AbstractWhat determines risk attraction or aversion? We experimentally examine three factors: the gain-loss dichotomy, the probabilities (0.2 vs. 0.8), and the money at risk (7 amounts). We find that, for both gains and losses and for low and high probabilities, the majority display risk attraction for small amounts of money, and risk aversion for larger amounts. Thus, when examining the risk attitudes of the majority, what matters is the amount of money at risk, and not the gain-loss dichotomy, or the probabilities. Yet the frequency of risk-attraction behavior does vary according to the gain-loss dichotomy and to the probabilities involved. Since Kahneman and Tversky, the literature has studied gain-loss reflections. We submit that a reflection can be decomposed into a "translation" and a probability "switch." We find (a) a translation effect for low probabilities of the bad outcome, but not for high ones; (b) a strong switch effect for gains, but not for losses, and (c) a strong reflection effect for high probabilities of gains, but not for low ones. We also argue that, while both the translation effect and the switch effect contradict the expected utility hypothesis, the translation effect implies a deeper violation of preference theory, invalidating non-paternalistic welfare economics.