We compare seven established risk elicitation methods and investigate how robustly they explain eleven kinds of risky behavior with 760 individuals. Risk measures are positively correlated; however, their performance in explaining behavior is heterogeneous and, therefore, difficult to assess ex ante. Greater diversification across risk measures is conducive to closing this knowledge gap. What we find is that performance increases considerably if we combine single-item risk measures to form multiple-item risk measures. Results are improved the more single-item measures they contain, and also if these single-item risk measures use different elicitation methods. Interestingly, survey items perform just as well as incentivized experimental items in explaining risky behavior.