Measuring identity multiplicity and intersectionality: Hierarchical Classes Analysis (HICLAS) of sexual, racial, and gender identities

Abstract

We propose an innovative approach for measuring identity multiplicity and intersectionality—Hierarchical Classes Analysis (HICLAS) of an Assessment of Multiple Identities. This method allows researchers to assess characteristics of individual identities and model implicit interrelationships between multiple identities held by an individual. We found support for the validity of this approach through analysis of sexual, racial, and gender identities among 40 lesbian, gay, or bisexually identified (LGB) participants. As hypothesized, poorer mental health indicators were significantly associated with greater negative valence of sexual identity and greater negative self-complexity in HICLAS. HICLAS also allowed us to discern subgroup differences indicative of intersectionality (e.g., in this LGB sample, more African American participants than White participants showed interconnected sexual and racial identities).

Acknowledgments

Study funded by NIMH (R01-MH066058, “Minority Stress, Identity, and Mental Health,” Principal Investigator: Ilan H. Meyer).

Notes

1. The methods employed in this study produce a data matrix [M], which represents an [i×a] identity attribute matrix where “i” is the number of identities and “a” is the number of attributes. HICLAS subsequently decomposes the M matrix into both an [i×r] binary matrix [S] and an [a×r] binary matrix [P] such that “M = SP”, with “r” being the rank of the solution. In many cases, more than one such decomposition exists, but it can be shown that a set-theoretical one always exists, and it is the latter that is selected by HICLAS (DeBoeck & Rosenberg, 1988; Rosenberg & Gara, 1985; Van Mechelen et al., 1995). In general, a minimization equation does not exist to obtain the best decomposition of M in any given rank r, so an estimate of S and P is calculated using an alternating least squares approach. Estimates of S and P are calculated for ranks 1, 2, 3 and so on, as is a measure of the goodness of fit of these decompositions of M in relation to the actual data. Goodness of fit is a monotonic function (never decreasing) of rank. Significance levels are not calculated in the HICLAS analysis of any given participant's data since the observations in his or her matrix are not independent of one another. In the present paper, any significance levels reported are based on between-subjects analyses of HICLAS based indices.

2. To determine the valence of the 70 attributes used in the identity assessment, we conducted a pretest survey of twelve project staff members (a group that was diverse in gender, race/ethnicity, and sexual orientation). These individuals rated the character of each attribute as negative (0), neutral (1), or positive (2). Based on this survey, we assigned attributes as negative, positive, or neutral. Attributes were judged to be negative if their mean ratings were between 0.0 to 0.666, neutral if their mean ratings were between 0.667 and 1.333, and positive if their mean ratings were between 1.334 and 2.0. Overall, 36 attributes were judged as negative (e.g., nervous, selfish, dishonest), 30 as positive (e.g., joyful, intelligent, friendly), and only 4 as neutral (e.g., introverted, reserved, shy). If the results of HICLAS indicated that a particular target identity was dropped from the participant's identity model (because the participant ascribed no or very few attributes to that identity), then target identity valence could not be directly calculated. In cases where this occurred, we replaced this missing data by assigning the sample's mean valence for that target identity.