Validating the Structure of Proximal and Distal Factors within the Sexual Minority Adolescent Stress Inventory

Participants

The procedures in the current study were conducted through a series of secondary analyses using data collected in two previous studies (Weeks et al., 2020; Weeks et al., 2021). A combined sample size of N = 445 was gathered through two independent instances of participant recruitment through online survey panels. Sample 1 included responses from 13–18-year olds, who were predominantly White (57.9%) and female (50%), and accounted for n = 152 participants. Sample 2 included responses from participants who were 13–17 years old, who were predominantly White (39.6%) and female (70.6%), and accounted for n = 293 participants. Full participant demographics, including age, gender, sexual orientation, and racial/ethnic identity, are outlined in Table 1.

OPEN IN VIEWER

Table 1.

Sample Demographics: Frequencies and Percentages (N = 445).

Demographic Variable	Count	%
Age
13	20	4.5
14	43	9.7
15	98	22
16	100	22.5
17	148	33.3
18	36	8.1
Gender
Woman	283	63.6
Man	83	18.7
Transgender woman	3	0.7
Transgender man	13	2.9
Gender fluid	49	11
I identify differently	14	3.1
Sexual orientation
Asexual	13	2.9
Bisexual	234	52.6
Fluid	6	1.3
Gay	44	9.9
Lesbian	57	12.8
Pansexual	41	9.2
Queer	10	2.2
Questioning	28	1.1
Demisexual	5	1.1
I identify differently	6	1.3
I prefer not to answer	1	0.2
Race/ethnicity
Asian	28	6.3
Multiracial	53	11.9
Black or African American	55	12.4
Hispanic or Latinx	75	16.9
Middle Eastern	8	1.8
American Indian or Native American	10	2.2
White or European	204	45.8
I prefer not to answer	11	2.5
I identify differently	1	0.2

Measures

Statistical Analyses

Preliminary analyses were run for observed scores from the three levels of the SMASI—including global minority stress (total scale), proximal and distal stress domains (composite scales), and the nine specific stressors (subscales)—to evaluate distribution of responses, central tendency, and internal consistency reliability. Pearson bivariate correlations were also calculated to assess strength of relationships, independence, and directionality between scores derived from these three levels of measurement (i.e., total scale, composite scales, and subscales). Given the study combined two samples from different online panels, a two samples t-test was conducted to determine potential mean differences in the two samples prior to aggregating the data.

After ensuring through preliminary analyses that the data met basic statistical assumptions, confirmatory factor analysis (CFA) was conducted on the SMASI’s measurement model using the weighted least squares estimator (WLSMV; Jöreskog et al., 2001) with mean and variance adjustments (see Figure 1). Goodness of fit was determined based on the chi-square test, an RMSEA of less than .06, a CFI of greater than .90, and an SRMR of less than .08 (Hu & Bentler, 1999). Indicators were determined to be representative of the latent variable based on their standardized loadings, with loadings > .70 considered ideal and > .40 deemed acceptable (Kline, 2014). These statistical procedures assessed factor loadings of SMASI items on the latent variables representing specific stressors: internalized homonegativity (proximal), identity management (proximal), negative expectancies (proximal), intersectionality (proximal), family rejection (distal), social marginalization (distal), homonegative communication (distal), homonegative climate (distal), and negative disclosure experiences (distal). According to the minority stress model, these first-order factors were expected to have acceptable to ideal loadings onto the second-order latent variables of distal and proximal stress (as previously indicated in the parentheses following each first-order factor).OPEN IN VIEWER

The factor variance of distal and proximal stress factors on a third-order latent variable representing global minority stress was also evaluated by constraining both second-order loadings on this factor to one and calculating the correlation. A nested model comparison index, comparing the second-order model with distal and proximal stress factors (see Figure 1) with the second-order model with a global minority stress factor (sans distal and proximal factors; see Figure 2), was evaluated to determine how well the updated multidimensional measurement model compared to the original and simpler SMASI measurement model with fewer factors. Based on model comparison using a likelihood ratio test with WLMSV corrections, we expected the model with proximal and distal second-order factors would demonstrate statistical value added (i.e., p < .05) compared to the original model with one second-order factor representing global minority stress (sans distal and proximal factors). Proposed CFA model pathways for the model with two second-order factors are outlined in Figure 1. In addition to the primary CFA, latent construct reliability analyses were conducted to explore the consistency of the latent variables (as opposed to the observed scores) across each of the three levels of measurement (i.e., first-order, second-order, and third-order factors). For latent construct reliability, which is an indicator of internal consistency at the factor level, H ≥ .70 were considered desirable, as they indicate a strong intrafactor correlation over repeated administrations (Hancock & Mueller, 2001). Factor analyses were conducted using the lavaan package (Rosseel, 2012) through R Statistical Environment (R Core Team, 2020).OPEN IN VIEWER

Descriptive Statistics

All subscale scores were assessed to determine central tendencies, range, distribution, and internal consistency reliability. Scale scores were generally normally distributed except for two subscales (i.e., social marginalization and internalized homonegativity). While distribution of scale scores is important within measurement, strict normality is not necessary for conducting confirmatory factor analyses with dichotomous items. Additionally, skewness and kurtosis values for all subscale scores were < |2|, suggesting the distributions could be interpreted as relatively normally distributed for our analytic purposes. Evidence of a difference in mean SMASI total scores between samples was found, t (433) = 3.544, p < .001, though the effect size was small (d = .35). Given this finding, sample group was included as a covariate in the CFA models to control for the potential effect of sampling on results from the current study.

Correlations

A bivariate correlation matrix was computed to assess the strength and direction of the relationship among SMASI scales sum scores (see Table 2). Correlations ranged from weak (r = .22) to very strong (r = .95). All variable relationships were positive, and strength of relationships generally aligned with minority stress theoretical frameworks. Interestingly, the distal stress domain score had a nearly perfect correlation with the global minority stress score (r = .95) and the proximal stress variable likewise had a very strong association with this global indicator (r = .85), possibly making differentiation among these variables difficult at the level of observed scale scores.

OPEN IN VIEWER

Table 2.

Bivariate Correlation Matrix (Pearson’s r) for SMASI Scale Scores.

Measures	GMS	Id Ma	Ne Ex	Ne Di	Fa Re	In Ho	Ho Co	Ho Cl	So Ma	In	DMSD	PMSD
GMS
Id Ma	.56*
Ne Ex	.75*	.49*
Ne Di	.69*	.30*	.38*
Fa Re	.81*	.31*	.53*	.52*
In Ho	.62*	.46*	.53*	.29*	.37*
Ho Co	.62*	.22*	.40*	.38*	.44*	.23*
Ho Cl	.69*	.34*	.55*	.43*	.35*	.33*	.46*
So Ma	.72*	.32*	.43*	.39*	.46*	.35*	.31*	.59*
In	.64*	.34*	.43*	.39*	.45*	.25*	.45*	.43*	.43*
DMSD	.95*	.40*	.63*	.73*	.85*	.44*	.65*	.68*	.74*	.57*
PMSD	.85*	.72*	.80*	.44*	.54*	.83*	.42*	.53*	.51*	.62*	.66*

Factor Analyses

Prior to conducting the primary analyses, a preliminary CFA was conducted to confirm the psychometric defensibility of the original SMASI measurement model with one second-order factor representing global minority stress. To determine the appropriateness of observed variable loadings on subscale latent variables, a CFA model was fit. Parameter estimates were conducted using the WSLMV estimator with mean and variance adjustments. Results for the second-order model with one higher-order factor (Figure 2) demonstrated good data–model fit (χ² (1166, N = 445) < .01, RMSEA = .045, CFI = .918, SRMR = .10), with first-order factor loadings ranging from λ = .64–.92 onto the higher-order global minority stress factor. Additionally, global minority stress exhibited strong internal consistency reliability at the observed level (α = .93) and the latent level (H = .95).

Following confirmation of the original SMASI measurement model, a CFA was fit to test the psychometric defensibility of the updated SMASI model with two second-order factors representing proximal and distal stress domains. Results for the second-order model with two higher-order factors (Figure 1) demonstrated good data–model fit (χ² (1164, N = 445) < .01, RMSEA = .044, CFI = .923, SRMR = .10) and are outlined in Table 3. With a moderately large scale (49 items), observed parameter estimates within their respective subscale groupings demonstrated consistent factor loadings (λ > .40). Many first-order factors had loadings between λ = .70 and .90, and all second-order latent variables had factor loadings > .69. The lowest coefficient was the loading for item 1 on identity management (first-order factor), which was still considered adequate (λ = .46). Although thresholds appeared to range in valence across the SMASI items, subscale item clusters generally matched in terms of directionality. This suggests that the Z-score each latent subscale must be above for the observed item to change responses are mostly positive or negative for a given group, offering additional information related to the directionality and sensitivity of an item response compared to the latent variable.

OPEN IN VIEWER

Table 3.

CFA Results for the Updated SMASI Model with Proximal and Distal Domain Factors.

Factor	Item	WLSMV		Factor	Item	WLSMV
		λ	Τ			λ	Τ
Id Ma	1	.46	−.56	Ne Ex	4	.81	−.55
	2	.83	−.18		8	.85	−.71
	3	.73	−.50		9	.85	−.27
Ho Co	21	.84	−.33	In Ho	14	.96	−.41
	42	.66	−.33		15	.85	.12
	45	.87	−.43		16	.97	.24
	48	.56	−.02		17	.88	.28
	49	.64	−.28		18	.83	−.25
Ne Di	5	.70	.12		19	.55	−.33
	6	.82	−.01		20	.86	−.31
	11	.54	−.36	Ho Cl	31	.84	−.27
	12	.77	.42		33	.76	−.76
	13	.78	.35		34	.90	−.65
Fa Re	7	.77	.80		35	.86	−.62
	10	.88	.40	So Ma	32	.86	.12
	22	.85	.18		36	.69	−.27
	23	.82	.48		37	.80	.34
	24	.89	.17		38	.67	1.07
	25	.65	.24		39	.72	−.22
	26	.85	.55		40	.89	−.16
	27	.74	1.02		41	.73	.14
	28	.69	.16		43	.86	−.42
	29	.83	.32	In	44	.89	.04
	30	.80	.57		46	.87	−.04
					47	.75	−.21
Proximal	Factor	λ	σ2	Distal	Factor	λ	σ2
	Id Ma	.80	.36		Ne Di	.80	.38
	Ne Ex	.98	.04		Fa Re	.81	.35
	In Ho	.69	.53		Ho Co	.81	.34
	In	.80	.37		Ho Cl	.79	.37
					So Ma	.83	.31
	ψ 1 1			0.88
	df			1164
	p value			< .001

Second-order loadings of first-order (subscale level) factors on the higher-order factor of proximal stress appeared to be more variable compared with the first-order factor loadings on the higher-order distal stress factor (see Table 3). Subscale (i.e., first-order factor) loadings on the proximal domain have a larger range, with internalized homonegativity furthest from the subscale grouping (λ = .69). On the other hand, subscale factor loadings on the distal domain are clustered above the .70 threshold, with only a range of .04 in parameter estimates (see Table 3). Overall, estimates indicated that the proximal and distal latent variables are likely accurate representations of the latent subscales that theoretically comprise them, but that there is more variability in the contribution made by first-order factors for the proximal domain compared to the distal domain. Furthermore, when considering the larger conceptual framework of the minority stress model, results showed that the third-order global minority stress factor can account for a very strong latent correlation between the proximal and distal stress domains, ψ = .88, suggesting a positive relationship among these factors that is consistent with theory.

Finally, the nested model comparison index assessed the value added of the second-order two-factor minority stress model accounting for proximal and distal domains. A likelihood ratio test with WLSMV corrections of the second-order two-factor model (see Figure 1) with the second-order one-factor model (see Figure 2) found a chi-squares difference of 10.92, df = 2, p = .004. This indicates that the more complex model with distal and proximal stress domains seems to be a better fit for the data than simply using a one-factor, second-order, global minority stress model to capture all subscales assessed using the SMASI. The difference between the nested models showed significant difference in terms of model fit, as determined by chi-square and degrees of freedom, relative to the additional parameters and degrees of freedom required to estimate them. Testing of latent construct reliability further supported the model comparison outcome, with both distal and proximal latent factors exhibiting good internal consistencies at the observed level, α = .91 and .86, and the latent level, H = .90 and .99, respectively.

Limitations

Limitations within the current study include the operationalization of the minority stress model within the measure used (SMASI). As stated in the introduction, various measures exist for assessing minority stress, including the Heterosexist Harassment, Rejection, and Discrimination Sale (Szymanski, 2009); the Gay-related Stressful Life Events Scale (Rosario et al., 2002); the Daily Heterosexist Experiences Questionnaire (Balsam et al., 2013); and the LGBT People of Color Microaggressions Scale (Balsam et al., 2011). While these measures are not currently validated with youth, they raise the point that minority stress can be operationalized in multiple ways in addition to how it is operationalized in the SMASI. The current study captured youth minority stress based on the combination of some of the most widely accepted minority stress frameworks (Goldbach et al., 2014; Hatzenbuehler, 2011; Meyer, 2003); however, future research would benefit from validating the levels of the minority stress model—specific stressors, proximal and distal stress, and global minority stress—using other measures and across diverse demographic groups (e.g., racial and gender). Additionally, the SMASI phrased items in a negative manner, with more items endorsed as “yes” indicating higher experiences of minority stress. Future research could further investigation in this area by measuring minority stress through strengths-based or coping-oriented item phrasing that aims to measure how well adolescents are navigating minority stress across specific, domain, and global levels. Adapting SMASI item response options to Likert-type scaling may also provide opportunity to investigate if results observed in this study would hold across alternative measurement methods.

Other methodological limitations were also present in the current study. Sampling using purposive procedures that required participants to engage via computer or smartphone and required internet connection likely limited participation by youth without access, especially those from families with lower socio-economic means. Other demographic considerations should be taken when generalizing results from the study, given the sample identified predominantly as White, bisexual, and female. Additionally, while the sample was large enough for the current analytic approach, it was comprised of a combination of two subsamples obtained at different timepoints of data collection occurring over a year apart. Therefore replicating these analyses with larger and more diverse cross-sectional samples obtained at single time point would account for limitations in the present sample while also lending greater credibility to the stability of these results. Last, although the third-order factor of global minority stress was included in the updated model (Figure 1) to represent the theoretical framework utilized in informing the SMASI, loadings for the second-order proximal and distal stress factors onto this third-order global minority stress factor could not be estimated due to limitations in the statistical software. Specifically, it is not currently possible to estimate loadings for a factor when there are only two indicators; three or more are required. Thus, our modeling approach was a workaround that indirectly accounted for the information of interest. Future research may try to account for this limitation by using alternative modeling approaches, such as bifactor CFA (Canivez, 2016). Yet we caution that bifactor modeling approaches may fundamentally alter the interpretation of SMASI factors—and therefore the interpretation of resulting scale scores for use in research and practice—compared to the original SMASI model (Schrager et al., 2018) and our updated model (see Bonifay et al., 2017).