A Meta-Analytic Investigation on the Construct Validity of Risk Propensity at Work: Insights from Decision Science and Large Language Models

Abstract Risk propensity is a central construct in personality, economic, and decision sciences, but its predictive utility in a work context remains underexplored. In this paper, we integrate psychometric meta-analysis with decision modeling to examine the construct validity of risk propensity for work. We also use a large language model (LLM) approach to examine how semantic representations of specific decision attributes (e.g., potential gains, losses, and uncertainty) reflected in survey items map onto the observed meta-analytic relationships between risk propensity and work behaviors. We found that risk propensity outperformed most Big Five traits, the dominant model of personality at work, in uniquely predicting various constructive (e.g., creativity, constructive deviance) and destructive (e.g., counterproductivity, safety non-compliance) work performance constructs above and beyond the Big Five. Using sentence transformers, we found that outcome item embeddings were highly predictive of the meta-analytic correlations between risk propensity and work outcomes. Meta-regression using LLM-derived decision attributes revealed that the relationship between risk propensity and work performance was stronger for behaviors that involve greater personal risk (e.g., safety non-compliance) and positive organizational valence (e.g., creativity). Together, this paper expands the constellation of workplace predictors while advancing a novel methodological approach that combines modern LLMs with decision modeling to test the theoretical underpinnings of trait-behavior relationships.