SUMMARY While making decisions in a semantically rich natural situation, people encounter information, such as in texts, that is often massive and at the same time imprecise. Yet the desired information is extracted and decisions are made. In an attempt to study the characteristics of performance in such a situation, we created a laboratory analogue of the stock market. Subjects acting as stockbrokers acquired a conjunctive decision rule for predicting the market performance of a fictitious stock. After receiving suitable training, subjects read 20 quarterly reports containing information about six market information categories, of which only two were relevant to correct decisions. In each trial, subjects recorded their hypotheses about the categories in the the decision rule. Feedback was given following each decision, and the subjects were able to reread each text in order to decide if and how to modify their hypothesis about the rule. On some trials, following the decisions, the subjects were asked to free recall the texts they had just read. Decision performance distinguished between learners (subjects who discovered the decision rule) and nonlearners. The characteristics of hypothesis-selection behavior were similar to those observed in simpler concept-learning problems. Learners displayed a more global approach to the problem, were more sensitive to the given feedback, and utilized the available information more effectively than nonlearners. The category-recall patterns reflected the observed hypothesis-selection and decision behavior and the subjects' overall category-identification strategies. These data were congruent with a model of text comprehension, a result suggesting an equivalence between some of the comprehension and the hypothesis-selection operators. Effective decision in this task was viewed as the ability to acquire an appropriate control schema to guide the comprehension and the analysis of complex, often unreliable, text inputs.

Abstract : Decision making based on information in texts was studied in a laboratory analogue of a complex, natural, information-analytic domain. Subjects acting as stock brokers acquired a conjunctive decision rule for predicting the market performance of a fictitious stock. Subjects read quarterly reports containing information on six market-information categories, of which only two were relevant to correct decisions. Decision performance differentiated between Learners (subjects who discovered the relevant categories) and Nonlearners. Hypothesis selection behavior was similar to that reported with simpler concept learning problems. The category recall pattern reflected hypothesis selection, decision behavior, and subjects' overall category identification strategies. Further, these data were congruent with a model of text comprehension. Effective decision making in this task was viewed as the ability to acquire an appropriate control schema to guide comprehension and analysis of complex, often unreliable text inputs. (Author)