Effects of familial and non-familial warmth during childhood and adolescence on sexual-orientation disparities in alcohol use trajectories and disorder during emerging adulthood

Abstract

Background:

We investigated sexual-orientation differences in typologies of self-reported familial and non-familial warmth in childhood (before age 11) and adolescence (ages 11-17); and tested whether warmth explained sexual minority emerging adults’ (ages 18-25) heightened odds of having heavier alcohol use trajectories (AUTs) and heightened risk for past-year alcohol use disorder (AUD) compared to completely heterosexuals.

Methods:

Using self-reported data from the U.S.-based Growing Up Today Study cohort, latent class analyses identified typologies of familial and non-familial warmth during childhood and adolescence. Multivariable regression models tested our objectives.

Results:

Six warmth classes emerged, including: High-High (i.e., high familial and high non-familial warmth, respectively); High-Moderate; Moderate-Moderate; Moderate-Occasional; Occasional-Occasional; and Low-Low. Among women, sexual minorities had higher odds than completely heterosexuals of being in the Moderate-Moderate, Moderate-Occasional, and Occasional-Occasional versus the High-High warmth class. There were not significant associations between sexual orientation and warmth classes for men. Lower warmth classes were generally associated with greater past-year AUD, and mediated heightened disparities in AUD for sexual minority women versus completely heterosexual women (4.3% mediated), but not among men. Warmth classes were generally unassociated with AUTs, and did not mediate sexual-orientation differences in AUTs.

Conclusions:

Lower warmth was associated with greater alcohol-related problems, but not alcohol use itself. Warmth explained a small proportion of AUD disparities for sexual minority women—but not for men.

1. Introduction

Sexual minority (e.g., gay/lesbian, bisexual, and mostly heterosexual) emerging adults (i.e., aged 18-25 years) are at greater risk for alcohol use, heavy alcohol use trajectories, and alcohol use disorder than their heterosexual peers, especially among women (Corliss et al., 2008; Coulter et al., 2018; Coulter et al., 2016; Eisenberg and Wechsler, 2003; Goldberg et al., 2013). Studies on the mechanisms of these disparities have largely focused on risk factors (e.g., bullying victimization), and much less attention has been given to protective factors (Kidd et al., 2018). Therefore, the current study will focus on the potential protective factor of warmth, and whether warmth serves as a mechanism (i.e., mediator) in the production of sexual-orientation disparities in alcohol use trajectories and alcohol use disorder.

1.1. Warmth and Sexual Orientation

Warmth refers to demonstrations and expressions of praise, closeness, love, and affection (Amato, 1990; Zhou et al., 2002), and can be provided by adult family members (e.g., parents) as well as non-family members (e.g., teachers, community leaders). The presence of warmth from adults is associated with healthy child and adolescent development as shown in both empirical research (Kincaid et al., 2012; Resnick et al., 1997; Williamson et al., 2017; Yap and Jorm, 2015) and theory (e.g., social development model; Hawkins and Weis, 1985).

Certain youth populations may experience less warmth than others. Based on the theories of minority stress and stigma as a fundamental cause of health (Hatzenbuehler et al., 2013; Meyer, 2003), sexual minority youth (SMY) are hypothesized to be less likely to experience warmth than heterosexual youth because SMY’s sexual orientation is often stigmatized. Empirical research found mixed results for sexual-orientation differences in familial warmth: most studies show that SMY report lower familial warmth than do heterosexual youth (Eisenberg and Resnick, 2006; Needham and Austin, 2010; Pearson and Wilkinson, 2013; Power et al., 2015; Stone et al., 2014); but other research found no differences in familial warmth by sexual orientation (Johnson et al., 2011; Martin-Storey and Crosnoe, 2012). Additionally, research suggests that boys report receiving less warmth (i.e., maternal affection) than girls (Murphy et al., 2010; Rosario et al., 2014a; Rosario et al., 2014b), and gender may modify the association between sexual orientation and familial warmth: one study found the presence of sexual-orientation disparities in parental closeness for girls, but not for boys (Fish and Russell, 2018).

Non-familial warmth is also associated with positive outcomes, such as self-esteem (Hurd et al., 2018; Snapp et al., 2015; Watson et al., 2019). However, few studies have examined sexual-orientation differences in non-familial warmth. Nevertheless, existing studies show that the presence of non-familial warmth is lower among SMY than heterosexual youth (Eisenberg and Resnick, 2006; Stone et al., 2014). Studies on both non-familial and familial warmth often only examine sexual-orientation differences in warmth from a single developmental period (i.e., adolescence)—and childhood warmth is often unexamined, despite warmth being known to vary across time (e.g., maternal warmth usually decreases from childhood to adolescence; Trentacosta et al., 2011; Wang et al., 2011).

To provide a more comprehensive view of warmth, it is worth identifying typologies of people with unique patterns of familial and non-familial warmth (e.g., low familial and low non-familial warmth), and whether there are sexual-orientation differences in memberships to these warmth typologies. One way to examine sexual-orientation differences in both familial and non- familial warmth is to use a combination of person-centered approaches (e.g., latent class analysis) and variable-centered approaches (e.g., regression analyses). Person-centered approaches can classify similar types of individuals into unique subgroups, which are unable to be captured by a single variable or scale (Howard and Hoffman, 2018). These approaches, including latent class analyses, can create emergent subgroup populations based on patterns derived from many variables (Howard and Hoffman, 2018). Therefore, person-centered approaches can be particularly useful for examining typologies of both familial and non-familial warmth across both childhood and adolescence. Subsequently, these person-centered approaches can be combined with variable-centered approaches (e.g., regression) to examine the associations between variables (Murdock and Miller, 2003), such as how sexual orientation is associated with different warmth typologies. Using a combination of person-centered and variable-centered approaches can increase our knowledge about how sexual orientation is related to warmth typologies, which can be based on multiple variables such as familial and non-familial warmth in both childhood and adolescence.

1.2. Warmth and Alcohol Use Trajectories/Disorder

Warmth can benefit youth in multiple ways: warmth can assist youth in developing healthy coping mechanisms and self-regulation (i.e., guiding one’s own cognitive, emotional, and behavioral processes to achieve goals), as well as decreasing their pro-substance use norms (e.g., social approval of substance use), thereby reducing consumption of alcohol (Donaldson et al., 2016; Eiden et al., 2009; Eiden et al., 2007). Cross-sectional studies of adolescents show that having familial and non-familial warmth is associated with lower alcohol use (Calafat et al., 2014; Johnson and Pandina, 1991; Kelly et al., 2011; Latendresse et al., 2008; Mogro-Wilson, 2013; Mongro-Wilson, 2007; Resnick et al., 1997; Vazsonyi et al., 2015). However, during emerging adulthood, cross-sectional analyses found mixed results for the associations between familial warmth and alcohol use: one study found null associations (Cleveland et al., 2014), and another found protective associations only for certain subgroups (Luk et al., 2015). Furthermore, longitudinal studies found that familial and non-familial warmth from earlier periods in the life-course were associated with lower alcohol use in later periods, including emerging adulthood (Donaldson et al., 2016; Eiden et al., 2016; McNeely and Falci, 2004). Yet, it remains unknown whether familial and non-familial warmth in childhood and adolescence are associated with alcohol use trajectories (AUTs; i.e., alcohol use over time) throughout emerging adulthood. No studies to our knowledge have examined how simultaneously having low familial and low non-familial warmth during early periods of the life course may work together to increase the risk of heavy AUTs later in life. If having lower warmth in both areas simultaneously are associated with heavier AUTs, then interventions that increase all types of warmth during childhood and adolescence (which currently do not exist; Coulter et al., 2019) may reduce the likelihood of having heavy AUTs throughout emerging adulthood. Such interventions may also reduce the many short- and long-term negative consequences of chronic moderate and heavy alcohol use.

A common sequela of heavy alcohol use is the development of an alcohol use disorder (AUD; e.g., abuse or dependence; Chassin et al., 2004; Nelson et al., 2015; Sher et al., 2004). However, few studies have examined the effects of warmth on AUD. One study found that low maternal warmth during adolescence was associated with increased odds of having comorbidity of AUD and mental health disorders at age 21 years (Salom et al., 2015). Yet most studies found that familial warmth was not associated with AUD or AUD-comorbidity in adolescence (Barnow et al., 2002; Greenfield et al., 2016) and emerging adulthood (Salom et al., 2016). However, previous studies have examined the effects of familial warmth from a single time period on AUD, thereby limiting knowledge about how warmth from multiple contexts (e.g., familial and non-familial) and during earlier time periods (e.g., childhood and adolescence) are associated with AUD in emerging adulthood. Such a study would provide a richer description of how warmth from earlier periods is associated with AUD in emerging adulthood.

Research has shown that sexual minority emerging adults evidence greater risk of alcohol use, heavy AUTs, and AUD than their heterosexual peers, especially among women (Corliss et al., 2008; Coulter et al., 2018; Coulter et al., 2016; Eisenberg and Wechsler, 2003; Goldberg et al., 2013). These differences may be partially explained by SMY having lower warmth than heterosexuals during childhood and adolescence. Support for this hypothesis would suggest that increasing warmth for SMY would mitigate sexual-orientation disparities in AUTs and AUD, thereby informing future intervention studies and prevention efforts.

1.3. The Current Study

This paper sought to address the aforementioned gaps in research on warmth and alcohol outcomes across the first three decades of the life-course using data from the Growing Up Today Study. First, we estimated latent classes of self-reported familial and non-familial warmth during childhood and adolescence. We hypothesized there would be different warmth classes characterized by high and low warmth. Second, we tested for sexual-orientation differences in warmth class memberships, and whether these differences varied by gender. We hypothesized that sexual minority populations would be more likely to be members of classes exemplifying less warmth, and these differences would be larger among girls. Third, we examined the effects of warmth classes on sexual-orientation differences in AUTs and AUD in emerging adulthood, and tested whether warmth classes mediated the sexual-orientation disparities in AUTs and AUD in emerging adulthood. We hypothesized that lower warmth classes would be associated with heavier AUTs and greater risk of AUD, and that warmth classes would mediate the sexual-orientation differences in AUTs and AUD.

2. Methods

We used data from the Growing Up Today Study (GUTS), a longitudinal cohort study. In 1996, GUTS enrolled 16,875 participants aged 9-14 years who were children of participants in the Nurses’ Health Study II—which is a cohort of 116,430 female registered nurses from 14 U.S. states begun in 1989. GUTS participants completed surveys, originally on an annual basis and every 2-3 years from 2001-2010. Additional GUTS information is reported elsewhere (Corliss et al., 2008; Field et al., 1999). The current study included participants who provided information on sexual orientation and at least 1 of the 4 familial or non-familial warmth items, measured in the 2007 survey wave when participants’ average age was 22.7 years (range: 19-27 years). Our analytic sample included 9,095 participants (5,783 women; 3,312 men), representing 53.9% of the cohort. Brigham and Women’s Hospital Institutional Review Board approved this study.

2.1. Measurements

Warmth was assessed on the 2007 questionnaire with 4 items adapted from the Childhood Trauma Questionnaire (CTQ; Bernstein et al., 1994; Bernstein et al., 2003). The original CTQ item assessed familial warmth, and GUTS added a similar item to assess non-familial warmth. GUTS adapted the items to assess 2 developmental periods—childhood (before age 11 years) and adolescence (11-17 years of age). For example, we assessed childhood familial warmth using the following question: “When you were a child (before age 11), how often did someone in your family make you feel that you were important or special?” We modified this item to measure childhood non-familial warmth (by stating “someone who was NOT a family member”), teenage familial warmth (by stating “when you were a teenager (ages 11–17)”), and teenage non-familial warmth. We ordinally coded all 4 items’ response options: never; rarely; sometimes; often; and very often.

Sexual orientation was assessed at each survey wave from 1999-2010: “Which one of the following best describes your feelings?” Response options included: completely heterosexual (attracted to persons of the opposite sex); mostly heterosexual; bisexual (equally attracted to men and women); mostly homosexual; completely homosexual (gay/lesbian, attracted to persons of the same sex); and unsure. We classified sexual orientation based on participants’ most recent report. Like prior research (Corliss et al., 2008), we removed participants who were “unsure” of their sexual orientation because of small sample size (n=6). For our primary analyses, we combined completely homosexual, mostly homosexual, bisexual, and mostly heterosexual into a single group (henceforth referred to as sexual minority) to increase statistical powers. Secondarily, we conducted analyses using a variable that highlights the subgroups of sexual minorities (results located in the appendices)^*.

AUT classes were derived using longitudinal latent class analyses (LLCA; Coulter et al., 2018). We conducted LLCA on data from the emerging adulthood period (18-25 years) with past-year average frequency of drinking, past-year average quantity of drinking per episode, and past-year frequency of binge drinking. Six AUT classes emerged for women, and five emerged for men: these included heavy, moderate, escalation-to-moderately-heavy, light (for women only), legal (drinking onset at age 21), and non-drinkers.

Probable past-year AUD was assessed in 2010 (when participants’ mean age was 25.3 years) with items assessing symptoms based on the DSM-IV (American Psychiatric Association, 1994), as adapted by the National Survey on Drug Use and Health (U.S. Department of Health and Human Services, 2009). The DSM-IV described two distinct disorders, alcohol abuse and alcohol dependence, which were present in 9.9% and 10.2% of participants, respectively. If participants had alcohol abuse or dependence, we coded them as having a probable AUD (producing a binary variable). We based this on extant research (Cochran et al., 2007; Hatzenbuehler et al., 2009) and because the DSM-5 (American Psychiatric Association, 2013) combines abuse and dependence into one solitary disorder.

Demographics included: gender (natal female versus male; measured at baseline); race/ethnicity (White vs. non-White; measured at baseline); region of residence (West vs. Midwest, Southwest, and Northeast; measured in 2007); and age in years (calculated based on participant’s birthdate and date of the 2007 questionnaire return). Covariates included lifetime college attendance (any college attendance versus none; measured in 2010) and lifetime pregnancy (yes/no) for women (measured prospectively from 1999-2010). For covariates, we used the missing indicator method (Horton and Kleinman, 2007), permitting analysis of all available data and preservation of statistical power.

2.2. Analyses

2.2.1. Warmth Classes.

Following Masyn’s guidelines (2013), we used a classify-analyze approach to characterize latent classes of familial and non-familial warmth and estimate independent and dependent variables associated with these classes. First, we estimated warmth classes using latent class analysis in Mplus version 7.2 (Los Angeles, CA), which is a person-centered approach that allows for the estimation of subgroups who differ across multiple indicators of warmth. We selected latent class (instead of latent profile) analysis because our variables were ordinal and skewed. We estimated the unconditional model with the 4 ordinal warmth variables using the robust maximum likelihood estimator (Muthén and Shedden, 1999; Yuan and Bentler, 2000). We used the complex survey analysis procedure to account for non-independence of familial clusters (Muthén and Muthén, 1998-2012).

We estimated 1- through 8-class solution models. To determine the best-fitting number of classes, we examined the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), the Bayes Factor (BF), the correct model probability (cmP), and the Vuong-Lo-Mendell-Rubin adjusted likelihood ratio test (VLMR LRT; Kass and Raftery, 1995; Masyn, 2013; Muthén and Muthén, 1998-2012; Nylund et al., 2007). We considered the best fitting model to have a low BIC, a significant improvement in fit over the previous model (based on VLMR LRT), and the highest interpretative validity (Masyn, 2013; Muthén and Shedden, 1999; Nylund et al., 2007). We examined entropy, and considered good latent class separation and assignment as classes with average posterior probabilities>0.7 and odds of correct classification>5 (Masyn, 2013; Nagin, 2005). We examined class patterns separately by gender but results were conceptually similar (results available upon request); therefore, we reported results from the total sample. We assigned participants to the class for which they had the highest posterior probability of membership.

2.2.2. Differences in Warmth Class Membership.

We used chi-square tests accounting for familial clusters (Rao and Scott, 1984) in SAS version 9.4 (Cary, NC) to examine the bivariate associations of warmth classes with sexual orientation. We used multinomial logistic regression models to examine sexual-orientation differences in warmth class membership, controlling for demographics, using generalized estimating equations (GEE) to account for familial clusters. We added sexual-orientation-by-gender interaction terms to test effect modification of sexual orientation by gender.

2.2.3. Associations with AUTs and AUD.

We engaged in a two-step model building process to examine whether warmth mediates sexual-orientation differences in AUTs and AUD. The first model contained the main effects of sexual orientation on the outcome (controlling for demographics and covariates); and the second model added warmth class as an independent variable to the first model. When warmth was significantly associated with sexual orientation and the outcome, we conducted a formal test of mediation using the %MEDIATE macro (Hertzmark et al., 2012). We stratified analyses by gender because AUTs differ by gender (Coulter et al., 2018) and sexual-orientation disparities in AUD are larger for women than men (Goldberg et al., 2013).

For AUTs, multinomial logistic regression models tested the associations with trajectory class membership, controlling for demographics and covariates, using GEE. Reference groups in the models were non-drinkers (versus other trajectory classes). For the binary AUD variable, we fit modified Poisson regression models (Zou, 2004) with GEE.

3. Results

Table 1 describes the sample composition by gender and sexual orientation. Table 2 shows the results of latent class analyses for the warmth variables. We selected the 6-class model based on interpretability and the fit indices. There was significant improvement in model fit from the 5- to 6-class model, but a non-significant improvement from the 6- to 7-class model. The final model had highest entropy, good separation, and adequate class assignment.

Table 1.

Characteristics of the sample by sexual orientation, stratified by gender: Growing Up Today Study

	Sexual Orientation
	Completely Heterosexual			Sexual Minority
	n	(%)		n	(%)	p
WOMEN
Total, row percentage	4,642	(80.3)		1,141	(19.7)
Race/ethnicity
White	4,362	(94.0)		1,042	(91.3)	0.004
Non-White	280	(6.0)		99	(8.7)
Region
West	708	(15.3)		235	(20.6)	<.001
Midwest	1,634	(35.2)		312	(27.3)
South	793	(17.1)		193	(16.9)
Northeast	1,507	(32.5)		401	(35.1)
College Attendance
Never Attended	95	(2.4)		46	(4.9)	0.001
Attended	3,859	(97.6)		902	(95.2)
Lifetime Pregnancy
No	3,158	(78.8)		748	(76.7)	0.163
Yes	849	(21.2)		227	(23.3)
Age, mean (sd)	22.7	(1.7)		22.7	(1.7)	0.602
MEN
Total, row percentage	2,930	(88.5)		382	(11.5)
Race/ethnicity
White	2748	(93.8)		337	(88.2)	0.001
Non-White	182	(6.2)		45	(11.8)
Region
West	498	(17.0)		76	(19.9)	0.205
Midwest	1,036	(35.4)		115	(30.1)
South	434	(14.8)		61	(16.0)
Northeast	962	(32.8)		130	(34.0)
College Attendance
Never Attended	85	(4.0)		9	(2.9)	0.322
Attended	2,058	(96.0)		299	(97.1)
Age, mean (sd)	22.6	(1.7)		22.6	(1.7)	0.583

Note. Column percentages are presented throughout the table except where noted. P-values were derived using Rao-Scott chi-squared tests for categorical variables and univariable models with generalized estimating equations for age, both of which adjusted for sibling clusters. Missing data for college attendance and pregnancy were excluded from this table, including the Rao-Scott chi-squared tests. sd=standard deviation. Sexual orientation was based on participants' last self-report from 1999-2010, wherein sexual minority includes mostly heterosexuals, bisexuals, and gays/lesbians; gender and race/ethnicity were assessed in 1996 at baseline; region was assessed in 2007; college attendance was assessed in 2010; lifetime pregnancy was assessed prospectively from 1999-2010; age was assessed in 2007.

Table 2.

Class enumeration fit indices and qualities for latent class analyses for familial and non-familial warmth during childhood and adolescence among total sample: Growing Up Today Study

Classes	Free parameters	Log- Likelihood	AIC	BIC	SSA -BIC	X2 model fit	X2 df	X2 p- value	Entropy	B F	cm P	AWE	VLM R LRT p- value
1	16	−46,607	93,246	93,360	93,310	10,123.11	600	<0.001	n/a	<1	0.00	93,554	n/a
2	33	−40,254	80,575	80,810	80,705	7,680.45	583	<0.001	0.889	<1	0.00	81,209	<0.001
3	50	−37,879	75,857	76,213	76,054	6,079.23	571	<0.001	0.849	<1	0.00	76,820	<0.001
4	67	−36,923	73,979	74,456	74,243	5,052.32	555	<0.001	0.852	<1	0.00	75,269	<0.001
5	84	−35,939	72,047	72,644	72,377	4,505.35	538	<0.001	0.908	<1	0.00	73,662	<0.001
6	101	−35,336	70,874	71,592	71,271	3,662.29	520	<0.001	0.932	<1	0.00	72,816	<0.001
7	118	−35,066	70,368	71,208	70,833	3,212.82	503	<0.001	0.907	<1	0.00	72,637	0.764
8	135	−34,832	69,935	70,895	70,466	2,658.99	486	<0.001	0.908	n/a	1.00	72,530	0.760

Note. Models were estimated with a sample of 5,786 women and 3,316 men. Warmth was assessed in 2007. Boldface indicates the model we selected. AIC=Akaike information criterion; BIC=Bayesian information criterion; SSA-BIC=Sample size-adjusted Bayesian information criterion; BF=Bayes factor; cmP=Correct model probability; AWE=Approximate weight of evidence criterion; VLMR LRT=Vuong-Lo-Mendell-Rubin likelihood ratio test; df=degrees of freedom; n/a=not applicable. Bootstrap Likelihood Ratio Test could not be estimated because we employed the COMPLEX command in Mplus to adjust for non-independence within sibling clusters.

Figure 1 depicts the 6 latent classes of warmth. Classes included: high familial and high non-familial warmth (henceforth, High-High); high familial and moderate non-familial warmth (henceforth, High-Moderate); moderate familial and moderate non-familial warmth (henceforth, Moderate-Moderate); moderate familial and occasional non-familial warmth (henceforth, Moderate-Occasional); occasional familial and occasional non-familial warmth (henceforth, Occasional-Occasional); and low familial and low non-familial warmth (henceforth, Low-Low).

Figure 1.

Latent class analysis profile plots for familial and non-familial warmth in childhood and adolescence: Growing Up Today Study.

3.1. Differences in Warmth Class Membership

Table 3 shows the unadjusted percentages of membership in each warmth class by sexual orientation for women and men separately. In a multivariable model including the main effects of gender and sexual orientation on warmth class membership (adjusting for demographics) showed that men had higher odds than women of being in the High-Moderate (OR=1.45; p<0.001), Moderate-Moderate (OR=1.93; p<0.001), Moderate-Occasional (OR=3.15; p<0.001), Occasional-Occasional (OR=2.64; p<0.001), and Low-Low (OR=1.94; p<0.001) warmth classes versus the High-High warmth class (results not shown). Multivariable models including gender-by-sexual-orientation interactions showed that all of the interaction effects were statistically significant with sexual-orientation differences in warmth class membership being smaller for men than women (ORs range: 0.43-0.64; all p-values<0.05; results not shown). Therefore, we present final models stratified by gender.

Table 3.

Bivariate associations between sexual orientation and warmth classes, stratified by gender: Growing Up Today Study

	Warmth Classes
	High Familial & High Non- Familial (High-High)			High Familial & Moderate Non-Familial (High- Moderate)			Moderate Familial & Moderate Non-Familial (Moderate- Moderate)			Moderate Familial & Occasional Non- Familial (Moderate- Occasional)			Occasional Familial & Occasional Non-Familial (Occasional- Occasional)			Low Familial & Low Non- Familial (Low-Low)
	n	(%)		n	(%)		n	(%)		n	(%)		n	(%)		n	(%)		p
WOMEN
Total	1,285	(22.2)		1,466	(25.4)		1,109	(19.2)		966	(16.7)		726	(12.6)		231	(4.0)
Sexual Orientation
Completely Heterosexual	1098	(23.7)		1218	(26.2)		888	(19.1)		729	(15.7)		515	(11.1)		194	(4.2)		<.001
Sexual Minority	187	(16.4)		248	(21.7)		221	(19.4)		237	(20.8)		211	(18.5)		37	(3.2)
MEN
Total	401	(12.1)		657	(19.8)		651	(19.6)		906	(27.4)		556	(16.8)		141	(4.3)
Sexual Orientation
Completely Heterosexual	352	(12.0)		599	(20.4)		576	(19.7)		803	(27.4)		468	(16.0)		132	(4.5)		0.002
Sexual Minority	49	(12.8)		58	(15.2)		75	(19.6)		103	(27.0)		88	(23.0)		9	(2.4)

Note. Row percentages are presented throughout the table. P-values were derived using Rao-Scott chi-squared tests adjusting for sibling clusters. Sexual orientation was based on participants' last self-report from 1999-2010, wherein sexual minority includes mostly heterosexuals, bisexuals, and gays/lesbians; gender was assessed in 1996 at baseline; warmth was assessed in 2007.

Among women (Table 4), sexual minorities had higher odds than completely heterosexuals of being in the Moderate-Moderate, Moderate-Occasional, and Occasional-Occasional warmth classes than the High-High warmth class. Among men, there were no significant differences in warmth class memberships for sexual minorities compared with completely heterosexual men.

Table 4.

Results of multinomial logistic regression models testing sexual-orientation differences in warmth class memberships, stratified by gender: Growing Up Today Study

	High-Moderate		Moderate-Moderate		Moderate-Occasional		Occasional-Occasional		Low-Low
	versus		versus		versus		versus		versus
	High-High Warmth Class		High-High Warmth Class		High-High Warmth Class		High-High Warmth Class		High-High Warmth Class
	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
WOMEN
Sexual Orientation
Completely Heterosexual	1.00 (referent)		1.00 (referent)		1.00 (referent)		1.00 (referent)		1.00 (referent)
Sexual Minority	1.18 (0.96, 1.46)	0.107	1.48 (1.20, 1.84)	<.001	1.92 (1.55, 2.38)	<.001	2.36 (1.88, 2.95)	<.001	1.12 (0.76, 1.64)	0.571
MEN
Sexual Orientation
Completely Heterosexual	1.00 (referent)		1.00 (referent)		1.00 (referent)		1.00 (referent)		1.00 (referent)
Sexual Minority	0.71 (0.47, 1.06)	0.096	0.94 (0.64, 1.38)	0.754	0.93 (0.65, 1.33)	0.676	1.34 (0.92, 1.96)	0.126	0.48 (0.23, 1.01)	0.053

Note. Boldface indicates statistical significance (p<0.05). Models were estimated with samples of 5,783 women and 3,312 men. Models adjusted for race/ethnicity (assessed in 1996 at baseline), age (assessed in 2007), and region (assessed in 2007). Sexual orientation was based on participants' last self-report from 1999-2010, wherein sexual minority includes mostly heterosexuals, bisexuals, and gays/lesbians; gender was assessed in 1996 at baseline; warmth was assessed in 2007. High-High=High familial & high non-familial warmth; High-Moderate=High familial & moderate non-familial warmth; Moderate-Moderate=Moderate familial & moderate non-familial warmth; Moderate-Occasional=Moderate familial & occasional non-familial warmth; Occasional-Occasional=Occasional familial & occasional non-familial warmth; Low-Low=Low familial & low non-familial warmth; OR=odds ratio; CI=confidence interval.

3.2. Unadjusted Associations with AUTs and AUD

Table 5 shows the unadjusted percentages depicting associations of sexual orientation and warmth on AUTs and past-year AUD stratified by gender. Sexual orientation was significantly associated with AUTs and AUD. Warmth was significantly associated with AUD but not AUTs.

Table 5.

Alcohol use trajectories and disorders by sexual orientation and warmth classes, stratified by gender: Growing Up Today Study

	Alcohol Use Trajectory Classes from Ages 18-25								Alcohol Use Disorder
	Non Drinkers	Heavy Drinkers	Moderate Drinkers	Legal Drinkers	Escalation-to- Moderately- Heavy Drinkers	Light Drinkers			Past-Year Prevalence
	%	%	%	%	%	%	p		%	p
WOMEN
Total	7.4	22.9	31.3	11.0	10.4	17.0			16.8
Sexual Orientation
Completely Heterosexual	8.2	21.6	30.9	12.0	10.3	17.0	<.001		14.0	<.001
Sexual Minority	4.1	28.1	33.2	6.8	10.8	17.0			28.3
Warmth Classes
High-High	7.7	23.4	30.5	10.8	9.6	18.0	0.207		13.5	<.001
High-Moderate	6.8	23.5	32.0	10.0	11.9	15.8			15.6
Moderate-Moderate	6.7	22.7	32.5	13.3	9.9	14.9			15.0
Moderate-Occasional	7.9	21.1	32.6	10.6	10.2	17.5			20.5
Occasional-Occasional	8.7	22.2	29.1	10.4	10.0	19.6			22.4
Low-Low	7.4	26.5	27.4	10.0	9.6	19.1			19.7
MEN
Total	9.2	35.3	25.4	16.6	13.5				27.3
Sexual Orientation
Completely Heterosexual	9.8	35.8	23.9	16.9	13.6		<.001		25.8	<.001
Sexual Minority	4.7	31.3	36.8	14.2	12.9				37.8
Warmth Classes
High-High	11.5	31.8	24.0	19.5	13.3		0.288		23.4	0.010
High-Moderate	9.0	36.9	26.5	15.0	12.7				24.7
Moderate-Moderate	9.2	34.5	25.8	16.7	13.8				30.8
Moderate-Occasional	7.4	38.1	24.1	16.1	14.4				30.7
Occasional-Occasional	10.3	30.9	27.5	17.9	13.4				26.0
Low-Low	11.4	40.4	22.7	13.5	12.1				17.1

Note. Row percentages are presented throughout the table. Univariable p-values were derived using Rao-Scott chi-squared tests adjusting for sibling clusters. Sexual orientation was based on participants' last self-report from 1999-2010, wherein sexual minority includes mostly heterosexuals, bisexuals, and gays/lesbians; gender was assessed in 1996 at baseline; warmth was assessed in 2007; alcohol use trajectories were derived using longitudinal latent class analyses from participants' past-year alcohol frequency, quantity, and heavy episodic drinking from 2003-2010 when they were aged 18-25 years (the light drinker class only emerged among women); past-year alcohol use disorder was assessed in 2010. For alcohol use trajectories, we analyzed data from 5,764 women and 3,284 men. For alcohol use disorder, we analyzed data from 4,620 women and 2,158 men. High-High=High familial & high non-familial warmth; High-Moderate=High familial & moderate non-familial warmth; Moderate-Moderate=Moderate familial & moderate non-familial warmth; Moderate-Occasional=Moderate familial & occasional non-familial warmth; Occasional-Occasional=Occasional familial & occasional non-familial warmth; Low-Low=Low familial & low non-familial warmth.

3.3. Multivariable Models for AUTs

Among women, sexual minorities had higher odds than completely heterosexual participants of being heavy, moderate, escalation-to-moderately-heavy, and light drinkers versus non-drinkers (Table 6; Model 1). Warmth classes were not significantly associated with AUTs (Model 2; controlling for sexual orientation, demographics, and covariates), thereby not mediating sexual-orientation differences in AUTs for women.

Table 6.

Results of multinomial logistic regression models testing the associations of sexual orientation and warmth classes on longitudinal alcohol use trajectories, stratified by gender: Growing Up Today Study

	Alcohol Use Trajectories (Referent=Non-drinkers)
	Heavy drinkers		Moderate drinkers		Legal drinkers		Escalation-to-Moderately- Heavy drinkers		Light drinkers
	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
WOMEN
Model 1: Controlling for Demographics
Sexual Orientation (Referent=Completely Heterosexual)
Sexual Minority	2.65 (1.89, 3.70)	<.001	2.16 (1.56, 3.00)	<.001	1.13 (0.76, 1.67)	0.547	2.11 (1.46, 3.04)	<.001	2.01 (1.42, 2.84)	<.001
Model 2: Controlling for Demographics and Warmth Classes
Sexual Orientation (Referent=Completely Heterosexual)
Sexual Minority	2.76 (1.97, 3.86)	<.001	2.22 (1.59, 3.09)	<.001	1.14 (0.77, 1.70)	0.500	2.17 (1.50, 3.14)	<.001	2.04 (1.44, 2.89)	<.001
Warmth Classes (Referent=High-High)
High-Moderate	1.11 (0.80, 1.53)	0.530	1.17 (0.85, 1.59)	0.333	1.04 (0.73, 1.50)	0.814	1.39 (0.96, 1.99)	0.079	0.98 (0.70, 1.37)	0.891
Moderate-Moderate	1.04 (0.73, 1.47)	0.844	1.16 (0.83, 1.62)	0.398	1.39 (0.95, 2.04)	0.092	1.12 (0.75, 1.68)	0.574	0.90 (0.63, 1.30)	0.583
Moderate-Occasional	0.79 (0.56, 1.13)	0.200	0.97 (0.69, 1.36)	0.874	0.96 (0.65, 1.43)	0.840	0.98 (0.65, 1.47)	0.917	0.89 (0.62, 1.28)	0.519
Occasional-Occasional	0.74 (0.50, 1.08)	0.121	0.77 (0.53, 1.11)	0.162	0.84 (0.55, 1.30)	0.439	0.85 (0.55, 1.32)	0.468	0.87 (0.59, 1.29)	0.502
Low-Low	1.11 (0.62, 2.00)	0.725	0.89 (0.50, 1.60)	0.699	0.93 (0.47, 1.84)	0.836	0.98 (0.49, 1.95)	0.955	1.06 (0.57, 1.95)	0.861
MEN
Model 3: Controlling for Demographics
Sexual Orientation (Referent=Completely Heterosexual)
Sexual Minority	1.84 (1.10, 3.09)	0.020	3.26 (1.96, 5.44)	<.001	1.71 (0.98, 2.97)	0.059	1.94 (1.11, 3.41)	0.021
Model 4: Controllina for Demographies and Warmth Classes
Sexual Orientation (Referent=Completely Heterosexual)
Sexual Minority	1.89 (1.13, 3.18)	0.016	3.30 (1.97, 5.52)	<.001	1.69 (0.97, 2.95)	0.0640	1.95 (1.11, 3.45)	0.021
Warmth Classes (Referent=High-High)
High-Moderate	1.58 (1.01, 2.46)	0.044	1.52 (0.96, 2.41)	0.073	1.01 (0.62, 1.63)	0.977	1.29 (0.77, 2.16)	0.338
Moderate-Moderate	1.42 (0.90, 2.22)	0.128	1.39 (0.87, 2.21)	0.168	1.08 (0.66, 1.75)	0.762	1.32 (0.78, 2.23)	0.299
Moderate-Occasional	2.00 (1.29, 3.11)	0.002	1.67 (1.05, 2.63)	0.029	1.35 (0.84, 2.16)	0.214	1.82 (1.10, 3.01)	0.019
Occasional-Occasional	1.12 (0.71, 1.79)	0.622	1.27 (0.79, 2.05)	0.320	1.06 (0.64, 1.73)	0.830	1.19 (0.70, 2.03)	0.523
Low-Low	1.32 (0.69, 2.54)	0.402	1.01 (0.50, 2.06)	0.968	0.72 (0.34, 1.52)	0.385	0.95 (0.43, 2.09)	0.904

Note. Boldface indicates statistical significance (p<0.05). Light drinkers were only present among women. Models were estimated with samples of 5,764 women and 3,284 men. All models adjusted for race/ethnicity, age, region, and lifetime college attendance. Models for women also adjusted for lifetime pregnancy. High-High=High familial & high non-familial warmth; High-Moderate=High familial & moderate non-familial warmth; Moderate-Moderate=Moderate familial & moderate non-familial warmth; Moderate-Occasional=Moderate familial & occasional non-familial warmth; Occasional-Occasional=Occasional familial & occasional non-familial warmth; Low-Low=Low familial & low non-familial warmth; OR=odds ratio; CI=confidence interval.

Among men, sexual minorities had higher odds than completely heterosexual participants of being heavy, moderate, and escalation-to-moderately-heavy drinkers versus non-drinkers (Table 6; Model 3). A few warmth classes were significantly associated with AUTs (Model 4). Men in the High-Moderate warmth class had higher odds than men in the High-High warmth class of being heavy drinkers versus non-drinkers. Men in the Moderate-Occasional warmth class had higher odds than men in High-High warmth class of being heavy, moderate, and escalation-to-moderately-heavy drinkers versus non-drinkers. However, warmth did not mediate sexual-orientation differences in AUTs because the sexual-orientation effects were not attenuated from Models 3 to 4.

3.4. Multivariable Models for AUD

Among women, sexual minorities were 2.03 times more likely than completely heterosexual participants to meet criteria for probable AUD (Table 7; Model 1). Moderate-Occasional, Occasional-Occasional, and Low-Low warmth classes were 1.44-1.56 times more likely to report probable AUD than High-High warmth class participants (Model 2). Warmth classes mediated sexual-orientation differences in AUD for sexual minorities compared to completely heterosexuals (mediated proportion=4.3%; p-values=0.003). After controlling for warmth classes, sexual minority women remained more likely than completely heterosexual women to evidence AUD (Model 2).

Table 7.

Mediational effects of warmth classes on sexual-orientation differences in alcohol use disorder, stratified by gender: Growing Up Today Study

	Alcohol Use Disorder
	Model 1: Controlling for Demographics		Model 2: Model 1 + Warmth Classes		Proportion Mediated for Model 1 vs 2
	RR (95% CI)	p	RR (95% CI)	p	%(p)
WOMEN
Sexual orientation
Completely Heterosexual	1.00 (referent)		1.00 (referent)
Sexual Minority	2.03 (1.77, 2.32)	<.001	1.97 (1.72, 2.25)	<.001	4.3 (0.003)
Warmth Classes
High-High			1.00 (referent)
High-Moderate			1.15 (0.94, 1.41)	0.162
Moderate-Moderate			1.07 (0.86, 1.33)	0.543
Moderate-Occasional			1.44 (1.17, 1.77)	<.001
Occasional-Occasional			1.56 (1.25, 1.94)	<.001
Low-Low			1.48 (1.08, 2.04)	0.016
MEN
Sexual orientation
Completely Heterosexual	1.00 (referent)		1.00 (referent)
Sexual Minority	1.46 (1.23, 1.73)	<.001	1.45 (1.22, 1.72)	<.001	not mediated
Warmth Classes
High-High			1.00 (referent)
High-Moderate			1.07 (0.82, 1.39)	0.631
Moderate-Moderate			1.32 (1.02, 1.71)	0.034
Moderate-Occasional			1.30 (1.02, 1.66)	0.035
Occasional-Occasional			1.08 (0.81, 1.43)	0.602
Low-Low			0.74 (0.44, 1.24)	0.250

Note. Boldface indicates statistical significance (p<0.05). Models 1 and 2 were estimated with 4,591 women and 2,155 men. All models adjusted for race/ethnicity (assessed in 1996 at baseline), age (assessed in 2007), region (assessed in 2007), and lifetime college attendance (yes/no; assessed in 2010). Models for women also adjusted for lifetime pregnancy (yes/no; assessed prospectively from 1999-2010). Sexual orientation was based on participants' last self-report from 1999-2010, wherein sexual minority includes mostly heterosexuals, bisexuals, and gays/lesbians; gender was assessed in 1996 at baseline; warmth was assessed in 2007; alcohol use trajectories were derived using longitudinal latent class analyses from participants' past-year alcohol frequency, quantity, and heavy episodic drinking from 2003-2010 when they were aged 18-25 years; past-year alcohol use disorder was assessed in 2010. High-High=High familial & high non-familial warmth; High-Moderate=High familial & moderate non-familial warmth; Moderate-Moderate=Moderate familial & moderate non-familial warmth; Moderate-Occasional=Moderate familial & occasional non-familial warmth; Occasional-Occasional=Occasional familial & occasional non-familial warmth; Low-Low=Low familial & low non-familial warmth; RR=relative risk; CI=confidence interval.

Among men, sexual minorities were 1.46 times more likely than completely heterosexual participants to meet criteria for probable AUD in 2010 (Table 7; Model 1). Moderate-Moderate and Moderate-Occasional warmth classes were 1.30-1.32 times more likely to report probable AUD than High-High warmth class participants (Model 2). Warmth classes did not significantly mediate sexual-orientation differences in AUD for men. After controlling for warmth classes, sexual minority men remained more likely than completely heterosexual men to evidence AUD (Model 2).

4. Discussion

We found distinct typologies of familial and non-familial warmth across childhood and adolescence, and membership in these typologies differed by sexual orientation. Our study adds unique contributions to the literature by simultaneously examining latent classes of warmth from familial and non-familial contexts. Our analyses extended previous literature by formally testing gender-by-sexual orientation interactions in warmth class membership by using multivariable models. Specifically, among women, we found that several sexual minority subgroups were more likely to be in lower warmth classes compared with completely heterosexuals; among men, sexual-orientation differences in warmth were minimal. These results align with the bivariate results from prior research that shows that sexual-orientation differences in warmth were slightly smaller for men than women (Eisenberg and Resnick, 2006; Needham and Austin, 2010; Pearson and Wilkinson, 2013).

Strong theoretical and empirical foundations explaining these gendered sexual-orientation differences in warmth are lacking. Nevertheless, because SMY are stigmatized (Hatzenbuehler et al., 2013; Herek, 1988), SMY may experience lower familial and non-familial warmth. Why this pattern may be stronger among women than among men remains unknown and under-theorized. Nevertheless, these differences are likely derived from bidirectional gender-specific processes involving both child-level factors (e.g., gender nonconformity) and adult-level factors (e.g., attitudes towards gay/lesbian and bisexual populations; Rosario, 2015; Rosario et al., 2014a; Rosario et al., 2014b). Since a constellation of factors likely influences these findings, qualitative research with youth and adults using grounded theory may help illuminate knowledge about existing gendered sexual-orientation differences in warmth.

Our study also found that warmth during childhood and adolescence was significantly associated with AUD—but not AUTs—in emerging adulthood. Alcohol use is highest in emerging adulthood (Centers for Disease Control and Prevention, 2015), and heavy and moderate drinking trajectories are normative during this developmental period (Coulter et al., 2018). Our results suggest that emerging adults will engage in heavy and moderate AUTs regardless of warmth provided in earlier periods. On the other hand, lower warmth was positively associated with probable AUD. Though our study’s findings of negative associations between warmth and AUD confirm some research (Salom et al., 2015), it contradict others (e.g., Greenfield et al., 2016). However, our study was unique in that it examined AUD as an outcome separate from other comorbidities (e.g., mental health disorders) and assessed warmth across multiple time periods, which may explain our novel findings.

Additionally, warmth explained a small albeit statistically significant proportion of the sexual-orientation differences in AUD for women—but not for men. This has implications for future epidemiologic and intervention research. Future research can examine whether warmth serves as a resiliency factor (Herrick et al., 2014; Masten, 2013) that buffers the numerous minority stressors faced by SMY, thereby reducing their risk of having problematic alcohol use despite facing adversity. In light of our current findings, though, interventions that aim to reduce sexual-orientation disparities in AUTs or AUD may focus on increasing warmth, but additional targets will also be needed.

There are limitations to our study. GUTS participants were not randomly sampled from the U.S., were primarily non-Hispanic White, and children of Nurses’ Health Study II participants; therefore, our results may not generalize to more racially, ethnically, socioeconomically, or globally diverse populations. Our study may be prone to recall bias. On average, participants were 23 years of age when they reported on warmth from childhood and adolescence, but previous research has validated adult responses to childhood and adolescent experiences for other measures (Murphy et al., 2010). Attrition bias may also be present if nonresponse was differentially related to warmth, sexual orientation, AUTs, or AUD; the extent of this bias is unknown. Additionally, we measured sexual orientation using each participant’s last report of sexual identity/attraction; thus, our findings may not generalize to other ways of operationalizing sexual orientation (e.g., sexual behavior, sexual orientation trajectories, extent to which one was “out”). Also, some warmth classes and sexual-orientation subgroups had small sample sizes, limiting statistical power. We measured past-year AUD using self-reported items based on the DSM-IV criteria, which was slightly revised in the DSM-5. Single items measured familial and non-familial warmth during each developmental period, which may not capture all important information (e.g., the number and kinds of people who provided warmth). We may also have residual confounding (e.g., parenthood), though we controlled for multiple confounding variables.

Overall, our paper used a life-course approach to examine how early life experiences influence health disparities for sexual minority populations later in life. Compared to completely heterosexual women, sexual minority women report having lower familial and non-familial warmth in childhood and adolescence, which mediated a small proportion of their elevated risk of AUD. However, warmth did not mediate sexual-orientation disparities in AUD for men. Warmth also had little effect on AUTs for all emerging adults. Epidemiologic research can consider the role warmth plays in combination with other factors that influence AUTs and AUD in emerging adulthood. Warmth is also protective against many other health problems, including other substance use (Resnick et al., 1997). Since these health problems also disproportionately burden SMY (Herrick et al., 2011; Marshal et al., 2011; Marshal et al., 2008), research can test if warmth helps to explain sexual-orientation disparities in these health areas, especially for women.

Highlights.

• We examined the associations of warmth on alcohol use trajectories and disorder.

• We investigated sexual-orientation differences in familial and nonfamilial warmth.

• Lower warmth was associated with alcohol use disorder but not alcohol use itself.

• Warmth mediated a small percent of alcohol use disorder for sexual minority women.