Abstract
Three studies provide convergent evidence that the inverse base-rate effect (CitationMedin & Edelson, 1988) is mediated by rule-based cognitive processes. Experiment 1 shows that, in contrast to adults, prior to the formal operational stage most children do not exhibit the inverse base-rate effect. Experiments 2 and 3 demonstrate that an adult sample is a mix of participants relying on associative processes who categorize according to the base-rate and participants relying on rule-based processes who exhibit a strong inverse base-rate effect. The distribution of the effect is bimodal, and removing participants independently classified as prone to rule-based processing effectively eliminates the inverse base-rate effect. The implications for current explanations of the inverse base-rate effect are discussed.
Acknowledgments
This research was supported by the Swedish council for Research in the Humanities and Social Sciences. We would like to thank John K. Kruschke, Henrik Olsson and Magnus Persson for helpful comments on previous versions of this manuscript and Ann-Margret Rydell for advice concerning children's learning abilities in general. We are also indebted to Erik Hellman for help with running Experiment 1, and to Vladimir Petrovich for running Experiment 3. Portions of this research were presented by Pia Wennerholm at the Experimental Psychology Society meeting in London, 4-5 of January, 2001.
Notes
The claim is not that young children can never analyse objects dimensionally or that dimensions never have psychological reality for them. The opposite view has been argued in several papers emanating from the integrality–separability framework. The claim is rather that nonanalytic processing is the preferred, or primary mode for children.
In this experiment, the learning criterion was slightly lower (i.e., to 22/24 correct responses) than in some previous studies on the inverse base-rate effect (e.g., in CitationJuslin et al., 2001, the learning criterion was 23/24 correct responses) in order to include a larger sample in the child group. It should be noted, however, that the response pattern with the smaller sample is very similar to the results presented here.
Note that a conflicting probe in the present experiment is constructed by a compound of any two perfect predictors (see also CitationKruschke, 1996), not just the pair having been presented with the same imperfect predictor. This allows a computation of four probes instead of two. Because a separate analysis of only the “really competing” predictors (i.e., those that share a common imperfect predictor) showed a highly similar pattern to the one reported, we chose not to provide separate data for those probes.
In the chi-square analyses, several responses of the same participant are not entered twice, to avoid dependency between the cells.
In the present paper, we rely on a definition of learning efficiency that involves general overall task performance. This ability may or may not depend on an ability to concentrate on learning the more common categories first, and it is possible that this learning efficiency is not reflected in performance on the rare categories.
Recent data from Juslin et al. (Citation2001, Exp. 3) indicate that the interpretation of responses to perfect predictors, such as PC and PR, is particularly problematic. When participants were allowed to give an “other disease” response, the choice proportion for the common disease on the PC probe was only 58%. This figure is considerably lower than the corresponding proportion for the children in the present experiment (71%). However, naturally we cannot conclude that the participants' learning performance in that experiment was particularly poor (they did pass a standard learning criterion).
Note that learning efficiency is only used as an indirect indicator of predisposition to engage in rule-based behaviour. All else being equal, if this behaviour did not vary as a function of learning efficiency we would actually expect the opposite; inefficient learners would have less knowledge and more room for eliminative inference.
A subset of the data in Experiment 2 has previously been reported in Juslin et al. (Citation2001, for details, see Experiment 1). In the present article, these data are reanalysed with regard to the predictions of different strategies of learning and generalization as discussed above.