The conjunction fallacy: The case for the existence of competing heuristic strategies

Abstract

A study was conducted to evaluate the effects of training on the incidence of the conjunction fallacy. One group received training in the extension rule (normative), the other training which stressed that judgments should be based on similarity or representativeness (non‐normative). Participants receiving the former made fewer errors, those receiving the latter made more errors. However, multiple regression analysis showed that under both training regimes in a majority of instances only the smaller component probability was statistically significant in determining the conjunction. A second study, omitting the training element, replicated this finding. Both studies highlight the fact that existing theories cannot account for the pattern of participants' responses under the training conditions employed. It is proposed that rather than choose between two competing strategies (Agnoli & Krantz, 1989), participants derive their estimate in two stages, first by selecting a reference point in the probability continuum, usually based on the ‘ surprise value’ of the smaller component event (Shackle, 1969) and then assigning a value to the conjunction relative to this point. During the second stage, training is hypothesized to produce its effect as participants weigh the available information deriving some compromise reflecting both normative and non‐normative tendencies.