Information Search, Information Purchase

One common way in which decisions are linked sequentially is when the outcomes of an earlier decision provide (part of) the information environment for a later decision. A doctor de­ciding on what laboratory tests to order for a patient is setting up the information environment in which she will make her subsequent diagnostic and treatment decisions. Similarly, a new product manager ordering a market survey is gathering information on which to base a later decision on whether to launch the product. In a shorter time frame these acquisition and use processes merge.

Research on these processes has varied in how explicit the cost of acquiring information is. Russo and Dosher (1983) recorded the subject's eye movements to study which items of information he or she extracts from a decision table and in what order. The cost of an information item is the cognitive effort involved in attending to an item. A related method­ology is the information board (Payne, 1976), in which decision-relevant information is displayed to the subject in a matrix of small envelopes that may be removed and opened. A computer-based analog called Mouselab has been exten­sively used (Payne et al., 1993) to explore underlying cogni­tive processes such as the combination rule being used by the subject.

Information cost is somewhat more explicit in work such as Wason (1960, 1968; Wason & Johnson-Laird, 1972), in which the subject makes an explicit request of the experi­menter to turn over a card based on whether an exemplar fits some unknown rule. In Wason and Johnson-Laird's (1972) experiment, for example, subjects were shown four cards dis­playing E, K, 4, and 7. They were told that each card had a letter on one side and a number on the other and were asked which cards they would turn over to test the rule, "If a card has a vowel on one side it has an even number on the other side." Only 4% of their subjects selected E and 7, the correct

choice. Almost half chose E and 4—an error because the obverse of the 4 card cannot invalidate the rule, and thus pro­duces, at best, evidence consistent with the rule but not test­ing it. This common finding has been interpreted as a general bias toward confirmatory search: seeking evidence that will confirm, rather than test, one's initial beliefs. However, a penetrating analysis by Klayman and Ha (1987) suggests that such search patterns are better understood as examples of a positive test strategy, a generally appropriate heuristic that fails only in relatively rare situations, such as the four-card problem.

Explicit treatments of sampling cost flow easily from the Bayesian inference task discussed earlier (see the section on heuristics and biases). Instead of being presented with a sample of poker chips drawn from the selected bag, subjects are allowed to buy them, at a fixed monetary cost per chip, ^ before making their bet on which bag was selected—a bet for W which they can win money. Findings from many such studies (see Einhorn & Hogarth, 1981, for a review) include the following:

1. Partial sensitivity to normatively relevant variables (e.g., Pitz, 1968, found increased buying when cost per chip was reduced and diagnosticity was increased, and Snapper & Peterson, 1971, found some sensitivity to variations in information quality).

2. Sensitivity to normatively irrelevant variables, such as information order (Fried & Peterson, 1969) and total in­formation available (Levine, Samet, & Brahlek, 1975).

3. Substantial losses (e.g., Kleiter & Wimmer, 1974), which persist with little or no learning over repeated trials (e.g., WaUsten, 1968).

| 4. Both over-purchase and under-purchase (e.g., Hershman & Levine, 1970; largely parallel results are reported in an alternative, regression-based model of information pur­chase by Connolly and colleagues; see Connolly, 1988, for an overview).

The evidence from both Bayesian and regression models °f information purchase suggests that subjects routinely and Persistently make costly errors in balancing the costs and benefits of their information purchases. This should not be surprising. Optimal information purchase requires the subject «> make accurate assessments of how accurate the different sources are, to select the best subset, and to combine the in-^0rmation acquired in an optimal way. Extensive evidence ^Suggests that all three subtasks are quite difficult. It is thus "ceiy that serious nonoptimalities will be found when the ^atance must be struck in practical settings. This is consistent 'rtn the reluctance of patients to seek second and third

Deciding: IB i Multiple Related Events 507

medical opinions before undertaking major courses of treat­ment, which, in our terms, represents a major underpurchase of decision-relevant information, It is also consistent with the huge body of evidence (Guion, 1975) on the predic­tive uselessness of unstructured job interviews—which are, nonetheless, still very widely used and represent a huge over-purchase of decision-irrelevant information. Wherever infor­mation costs and benefits need to be brought into balance, then, there is good reason to suspect significant departures from optimality (March & Feldman, 1981).

Date: 2016-03-03; view: 780