Self-signaling and diagnostic utility in everyday decision making1

A self-signaling action is an action chosen partly to secure good news about one’s traits or abilities, even when the action has no causal impact on these traits and abilities. We discuss some of the odd things that happen when self-signaling is introduced into an otherwise rational conception of action. We employ a signaling game perspective in which the diagnostic signals are an endogenous part of the equilibrium choice. We are interested (1) in pure self-signaling, separate from any desire to be regarded well by others, and (2) purely diagnostic motivation, that is, caring about what an action might reveal about a trait even when that action has no causal impact on it. When diagnostic motivation is strong, the person’s actions exhibit a rigidity characteristic of personal rules. Our model also predicts that a boost in self-image positively affects actions even though it leaves true preferences unchanged — we call this a “moral placebo effect.” 1 The chapter draws on (co-authored) Chapter 3 of Bodner’s doctoral dissertation (Bodner, 1995) and an unpublished MIT working paper (Bodner and Prelec, 1997). The authors thank Bodner’s dissertation advisors France Leclerc and Richard Thaler, workshop discussants Thomas Schelling, Russell Winer, and Mathias Dewatripont, and George Ainslie, Michael Bratman, Juan Carillo, Itzakh Gilboa, George Loewenstein, Al Mela, Matthew Rabin, Duncan Simester and Florian Zettelmeyer for comments on these ideas (with the usual disclaimer). We are grateful to Birger Wernerfelt for drawing attention to Bernheim's work on social conformity. Author addresses: Bodner – Director, Learning Innovations, 13\4 Shimshon St., Jerusalem, 93501, Israel, learning@netvision.net.il; Prelec — E56-320, MIT, Sloan School, 38 Memorial Drive, Cambridge, MA 02139, dprelec@mit.edu. 1 Psychological evidence When we make a choice we reveal something of our inner traits or dispositions, not only to others, but also to ourselves. After the fact, this can be a source of pleasure or pain, depending on whether we were impressed or disappointed by our actions. Before the fact, the anticipation of future pride or remorse can influence what we choose to do. In a previous paper (Bodner and Prelec, 1997), we described how the model of a utility maximizing individual could be expanded to include diagnostic utility as a separate motive for action. We review the basic elements of that proposal here. The inspiration comes directly from signaling games in which actions of one person provide an informative signal to others, which in turn affects esteem (Bernheim, 1994). Here, however, actions provide a signal to ourselves, that is, actions are selfsignaling. For example, a person who takes the daily jog in spite of the rain may see that as a gratifying signal of willpower, dedication, or future well being. For someone uncertain about where he or she stands with respect to these dispositions, each new choice can provide a bit of good or bad "news.” We incorporate the value of such "news" into the person's utility function. The notion that a person may draw inferences from an action he enacted partially in order to gain that inference has been posed as a philosophical paradox (e.g. Campbell and Sawden, 1985; Elster, 1985, 1989). A key problem is the following: Suppose that the disposition in question is altruism, and a person interprets a 25¢ donation to a panhandler as evidence of altruism. If the boost in self-esteem makes it worth giving the quarter even when there is no concern for the poor, than clearly, such a donation is not valid evidence of altruism. Logically, giving is valid evidence of high altruism only if a person with low altruism would not have given the quarter. This reasoning motivates our equilibrium approach, in which inferences from actions are an endogenous part of the equilibrium choice. As an empirical matter several studies have demonstrated that diagnostic considerations do indeed affect behavior (Quattrone and Tversky, 1984; Shafir and Tversky, 1992; Bodner, 1995). An elegant experiment by Quattrone and Tversky (1984) both defines the self-signaling phenomenon and demonstrates its existence. Quattrone and Tversky first asked each subject to take a cold pressor pain test in which the subject's arm is submerged in a container of cold water until the subject can no longer tolerate the pain. Subsequently the subject was told that recent medical studies had discovered a certain inborn heart condition, and that people with this condition are “frequently ill, prone to heart-disease, and have shorter-than-average life expectancy.” Subjects were also told that this type could be identified by the effect of exercise on the cold pressor test. Subjects were randomly assigned to one of two conditions in which they were told that the bad type of heart was associated with either increases or with decreases in tolerance to the cold water after exercise. Subjects then repeated the cold pressor test, after riding an Exercycle for one minute. As predicted, the vast majority of subjects showed changes in tolerance on the second cold pressor trial in the direction correlated of “good news”—if told that decreased tolerance is diagnostic of a bad heart they endured the near-freezing water longer (and vice versa). The result shows that people are willing to bear painful consequences for a behavior that is a signal, though not a cause, of a medical diagnosis. An experiment by Shafir and Tversky (1992) on "Newcomb's paradox" reinforces the same point. In the philosophical version of the paradox, a person is (hypothetically) presented with two boxes, A and B. Box A contains either nothing or some large amount of money deposited by an "omniscient being." Box B contains a small amount of money for sure. The decision-maker doesn’t know what Box A contains choice, and has to choose whether to take the contents of that box (A) or of both boxes (A+B). What makes the problem a paradox is that the person is asked to believe that the omniscient being has already predicted her choice, and on that basis has already either "punished" a greedy choice of (A+B) with no deposit in A or "rewarded" a choice of (A) with a large deposit. The dominance principle argues in favor of choosing both boxes, because the deposits are fixed at the moment of choice. This is the philosophical statement of the problem. In the actual experiment, Shafir and Tversky presented a variant of Newcomb’s problem at the end of another, longer experiment, in which subjects repeatedly played a Prisoner’s Dilemma game against (virtual) opponents via computer terminals. After finishing these games, a final “bonus” problem appeared, with the two Newcomb boxes, and subjects had to choose whether to take money from one box or from both boxes. The experimental cover story did not mention an omniscient being but instead informed the subjects that "a program developed at MIT recently was applied during the entire session [of Prisoner’s Dilemma choices] to analyze the pattern of your preference.” Ostensibly, this mighty program could predict choices, one or two boxes, with 85% accuracy, and, of course, if the program predicted a choice of both boxes it would then put nothing in Box A. Although it was evident that the money amounts were already set at the moment of choice, most experimental subjects opted for the single box. It is “as if” they believed that by declining to take the money in Box B, they could change the amount of money already deposited in box A. Although these are relatively recent experiments, their results are consistent with a long stream of psychological research, going back at least to the James-Lange theory of emotions which claimed that people infer their own states from behavior (e.g., they feel afraid if they see themselves running). The notion that people adopt the perspective of an outside observer when interpreting their own actions has been extensively explored in the research on self-perception (Bem, 1972). In a similar vein, there is an extensive literature confirming the existence of “self-handicapping” strategies, where a person might get too little sleep or under-prepare for an examination. In such a case, a successful performance could be attributed to ability while unsuccessful performance could be externalized as due to the lack of proper preparation (e.g. Berglas and Jones, 1978; Berglas and Baumeister, 1993). This broader context of psychological research suggests that we should view the results of Quattrone and Tversky, and Shafir and Tversky not as mere curiosities, applying to only contrived experimental situations, but instead as evidence of a general motivational “short circuit.” Motivation does not require causality, even when the lack of causality is utterly transparent. If anything, these experiments probably underestimate the impact of diagnosticity in realistic decisions, where the absence of causal links between actions and dispositions is less evident. Formally, our model distinguishes between outcome utility — the utility of the anticipated causal consequences of choice — and diagnostic utility — the value of the adjusted estimate of one’s disposition, adjusted in light of the choice. Individuals act so as to maximize some combination of the two sources of utility, and (in one version of the model) make correct inferences about what their choices imply about their dispositions. When diagnostic utility is sufficiently important, the individual chooses the same action independent of disposition. We interpret this as a personal rule. We describe other ways in which the behavior of self-signaling individuals is qualitatively different from that of standard economic agents. First, a self-signaling person will be more likely to reveal discrepancies between resolutions and actions when resolutions pertain to actions that are contingent or delayed. Thus she might honestly commit to do some worthy action if the circumstances requiring t