Succinct and efficient: Optimizing risk-taking propensity measurement (DOSPERT) with artificial intelligence techniques

The Domain-Specific Risk-Taking (DOSPERT) scale is a widely used psychometric tool for assessing individuals' risk-taking attitudes across different domains. The DOSPERT scale includes five dimensions, totaling 30 items, which may impose a cognitive load on respondents. Therefore, shortening this measurement tool could significantly improve its usability. This study is the first to optimize or abbreviate this measure. Using an automatic genetic algorithm (GA) approach with data from two waves of surveys from Taiwanese respondents, we generated two shortened forms of DOSPERT for behavioral decision scientists: a l5-item version that retains the five domains in the original scale, and a very short 5-item version. Both abbreviated versions demonstrate satisfactory psychometric properties and reduce the number of items by 50 % and 83 %, respectively. In addition to essential psychological variables like the Big Five personality traits or self-efficacy, these abbreviated versions can also predict significant information-seeking and political behaviors, such as the intention to share misinformation and willingness to fight for one's country. • Shortening the DOSPERT Scale: AI algorithms created two shortened versions. • Evaluation: Both versions demonstrated satisfactory reliability and validity. • Convergent Validity: Both versions correlated with many psychometric measures. • The 15-item scale predicted fake news sharing among risk-takers. • Higher risk-takers were more likely to seek potentially undesirable information.