A Psychometric Review of Measures Assessing Discrimination Against Sexual Minorities

ABSTRACT

Discrimination against sexual minorities is widespread and has deleterious consequences on victims’ psychological and physical wellbeing. However, a review of the psychometric properties of instruments measuring lesbian, gay, and bisexual (LGB) discrimination has not been conducted. The results of this review, which involved evaluating 162 articles, reveal that most have suboptimal psychometric properties. Specifically, myriad scales possess questionable content validity as (1) items are not created in collaboration with sexual minorities; (2) measures possess a small number of items and, thus, may not sufficiently represent the domain of interest; and (3) scales are “adapted” from measures designed to examine race- and gender-based discrimination. Additional limitations include (1) summed scores are computed, often in the absence of scale score reliability metrics; (2) summed scores operate from the questionable assumption that diverse forms of discrimination are necessarily interrelated; (3) the dimensionality of instruments presumed to consist of subscales is seldom tested; (4) tests of criterion-related validity are routinely omitted; and (5) formal tests of measures’ construct validity are seldom provided, necessitating that one infer validity based on the results obtained. The absence of “gold standard” measures, the attendant difficulty in formulating a coherent picture of this body of research, and suggestions for psychometric improvements are noted.

KEYWORDS:

• Discrimination
• gay men
• lesbian women
• psychometric
• sexual minority
• stigma

Notes

^1. The computation of total scores is potentially problematic as doing so assigns equal weight to forms of victimization that potentially differ in severity. For example, is “being called a homophobic name” such as fag or sissy comparable to being “pushed, shoved, hit or threatened with harm” (e.g., McGarrity et al., 2013)? Similarly, does “being treated unfairly by strangers” because of one’s sexual orientation have the same potential to cause harm as “being treated unfairly by an employer, boss, or supervisor” (e.g., Szymanski, 2006)?

^2. If the authors were using a measure that had been created previously, the source article was reviewed for evidence of content validity and criterion-related validity.

^3. We endorse the contemporary perspective that validity is a unitary concept requiring an “evidentiary chain which clearly links the interpretation of the assessment scores or data to a network of theory, hypotheses, and logic which are presented to support or refute the reasonableness of the desired interpretations” (DeVon et al., 2007; Downing, 2003, p. 831). However, for ease of presentation, we have differentiated between criterion-related and construct validity.

^4. We recognize that, in the event of a multifactor solution, scales consisting of three items would be problematic as one or more factors would be underrepresented (i.e., they would contain an insufficient number of items). However, for unidimensional solutions, three items would be adequate (see Costello & Osborne, 2005; Floyd & Widamen, 1995).

^5. Dimensionality is not a “fixed” property of a scale and, consequently, should be explored with each new sample.

^6. Sexual Orientation Factor 2 was represented by a single item (“Seriously hurt or hospitalized because of your sexual orientation”).

^7. Huebner and colleagues (2005) also used the SRE and SSE for the purposes of creating a measure of perceived anti-gay discrimination. The same concerns, expressed for Szymanski’s (2006) scale, also apply.

^8. This feature of the SRE was omitted in the measures created by Huebner and colleagues (2005) and Szymanski (2006). The authors do not provide a rationale for this omission.

^9. In both studies, criterion-related validity was erroneously identified as construct validity.

^10. This indicator was not applicable for two articles in which novel constructs (e.g., bisexual-specific forms of discrimination—Brewster, Moradi, DeBlaere, & Velez, 2013) were measured.

^11. Wright and Wegner (2012) framed these hypotheses in terms of criterion-related validity; however, in actuality, they embody convergent validity, which falls under the rubric of construct validation.