This nine-item scale includes items measuring victimization by verbal aggression, confrontive physical aggression, and ostracism or relational aggression, rated on a scale ranging from never (1) to a few times per week (5) in the preceding year. Reliability for this scale in the current sample across data collection cohorts was high (α range = .777–.787).
Harsh parenting at age 17 was assessed using the child-report version of the 19-item Conflict Tactics Scale–Parent/Child Version (Straus, 1979), which measures the level of child-directed harsh parenting practices from the parent. Subscales of this measure included items reflecting psychological aggression and spanking. Straus and colleagues (Straus, 1998) reported adequate discriminant and construct validity on all subscales, and adequate reliability coefficients are reported in the current sample across data collection cohorts (α range = .742–.753).
Discrimination at age 17 was measured using the Everyday Discrimination Scale (EDS; Williams et al., 1997). The EDS is a nine-item measure of chronic and routine experiences of unfair treatment, such as being treated with less respect than others, experienced in the preceding year. The EDS was computed as a sum and demonstrated high reliability across cohorts (α range = .837–.839).
We combined victimization, harsh parenting, and discrimination using a latent variable specification. We used confirmatory factor analysis (CFA) to assess whether these measures could be reasonably combined in a latent measurement model. Noting that fit measures are not informative in saturated structural equation models (i.e., a CFA with three manifest variables), we examined factor loadings produced by this model to determine whether this specification was feasible as a latent variable model. Standardized factor loadings were 0.66 for peer victimization, 0.80 for discrimination, and 0.32 for harsh parenting practices. This suggested that each social stress measure was a reasonable indicator of a latent social stress construct at age 17.
Emotion dysregulation
Annual assessments of emotion dysregulation from ages 19 to 21 were measured using two subscales from the Difficulties in Emotion Regulation Scale (DERS; Gratz & Roemer, 2004) and the Rumination subscale of the Perfectionism Inventory (Hill et al., 2004). The DERS is designed to assess clinically relevant difficulties in emotion regulation with six subscales, although only two (the six-item Awareness subscale and the eight-item Strategies subscale; Gratz & Roemer, 2004) were available in the present study. Both Awareness (α range = .868–.939) and Strategies (α range = .823–.907) subscales demonstrated high reliability across the study period and across data collection cohorts. The seven-item Rumination subscale from the Perfectionism Inventory measures intrusive worrying about past errors, imperfect performance, or future mistakes and has adequate reliability across study ages and cohorts (α range = .908–.916).
Similar to our social stress measure, we used CFA to justify a latent emotion-dysregulation measure comprising the Awareness, Strategies, and Rumination subscales, particularly because some prior work has demonstrated poor psychometric properties of several DERS subscales (especially the Awareness subscale; e.g., Hallion et al., 2018; Osborne et al., 2017). We specified CFA models using the observed data in which Awareness, Strategies, and Rumination were indicated by latent factors, conducted at each respective time point (ages 19, 20, and 21). Standardized factor loadings indicated that whereas Rumination and Strategies were strongly and similarly correlated with each latent factor (range = 0.56–0.91), Awareness correlated weakly with these factors at each time point (range = 0.11–0.15). Given comparable factor loadings and moderate correlation between Rumination and Strategies at each time point (range: r = .45–.50), we therefore quantified emotion dysregulation as a sum of the standardized Rumination and Strategies subscales at each time point. Before summing, each of these variables had been standardized across all time points to ensure that standardization did not remove time trends in these variables (Bollen & Curran, 2006; King et al., 2018).
Substance use
Preceding-year alcohol, cigarette, and marijuana use was assessed annually from age 17 to age 22 using the Nicotine, Alcohol and Drug Substance Use Scale, adapted from an instrument of the Rutgers Health and Human Services Project (Pandina et al., 1984). This scale contains items measuring frequency of use of each substance within the preceding year, assessed on an 8-point scale ranging from 0 (I did not within the past year) to 7 (every day or more than once a day).
Analytic strategy
Missing data
We used multiple imputation to estimate missing data (Graham et al., 2006; Little & Rhemtulla, 2013) using Blimp (Enders et al., 2016; Keller et al., 2017) to generate a total of 20 imputed data sets. We conducted imputations on two Markov chains simultaneously and computed potential scale reduction factors (Gelman & Rubin, 1992) as a means of diagnosing the convergence of the estimation procedure. Potential scale reduction factors assess convergence by comparing the estimated between-chains and within-chains variances for each model parameter to determine whether both chains are producing similar means and variances for a given parameter (i.e., a posterior distribution).
We specified alcohol, cigarette, and marijuana use as ordinal; sexual-minority status as nominal; and all remaining variables with missingness as continuous. This produced a maximum potential scale reduction factor of 1.070 across all fixed and error parameters, well within the acceptable range (Enders, 2017). To accommodate the ordinal nature of our substance-use outcomes, we used diagonally weighted least squares estimation in all analyses. This provided more accurate and robust parameter estimation for discrete outcomes compared with standard maximum likelihood estimation and alternatives (e.g., weighted least squares; Mîndrilă, 2010; Rhemtulla et al., 2012; Schumacker & Beyerlein, 2000).
Primary analyses
All analyses were conducted in the R software environment (Version 4.0.2; R Core Team, 2020) using lavaan (Version 0.6-8; Rosseel, 2012). We conducted these in parallel across imputed data sets using the semTools package (Version 0.5-3; Jorgensen et al., 2020), which pooled parameter estimates by computing arithmetic means of these values across data sets (e.g., Baraldi & Enders, 2010), and applied Rubin’s rules (Rubin, 2004) to compute the standard errors for each pooled estimate.
Primary hypotheses were addressed using parallel-process growth-curve modeling (Cheong et al., 2003; Preacher et al., 2011; Singer & Willett, 2009). We evaluated model fit using the adjusted χ2 difference test, in which a nonsignificant result was an indication of adequate model fit. We supplemented this test with a number of alternative fit indices (Chen, 2007; Cheung & Rensvold, 2002; Meade et al., 2008), including the comparative fit index (CFI), the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). Recommended approximate cutoffs for these fit indices were CFI values greater than .95, TLI values greater than .95, RMSEA values less than 0.06, and SRMR values less than 0.08 (Chen, 2007; Hu & Bentler, 1999; Yu, 2002). In our comprehensive final model, we included race as a covariate to control for confounding effects of racial majority identity. This was represented as a set by a dummy variable indicating Black and a dummy variable indicating other or mixed race, each coded relative to a White reference group.
We used the product-of-the-coefficients approach (MacKinnon et al., 2002) to test two-path and three-path (e.g., Taylor et al., 2008) indirect effects. To establish temporal precedence for mediation effects, we set the intercept of emotion dysregulation at age 19 and the intercept for alcohol, cigarette, and marijuana use at age 20 and used self-report sexual-minority status at age 18 so that indirect effects reflected appropriate temporal sequencing to the extent possible (e.g., age-18 sexual-minority status → age-17 social stress → age-19 emotion dysregulation→ age-20 substance use). 2 We used the Monte Carlo method (MacKinnon et al., 2004) to compute confidence intervals (CIs) for each effect to evaluate whether these effects were significantly different from zero, carried out using the Rmediation package (Version 1.1.4; Tofighi & MacKinnon, 2011). Given the large number of indirect effects examined, we used 99% CIs to conduct more stringent tests of statistical significance for each effect.
Exploratory analyses
After estimating the above models using the entire sample, we conducted follow-up analyses to ensure the robustness and generalizability of our findings among racial and sexual-minority subgroups in our sample. First, we conducted a sensitivity analysis in which we limited our final model to heterosexual and bisexual women only. Sensitivity analyses were limited to bisexual women (n = 182) because only 64 women identified as lesbian in the sample. This was done to ensure that combining lesbian and bisexual women into a single group was a tenable decision for our data.
Second, we explored differences in key parameter estimates in our final growth-curve model across White and Black women by estimating our final model separately by race. We limited these analyses to White and Black women given the sample-size limitations among other/mixed race participants (5.8%). These post hoc analyses were pursued to ensure that model estimates were generalizable to both White and Black participants and to describe differences between these racial groups in the pattern of results. We estimated this model by stratifying our final growth-curve model by White and Black participants using a multigroup framework, in which all parameters were estimated freely across groups. To test for differences in path coefficients by race, we defined parameters within this model quantifying the difference in path estimates between Black and White women. Significance of these difference parameters provided a test of whether these parameters were equal across groups (i.e., were moderated by race). We limited these analyses to direct effects given the exploratory nature of these analyses and given that we are unaware of reliable methods for computing asymmetric CIs on indirect effect difference scores. Parameter estimates provided by these analyses are provided in Tables S5 and S6 in the Supplemental Material and are described below.
Results
Descriptive statistics
Table 1 provides a summary of key study measures based on the observed data, stratified by sexual orientation group. Table S2 in the Supplemental Material also provides a sample correlation matrix and descriptive statistics of the study variables. Here, we describe general trends present in the observed data, although results reported in the sections below provide a more detailed account based on the imputed data set solutions.
Descriptive Statistics for Study Variables by Sexual Minority Status and Age
Note: DERS = Difficulties in Emotion Regulation Scale (Gratz & Roemer, 2004).
Overall, lesbian and bisexual women reported higher mean levels of stress and emotion dysregulation across study measures, and none of the emotion-dysregulation measures appeared to change substantially over time. Although alcohol, cigarette, and marijuana use tended to increase over time, little evidence indicated that growth differed across sexual orientation groups, although lesbian and bisexual women tended to report higher levels of cigarette and marijuana use at each age. There was evidence of zero inflation and positive skew in alcohol- and marijuana-use variables, particularly at younger ages (see Table S2 in the Supplemental Material), suggesting modeling approaches that accommodate these data features may be most optimal. Although modeling strategies such as zero-inflated count models may be appropriate (e.g., Atkins et al., 2013), they are presently unavailable for the analytic approaches used here. We alternatively used diagonally weighted least squares estimation procedures that can adequately accommodate nonnormal continuous data (Mîndrilă, 2010; Rhemtulla et al., 2012; Schumacker & Beyerlein, 2000).
Unconditional growth-curve models
Analyses began with unconditional latent growth-curve models for emotion dysregulation and substance use to describe whether and how each changed over time. Factor loadings for alcohol, cigarette, and marijuana use at age 21 and 22 were freed to account for anticipated nonlinear increases in use from age 20 to age 21 because of participants in the sample reaching legal drinking age. Linear unconditional growth models for emotion dysregulation and substance use fit well (see Models 1–4 of Table 2). Results are summarized in Table S3 in the Supplemental Material and described below.
Summary Fit Measures for Estimated Growth-Curve Models
Note: CI = confidence interval; GCM = growth-curve model; PPGCM = parallel-process growth-curve model; CFI = comparative fit index; TLI = Tucker-Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual.
The estimated level of emotion dysregulation at age 19 was −0.038; there was no evidence indicating that emotion dysregulation changed over time (b = 0.022, 95% CI = [−0.029, 0.073]). Emotion dysregulation varied in intercept (ψ11 = 2.121, 95% CI = [1.768, 2.475]) but not in slope (ψ22 = 0.058, 95% CI = [−0.137, 0.253]). There was no evidence of covariation between the intercept and slope of emotion dysregulation (ψ21 = 0.048, 95% CI = [−0.174, 0.271], r = .137).
Frequency of alcohol use was indicated by a mean score of 1.790 at age 17 (slightly fewer than “less than 5” instances of alcohol use per year) and increased by 0.381 categories each year from age 17 to 22 (95% CI = [0.355, 0.408]), and there was substantial variation in the intercept (ψ11 = 1.356, 95% CI = [1.23, 1.48]) and slope (ψ22 = 0.097, 95% CI = [0.082, 0.112]). Nonzero covariance between the intercept and slope (ψ21 = −0.149, 95% CI = [−0.182, −0.116], r = −.411) suggested that participants with higher levels of alcohol use at age 17 exhibited less growth through age 22, reflecting that participants with greater use at age 17 remained high through age 22 and that participants lower at age 17 “caught up” by age 22 (see Fig. 1).

Slope-intercept correlations of frequency of alcohol, cigarette, and marijuana use from age 17 to age 22.
Frequency of cigarette use was indicated by a mean score of 2.11 at age 17 (slightly more than “less than 5” instances of cigarette use per year) and increased by 0.175 categories from age 17 to age 22 (95% CI = [0.134, 0.216]). As with alcohol use, there was substantial variation in the intercept (ψ11 = 4.792, 95% CI = [4.237, 5.348]) and slope (ψ22 = 0.350, 95% CI = [0.235, 0.465]) and a negative correlation between the slope and age 17 intercept (ψ21 = −0.326, 95% CI = [−0.547, −0.105], r = −0.252). Participants reporting lower use at age 17 caught up in use by age 22 (see Fig. 1).
Frequency of marijuana use was indicated by a mean score of 1.845 at age 17 (slightly fewer than “less than 5” instances of marijuana use per year) and increased by 0.173 categories from age 17 to age 22 (95% CI = [0.142, 0.204]). There was substantial variation in the intercept (ψ11 = 2.553, 95% CI = [2.204, 2.902]) and slope (ψ22 = 0.260, 95% CI = [0.194, 0.326]) and a negative correlation between the slope and age 17 intercept (ψ21 = −0.268, 95% CI = [−0.390, −0.146], r = −.329), suggesting that participants reporting lower use at age 17 caught up in use by age 22 (see Fig. 1).
Parallel-process growth-curve models
For these models, we set the intercept of emotion dysregulation to age 19 (i.e., growth-rate factor loadings set to [0 1 2]) and the intercept of both substance-use variables to age 20 (i.e., growth-rate factor loadings set to [−3 −2 −1 0 * *]) to ensure that the indirect effects reflected an appropriate temporal sequencing to the extent possible (i.e., sexual-minority status at age 18 predicting social stress at age 17, in turn predicting emotion dysregulation at age 19 and substance use at age 20). Model fit remained adequate for all parallel-process growth-curve models (see Models 5 and 6 of Table 2).
Emotion dysregulation as a link between sexual-minority status and substance use
We first specified a parallel-process growth-curve model regressing level and change in emotion dysregulation and level and change in substance use on sexual-minority status (see Table S4 in the Supplemental Material). This was done to evaluate correlations between these variables while examining differences across sexual orientation groups. Emotion dysregulation at age 19 was modestly associated with more marijuana (b = 0.217, 95% CI = [0.178, 0.257], β = 0.180), cigarette (b = 0.170, 95% CI = [0.119, 0.221], β = 0.103), and alcohol use (b = 0.086, 95% CI = [0.060, 0.112], β = 0.111) at age 20 as well as more rapid growth in marijuana use (b = 0.025, 95% CI = [0.005, 0.045], β = 0.076). Relative to White women, Black women reported higher levels of emotion dysregulation (b = 0.202, 95% CI = [0.056, 0.348], β = 0.068), faster marijuana-use growth (b = 0.061, 95% CI = [0.004, 0.119], β = 0.063), and higher levels of marijuana use by age 20 (b = 0.350, 95% CI = [0.237, 0.463], β = 0.098) but reported lower levels of alcohol (b = −0.680, 95% CI = [−0.756, −0.603], β = −0.294) and cigarette use (b = −0.478, 95% CI = [−0.657, −0.300], β = −0.098). Participants identifying as other or mixed race reported more marijuana use (b = 0.330, 95% CI = [0.056, 0.603], β = 0.043) and less alcohol use (b = −0.467, 95% CI = [−0.666, −0.267], β = −0.094) compared with White women.
Relative to heterosexual women, SMW reported about a half a standard deviation greater emotion dysregulation at age 19 (b = 0.509, 95% CI = [0.278, 0.739]), about one category more frequent marijuana use at age 20 (b = 1.028, 95% CI = [0.806, 1.252], β = 0.182), and about 1.3 categories more frequent cigarette use at age 20 (b = 1.353, 95% CI = [1.010, 1.697], β = 0.174). Regarding group differences in growth for each construct, SMW reported slightly less growth in alcohol use through age 22 (b = −0.063, 95% CI = [−0.121, −0.005], β = −0.064), although they were no different from their heterosexual peers in growth in marijuana or cigarette use. Tests of indirect effects suggested that greater emotion dysregulation partially mediated relations between sexual-minority status and alcohol (a × b = 0.044, 99% CI = [0.017, 0.078]), cigarette (a × b = 0.086, 99% CI = [0.036, 0.145]), and marijuana use (a × b = 0.110, 99% CI = [0.046, 0.176]).
Stress and emotion dysregulation as serial mediators linking sexual-minority status and substance use
We then included sexual-minority status as an exogenous predictor of social stress and growth factors for emotion dysregulation and alcohol, cigarette, and marijuana use. These analyses allowed us to examine direct and indirect effects on these outcomes with the inclusion of social stress. Results are provided in Table 3 and Figure 2. Results were largely identical when analyses were limited to only heterosexual and bisexual women (see Table S5 in the Supplemental Material).
Sexual Minority Status and Social Stress Predicting Alcohol, Marijuana, and Cigarette Use Growth Factors
Note: Fixed parameter estimates, covariate effects, and residual variances are omitted for parsimony. Confidence intervals (CIs) provided are at the 95% level for parameter estimates and the 99% level for indirect effects. Boldface type indicates significant effects of substantive interest are indicated in bold. b = unstandardized estimate; β = standardized estimate.

Final parallel-process growth-curve model regressing growth factors on stress and sexual-minority status. Coefficients presented are unstandardized estimates and their standard errors (in parentheses). Nonsignificant paths, manifest indicators of age 17 stress, and latent variable covariances were omitted for parsimony. SM = sexual minority; MJ = marijuana; Alc. = alcohol; Cig. = cigarette; ED = emotion dysregulation.
Direct effects indicated that Black women reported 0.95 SD greater social stressors compared with White women. Moreover, each standard deviation increase in social stress was associated with a 0.195-SD increase in marijuana use, a 0.198-SD increase in cigarette use, and a 0.229-SD increase in alcohol use at age 20. Tests of two-path indirect effects indicated mediating effects of social stress in the relation between sexual-minority status and alcohol use, marijuana use, and emotion dysregulation. Tests of three-path effects, in turn, indicated a serial mediation pathway of sexual-minority status on marijuana use at age 20 via social stress and emotion dysregulation. Specifically, sexual-minority status was associated with a 0.182-SD increase in social stress at age 17, which in turn was associated with a 0.515-SD increase in emotion dysregulation at age 19. Emotion dysregulation was in turn associated with a 0.080-SD increase in marijuana use at age 20, and this serial pathway explained approximately 2.2% of the effect of sexual-minority status marijuana use, suggesting a modest effect. Sexual-minority status remained a direct predictor of cigarette and marijuana use when social stress was included as a mediator (see Table 3), suggesting partial mediation of these associations via social stress. In contrast, the relation between sexual-minority status and emotion dysregulation was fully mediated by social stress, and status was related only to greater alcohol use through its associations with social stress.
Examining this model separately for White and Black women provided comparable fit among the imputed data sets used in analysis, 3 suggesting this model fit appropriately for both groups. We noted two differences in path estimates across groups (see Table S6 in the Supplemental Material). First, although there was no association between sexual-minority status and alcohol use in the full sample, this association differed across races such that sexual-minority status predicted greater alcohol use among Black women (b = 0.216, 95% CI = [0.045, 0.388], β = 0.069) but was associated with less alcohol use among White women (b = −0.299, 95% CI = [−0.530, −0.068], β = −0.074). Second, emotion dysregulation (and not social stress) predicted cigarette use among Black women (b = 0.167, 95% CI = [0.062, 0.273], β = 0.110), whereas social stress (and not emotion dysregulation) predicted cigarette use among White women (b = 0.902, 95% CI = [0.511, 1.293], β = 0.276). There was no evidence that the magnitude of any other direct effects differed between Black and White women.
Discussion
We examined whether and how stress-related mechanisms of psychopathology—including experiences of social stressors and emotion dysregulation—explained relations between sexual-minority status and substance use in a population-based sample of young women. Prior studies found that emotion dysregulation mediated relations between sexual-minority status and internalizing symptoms (Hatzenbuehler et al., 2008) and stigma-related stress and psychological distress among sexual-minority populations (Hatzenbuehler, 2009; Szymanski & Henrichs-Beck, 2014). The current study extended this work to prospectively examine stress as a link between sexual-minority identity and later emotion dysregulation among youths and to test whether and how these associations explained substance use through young adulthood. Note that our results showed that stress and emotion dysregulation were intervening factors in the relation between sexual-minority status and substance use. Our results suggested that social stressors explained the relation between sexual-minority status and alcohol, cigarette, and marijuana use. Above and beyond this pathway, stress-related increases in emotion dysregulation further explained elevations in the levels (but not growth rates) of marijuana use among SMW in adolescence and young adulthood.
Although alcohol, cigarette, and marijuana use increased through young adulthood in the sample, SMW reported higher levels of cigarette and marijuana use, and only Black SMW reported higher levels of alcohol use. Moreover, we found no evidence that growth rates across substances differed among sexual-minority and heterosexual women from ages 17 to 22. These findings clarify some equivocal results in prior longitudinal studies examining rates of substance use growth across sexual orientation populations, particularly among SMW, across this developmental period. First, prior work using the present sample suggested alcohol, cigarette, and marijuana use increased more rapidly among SMW from ages 13 to 20, and the growth-rate disparity tended to lessen through later adolescence (particularly for alcohol and cigarette use; Dermody et al., 2020). Given that this study focused on a later age range (ages 17–22), our results were consistent with these estimates, adding that little evidence supports a growth disparity between groups in later adolescence and young adulthood.
We note further that our results clarify disparate findings addressing substance-use trajectories among sexual-minority youths reported in prior studies. Several studies using three waves of data from the National Longitudinal Study of Adolescent Health have suggested that individuals who identified as a sexual minority at age 28 reported more alcohol, cigarette, and marijuana use at age 15 and increased use of these substances more rapidly through the follow-up at age 21 compared with heterosexual peers (Marshal et al., 2009). Subsequent work has shown further that SMW specifically showed more rapid increase in alcohol and tobacco use from ages 15 through 28 (Marshal et al., 2012). However, although other research using the same data set has replicated that SMW reported higher levels of heavy drinking, smoking, and marijuana use, this work found no such differences in growth rates for these outcomes (Needham, 2012). Taking these results together with ours and prior estimates provided using the Pittsburgh Girls Study data (Dermody et al., 2020), evidence tends to indicate that growth disparities in substance use may be most pronounced in earlier adolescence (e.g., ages 13–17) compared with later adolescence through young adulthood (e.g., beyond age 17). Future longitudinal work should therefore focus on the early-adolescence to midadolescence period to optimally study the antecedents and consequences of the initiation and escalation of substance use among sexual-minority youths.
Exploring differences across race highlighted several disparities in rates of substance use. First, non-White women (Black and other/mixed race women) relative to White women reported higher levels of marijuana use but lower levels of alcohol and cigarette use. Consistent with some prior work (Keyes et al., 2015), Black women also increased relatively more rapidly in marijuana use compared with White women through young adulthood. These findings highlight that women of color may be at greater risk for certain substance-related problems despite general trends suggesting greater substance use among Whites (e.g., S. E. McCabe et al., 2007). Our findings add that these associations may further differ across substances such that the nature of risk for specific substance disorders may vary as a complex function of intersecting race, sex, and substance (Evans et al., 2017; Vu et al., 2019).
We note further that the intersection between sexual-minority status and race differentiated risk for alcohol use among women such that sexual-minority status was a risk factor among Black women and a protective factor among White women in predicting alcohol use by young adulthood. This finding is consistent with intersectional theories of substance-use risk, highlighting in particular that the confluence of Black, sexual-minority, and female identity may represent a unique risk context (Bowleg, 2008; Crenshaw, 1991; Greene, 1996). We emphasize that these analyses were conducted post hoc in an effort to ensure the generalizability of our findings for majority racial groups in our sample (i.e., White and Black women). We also note that the complexity of the models employed in this work raise caution in drawing definitive race moderation effects. Nonetheless, we consider intersectionality an essential extension of this research and suggest testing this more formally using methods and design most optimally suited for addressing intersectional theories of risk (Bauer, 2014; Else-Quest & Hyde, 2016). Specifically, we encourage that future work use samples with more equal sample sizes among racial- and sexual-minority subgroups to bolster statistical power in detecting interactions quantifying these intersecting identities (Aguinis, 1995). We further advocate that researchers use stress measures that more directly quantify the unique experiences of SMW of color (Bowleg, 2008).
To our knowledge, this research is the first to prospectively investigate the mechanistic roles of social stress and emotion dysregulation in the relation between sexual-minority status and later substance use. Results generally supported intervening processes of risk via social stress and emotion dysregulation, suggesting that interventions promoting stress resilience and emotion regulation may be crucial targets in the remission of problematic substance-use behaviors among SMW with SUD. Several behavioral interventions such as cognitive behavioral therapy (McHugh et al., 2018) or dialectical behavior therapy (Linehan, 1993) provide skills designed explicitly to improve emotion regulation. Given that such interventions have been shown effective in treating SUDs in adolescence (e.g., Babowitch & Antshel, 2016; Dimeff & Linehan, 2008), these may be particularly well-suited treatment modalities for the remission of substance use among young SMW. Moreover, one particular emotion-regulation resource is access to positive social supports, which has been established as among the most robust factors that have buffered mental health risk among sexual-minority youths (Kwon, 2013). Several studies have shown a mitigating role of social support on substance-use risk in sexual-minority samples, including reducing the effect of rejection experiences on substance use (Rosario et al., 2004) and buffering the relation between psychological distress and cigarette smoking (Rosario et al., 2011). Interventions that facilitate social connectedness (e.g., Tsai et al., 2014) may therefore play an especially crucial role in treatment among sexual-minority youths with presenting substance-use problems. Relatedly, community resources promoting social support, including the availability of community resources supporting sexual-minority health, may help reduce substance-use involvement among sexual-minority youths (e.g., Heck et al., 2014).
Limitations
There were several limitations in the current study. Because of limited statistical power, we were unable to test our hypotheses evaluating lesbian and bisexual women as separate groups, although there may be heterogeneity in substance-use behaviors across sexual-minority subgroups. For instance, prevalence rates of alcohol and other drug-use behaviors among sexual-minority subpopulations indicate that bisexual individuals may be the sexual-minority group at highest risk for substance-related problems (e.g., Schuler & Collins, 2020), given that multiple studies have suggested that both bisexual identity and behavior increased risk for alcohol and drug-use problems (Green & Feinstein, 2012). Further evidence is needed to determine whether and how the effects reported in the current study differ when examined within each of these sexual-minority subgroups.
Moreover, a careful consideration of how and when minority status is quantified is essential in addressing these questions with greater nuance in future research. The majority of prior studies (as well as the current study) used a binary indicator sexual identity (e.g., endorsing “not 100% heterosexual”) at a single time point, yet different methods of quantifying sexual-minority identity may produce differing results (e.g., Marshal et al., 2012; Needham, 2012). For instance, one prior study (Talley et al., 2010) tested changes in substance use through adulthood using three separate measures of orientation (identity, attraction, and behavior) reported at either the initial or the final time point of the study and found that different measures and time points revealed different patterns of how marijuana and alcohol use changed over time. We acknowledge that using sexual-minority identity at the time point following our measurement of social stress may have biased our results somewhat compared with the use of an indicator of status at earlier ages (i.e., we could not determine definitively whether sexual-minority participants had come out before reporting social stressors). Indeed, some research has suggested that coming out as a sexual-minority earlier in development may confer positive health benefits in some contexts but health risk factors in others (D’Augelli, 2006; Legate et al., 2012); it is possible that the age of coming out may have a significant impact on both the experiences of minority stress and substance-use risk.
Finally, we note several potential limitations to our study measures and sample that we hope will be addressed in future research. First, the current study used general social stress measures such that none of the items within these measures were stressors due specifically to sexual-minority status or gender. Although we found that SMW reported higher levels of social stress relative to heterosexual women, studies using measures that more directly quantify stressors related to sexual-minority identity will provide a more direct test of mediation proposed in sexual-minority-stress models. Second, although assessing change over time in emotion dysregulation was central to our study hypotheses, we were able to examine only a relatively narrow developmental period of emotion dysregulation (ages 19–21). This may have limited our power to detect individual differences in change over time in this measure as it corresponds with developing substance use. Third, this study focused on analysis of past-year frequency of alcohol and marijuana use, yet whether these findings generalize to other forms of substance use—including problem or disordered use—is an essential avenue of future research. Fourth, this study focused on an urban population-based sample of women. However, it remains an empirical question whether and how the findings reported here generalize to other sexual-minority populations at risk for SUD, including SMW living in rural settings, gay and bisexual men (e.g., S. E. McCabe et al., 2009), and transgender youths and adults (e.g., Day et al., 2017).
Future directions
Although we found evidence of a direct mediating role of emotion dysregulation for all substance-use outcomes in building to our final model, we highlight that social stressors explained this relation when it was included as a mediator for alcohol and cigarette use. We note further that the serial mediation effect for stress and emotion dysregulation was found for only marijuana use, suggesting that social stressors may remain a central intervening factor that explains the link between sexual-minority status and substance use. For instance, it may be that SMW self-medicate using substances in direct response to social stressors (e.g., Johnson et al., 2013). Although our findings support this link at a protracted time course, this question may be most optimally addressed in the context of an ecological momentary assessment design (e.g., Dworkin et al., 2018). For instance, a crucial extension of our findings may be examining whether heightened substance use immediately following experiences of social stressors explains disparity in use among SMW relative to heterosexual women.
These findings also imply that the link between stress and substance-use risk among SMW may be explained by other stress-related mechanisms of risk. The current study focused on emotion dysregulation as a mechanism of risk, although theory describing the risks of early substance use in the general youth population (Hussong et al., 2011) and among sexual-minority populations (e.g., Hatzenbuehler, 2009; Pachankis et al., 2015) have proposed several other stress-related pathways. In addition to addressing limitations described above, we encourage future work to examine these additional pathways, including the presence of co-occurring psychopathology (e.g., internalizing symptoms) or social factors influencing substance-use propensity. We note that to date, prospective research examining these mediators remains sparse (C. J. McCabe & King, 2021). For instance, two prior studies with disparate findings (Dermody et al., 2016; Huebner et al., 2015) tested whether affiliation with deviant peers explained the relation between victimization and substance use among sexual-minority youths, and we know of only two studies (Feinstein & Newcomb, 2016; Marshal et al., 2013) that tested negative affect mechanisms (e.g., coping, enhancement motives, and/or depressive symptoms) as mediators between sexual-minority stress and substance use. Given the dearth of mediation studies addressing these intervening processes, it is essential that future research examine the confluence of these additional mediating processes to best characterize the progression of substance use across development. Nonetheless, the current study highlighted that stress-related increases in emotion dysregulation may serve as one mediating process in the relation between sexual-minority status and marijuana use.
Supplemental Material
sj-pdf-1-cpx-10.1177_2167702621999359 – Supplemental material for Substance Use and Sexual-Minority Status: Examining the Mediating Roles of Stress and Emotion Dysregulation in Young Adult Women
Supplemental material, sj-pdf-1-cpx-10.1177_2167702621999359 for Substance Use and Sexual-Minority Status: Examining the Mediating Roles of Stress and Emotion Dysregulation in Young Adult Women by Connor J. McCabe, Alison E. Hipwell, Kate Keenan, Stephanie Stepp, Tammy Chung and Kevin M. King in Clinical Psychological Science
Footnotes
Acknowledgements
We are grateful to all the families who took part in this study and to the Pittsburgh Girls Study team, which includes interviewers and their supervisors, data managers, student workers, and volunteers.
Transparency
Action Editor: John Curtin
Editor: Kenneth J. Sher
Author Contributions
This publication is the work of the authors, and C. J. McCabe will serve as guarantor for the contents of this article. C. J. McCabe developed the study concept. All authors contributed to the study design. Data collection was performed by A. E. Hipwell, K. Keenan, S. D. Stepp, and T. Chung, and/or their research team. C. J. McCabe performed the data analysis and interpretation under the supervision of K. M. King. C. J. McCabe drafted the manuscript, and all authors provided critical revisions. All of the authors approved the final manuscript for submission.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
