Measurement invariance of the Domain-Specific Risk-Taking (DOSPERT) scale

Abstract

Group-level risk attitudes are often studied across psychology domains (e.g., binge drinking among college students, and driving risk by gender). In measuring these differences by self-report, such work relies on the assumption that those measures of risk attitude function equivalently across demographic groups—that is, that the measure employed has the property of measurement invariance. Here, we examine the measurement invariance properties of a widely used risk measure, the Domain-Specific Risk-Taking (DOSPERT) scale across different demographic groups. A secondary goal was to determine whether a hierarchical or bifactor model better fits the data. Data were collected from Prolific using a stratified sampling approach to ensure sufficient and unconfounded sampling of sex, socioeconomic status (SES), and race (N = 412). Sample groups consisted of approximately 50 participants each, based on the intersection of three dichotomized demographic groups (high vs. low SES, White vs. non-White, and female vs. male). Subjects completed the 30-item form of the DOSPERT assessing likelihood, perceived benefit, and riskiness of the same 30 behaviors. The bifactor models showed a superior fit to the hierarchical models and were used in subsequent analyses. These analyses demonstrated that no models fit generally acceptable criteria for configural fit, and many models additionally fail cutoffs for metric and scalar invariance. This study adds to findings that the DOSPERT does not perform equivalently across demographic groups. We suggest development of a scale of risk that is invariant across commonly assessed demographic factors.