Order effects in diagnostic reasoning with four candidate hypotheses

Abstract

Order effects in diagnostic reasoning with four candidate hypotheses Felix G. Rebitschek (felix.rebitschek@uni-greifswald.de) University of Greifswald, Department of Psychology Agnes Scholz (agnes.scholz@psychologie.tu-chemnitz.de) Chemnitz University of Technology, Department of Psychology Franziska Bocklisch (franziska.bocklisch@psychologie.tu-chemnitz.de) Chemnitz University of Technology, Department of Psychology Josef F. Krems (josef.krems@phil.tu-chemnitz.de) Chemnitz University of Technology, Department of Psychology Georg Jahn (georg.jahn@uni-greifswald.de) University of Greifswald, Department of Psychology alternatives, but none is decisive by itself. The order in which symptoms are encountered can influence the final diagnosis because the initial diagnostic hypotheses may affect how the subsequent symptoms are weighed and integrated (e.g., Chapman, Bergus & Elstein, 1996). If symptoms are observed in sequence, the initially encountered symptoms trigger diagnostic hypotheses (Mehlhorn, Taatgen, Lebiere, & Krems, 2011). Sequential symptom processing towards the initial hypothesis demonstrates a confirmation bias (Nickerson, 1998), which would be overcome if all alternative diagnostic hypotheses could be considered in parallel. In previous studies such impartial symptom integration sometimes succeeded for two alternative diagnoses (McKenzie, 1998), but doubts have been raised whether more than two alternative diagnoses can be considered impartially in parallel (Dougherty & Hunter, 2003). According to normative Bayesian information integration, the order of symptom presentation should not matter. Symptom patterns equally supportive of two alternative diagnoses should produce equal proportions of these diagnoses. However, already updating of a single hypothesis can be biased by the order, in which pieces of evidence are encountered (Wang, Johnson, & Zhang, 2006). Hogarth and Einhorn (1992) specified circumstances under which normative updating of a single belief is possible and no order effects should occur (e.g. stepwise simple evaluation of short and consistent sequences). Yet, models of sequential information integration including the belief adjustment model of Hogarth and Einhorn (1992) typically postulate a disproportionately large influence of early encountered information resulting in a primacy effect. Abstract Sequentially observed symptoms in diagnostic reasoning have to be integrated to arrive at a final diagnosis. In our experiments employing quasi-medical problems, four sequentially presented symptoms were consistent with multiple diagnostic hypotheses. We tested whether symptom order creates biases in symptom evaluation. Early symptoms induced a bias towards the initial hypothesis even though an alternative hypothesis was equally supported. In two experiments, stepwise ratings were prompted to explicitly highlight alternative hypotheses. Explicit highlighting eliminated the bias towards the initial hypothesis if only two hypotheses competed, but the bias remained if more than two hypotheses were associated with symptoms in a sequence. Our results are consistent with process models of information integration that specify how early information can frame the processing of later information. Extending previous results obtained with fewer contending hypotheses, we show limits in impartially considering more than two hypotheses. Keywords: Order Effects; Diagnostic Reasoning; Multiple Candidate Hypotheses; Construction Integration Theory Introduction When humans explain observations in their environment, they apply knowledge about possible causes and the effects that each cause can bring about. Explaining observed symptoms by a diagnosis that specifies the most probable cause can be difficult for symptoms which are ambiguous and thus consistent with multiple diagnoses or inconsistent and hard to subsume under a single diagnosis (Johnson & Krems, 2001). Imagine the sequential integration of symptoms in medical diagnosis in its simplest form: You, a physician, become aware of a symptom pointing towards different possible diseases your new patient might have caught. Bit by bit you take notice of a second, third and a fourth symptom. Some of them are unspecific, others strengthen your belief in a diagnosis and weaken