Validation of the HIV Scale to the Rasch Measurement Model, GAIN Methods Report 1.1

Purpose. The purpose of this report is to provide a brief psychometric analysis of the HIV Risk Scale, using the Rasch measurement model. The 35-item HIV Risk Scale consists of 31 yes/no items and four recency items related to the risk of contracting HIV. Methods. Data were analyzed on 6269 persons who presented for substance problem screening. Rasch analysis included an examination of: person and item reliabilities; dimensionality; construct validity including item and person fit statistics; and differential item functioning (DIF) across subgroups. DIF analysis allowed us to determine if the relative item estimates (i.e., item difficulty estimates) remained invariant across subgroups of persons. Results. The HIV Risk Scale performs well as a measure of the construct of HIV risk. Both items and scales form the theoretically expected hierarchies with a Rasch person internal consistency reliability of .79 and an item reliability of 1.00. Cronbach's alpha is .86. The persons' responses generally conformed to the expectations of the Rasch model. Of the 35 items in the HIV Risk Scale, a DIF > .5 SD occurred in 10 items for males vs. females, 3 items for youth vs. adults, 12 items for race when using African American as the reference group, and 15 items for primary substances when using opiates as the referent and in 5 items involving needle use when alcohol was the referent. For the above contrasts, 22 of the 35 items exhibited DIF for one or more groups. While such DIF usually cancels out in the overall scale, it should be taken into account when interpreting items, for tailoring interventions to these groups, and in some cases corrected when the goal is to compare these groups. In terms of person fit, 81% of the sample exhibited person infit and outfit that was low or moderate (L/M) and are thus, regarded as fitting the Rasch model expectations well from a clinical perspective. Five percent (5%) of the persons had a L/M infit and high (HI) outfit and are referred to as Atypical Type 1, where the overall score may under- estimate severity since these tended to be persons engaging in high risk sex who are not reporting lower risk victimization. Six percent (6%) of the persons had HI infit and L/M outfit (called Atypical Type 2) and tended to be valid low scorers who endorsed a few victimization items. The remaining 8% of persons had HI infit and HI outfit (called Atypical Type 3) and tended to be valid high scorers who also had an overall score that may underestimate severity slightly and tended to be needle users with less risky sex than expected. Citation. Conrad, K. J., Conrad, K. M., Dennis, M. L., Riley, B.B., & Funk (2009). Validation of the HIV Scale to the Rasch Measurement Model, GAIN Methods Report 1.1. Chicago, IL: Chestnut Health Systems. Retrieved from http://www.chestnut.org/li/gain/psychometric_reports/Conrad_et_al_2009_HIV_Rasch_Report.pdf