<p>Understanding whether risk preference represents a stable, coherent trait is central to efforts aimed at explaining, predicting and preventing risk-related behaviours. We help characterize the nature of the construct by adopting a systematic review and individual participant data meta-analytic approach to summarize the temporal stability of 358 risk preference measures (33 panels, 57 samples, 579,114 respondents). Our findings reveal noteworthy heterogeneity across and within measure categories (propensity, frequency and behaviour), domains (for example, investment, occupational and alcohol consumption) and sample characteristics (for example, age). Specifically, while self-reported propensity and frequency measures of risk preference show a higher degree of stability than behavioural measures, these patterns are moderated by domain and age. Crucially, an analysis of convergent validity reveals a low agreement across measures, questioning the idea that they capture the same underlying phenomena. Our results raise concerns about the coherence and measurement of the risk preference construct.</p>