Diagnostic Reasoning With Circumstantial Evidence (problem-Solving)

185 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1984.Previous theories of diagnostic reasoning have focussed on the role of direct, covariation evidence of causal relations. Here, the use of circumstantial evidence was examined. Subjects worked with a prototypical case of diagnostic reasoning with circumstantial evidence: solving fictional murder mysteries.Dulany's (unpublished manuscript, 1979) theory of propositional learning (TPL) was used to develop a quantitative model of the process by which belief is transferred from evidential premises to causal hypotheses. The reasoning process is driven by the two goals of identifying the murderer and explaining the pattern of clues. In evaluating clues, individuals consider two types of links within the causal scenario: an explanatory linkage, the relative sufficiency of guilt or innocence for explaining the clue; and a diagnostic linkage, the relative necessity of predicting a clue or its denial of the murderer. The model describes a cascaded reasoning process including intermediate conclusions.Directed belief assessments were used to trace subjects' diagnostic reasoning in several experiments. Experiment 1 provided a detailed, quantitative picture of the development of belief over a series of clues. Experiment 2 replicated Experiment 1, with a change in assessment order. Experiment 3 examined inference to an intermediate conclusion, subjective evidence value. Experiment 4 contrasted TPL and Bayesian models of belief revision. In Experiments 1, 2, and 4, two aspects of the causal scenario--the scope of evidence, and the position of evidence in sequence--were manipulated. Results consistently supported the TPL reasoning model, the hypothesis that subjects develop a causal scenario, and the hypothesis that the reasoning process is invariant over experimental conditions and levels of belief. Precise quantitative predictions of belief in causal hypotheses were achieved, without empirically estimated parameters. Memory data collected in Experiments 1 and 4 also supported the scenario hypothesis.The main features of the theoretical analysis were strongly supported. The TPL reasoning model is superior to alternatives in adequacy and detail, and the directed process tracing methodology provided a detailed picture of the reasoning process. Limitations of the present approach and suggestions for possible extensions are discussed. The role of consciousness in diagnostic reasoning is also discussed.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD