Baron, J. (1992). The effect of normative beliefs on anticipated emotions. Journal of Personality and Social Psychology, 63, 320-330.

## 1  The effect of normative beliefs on anticipated emotions

Jonathan Baron
Department of Psychology University of Pennsylvania

## 2  Acknowledgment

A version of this paper was presented at a conference on the role of anticipation and regret in decision making, La Jolla, January 7-8, 1991, sponsored by the John D. and Catherine T. MacArthur Foundation. I thank the participants, David M. Clark, Jon Haidt, Daniel Kahneman, the editor, and an anonymous reviewer for comments. The work was supported by a grant from the National Science Foundation (SES-8809299) and by the MacArthur Foundation.

## 3  Abstract

In three experiments, subjects were asked how they would or should make hypothetical decisions, and how they would react emotionally to the options or outcomes. The choices were those in which departures from proposed normative models had previously been found: omission bias; status-quo bias; and the person-causation effect. These effects were found in all judgments, including judgments of anticipated emotion. Arguments against the departures affected judgments of anticipated emotion as well as decisions, even though the arguments were entirely directed at the question of what should be done. In all but one study, effects of these arguments on anticipated emotion were as strong as their effects on decisions or normative beliefs. Thus, in many situations, people think that their emotional reactions will fall into line with their normative beliefs. In other situations, some people think that their emotional reactions have a life of their own. It is suggested that both normative beliefs and anticipated emotions affect decisions.
I take emotions to be states that are subjectively experienced, that have some hedonic component, and that drive or motivate certain kinds of behavior specific to the emotion. Examples of hedonically negative emotions are guilt feelings, regret, sadness, anger, and fear. Anticipation of emotions can affect decisions even when the emotion is not an inherent part of the desired or undesired outcomes. This effect has been widely recognized by clinical psychologists and social psychologists. For example, fear of having a panic attack keeps some agoraphobics from going out in public (Chambless & Gracely, 1989), even when the anticipation of an attack is not related to reality (Cox, Swinson, Norton, & Kuch, 1991). Fear of embarrassment could account for a variety of phenomena in which choices are manipulated in the social-psychology laboratory, such as the Milgram obedience effect (Sabini, 1992).
Decision theorists have also begun to recognize the role of anticipated emotions in decision making. Kahneman and Lovallo (1990) describe the role of fear in decisions under risk, and Lopes (1987) discusses hope as well. Bell (1982, 1985) and Loomes and Sugden (1982; Loomes, 1987) proposed that certain apparent departures from expected-utility theory were the result of anticipated regret, rejoicing, disappointment, and elation. We sometimes avoid taking risks because we anticipate feeling regretful if worse comes to worse. Such anticipated regret could affect many real decisions, such as those made in medicine (Hershey & Baron, 1987). Loewenstein (1987) and Elster (1985) likewise called attention to the role of savoring and other emotions in intertemporal choice. We sometimes put off pleasant events so that we can savor our anticipation of them, and we sometimes get aversive events over with so that we don't have to put up with dread.
In these cases, what matters for decision making is the anticipated emotion, not the real one. People might be wrong to think that they will experience pleasant anticipation rather than impatience, for example. Kahneman and Snell (1990) have documented systematic errors in people's predictions of their hedonic experiences. The present study is concerned with anticipated emotions only, not real ones. Subjects are asked to compare how they would feel in certain situations. Anticipated emotions are important to the extent to which they affect decisions. Real emotions are important for decision making to the extent to which people learn to adjust their anticipations to reality, and real emotions are also important, of course, as outcomes that affect behavior and experience.
Decisions can be described in terms of options leading to outcomes, with the outcome depending on the option chosen as well as on the state of the world, which is often uncertain when the decision is made (Baron, 1988). These consequences include emotions, and the anticipation of these emotional consequences can affect the choice of options.
Emotions are associated with decisions in four ways. The first three concern emotions as consequences: 1. Emotions can depend directly on the expected consequences themselves. Skydivers seek the excitement that is part of skydiving. The rest of us avoid the fear. 2. Emotions can depend on comparison of an outcome to outcomes of different options. If I decide to buy shares of stock and the value of the stock goes down, I will experience regret (Bell 1982, Loomes and Sugden 1982). Such regret depends on my having had another option. 3. Emotions can depend on comparison of an outcome to outcomes of the same option in different states, to what might have happened,' regardless of whether a decision is made or not (Kahneman & Miller, 1986; Bell, 1985; Loomes, 1987). I can be disappointed that my stock went down even if I do not feel regret about buying it. 4. Finally, emotions can also affect decisions directly. Sympathy for a plaintiff in a lawsuit or anger at a defendant can make us choose to provide the plaintiff with greater compensation for an injury, for example.
Decisions are affected by normative beliefs, beliefs about what ought to be done in certain cases or about how decisions ought to be made. For example, people might believe that hurting others through acts is worse than causing the same hurt through omissions (Spranca, Minsk, & Baron, 1991), or that injuries caused by people are less deserving of compensation than those caused by people (Baron, in press a). Normative beliefs include beliefs about personal virtue as well as about interpersonal morality.
At issue in this paper is whether these normative beliefs also affect anticipated emotions. How could such effects occur? In principle, such beliefs could strengthen, weaken, or reverse all four of the emotional effects just listed, and decision makers could anticipate such effects when they make decisions: 1. A belief that smoking is wrong, for example, could reduce the expectation of pleasure from smoking. 2. A belief that harmful acts are worse than harmful omissions could cause us to expect to regret acts that cause harm more than we regret omissions that cause the same harm. 3. A belief that risk-taking is wrong can lead us to expect greater disappointment from taking a risk and losing. 4. A belief that smoking is wrong could cause us to get angry at smokers.
If such effects on anticipated emotions occur, then beliefs could affect decisions in two ways, first, through direct effects on which option is chosen, and, second, through effects on anticipated emotion. If these effects agree, decisions are overdetermined. For example, people could choose an option either because they believe it is morally right or because the want to avoid the guilt feelings that would result from choosing it. The two effects do not necessarily agree, however. People could think that an act is right yet anticipate feeling guilty if they do it. Such a situation can cause difficulty in making decisions. We might expect people to adopt standard ways of resolving such conflicts, some choosing to side with their head,' others with their (anticipated) heart.'
What happens when people are persuaded to change their normative beliefs? Will their tendencies to choose certain options change accordingly? Or, alternatively, will they still tend to choose the options they would have chosen before the change? Persistence of choice could result from an expectation that, although normative beliefs have changed, emotions will not change. A person who comes to believe that premarital sex is not immoral, for example, could still be reluctant to engage in it because of anticipated guilt feelings.
People who believe that their emotions are autonomous are likely to be skeptical of rational thought as a means of making decisions. And people who believe that emotions will fall into line with beliefs automatically are more likely to allow their decisions to be governed by reasons, without fear of negative emotional consequences of doing so. These different beliefs about the autonomy of emotions might also be self-fulfilling. The expectation of guilt feelings could lead to guilt feelings even in the absence of a corresponding moral belief. (I know what I did was not wrong, but I still feel guilty about it.')
The experiments reported here attempt to change (at least briefly) the beliefs of the subjects about how they should choose. Subjects are then asked how they would feel in various cases. Assuming that such arguments are effective in changing beliefs, do they change anticipated emotions too? If so, subjects believe that their emotions are largely determined by their beliefs about what they ought to do. If, on the other hand, otherwise effective arguments do not affect anticipated emotions, subjects must believe that their emotions are not under the control of the reasons that they consciously accept. If these anticipated emotions affect decisions, then decisions, too, are not entirely under the control of non-emotional reasons.
The experiments address other questions, if only indirectly: Do anticipated emotions affect norms and choices? Do emotional reactions maintain beliefs that would otherwise be weakened by argument? Are people whose emotions and beliefs agree more resistant to arguments concerning their beliefs?
The examples chosen for these experiments are drawn from the literature on biases in decision making (e.g., Baron, 1988). Spranca et al. (1991) have argued that many subjects make an unjustified distinction between acts and omissions in deciding what to do or in judging the decisions of others. In one study, subjects were asked to evaluate medical policies for treating a serious disease. In one condition, 20% of the patients with the disease would be brain damaged, and the treatment (which completely cured the disease) would cause brain damage in 15%; in another condition, the proportions were reversed. Across both conditions, subjects rated the omission (no treatment) as a better policy. Some subjects even rated no treatment with 20% brain damage as better than treatment with 15%. Ritov and Baron (1990) likewise found that many subjects prefer not to vaccinate children when the vaccine can kill the children, even though the death rate from the vaccination is a mere fraction of the death rate from the disease it prevents. Subjects often said that they would feel more guilty if a child died from the vaccine than if the child died from the disease.
Related phenomena are the endowment effect, in which subjects require more money to give up something than they are willing to pay for it (Knetsch & Sinden, 1984; Kahneman, Knetsch, & Thaler, 1990), and the status quo effect, in which subjects prefer the option they have (e.g., a retirement plan) to one they do not have, even though they regard the latter as superior when they have neither (Samuelson & Zeckhauser, 1988). Ritov and Baron (1992) present evidence that these effects result in part from a bias toward inaction rather than a bias toward the status-quo itself.
Another effect considered here is the tendency to compensate victims more when they are harmed by people than when they are harmed by nature. We want to compensate a person who suffers from a severe reaction to a vaccine, although we are less prone to compensate someone who suffers just as much from a natural disease (Baron, in press a). Such differences are difficult to justify (Baron, in press a). They may, however, be related to such emotions as anger at the injurer in the cases of human-caused misfortune. In a sense, we compensate the victim to assuage our own emotional reactions.
The arguments given to subjects here are those that might be used in experiments on debiasing' of the sort done by Larrick, Morgan, and Nisbett (1990). An argument designed to remove a bias might be ineffective if it does not changes anticipated emotions as well as beliefs. In such a case, the purported bias might not even be nonnormative (Baron, 1985, pp. 62-63, 1988; Hershey & Baron, 1987). A strong negative emotion associated with one option can reduce the expected utility of that option enough so that it is no longer optimal. (However, we might still do better by trying to control our emotion instead of giving in to it. And when we make decisions for others, as in the vaccination example, we are selfish if we weigh our own emotions too heavily.) The effect of normative arguments on anticipated emotions is therefore relevant both to the success of debiasing in changing decisions and to the normative status of those decisions.

## 4  Experiment 1

In the first experiment, subjects were presented with two cases involving omission bias. The first case involved a choice of whether to vaccinate a child against a disease, given that the vaccine might kill the child, although unvaccinated children were twice as likely to die as vaccinated children. The second case involved a choice of whether to kill one innocent prisoner in order to save two. The two cases are analogous except that death is certain in the prisoner case and the alternative cause of death is natural in the vaccination case. Subjects were asked what they should do in each choice, what they would do, and which outcome would lead to greater feelings of guilt. Some of the subjects were then presented with an argument against the relevance of the act-omission distinction. The argument was addressed to the intellect, not the emotions. The effect of the argument on Should, Would, and Guilt responses could then be examined. The Should response serves as a manipulation check on the effectiveness of the argument.
The main hypothesis is that the argument will affect anticipated guilt feelings if it affects normative beliefs, at least in some subjects. A secondary hypothesis is that some subjects will show effects only in normative beliefs, and not in anticipated emotions. In addition, we can examine the extent to which the Would response is related both to Should and Guilt responses.
I assume that the correct response in both cases is to act. The argument presented to subjects in the debiasing' condition advocated action in both cases. This is a controversial assumption, especially in the prisoner case (where very few subjects agreed with it, even after the argument). Good arguments can be made on the other side of the prisoner case. For example, in any realistic version of this case it would be impossible to know that the expected long-run consequences of shooting are better than those of not shooting. Even from a consequentialist perspective, then, one might do better to follow the rule never shoot anyone,' even in situations in which that rule appears to lead to worse consequences (Hare, 1981). Or, even if shooting has better consequences, a person who refused to shoot might be a better person, a person with the kind of character that will lead to more good in the long run (Hare, 1981). Finally, a person could regard the guilt feelings from shooting the prisoner as worse than death, in which case the consequences of shooting would be clearly worse on the whole than those of not shooting. For subjects who thought of these arguments, the persuasive argument given should be ineffective in changing normative belief.
In assuming that the correct answer is to shoot, I assume that the case is not realistic, that subjects should accept the case as stated, that they should consider the choice (as requested) rather than the character of the decision maker, that decisions should be judged by their expected consequences, and that subjects do not regard the guilt feelings from killing as worse than death. Although these assumption are questionable, they do not affect the conclusions that I draw, which do not depend on the normative status of the argument. (For recent discussion of the act-omission issue, see: Baron, in press b, ch. 7; Bennett, 1966, 1981; Gorr, 1990; and Kuhse, 1987, ch. 2.) From here on, then, I shall speak as though the act were normatively correct, purely for brevity of expression.

Subjects were solicited by placing a sign on a prominent campus walkway at the University of Pennsylvania. Most subjects were undergraduate students there or at nearby Philadelphia College of Pharmacy Science. Subjects were paid $5 per hour for completing this questionnaire and others. 120 subjects were given the following questions:  1. A kind of flu can be fatal to children under 3. A vaccine for this kind of flu has been developed and extensively tested. The vaccine prevents the flu, but it sometimes causes side effects that can be fatal. The death rate for unvaccinated children is 10 out of 10,000 children under 3. These children die from the flu. The death rate for vaccinated children is 5 out of 10,000 children under 3. These children die from the vaccine. A. If you had a child under 3, should you vaccinate your child? B. Do you think that you would vaccinate your child? C. Consider two situations: 1. You vaccinate and your child dies from the vaccine. 2. You do not vaccinate and your child dies from the flu. In which of these situations would you feel guiltier about your decision? 2. Imagine that you and three others are held prisoner in the Middle East. All four of you are innocent. Your captors are planning to murder two of the other prisoners. You do not know which two. They will spare these two if you will murder the single remaining prisoner. (You believe them when they say this.) Nobody except you and your captors will know about your decision. A. Should you kill the one to save the two? B. Do you think that you would kill the one to save the two? C. Would you feel guiltier about your decision if you: 1. committed the murder; 2. did not commit the murder. (Answer 1 or 2.)  Then 63 subjects in a debiasing' condition were given the following argument: The questions you have just answered concern the distinction between actions and omissions. We would like you now to reconsider your answers. First, read this page and think about it. Then answer the questions again on the next page. (Do not go back and change your original answers.) Feel free to comment as well. When we make a decision that affects mainly other people, we should try to look at it from their point of view. What matters to them, in these two cases, is whether they live or die. In the vaccination case, what matters is the probability of death. If you were the child, and if you could understand the situation, you would certainly prefer the lower probability of death. It would not matter to you how the probability came about. The case of the prisoners is much the same. Imagine that each prisoner knows about the choice you must make but does not know whether he is the one that would be shot by you or whether he is one of the two that would be shot by your captors. Each of the other prisoners would certainly want you to do the shooting. If you shoot, only one of the three will die, so the chance of death is 1/3. If you refuse to shoot, then two of the three will die, so the chance is 2/3. In cases like these, you have the choice, and your choice affects what happens. It does not matter what would happen if you were not there. You are there. You must compare the effect of one option with the effect of the other. Whichever option you choose, you had a choice of taking the other option. If the main effect of your choice is on others, shouldn't you choose the option that is least bad for them?  These subjects were given the original two cases again. 42 of the 63 were also asked, Please comment on how the argument affected, or did not affect, your answers.' (One of these subjects did not do so.) An additional 38 subjects in a control condition were asked, after the first two questions: Here are the same questions again. Please think through your answers. Do not simply repeat what you said before unless, after further thinking, you still think that you gave the best answers.' After the repeated questions, these subjects were asked, Please comment on why you changed, or did not change, your answers.' In sum, 120 subjects did the first two questions, 63 of these were in the debiasing condition and answered the questions again after reading an argument (with 42 asked to comment), and 38 were in a control condition that required simply answering the questions again. ### 4.2 Results and discussion Table 1 shows the proportions of subjects who answered each question favoring the act (vaccinate or shoot) before the argument (initial') and after it (final'). Of primary interest, all responses in the debiasing condition, including Guilt, favored acts more strongly after subjects read the argument (p<.05 for each comparisons by a sign test). The control condition showed no significant improvement, although 15 of the 38 subjects did change at least one answer in one direction or the other. The debiasing condition showed more change toward the act option than the control condition (across both situations) for Should (p=.000, Mann-Whitney U test), Would (p=.005), and, most importantly, Guilt (p=.014). The hypothesis that the argument would affect anticipated emotion was therefore supported. Table 1. Number of subjects favoring the act (which was less harmful) for each question, out of the total number who answered the relevant question (to the right of each slash). Change' is the net change toward favoring the act, out of the number who disagreed with the argument initially.  All Ss Debias condition Control condition Initial Initial Final Change Initial Final Change Vaccination Should 98/120 51/63 59/63 8/12 31/38 32/38 1/7 Would 88/119 47/63 57/63 10/16 26/37 26/38 0/11 Guilt 83/118 44/63 51/63 7/19 25/36 26/36 1/11 Prisoners Should 29/120 14/63 36/63 22/49 8/38 8/38 0/30 Would 15/120 9/63 17/63 8/54 2/38 2/37 0/36 Guilt 8/120 4/63 14/63 10/59 3/38 1/38 -2/35  The effects of the argument on the three responses did not differ significantly in the vaccination case. The effect on Guilt was therefore just as strong as that on Should or Would. In the prisoner case, however, the number of subjects who changed in response to the argument was greater for Should than for Would (p=.017, sign test) or Guilt (p=.007). (These results were substantially identical when the analysis took into account the number of opportunities for change.) Importantly, then, the effect on Guilt is not found in all subjects. Some subjects change their normative belief (Should) but not their anticipated emotion in response to the argument. Typical written responses of subjects who accepted the argument for Should but not for Would or Guilt were: ... after reading the passage ... I realize that I should kill the one to spare the 2 lives, but I still feel that I would be unable to commit a murder.' I agree with the argument but could not do it.' I feel irrational that I would rather have more people killed ... but my mental blocks against actively killing things are so strong that I don't think I could do it, even for the greater good.' I am aware that this is actually a selfish decision.' Table 2 shows the data relevant to questions about contingency between changes in Should and Guilt in the debiasing condition. The main result here is that Guilt did not favor the act (vs. the omission) after the argument unless Should favored the act after the argument. Subjects who were still unconvinced that they should act would still feel more guilty from acting. The argument therefore had no effect on emotion as long as normative belief favored the omission. When belief initially favored the act, additional argument might have strengthened the belief either by reminding the subject of old arguments or presenting new ones. The additional strength could have led the subject to anticipate a change in emotion. Table 2. Number of subjects in the debiasing condition of Experiment 1 who favored the act or omission before/after the presentation of the debiasing argument on the Should and Guilt questions. For example, subjects labeled omit/act' favored the omission before the argument but the act after it. Vaccination Guilt omit/omit omit/act act/omit act/act Should omit/omit 2 1 0 1 omit/act 2 2 0 4 act/omit 0 0 0 0 act/act 7 5 1 38 Prisoners Guilt omit/omit omit/act act/omit act/act Should omit/omit 26 0 0 1 omit/act 16 5 0 1 act/omit 0 0 0 0 act/act 7 5 0 2  Responses to the two cases were not significantly correlated. Table 1 shows that the act was less favored in the prisoner case. For both cases, Should favored the act more strongly than Would and Guilt. The Should percentage exceeded that of both Would and Guilt significantly (p<.001, Wilcoxon test) overall for the initial answers, but the difference between Would and Guilt was not significant. Correlations suggest that anticipated behavior is driven by a compromise between normative beliefs (Should) and anticipated emotions (Guilt), which are themselves in the greatest conflict: questions within a case (before the debiasing argument) were significantly correlated, with the strongest correlations between Should and Would (Goodman-Kruskal gamma .97 for vaccination and .86 for prisoners) and the weakest between Should and Guilt (.66 for vaccination, .55 for prisoners). A log-linear analysis on the initial responses led to the same conclusion. It found significant associations (p<.005) between Would and Guilt for both cases and between Would and Should for the prisoner case (p<.001; p<.10 for the vaccination case). No separate association was found between Should and Guilt. (The basic model was Should*Would + Should*Guilt + Would*Guilt'; in this notation, each interaction term includes its main effects. Tests were carried out by replacing each interaction term with its corresponding main effects only and then comparing the likelihood-ratio chi-squares.) An additional log-linear analysis suggested that the effect of the argument on Would was associated with its effect on Guilt, in the debiasing condition. Taking into account the effect of initial responses, Would2, the Would response after the argument, was associated with Guilt2 both for vaccination (p<.05) and for prisoners (p<.001), and it was associated with Should2 for prisoners (p<.005) but not for vaccination. (The basic model was Should2*Should1 + Would2*Would1 + Guilt2*Guilt1 + Would1*Should1 + Would1*Guilt1 + Should1*Guilt1 + Would2*Should2 + Would2*Guilt2.' Each of the last two interaction terms was replaced with its main effects for the test.) To test more directly the role of anticipated guilt in determining Would and Should responses an additional experiment was carried out on 28 subjects from a class in thinking and decision making (which had been assigned some reading about utilitarianism). The prisoner story was modified by using the word kill' instead of murder' and by adding, You are absolutely convinced that they will do what they say.' For 14 subjects, the number of other prisoners was raised from 3 to 39 (with 38 slated to be killed instead of 2). Subjects could indicate indifference for each question. The Should item favored killing for 20 subjects and favored not killing for 5 (the rest being indifferent). The Guilt item favored killing for only 9 subjects (vs. 10), and Would favored killing for only 9 (vs. 11). These both differed significantly from Should (sign test, p=.035 and .031, respectively), and Would and Guilt were correlated (gamma=.78, p<.001). (Should and Guilt favored killing significantly more for 38 other prisoners than for 2.) Subjects were also asked, If you were sure that you would not feel guilty from killing the one to save the others, do you think that you would do it?' Here, 22 said they would kill (vs. 3); significantly more than the original Would response (p=.000). Subjects were then asked: Suppose that someone else were in this position instead of you. This person would feel guiltier killing the one than not killing [not killing the one than killing the one]. Should this person kill the one to save the others?' 15 (vs. 10) said that the person who felt guiltier from killing should kill, and 21 (vs. 4) said that the person who felt guiltier from not killing should kill. The difference is significant (p=.039, sign test). Subjects think that anticipated guilt feelings affect not only action but also what a person should do. Log-linear analyses suggested that initial Guilt did not predict susceptibility to argument for Would or Should, but some important cells contained few subjects, so these analyses are inconclusive. (For example, of those whose initial Guilt favored vaccination, Should favored nonvaccination in only five subjects.) In sum, an argument about the irrelevance of the act-omission distinction affected anticipated emotion as well as moral belief. It seems that anticipated emotion is driven partly by belief. The effect on anticipated emotion is not as strong as that on belief in the prisoner case, however. Changes in expected action seem to occur in combination with changes in anticipated emotion, and subjects in an ancillary study believed that both expected action and the normatively correct response are affected by anticipated guilt feelings. ## 5 Experiment 2 This experiment concerned judgments of appropriate compensation for a misfortune, in particular, blindness. Baron (in press a) reports evidence that judged compensation is greater when a misfortune was caused by a person than when it was caused by nature. Ritov, Hodes, and Baron (1989) also found that people wanted to have more insurance against misfortunes caused by a person than against those caused by nature. Experiment 2 seeks to replicate those results and to examine their relation to emotions. In particular, judged compensation for misfortune could be related either to sadness at the person's suffering or to anger that the person had to suffer. (In principle, anger could be directed against nature as well as against people.) As in Experiment 1, we provide an argument against differences in compensation, and we examine the effect of the argument on both judgments and emotions. ### 5.1 Method Fifty-nine subjects, solicited as in Experiment 1, were given the following questionnaire (the debiasing condition): Suppose that two equally serious viral diseases, A and B, begin to spread around the world. Both diseases are spread through contact, much like flu. Neither disease can be detected early enough to quarantine those who carry it. Both diseases cause permanent blindness in those who get them. It is expected that 5 out of every 10,000 people (that is, 0.05\% of the population) will get disease A. It is expected that at least 50 out of every 10,000 people (that is, 0.5\% of the population) would get disease B if nothing is done. However, a pharmaceutical company developed a vaccine for disease B, and the government initiates a crash program which succeeds in getting everyone vaccinated. The vaccine completely prevents the disease, but it also causes permanent blindness in 5 out of 10,000 people (that is, 0.05\%) who are vaccinated, a reduction of 90\% in the amount of blindness. This was absolutely the best that the company could do in the time available to develop the vaccine. 1. As part of the program, the government sets aside a special fund to compensate the victims. How should the fund be used? Pick one and explain: a. More money should be spent to compensate those made blind by disease A. b. More money should be spent to compensate those made blind by the vaccine for disease B. c. Equal amounts should be spent on both kinds of victims. 2. Suppose that no special fund is set aside but insurance is available for blindness caused in either way. Both kinds of insurance cost the same. Which would you be more inclined to buy? Pick one and explain: a. More insurance against blindness caused by disease A. b. More insurance against blindness caused by the vaccine for disease B. c. Equal amounts of both kinds of insurance. 3. Which would you feel sadder: a. Hearing about someone who became blind from disease A. b. Hearing about someone who became blind from the vaccine for disease B. c. Equally sad. 4. Which would make you more angry. a. Becoming blind from disease A. b. Becoming blind from the vaccine for disease B. c. Equally angry. [New page} Now consider the following argument before answering the original questions again: Blindness is the same regardless of its cause. It does not matter whether blindness was caused by a natural disease or by a vaccine. The main functions of compensation are to pay for extra expenses (such as seeing-eye dogs, medical care, etc.), make up for losses (lost income), and to give people extra money so that they can try to make up for their loss of vision (e.g., by hiring help). The need for these things is great, and it is just as great regardless of the cause of the blindness. Any gain to one group of victims means that the other group will be receiving less compensation. The compensation comes from the same source regardless of the cause of the blindness. The point of the compensation is not to punish those that caused the blindness. The government, in any case, knew that it would have to compensate both kinds of victims.]  Subjects were asked to answer the original questions again, and they were then asked: Explain how the argument affected your responses, or, if it did not, why it did not.' An additional 58 subjects did a control condition, in which they were instructed, after answering the first four questions, Here are the same questions again. Please think through your answers. Do not simply repeat what you said before unless, after further thinking, you still think that you gave the best answers.' At the end, they were asked to comment on their reasons for changing or not changing. ### 5.2 Results Table 3 shows the number of subjects giving each answer to each question initially and finally. Initially, as predicted, Compensation, Sadness, and Anger showed a bias toward the vaccine (p=.000 by a sign test comparing responses favoring the vaccine to those favoring the disease), but Insurance did not (20 vs. 14). However, Insurance was significantly correlated with Compensation (gamma=.619, p=.001), and those subjects who wanted more compensation for vaccine-caused injury also tended to want more insurance for this (10 wanting more, one wanting less). So some subjects show a bias even for insurance. Some subjects may show the opposite bias, toward the disease. Table 3. Number of subjects favoring each response on each question in Experiment 2.  {\it Final Debiasing condition (N=59) Control condition (N=58)} more for more for more for more for vaccine equal disease vaccine equal disease {\it Initial} Compensation vaccine 3 14 0 12 5 0 equal 0 41 0 3 33 1 disease 0 1 0 0 2 2 Insurance vaccine 4 5 1 5 4 1 equal 0 42 0 6 35 0 disease 0 4 3 1 2 4 Sadness vaccine 10 7 0 18 7 0 equal 0 39 0 2 25 2 disease 0 1 2 0 1 3 Anger vaccine 30 11 0 40 3 0 equal 1 17 0 1 12 2 disease 0 0 0 0 0 0  Just as the two decision items, Compensation and Insurance, were correlated, the two emotion items, Sadness and Anger, were correlated as well (gamma=.700, p=.000). The sum of the two decision items was not correlated significantly with the sum of the two emotion items, however (gamma=.176). The argument seemed to reduce the bias toward the vaccine. Favoritism toward the vaccine - that is, more compensation or stronger emotion - decreased in Compensation (p=.001, one-tailed sign test), Anger (p=.003), and Sadness (p=.035). None of the corresponding changes in the control condition was significant. The difference between the two conditions in the reduction of favoritism was significant only for Compensation (p=.009, one-tailed Mann-Whitney U test), however. The argument was more clearly effective in leading subjects to treat the two causes equally, ignoring whether inequality favored the vaccine or the disease. In the debiasing condition, the number of judgments of equality increased significantly for Compensation (p=.000), Insurance (p=.002), Sadness (p=.004), and Anger (p=.003). No increases were significant for the control condition. The difference between the debiasing condition and the control condition in the change in the number of judgments of equality was significant for Compensation (p=.009), Insurance (p=.002), and Anger (p=.009), but not for Sadness. In the debiasing condition, the effect of the argument on the number of unequal emotion judgments (Sadness or Anger; 0, 1, or 2 of these could be unequal) was not significantly greater than its effect on the number of unequal normative judgments (Compensation or Insurance; p=.239, Wilcoxon test). These two effects were correlated in the debiasing condition (gamma=.551, p=.003). Initial Sadness and Anger judgments were completely uncorrelated with the effectiveness of the argument in changing compensation judgments. As in Experiment 1, the argument affected anticipated emotion (toward equality) both in subjects whose compensation judgments were already equal and in those who changed from inequality to equality. Of those whose compensation judgments were already equal, 12 changed toward equality in emotion judgments, out of 31 who gave initially unequal judgments for at least one emotion. (One subject in this group changed away from equality in emotion judgments.) Of those whose compensation judgment changed toward equality, 6 changed toward equality in emotion judgments, out of 10 who gave initially unequal judgments. Only three subjects gave unequal compensation judgments after the argument, and the unequal emotion judgments of all three also remained unequal. Experiment 2 agrees with Experiment 1 in showing effects of debiasing on anticipated emotion as well as normative beliefs as expressed in decisions. The effect on decisions was not significantly stronger overall than the effect on emotion. ## 6 Experiment 3 Experiment 3 examines the tradeoff of money and risk. Subjects are often asked to make judgments of their willingness to pay (WTP) for the reduction of risk or their willingness to accept (WTA) money in return for an increase in risk. These kinds of judgments are used in measuring the value of public goods by contingent valuation' (Mitchell & Carson, 1989). In these studies, WTA is generally higher than WTP for the same risk, as is consistent with the endowment effect (e.g., Kahneman et al., 1990). The WTA-WTP discrepancy could in principle result from the shape of the utility function for money. The utility gain from accepting$1,000,000 might equal the utility loss from paying only $100,000. In this case, it would be normatively correct to show a large discrepancy. Although WTA-WTP effects for small amounts of money are not thought to be understandable in terms of the utility function (Kahneman et al., 1990), they could arise from such effects applied to limited mental accounts' (Thaler, 1985). If I have a limited account for tradeoffs of money and risk (Thaler, 1985), I might think of accepting$100 as equivalent to losing \$50. Such an explanation of the WTA-WTP difference would not render the effect trivial, but the effect is usually interpreted as a bias toward the status-quo, not as the result of the subjective utility function.
In the present experiment, the debiasing argument needed to be directed toward a single explanation. Accordingly, this experiment compared WTP to a different measure, willingness to forgo loss' (WTFL). In WTFL, the subject is asked what the monetary loss would have to be in order to make her indifferent between the monetary loss and the risk. The only difference is in the assumed status-quo, which is less risk and more money for WTP and the reverse for WTFL. The future consequences are identical. Kahneman et al. (1990, Experiments 6 and 7) found that WTA was higher than willingness to forgo gains (which they called choosing'). This result indicates a pure status-quo bias, independent of wealth or income effects. (McClelland & Schulze [1991] make a similar comparison, but without holding future outcomes constant.) Here, we extend this to WTP and WTFL, and we examine emotional concomitants. WTP is of practical interest because it is often impossible to use WTA in risk evaluation: many subjects say that no amount of money is enough' (a reasonable response if their utility for monetary gains is bounded).
Experiment 3 addresses another issue concerning the finding of Experiments 1 and 2 that the debiasing argument does not always affect anticipated emotion even when it affects a decision. One possible interpretation of this incompleteness is that the questionnaires measured the wrong emotions. This is unlikely: in Experiment 1, no subject mentioned any other relevant emotion aside from guilt feelings; in Experiment 2, the high correlation between Sadness and Anger suggests that emotion judgments in these cases are subject to a halo effect in any case. In Experiment 3, as an additional test, subjects were asked to rate the strength of their emotional reaction,' without any specific mention of the type of emotion.

### 6.1  Method

In the debiasing condition, 91 subjects were solicited as in previous experiments. Eleven were eliminated for failing to follow instructions (e.g., writing only comments when numerical values were requested), leaving 80 for analysis. Subjects were given the following questionnaire:

In each question, you will be asked to name a dollar figure and
to indicate the strength of your emotional reaction to a
hypothetical outcome.  In rating the strength of your reaction,
use a scale of numbers from 0 to 100, in which 0 represents not
upset at all' and 100 represents as upset as one can be about
this sort of situation.'  Please try to use consistent numbers
throughout the questionnaire.  That is, stronger negative
reactions should get larger numbers.  If you must go beyond 100
in order to be consistent with an earlier response, feel free to
do so.

All questions concern nuclear waste.  As you may know,
repositories for high-level nuclear waste from nuclear plants are
built according to Federal safety standards in places designated
as geologically safe.  The wastes are stored more than 100 feet
below the earth's surface, in specially sealed canisters,
designed to last for thousands of years.

1. Suppose that the Federal government had planned to put a high
level nuclear waste repository 500 miles from your home.  The
decision was appealed, and more studies were conducted.  The
government recently proposed to locate the repository 50 miles

You and others in your area expected your taxes to increase.  If
you and others accept the change in the site of the repository
(from 500 miles to 50 miles away), the government will cancel the
increase for as long as you live near the repository.  {\it How
much would the planned increase have to be, in dollars, so that
you are indifferent between:}

* having the site 50 miles from your home without the increase;

* having it 500 miles from your home with the increase?

Suppose that the planned tax increase is more than the number you
just wrote.  You should prefer having the repository 50 miles
away without the increase.

Suppose that the increase is less than the number you wrote.  You
should prefer having the repository 500 miles away with the increase.

If these statements are not true, change your answer so that they
are true.  Finally, place a check mark next to your answer to
indicate that you have checked it.

of question 1:

A.  Your taxes do not increase, and the repository is located

B.  Your taxes increase by the amount you wrote in question 1,
and the repository is located 500 miles from your home.

3. Suppose that the Federal government had planned to put a high
level nuclear waste repository 50 miles from your home.  The
decision was appealed, and more studies were conducted.  The
government recently proposed to locate the repository 500 miles

If you and other accept this change, your taxes will increase for
as long as you live far away from the repository.  {\it What tax
increase would you be willing to pay, in dollars, so that you are
indifferent between:}

* having the site 50 miles from your home without the increase;

* having it 500 miles from your home with the increase?

[Subjects again checked their answer and, in question 4, rated
their emotional reactions as before.

[New page.]  Now, before answering the original questions again,
consider the following argument:

distance from the depository and money.  You can be close to or
far from the repository.  You can have less money than you have
now or you can have the same amount.  When you answer these
questions, you are saying that you are indifferent between having
less money, by a certain amount, and being far from the
repository.

{\it What matters for decisions like these is the future, not the
past, and not what was expected.'  The future is what you can
affect.  It does not matter whether the government initially
planned to locate the repository 50 miles or 500 miles from your
home.}

Therefore, the amount of money you indicate should be the same in
the two cases.  In both cases, you have the same choice with
respect to the future.

[Subjects answered questions 1-4 again and then commented on how
the argument affected or did not affect their answers.]]


In the version just presented, question 1 is WTFL and question 3 is WTP. Forty-two of the analyzed subjects received the questions in this order. For 38 subjects, the order was reversed, with WTP as question 1 and WTFL as question 3. The two orders did not differ significantly in any results, so order will be ignored.
In the control condition, 61 subjects were given the same questionnaire with instructions to rethink (and to explain any changes) as in the control conditions in Experiments 1 and 2. Fourteen subjects were eliminated for failure to follow instructions, leaving 47 (26 in the order given above, 21 in the reverse order).

### 6.2  Results

Table 4 shows the main results. In the answers to questions 1 and 3, WTP was smaller than WTFL in 53 subjects. These subjects showed a status-quo bias. Thirteen subjects showed the opposite effect, and 61 subjects gave equal dollar values. The status-quo bias (53 vs. 13) was significant by a sign test (p=.000).

Table 4. Number of subjects who responded in each way for Experiment 3. (Ns for emotion questions are smaller because some subjects did not answer it properly.)

{\it Final
Debiasing condition (N=80)  Control condition (N=47)}
WTP             WTFL        WTP             WTFL
worse   equal   worse       worse   equal   worse
{\it Initial}
Monetary question
WTP worse       2       7       0           0       2       2
equal           0      40       1           2      12       6
WTFL worse      2      17      11           1       8      14

Emotion question
WTP worse       4       4       1           6       2       2
equal           0      36       1           5      14       3
WTFL worse      5      15       9           3       3       8



The same bias was found in the emotion questions. Subjects were more upset by change than by keeping the status-quo. This was assessed by comparing the sum of the emotion ratings to question 2 with the sum of the ratings to question 4. Forty-three subjects gave higher ratings for WTP than for WTFL, and 19 gave the reverse (p=.002). (Five subjects did not give numerical emotion ratings.) The responses to the emotion questions were correlated with those to the monetary question (gamma=.337, p=.01).
The argument did not significantly reduce the status-quo bias, but it did move subjects toward equality for WTP and WTFL, regardless of which judgment was initially greater. Twenty-four subjects in the debiasing condition moved from inequality to equality for the two monetary values, and only one subject moved from equality to inequality (p<.001). The control condition had no effect, and the difference between experimental and control conditions in the increase in the number of equal judgments was significant (p=.013, Mann-Whitney U).
The same effect occurred for the emotions questions: in the debiasing condition, 19 moved from inequality to equality, and 1 moved from equality to inequality (p<.001). Although the status-quo bias was still present after the argument for the monetary questions (12 vs. 4), it was essentially absent for the emotion questions after the argument (11 vs. 9). The difference between the experimental and control conditions in the increase in the number of equal judgments was significant (p=.001) for the emotion question.
The effect of the argument on the number of equality judgments was essentially as strong for the emotion questions as for the monetary questions. A Wilcoxon test yielded no significant difference in the debiasing condition.
In the debiasing condition, once again, the effect on emotion was limited to those subjects who either changed from inequality to equality on the monetary question or who indicated equality both before and after the argument. Of those who changed to equality on the monetary question, 11 changed to equality on the emotion question, out of 18 who answered with inequality initially. Of those who answered with equality on the monetary question before and after the argument, 7 changed to inequality on the emotion question, out of 10 who answered with inequality initially. (One subject changed from equality to inequality.) Of those subjects who answered the monetary question with inequality both times, only one changed from inequality to equality on the emotion question, and 9 answered this question with inequality both times.
Although the correlation between change in the emotion question and change in the monetary question was high (gamma .70, p<.01), it was not perfect. A total of 11 subjects had unequal anticipated emotions after the argument even though their responses to the monetary question were equal. (This number can be compared to the 18 who changed to equality.) Some subjects change only their decision without changing their anticipated emotion. This incompleteness, found in Experiments 1 and 2 as well, cannot here be explained in terms of asking about the wrong emotion, for Experiment 3 did not specify the emotion.
No apparent relationship was found in the debiasing condition between change in the monetary question and the initial answer to the emotion question.
In sum, the normative argument affected emotion just as strongly as it affected judgment, but the two effects were imperfectly correlated. Again, the effect on emotion is found only in subjects who ultimately agree with the argument, if only because they agreed with it at the outset.

## 7  Discussion

Normative arguments affected anticipated emotions essentially as strongly as they affected normative beliefs or hypothetical decisions, except for the prisoner scenario in Experiment 1. In many subjects whose beliefs or decisions agreed with the normative argument initially, arguments affected anticipated emotions, bringing emotions into accord with beliefs or decisions. In other subjects, the arguments affected beliefs or decisions without affecting emotion.
Normative beliefs could affect anticipated emotions and decisions in three ways. First, subjects could believe that normative beliefs determine emotions. The change in anticipated emotion, in turn, could allow subjects to change their decisions. In this case, anticipated emotions are a determinant of decisions. The ancillary experiment following Experiment 1 suggests that some subjects do believe that decisions are affected by anticipated emotions, thereby supporting this mechanism (but not impugning the others). Second, a change in normative beliefs could affect the decision, and subjects could then believe that they could bring their own emotions into line with their decision. (See Ainslie, 1992, for discussion of the control of emotions.) Third, the normative beliefs could have independent effects on decisions and beliefs about future emotions. We cannot rule out any of these possibilities. At first blush it might seem that a correlation between emotion change and decision change would rule out the third account, but it is possible that the correlation is induced by differential effectiveness of the argument in changing normative belief. Further investigation is needed.
In most of the situations studied here, people seem to assume that their emotions will be justified and reasonable, so that they can do what they believe to be right without interference from their emotions. In the prisoner scenario, however, some people think that their anticipated emotions are unaffected when normative beliefs change. In situations like this, a change in normative belief will not necessarily lead to behavior change, and it could lead to conflict between belief and emotion. This emotional inertia' could complicate efforts to change decision making through purely cognitive means, such as cognitive therapy (Clark, 1986), debiasing (Larrick et al., 1990), or education (Baron & Brown, 1991).
Our results point to two different beliefs about the relation between emotion and normative judgment. By one belief, emotions fall into line (or can be brought into line) with rational judgment. The other belief, a sort of Freudian theory, which holds that emotion is not affected by rational argument. When people hold this theory about their own emotions, they may resist rational appeals concerning their decisions, for - even if they are intellectually persuaded - they will be held back by the belief that their emotions will not follow along. Even if they are convinced that radon is more dangerous than pesticides, they will not favor a change in governmental priorities away from pesticides and toward radon, because of the fear that they will worry more if the change is made.
No evidence, however, indicated that this sort of belief in the non-malleability of emotion causes resistance to the intellectual argument itself. We might have expected such resistance if people believe that emotions and normative beliefs should be consistent and that emotions would not change. The role of folk-psychological theories in belief change is another topic for further investigation (see Baron, 1991).
The danger of holding the view that emotions will coincide with normative beliefs is that the Freudian view could be correct. A young man or woman could be convinced that premarital sex is not immoral and then engage in it, thinking that no guilt feelings will follow, but then experience the guilt feelings anyway.
Does this sort of thing happen? It's hard to tell. In the real world, belief change itself is rarely complete and stable. I am impressed, however, with the findings of cognitive therapists such as Salkovskis, Clark, and Hackmann (1991), who found that panic attacks can be controlled by pure change in belief about their origin. Patients who believe that incipient attacks represent an immediate medical crisis, when convinced otherwise, cease having full blown attacks. The fear that leads to the full blown attack is, in this case, controlled by a belief that the fear is unjustified.

## 8  References

Ainslie, G. (1992). Picoeconomics: The interaction of successive motivational states within the individual. New York: Cambridge University Press.
Baron, J. (1985). Rationality and intelligence. New York: Cambridge University Press.
Baron, J. (1988). Thinking and deciding. New York: Cambridge University Press.
Baron, J. (1991). Beliefs about thinking. In J. F. Voss, D. N. Perkins, & J. W. Segal (Eds.), Informal reasoning and education, pp. 169-186. Hillsdale, NJ: Erlbaum.
Baron, J. (in press a). Heuristics and biases in equity judgments: a utilitarian approach. In B. A. Mellers and J. Baron (Ed.), Psychological perspectives on justice: Theory and applications. New York: Cambridge University Press.
Baron, J. (in press b). Morality and rational choice. Dordrecht: Kluwer.
Baron, J. & Brown, R. V. (Eds.) (1991). Teaching decision making to adolescents. Hillsdale, NJ: Erlbaum.
Bell, D. E. (1982). Regret in decision making under uncertainty. Operations Research, 30, 961-981.
Bell, D. E. (1985). Disappointment in decision making under uncertainty. Operations Research, 33, 1-27.
Bennett, J. (1966). Whatever the consequences. Analysis, 26, 83-102 (reprinted in B. Steinbock, ed., Killing and letting die, pp. 109-127. Englewood Cliffs, NJ: Prentice Hall).
Bennett, J. (1981). Morality and consequences. In S. M. McMurrin (Ed.), The Tanner Lectures on human values (vol. 2, pp. 45-116). Salt Lake City: University of Utah Press.
Chambless, D. L., Gracely, E. J. (1989). Fear of fear and the anxiety disorders. Cognitive Therapy & Research, 13, 9-20.
Clark, D. M. (1986). A cognitive approach to panic. Behavior Research and Therapy, 24, 461-470.
Cox, B. J., Swinson, R. P., Norton, G. R., Kuch, K. (1991). Anticipatory anxiety and avoidance in panic disorder with agoraphobia. Behaviour Research & Therapy, 29, 363-365.
Elster J. (1985). Weakness of will and the free-rider problem. Economics and Philosophy, 1, 231-265.
Gorr, M. (1990). Thomson and the trolley problem. Philosophical Studies, 59, 91-100.
Hare, R. M. (1981). Moral thinking: Its levels, method and point. Oxford: Oxford University Press (Clarendon Press).
Hershey, J. C., & Baron, J. (1987). Clinical reasoning and cognitive processes. Medical Decision Making, 7, 203-211.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1990). Experimental tests of the endowment effect and the Coase theorem. Journal of Political Economy, 98, 1325-1348.
Kahneman, D., & Lovallo, D. (1990). Timid decisions and bold forecasts: A cognitive perspective on risk taking. Manuscript, Department of Psychology, University of California, Berkeley.
Kahneman, D. & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93, 136-153.
Kahneman, D., & Snell, J. (1990). Predicting utility. In R. Hogarth (Ed.), Insights in decision making. Chicago: University of Chicago Press.
Knetsch, J. L., & Sinden, J. A. (1984). Willingness to pay and compensation: Experimental evidence of an unexpected disparity in measures of value. Quarterly Journal of Economics, 99, 508-522.
Kuhse, H. (1987). The sanctity of life doctrine in medicine: A critique. Oxford: Oxford University Press.
Larrick, R. P., Morgan, J. N., & Nisbett, R. E. (1990). Teaching the use of cost-benefit reasoning in everyday life. Psychological Science, 1, 362-370.
Loewenstein, G. (1987). Anticipation and the value of delayed consumption. Economic Journal, 97, 666-684.
Loomes, G. (1987). Testing for regret and disappointment in choice under uncertainty. Economic Journal, 97, 118-129.
Loomes, G., & Sugden, R. (1982). Regret theory: An alternative theory of rational choice under uncertainty. Economic Journal, 92, 805-824.
Lopes, L. L. (1987). Between hope and fear: The psychology of risk. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 20, pp. 255-295). New York: Academic Press.
McClelland, G. H., & Schulze, W. D. (1991). The disparity between wilingness-to-pay versus willingness-to-accept as a framing effect. In D. R. Brown & J. E. K. Smith (Eds.), Frontiers of mathematical psychology: Essays in honor of Clyde Coombs, pp. 166-192. New York: Springer-Verlag.
Mitchell, R. C., & Carson, R. T. (1989). Using surveys to value public goods: The contingent valuation method. Washington: Resources for the Future.
Ritov, I., & Baron, J. (1990). Reluctance to vaccinate: omission bias and ambiguity. Journal of Behavioral Decision Making, 3, 263-277.
Ritov, I., & Baron, J. (1992). Status-quo and omission bias. Journal of Risk and Uncertainty, 5, 49-61.
Ritov, I., Hodes, J., & Baron, J. (1989). Biases in decisions about compensation for misfortune. Manuscript, Department of Psychology, University of Pennsylvania.
Sabini, J. (1992). Social Psychology. New York: Norton.
Salkovskis, P. M., Clark, D. M., & Hackmann, A. (1991). Treatment of panic attacks using cognitive therapy without exposure or breathing retraining. Behaviour Research and Therapy, 29, 161-166.
Samuelson, W., & Zeckhauser, R. (1988). Status quo bias in decision making. Journal of Risk and Uncertainty, 1, 7-59.
Spranca, M., Minsk, E., & Baron, J. (1991a). Omission and commission in judgment and choice. Journal of Experimental Social Psychology, 27, 76-105.
Thaler, R. (1985). Mental accounting and consumer choice. Marketing Science, 4, 199-214.

File translated from TEX by TTH, version 3.40.
On 17 Sep 2003, 20:06.