Theideas we have reviewed to this point all associate preference or value with an outcome in isolation from others. That is, they suppose that a specific outcome has a specific utility to a specific decision maker. Both casual reflection and careful research show that this assumption is false. One's feelings about a $3,000 pay raise, for example, might shift significantly if one discovered that a rival had made more, or less; if one expected nothing, or $5,000; or if it was given for merit rather than as a cost of living adjustment. Comparison processes of various sorts influence the value we attach to options and outcomes.
Relatively recently, theories of preference have attempted to integrate the emotions of regret and disappointment through the use of comparisons. Regret theory (Bell, 1982; Loomes & Sugden, 1982) posits that the utility of a risky op-t'on depends not only on outcomes and associated probability but also on comparisons between the outcomes of the chosen and unchosen options. The modified utility function is "e expected utility of the option considered in isolation plus aregret-rejoice component that adjusts for comparison with what might have been if another option had been selected. A fruiter feels good about the success of a new employee she
has selected; perhaps she feels an additional pleasure when she learns that the rejected candidate is performing poorly at another firm. Thus, the feeling of regret (rejoicing) occurs when you would have been better (worse) off if you had made another choice. Studies have shown that anticipated regret leads to regret-minimizing choices (Josephs, Larrick, Steele, & Nisbett, 1992; Ritov, 1996; Zeelenberg, Beattie, van der Pligt, & de Vries, 1996). changes attitudes about behavior (Parker, Stradling, & Manstead, 1996), and changes future behavior (Richard, van der Pligt, & de Vries, 1996). Anticipation of regret about poor outcomes has been used to explain why consumers purchase higher priced but well-known brands (Simonson, 1992) and why they are reluctant to trade equivalent lottery tickets, even with an added cash incentive (Bar-Hillel, 1996).
Where regret theory compares outcomes across alternatives, disappointment theory (Bell, 1985; Loomes & Sugden, 1986) compares outcomes across different states of nature. In the hiring example, the recruiter whose chosen candidate turns in a poor performance feels bad both because of the poor performance itself and because she is disappointed in her expectations of a good performance. As with regret theory, disappointment theory adjusts the basic expected utility of an option according to the comparisons across the different possible outcomes. Finally, decision affect theory (Mellers, Schwartz, Ho, & Ritov, 1997; Mellers, Schwartz, & Ritov, 1999) integrates both disappointment and regret feelings with the utility of an option in order to determine the overall anticipated pleasure of an option.
Regret, disappointment, and decision affect theories all start with the basic expected utility of an outcome and make an adjustment to reflect comparisons with other outcomes, real or imagined. Comparisons are also central to equity theory (Adams, 1965; Walster, Berscheid, & Walster, 1973), in which an outcome's value is modified by the recipient's judgment of whether it was fair. According to equity theory, equity is achieved when the ratio of outputs (e.g., salary, benefits, rewards, punishment) to inputs (e.g., hours worked, effort, organizational citizenship behaviors) are the same for all individuals being considered. Thus, in order to determine if equity is achieved, a comparison other (e.g., a coworker) is required. Early studies investigated equity theory by placing subjects in an experimental work context in which they received payment for the amount of work completed. Subjects were informed about the pay given to other similar workers. Research results have strongly supported equity theory predictions (Greenberg, 1982). Equity imbalance was restored in a manner consistent with equity theory: Underpaid workers decreased their performance (i.e., lowered their inputs), whereas overpaid workers increased their performance (increasing inputs). In an interesting field study (Greenberg, 1988), workers
502 Judgment and Decision Making
were temporarily reassigned to offices that were either of higher or lower status than their regular offices. Consistent with equity theory, those assigned to higher status offices increased their performance, whereas those in lower status offices decreased their performance.
Inalmost every practical choice situation, each of the options being considered has a number of features, attributes, or dimensions that affect its worth to the decision maker. A job, for example, might be defined in terms of such dimensions as salary, location, interest of work, promotion possibilities, and so on. Researchers have proposed a number of alternative models to describe the process by which decision makers choose between such multiattribute alternatives.
Multiattribute utility theory (MAUT) models suppose that what people do (or, in the prescriptive use, should do) is to evaluate each attribute of each alternative, add the resulting utilities into an overall utility for each alternative, and choose the alternative with the highest total. This is referred to as a compensatory model, in the sense that an improvement on one attribute can compensate for or trade off against a loss on another. (We discuss decision-aiding procedures for making these tradeoffs in the following section.) Some authors (e.g., Edgell & Geisler, 1980) have proposed modifications of the basic MAUT models, called random utility models, to reflect the fact that subjects' preferences are not always stable from one occasion to another.
Conjunctive models reflect preferences of the screening type, such as an army physical examination. A candidate with flat feet, for example, would be rejected regardless of how well he or she scores on other measures of physical fitness. These models are thus noncompensatory, in the sense that more on one attribute may not make up for less on another: Any low attribute value makes the entire alternative low value. An early conjunctive model, the satisficing rule, was proposed by Simon (1955). Simon argued that in real settings MAUT procedures make unrealistic demands on a decision maker's time and attention. Instead, decision makers search for an alternative that is acceptable on all important dimensions and stop their search with the first such alternative. Note that this again introduces an element of probabilism into the choice, in that the order in which alternatives are considered may determine which of several acceptable options is found first. (Simon, 1955, also argued that aspiration levels may change as search proceeds, adding a second element of probabilism.)
Lexicographic (dictionary-like) models rely on sequential comparisons between alternatives. Options are compared first
on the most important attribute, and if they differ, the winnino option is chosen. If they tie, the next most important attribute is considered, and so on, until a winner is found. Another ver sion of this, called the elimination by aspects (EBA) model selects an attribute (or '"aspect") at random and eliminates from consideration any option that fails to reach threshold on this attribute. The process continues until only one option remains, and it is then chosen. (Note that neither of these processes is compensatory: Overall attractive options may be eliminated by a loss on an early comparison.)
Additive difference models (Tversky, 1969) assume that the decision maker compares alternatives along one dimension at a time, storing the sum of the differences favoring one alternative over the other. Probabilistic versions of this rule have also been proposed, in which comparison terminates when one alternative reaches some threshold of cumulative advantage over the other.
A number of authors (Beach & Mitchell, 1978; Payne, Bettman & Johnson, 1993) have suggested that the combination rule that a decision maker uses represents a tradeoff between effort and accuracy. The fully compensatory MAUT rule allows the fullest consideration of all attributes and values, but requires extensive information-processing effort. Other rules are less effortful, but do not guarantee that the best option will be chosen.