Academic journal article Psychonomic Bulletin & Review

Clinical Expertise and Reasoning with Uncertain Categories

Academic journal article Psychonomic Bulletin & Review

Clinical Expertise and Reasoning with Uncertain Categories

Article excerpt

Expert clinical psychologists, clinical psychology graduates, and nonclinical graduate students were presented with clinical and nonclinical cases in which the diagnosis or category membership of a character was uncertain; they then made feature predictions about the character. For each case, there was a diagnosis or category that was highly probable and a less likely alternative that either did (relevant condition) or did not (neutral condition) alter predictions. For clinical cases, clinical experts and graduate clinicians gave different predictions for the relevant and neutral conditions, indicating that they had considered the uncertain nature of the diagnosis in their predictions. Although they acknowledged that the diagnosis was uncertain, nonclinical students ignored the less likely diagnostic alternatives when making predictions. For the nonclinical cases, all three groups made predictions based on only the most likely category alternative. The results showed that clinical training and/or experience promote multiple-category reasoning, but that this effect is domain specific.

Category-based induction is inherently probabilistic, and new observations increase or decrease confidence in the belief that a property generalizes across category members. Murphy and Ross (1994) highlighted an additional complexity in category-based inference: Predictions may have to be made about objects whose category membership has yet to be determined. For example, a patient's symptoms may be consistent with a number of alternative diagnoses, but predictions about the future course of their illness may have to be made before a specific diagnosis has been reached.

An important question, therefore, is how people make inferences when category membership is uncertain. One answer comes from models that assume that induction follows Bayesian principles. According to Anderson's (1991) rational model, people estimate the probabilities of a predicted property for each category alternative and combine these estimates. Malt, Ross, and Murphy (1995) tested this prediction by presenting undergraduates with stories about a character who was most likely a realtor, but who may have belonged to a different category (cable TV technician or burglar). The predicted probability of a future behavior (the man ringing the doorbell) was lower when the category alternative was "burglar" than when it was "technician." According to the Bayesian prescription, people consider multiple categories and their predictions will differ across these conditions. Malt et al., however, found that predictions were based on a consideration of only the most probable category, and that subjects ignored less certain alternatives. Such "single-category reasoning" has been found in a range of induction tasks that utilize real-world and artificial categories (see, e.g., Murphy & Ross, 1994; Ross & Murphy, 1996).

One important extension of this work has been the study of predictions about objects that belong to more than one category (e.g., muffins are both "breads" and "breakfast foods"). Predictions of the probability that such an object has a particular property can vary depending on which category is considered. Although Anderson's (1991) model does not apply in this case, related Bayesian approaches (e.g., Tenenbaum, Kemp, & Shafto, 2007) suggest that the multiple categories to which an instance belongs should be considered in property inference. Murphy and Ross (1999), however, found that people usually make predictions about cross-classified objects based on only the category judged to be the most appropriate description in a given context. Moreover, single-category reasoning was equally prevalent for cross-classified objects and for those whose category membership was uncertain.

Although there appears to be broad support for the prevalence of single-category reasoning, it is interesting to consider whether it is affected by expertise when one makes inferences with multiple (uncertain or crossclassified) categories. …

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.