Assessing the Feasibility and Psychometric Properties of a Short Version of the Gay Affirmative Practice Scale

Background: Measurement instruments of LGBTQ affirming practice have not progressed at the pace of scholars’ understanding on that topic. Researchers working to create new or updated versions of LGBTQ affirming practice often rely on older measures as a source of convergent validity by including those measures as a source of comparison. One such measure is the Gay Affirmative Practice Scale (GAP) (Crisp, 2002, 2006), which has been used in several studies. If the GAP is to be included in efforts to create updated measures, the items in the scale should be reduced while maintaining adequate psychometric properties.  Thus, the objective of this pilot study was to determine whether the Gay Affirmative Practice Scale (GAP) could be shortened without significantly compromising its internal consistency.

Methods: Data from two prior studies were used for the purposes of this analysis. Samples for both studies were obtained by randomly sampling members of NASW and APA who engaged in direct practice with clients. The N for Study 1 was 488 while the N for Study 2 was 751. After eliminating those who did not answer all 30 GAP items, the N was 902 responses.

Cronbach’s alpha coefficient is used to test for the internal consistency of a measure, and indicates the amount of covariance items in a measure share. The Standard Error of Measurement (SEM) is an estimate of standard deviations of measurement and is less influenced by differences in variance and standard deviation and Cronbach’s alpha. Microsoft Excel Solver program was used to maximize Cronbach’s alpha while keeping the number of items used to determine Cronbach’s alpha a set constant. The program selects items that can be used to obtain the alpha set by the researcher. After running the Excel Solver program to maximize alpha, the SEM was computed for the items selected by the program.

Results: The 5 item/domain version resulted in an α < .251, well below the acceptable range. Consequently, the program was then run with 10 items and achieved an α = .85 in the belief domain and α =.84 in the behavior domain, both of which are close to the “acceptable” range by Springer et al (2002). SEM was then calculated for the  two domains: belief SEM=2.241 and behavior SEM=3.219.

Implications: The initial psychometric analyses of the shortened GAP maintained respectable internal consistency as evidenced by retaining an alpha close to .85; however, the standard error of the mean suffered. Future investigations will follow this to determine if the scale can be shortened further by applying Item Response Theory methods (Levine & Leucht, 2013). Item response theory is a useful method for assessing the psychometric properties of scales by looking at each item as a standalone indicator of the latent construct (Osteen, 2010). When applied to Likert-type scales, either a multi-dimensional IRT model can be used or the items can be converted to dichotomous data for simple Rasch models. Item response theory analysis is attractive because it helps the researcher to produce items that are not sample dependent, thus improving generalizability.