Table 2.

Quantitative papers explicitly applying latent variable or clustering analysis methods within an intersectionality framework

Method used	Paper	Sample	Variables used in LCA/LPA/clustering	Clusters	Analyses conducted with resulting cluster categories
Latent class analysis (LCA)	Garnett, et al., 2014 [39]	965 high school students	8 dichotomous variables. Past 12 months discrimination because of (yes/no for each): Race, ethnicity or color; Self or family from another country; Someone thought you were gay, lesbian, or bisexual; Weight. Past 12 months bullied or assaulted because of (yes/no for each): Race, ethnicity or color; Self or family from another country; Someone thought you were gay, lesbian, or bisexual; Weight.	1: Low discrimination 2: Racial discrimination 3: Sexual orientation discrimination 4: Racial and weight discrimination with high bullying (intersectional class)	Multinomial logistic regression with sociodemographic characteristics as exposure, and latent class membership as outcome. Linear and logistic regression models with latent classes as exposures, controlling for sociodemographics, to assess three mental health outcomes.
	Bécares and Priest, 2015 [45]	10,115 eighth-grade students	42 dichotomous variables, with the following each measured at 3–5 time points from kindergarten to 8th grade. Individual-level variables: Child lived in a single-parent household; Household below the poverty threshold level; Food insecurity; Parental expectations of child’s future academic achievement; Whether parents had moved since previous interview; Household-level composite index of socioeconomic status. School-level variables: % students eligible for free school meals; % students from a racial/ethnic background other than white non-Hispanic. Neighborhood variables: Neighborhood safety level	1: Individually and contextually disadvantaged 2: Individually wealthy, contextually disadvantaged 3: Individually and contextually wealthy 4: Individually disadvantaged, contextually wealthy	Looked at differences in each of the eight academic and non-academic outcomes between cross-classified race/ethnicity*gender*SES groups, where SES was defined by latent class membership.
	Budge et al., 2016 [35]	442 trans-identified individuals	1 dichotomous and 4 categorical variables. Dichotomous: Race—person of color. Categorical: Gender (4 categories); Sexual orientation (8 categories); Education (5 categories); Household income (8 categories).	1: Socioeconomic and racial privilege 2: Educational privilege	T-tests with latent classes as exposures, to assess differences in anxiety and depression.
	Byrd and Carter Andrews, 2016 [36]	1468 middle and high school students	22 dichotomous variables. Source of discrimination (yes/no for each): Peers; Teachers; Administrators; Front office personnel; counselors/social workers; Others; No discrimination. Perceived reason for discrimination (yes/no for each): Race/ethnicity; Gender; Religion; Social class/family’s financial status; Sexual orientation; Disability; Other attribution. Form of discrimination (yes/no for each): Don’t get called on in class by the teacher; Experience name calling; Fewer opportunities to access people in the school that can help me succeed; Excluded from certain social groups; Excluded from academic opportunities; Excluded from social opportunities; Punished more than other students; Other form.	1: Adult cluster 2: Multiple cluster 3: Peer cluster 4: Low cluster	Chi-square tests to assess differences in sociodemographic characteristics among latent class memberships. One-way ANOVA and Tukey post-hoc tests with latent class membership as exposure, to assess differences in discrimination frequency and school-related experiences and perceptions.
	Landale et al., 2017 [31]	1789 Latino and non-Latino white adults	6 dichotomous and 1 categorical variable. Dichotomous: Latino ethnicity; Age—30 years or older; Education—high school or higher; Speaks English; Employment status; Below poverty threshold. Categorical: Immigration status (3 categories).	1: U.S. born and advantaged whites 2: U.S. born and advantaged Latinos 3: U.S. born, young, jobless Latinos 4: Documented and older Latinos 5: Undocumented and disadvantaged Latinos	Linear and logistic multilevel models with latent class membership as the exposure (individual-level), and controlling for neighborhood- and individual-level covariates, for three discrimination-related outcomes.
	Goodwin et al., 2018 [37]	1052 community members	3 dichotomous and 7 categorical variables. Model 1 (only SES variables): Categorical: Social occupational class (4 categories); Employment status (4 categories); Household income (3 categories); Housing tenure (4 categories); Education (3 categories). Dichotomous: Receiving benefits (excluding pension or child benefits); Past year non-mortgage related debt; Moved twice or more in past 2 years. Model 2 (includes Model 1 variables plus the following): Categorical: Immigration status (4 categories); Race/ethnicity (6 categories).	Model 1: 1: Professional, homeowners 2: Professional, renters 3: Skilled, renters 4: Students, renters 5: Economically inactive, renters 6: Economically inactive, homeowners Model 2: 1: Professional, homeowners, White British 2: Economically inactive, renters, White British 3: Students, mixed tenure, non-migrant, mixed ethnicity 4: Skilled, renters, non-migrant, mixed ethnicity 5: Economically inactive, homeowners, mixed migration status, mixed ethnicity 6: Professional, renters, migrant, mixed ethnicity 7: Economically inactive, renters, migrant, mixed ethnicity	For both model 1 and model 2, logistic regression models with latent class membership as exposure, adjusting for age and gender, for the outcome “common mental disorder”.
	Earnshaw et al., 2018 [40]	1293 adults living in low-resource communities	9 dichotomous variables. Discrimination (ever experienced, never experienced). Discrimination attributions (binary ever/never for each): Race/ethnicity; Income; Age; Gender; Appearance; Language; Weight; Sexual orientation.	1: Single attribution 2: No discrimination 3: Several attributions 4: Most attributions	ANOVA and chi-square analyses with latent classes as exposure, assessing differences in sociodemographics, discrimination experiences, and health outcomes.
	Gazard et al., 2018 [32]a	1052 community members	Same variables as Model 2 from Goodwin et. al., 2018 [37]	1: Migrant, mixed ethnicity, low SES 2: White British, low SES 3: Non migrant, mixed ethnicity, student 4: Non migrant, mixed ethnicity, skilled 5: Mixed migration status, mixed ethnicity, economically inactive 6: Migrant, mixed ethnicity, high SES 7: White British, high SES	Chi-square tests to look at differences in discrimination experiences, mental/physical disorder and long standing illnesses between the latent classes. Logistic regression models with health service use as the outcome, and latent class membership as the exposure. Logistic regression models with health service use as the outcome, and discrimination as the exposure. Unadjusted and adjusted models were presented. Here the latent classes were used in the adjusted models, alongside other variables.
Latent profile analysis (LPA)	Shramko et al., 2018 [44]	219 Latinx sexual minority youth	6 continuous variables. Scales: Perceived bias-based victimization due to sexual orientation or gender identity; Perceived bias-based victimization due to Latinx identity; Perceived discrimination due to sexual orientation or gender identity; Perceived discrimination due to Latinx identity; Ethnic centrality; Sexual orientation centrality.	1: Low perceived discrimination and victimization and low identity centrality 2: Low perceived discrimination and victimization, but high identity centrality 3: Moderate perceived discrimination and victimization, and moderate centrality 4: High perceived discrimination and victimization, but moderate identity centrality	ANOVAs and chi-square tests to assess differences in demographic characteristics between latent profiles. Structural equation modeling looking at the association between the profiles and the following outcomes: GPA, depressive symptoms, self-esteem.
	Taggart et al., 2019 [38]	1170 African-American and Caribbean Black adolescents	7 continuous variables. Religiosity (4 subscales): Organizational participation; Religious support; Nonorganizational participation; Subjective religiosity. Racial identity (3 subscales): Racial centrality; Racial public regard; Racial private regard.	1: Low intersected identity 2: High intersected identity 3: High racial identity 4: High religiosity	Chi-square analyses to determine differences in class membership by demographic variables. Cox proportional hazards survival analysis to model the association between profile membership and sexual initiation.
Cluster analysis	Stirratt et al., 2008 [51]	40 lesbian, gay or bisexual adults	Each participant was asked to select up to 12 identities that best represented them. Had to include at least gender, racial/ethnic group, and sexual identities. Then for each identity selected, rated how it applied to a set of 70 attributes (binary variable for each of the 70).	N/A (cluster analysis conducted for each participant)	Separate cluster analysis for each respondent. Then used results/structure of each respondent’s cluster analysis to create identity measures. Then: 1. Correlated identity measures with other measures (psychological well-being, social well-being, collective self-esteem, internalized homophobia, depression). 2. Chi-square and ANOVA analyses to assess differences in identity measures between race*gender subgroups.
	Aspinall and Song, 2013 [33]	326 mixed-race individuals	16 dichotomous variables. Salient identities based on (yes/no for each): Age or life-stage; Kind of study or work; Level of education; Level of income; Political belief; Family; Ethnic group or cultural background; Country family came from originally; Regional identity; Nationality; Religion; Skin color; Social class; Gender; Disability; Sexuality/sexual orientation.	N/A, clusters not named (cluster analysis presented visually on dendrogram)	Description of how attributes were clustered together based on the dendrogram visualization. No further analyses conducted.
	Brown et al., 2018 [41]	116 African-American female college students	Model 1: 2 continuous variables: Racial socialization; Gender socialization. Model 2: 2 continuous variables: Ethnic socialization; Gender socialization.	Model 1 (racial-gender clusters): 1: Racial silence and low gender tradition cluster 2: High racial influence and high gender tradition 3: Minimal racial coping and low gender tradition Model 2 (ethnic-gender clusters): 1: High ethnic influence and high gender tradition 2: Minimal ethnic pride and minimal gender tradition 3: High ethnic influence and low gender tradition 4: Ethnic silence and low gender tradition	For each cluster set, MANCOVA analyses used to examine if identity cluster profiles were associated with the outcomes of sexual assertiveness and safer sex behaviors. Mediation analysis used to examine mediating role of sexual assertiveness on the association between the ethnic-gender clusters and the outcome of safer sex behavior.
	Whaley and Dubose, 2008 [34]	322 undergraduate students	16 dichotomous variables. Ever received mental health care or counseling for (yes/no for each): Mood disorder; Anxiety disorder; Substance-related disorder; Eating disorder; Adjustment disorder; Impulse control disorder; Stereotypic movement disorder. Ever used medications to treat the previous conditions. Currently receiving mental health care or counseling for (yes/no for each): Mood disorder; Anxiety disorder; Substance-related disorder; Eating disorder; Adjustment disorder; Impulse control disorder; Stereotypic movement disorder. Currently using medications to treat the previous conditions.	1: History of psychiatric treatment for emotional disorders 2: Addictive behaviors to cope with depression 3: Loss of control over eating behavior 4: Pharmacotherapy for clinical depression	For four ethnicity/race*gender intersectional groups, separate regression analyses conducted to assess the effect of the four profiles on the outcome cumulative frequency percentages of psychological problems.
	Price et al., 2019 [43]	946 high school students	2 dichotomous and 1 categorical variable. Dichotomous: Race—youth of color; Sexual orientation—LGBQ. Categorical: Sex and gender (3 categories).	1: LGBTQ youth 2: Heterosexual youth of color 3: Heterosexual white youth	ANOVA and logistic regression to assess differences in school and well-being related outcomes across clusters. ANOVA and chi-square tests to assess differences in discrimination and bullying across racial and gender.subgroups within LGBTQ cluster. Regression mediation analysis to examine the mediating role of discrimination on the association between cluster membership and the outcomes—well-being, depression and grade point average.
	Wanka et al., 2019 [42]	400 Turkish adults	4 continuous variables. Neighborhood discrimination due to: Ethnic origin; Age; Religion; Gender.	1: Cluster 1 2: Cluster 2 3: Cluster 3 4: Cluster 4	Descriptive statistics regarding how clusters differed by social-spatial variables. ANOVA to analyze differences between the clusters with regard to neighborhood satisfaction and sense of home. Linear regression models used to examine the association between cluster membership and the outcomes neighborhood satisfaction and sense of home.

^aUses model 2 LCA results from Goodwin et. al., 2018 [37], to assess different outcome in the same sample