Some violations of the expected utility hypotheses were explicated by Machina (1982) in a more general framework using the notion of local utility function and postulating that: (a) each local utility function displays decreasing absolute risk aversion and (b) the measure of absolute risk aversion is everywhere nondecreasing with respect to shifts towards first-order stochasticallY dominating distributions. In this paper we show that the evidence produced in the course of testing the "preference reversal" phenomenon, in conjunction with the assumption that individuals are disinclined to participate in fair symmetric bets, lends support to Machina's hypotheses.