Empirical investigation of some properties of the perceived riskiness of gambles

Abstract Empirical tests of some properties of the perceived riskiness of gambles are reported. In experiments conducted with U.S. and German subjects, we observed a remarkable consistency in risk judgments. Four possible measures of risk, derived by R. Duncan Luce, were examined. We found that risk decreases as a constant amount is added to all outcomes of a gamble. Two of Luce's measures require that risk not change with the addition of a constant, and thus these measures are not appropriate for describing perceived risk. We also found that Luce's logarithmic measure is not empirically valid. Luce's fourth measure (the expectation of the absolute value of the outcomes raised to a parameter θ) seems to have more promise than his other three measures. These results provide some necessary conditions that a new theory or extension of Luce's measures must satisfy.