In my paper 'Preference reversal in equal probability gambles: A case for anchoring and adjustment' (this issue), I report some results concerning preference reversal in equal-probability gambles, and argue that these results can be explained by anchoring and adjustment, but not by compatibility. Caverni (this issue) argues that these results can be explained by compatibility. For example, according to Caverni, the finding that there is a higher preference for high-variance equal-probability gambles in pricing than in choice can be explained by compatibility because 'the payoffs are compatible with the cardinal response scale, but they are not compatible with the 0-1 scale used in the choice task'. In my view, however, this argument cannot explain the observed difference in preference, since all the payoffs are equally compatible with the response scale. To explain this difference in preference, one needs to assume that the higher payoff is 'more compatible' with the response scale than the rest of the payoffs. But this argument cannot be derived from the principle of compatibility, that is, from the principle that 'the weight of any input information is enhanced by its compatibility with the output' (Tversky, Sattath, Slovic, 1988, p. 376).
A more elaborate argument based on compatibility is that in the process of evaluating equal-probability gambles, the information is encoded along two dimensions -expected value and riskand that expected value weighs more in pricing than in choice (or, equivalently, that risk weighs more in choice than in pricing). This argument may explain the observed preference reversals between pricing and choice, both for attractive and unattractive gambles, under the assumption (which is not necessary for the explanation of preference reversal in unequal-probability gambles) that people are risk averse in the domain of gain and risk seeking in the domain of loss . However, this argument cannot explain why in the pricing task preference for high-variance gambles is much higher in attractive gambles than in unattractive gambles. Anchoring and adjustment, on the other hand, can explain this difference. Furthermore, this argument cannot explain why the difference in preference for high-variance gambles in equal-probability gambles is different in rating and choice (Ganzach, this issue, Experiment 3), unless one makes the assumption that rating is more compatible with the expected value dimension than with the risk dimension. But there is no evidence to support this assumption.'
A short historical perspective is relevant here. The original research that led to the discovery of the preference reversal phenomenon was conducted by Slovic and Lichtenstein in the late 1960s. One of the major findings of this research was that the amount to win has the largest weight in the pricing of attractive bets and the amount to lose has the largest weight in the pricing of unattractive bets (Slovic and Lichtenstein, 1968) -a phenomenon that cannot be explained by compatibility but can be explained by anchoring ad adjustment. Based on an anchoring and adjustment hypothesis, Slovic and Lichtenstein predicted -and confirmedpreference reversal in attractive unequal-probability gambles , as well as unattractive unequalprobability gambles . Now, while compatibility can explain the findings of Lichtenstein and Slovic (1971, 1973) equally as well as anchoring and adjustment, it cannot explain the findings of Slovic and Lichtenstein (1968), which were the basis for the prediction of preference reversal in attractive as well as unattractive gambles. Anchoring and adjustment, on the other hand, can explain the findings of . Thus, at least when these sets of results are considered (i.e. Lichtenstein and Slovic, 1971, 1973), compatibility appears to be an ad hoc explanation for preference reversal, while anchoring and adjustment does not.
Two other points, which are relevant to the comparison between anchoring and adjustment and compatibility as explanations for preference reversal, are worth mentioning. First, anchoring and adjustment could be viewed as a more primitive process than compatibility. Indeed, to a large extent, Tversky et al. (1988) trace the effects of