Academic journal article Memory & Cognition

No Evidence for Rule-Based Processing in the Inverse Base-Rate Effect

Academic journal article Memory & Cognition

No Evidence for Rule-Based Processing in the Inverse Base-Rate Effect

Article excerpt

The inverse base-rate effect in categorization (Medin & Edelson, 1988) arises when participants assign an ambiguous stimulus to a category that occurred less frequently than an alternative category, against the principles of Bayesian decision making. In the experiment reported in this article, rule-based and attention-shifting accounts of the inverse base-rate effect were evaluated. Participants completed a categorization task, known to produce the inverse base-rate effect, under standard conditions, under time pressure, and with a secondary task load. The inverse base-rate effect persisted under severe time pressure and under secondary task load. The results provided no evidence for the role of rule-based processes in producing the inverse base-rate effect. The data from the experiment are compatible with an attention-shifting account.

The inverse base-rate effect is the finding tiiat participants in a categorization task sometimes assign an ambiguous stimulus to a category that occurred less frequently dian other categories, against the normative specification of Bayesian decision tiieory. The effect was first reported by Medin and Edelson (1988) in a simulated medical diagnosis task. The structure of the training stimuli used in the task is shown in Table 1. Symptoms PCa, PCb, and PCC were perfect predictors of a common disease, symptoms PRa, PRb, and PRC were perfect predictors of a rare disease, and symptoms Ia, Ib, and Ic were imperfect predictors (associated with both a common disease and a rare disease). In the transfer stage that followed the training task, die use of base-rate information was investigated using tests with the imperfect predictors (Ia, Ib, Ic), with two conflicting predictors (one of a common disease and one of a rare disease-i.e., PCaJ>Ra, PCb.PRb, or PCC.PRC), and with combined predictors (Ia.PCa.PRa, Ib.PCb.PRb, or IC.PCC. PRC). Medin and Edelson reported the surprising result that participants preferred the less frequent category for die conflicting predictors (against the base rate), whereas they used base-rate information correctly for the imperfect predictors and for the combined predictors. The inverse base-rate effect has since been replicated in a number of studies, using different stimuli and procedures (e.g., Kalish, 2001; Kruschke, 1996; Winman, Wennerholm, Juslin, & Shanks, 2005).

Whereas the robustness of the inverse base-rate effect is not disputed, there is disagreement about the processes that underlie the effect. Two different kinds of explanations have been put forward. First, the effect has been attributed to context-dependent selective attention, operating within an otherwise normative learning and decision system. Probably the best example of such an account is Kruschke's (1996) ADIT model. ADIT has been developed further since (e.g., Kruschke, 2001 b), but the original model is appropriate for our present purposes. According to ADIT (which is an extension of Gluck & Bower's, 1988, component-cue model), the inverse base-rate effect arises tiirough selective attention during learning. Because the common disease occurs more frequently in the training phase, the symptoms associated with the common disease tend to be learned before those associated with the rare disease. Therefore, both the imperfect predictor / and the perfect predictor PC become associated with the common disease. When participants subsequently encounter the rare disease, attention moves away from the imperfect predictor I (which is shared by both diseases) and becomes focused on the perfect predictor PR (which is diagnostic for the rare disease). As a result, the representation of the common disease involves associative links from both I and PC, whereas the rare disease will primarily be associated with the predictor PR. This asymmetry in the representation of the disease categories is sufficient to produce the inverse base-rate effect (Kruschke, 1996).

An alternative explanation of the inverse base-rate effect has been proposed by Justin, Wennerholm, and Winman (2001). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.