Extensive evidence indicates that subjective evaluations of public goods systematically violate the predictions of the economic model. The proposed research follows up on two research programs addressing the behavioral basis for some violations: one viewing judgement primarily as an expression of a spontaneously evoked reaction to a typical affected individual (Ritov and Kahneman, 1997; Kahneman, Ritov, and Schkade, 1998) and a second suggesting that evaluation is sometimes based on absolute rules (Baron and Spranca, 1997). In particular, people hold values which are protected from being traded off with any other values ('protected values').
The present research seeks to determine the role of these two modes of evaluation in propagating violations of economic principles. Specifically, we shall examine preference for complete solutions, insensitivity to quantity, and omission bias. We hypothesize that the omission bias will increase when human life, protected values, and identified (vs. statistical) cases are at issue. Quantity insensitivity and preference for complete solutions will also increase with these factors. The research is relevant to explanations of the biases at issue. The existence of moderating factors rules out a number of simple accounts.
We expect our research to integrate the above programs, and provide new insight regarding the different sources of rationality violations in evaluation of public endeavors.
Section 2
Detailed Description of the Research Plan
1. Brief description of the subject and of the scientific background
Public actions involve a wide range of activities, varying in kind, scope, effort, and importance. Ideally, public actions are designed to best serve the public interest. The individuals' subjective evaluation of a public action is an important determinant (even if not the only one) of the degree to which it serves the public interest. The present proposal focuses on the individuals' evaluation of public actions. For our purposes, it does not matter whether the individuals are citizens or politicians. We assume that the factors we examine affect both.
Consider the following examples, in which evaluation of a public action is requested: (1) to what extent would you support an intervention to completely eliminate the risk of leukemia due to power lines? (2) Would you choose to provide cure to 1000 cancer patients if it meant that 100 other patients would die, because their treatment program was terminated? (3) How much would you be willing to contribute to a special fund to save a little girl who was critically wounded in a rare accident and is in need of a costly emergency treatment? Would you be willing to contribute an equal amount if the fund was targeted at victims of such accidents in general? (4) How much personal satisfaction would you expect to experience as a result of your support for enforcing the ban against ivory trade? The responses to all these questions reflect the respondent's evaluation of the different public actions described above.
A large body of recent research demonstrates that the interpretation of responses to such evaluation questions do not conform to economic principles as applied to public goods. In particular, responses are not sufficiently sensitive to quantitative aspects, either of the problem or of the proposed action (reviewed in Baron, 1997; Kahneman, Ritov, and Schkade, 1998). On the other hand, responses are affected by factors which, normatively, should not be considered, such as the context in which the question is embedded.
1.1 Two processes
The core of the research proposed here is the presumed existence of two kinds of processes which may both yield the documented violations of rational evaluation. One process proposed by Kahneman and Ritov (Kahneman and Ritov, 1994; Ritov and Kahneman, 1997; Kahneman, Ritov, and Schkade, 1998) is a process of spontaneous reaction to a mental representation of the intervention's target. The other kind of process, studied by Baron and his associates (Baron and Spranca, 1997; Baron and Leshner, 1999; Lim and Baron, 1997; Ritov and Baron, 1999), involves absolute decision rules, or protected values. We briefly discuss, next, each of the two processes, or modes, of evaluation. These processes may be related to each other, in that both are concerned with the role and feelings of the decision maker rather than the consequences of the decision.
The psychological model of the individual subject's responses to questions about public goods, proposed by Kahneman and Ritov, interprets the responses as an expression of an immediate emotional reaction to a spontaneously evoked prototypical object (Kahneman and Ritov, 1994; Ritov and Kahneman, 1997; Kahneman, Ritov, and Schkade, 1998). The response evoked by a prototypical object is naturally not affected by the quantitative aspects of the action. For instance, support for an intervention to eliminate the risk of leukemia due to power lines may be unaffected by the relatively minute impact it might have. Furthermore, a basic tenet of the model is that when a problem is judged on its own, as is often the case when an evaluation of a public action is called for, the problem is implicitly compared to other problems of its natural category. Thus, a human health problem is evaluated by comparison to other human health problems, whereas an endangered animal species problem is naturally compared to other endangered species problems. Consequently, although people may consider public health problems more important than a threat to some animal species, the image of an elephant being hunted for its tusks would yield higher expected satisfaction from supporting the ban on ivory trade, for example, than from instating a medical check up program for farm workers (Kahneman and Ritov, 1994).
The second process is the use of absolute rules. Most people have some values that they think of as absolute, not to be traded off for anything else. Baron and Spranca (1997) called these protected values (PVs) and proposed that these values arise from deontological rules, rules concerning actions rather than consequences. Several questionnaire studies have supported this proposal by finding that PVs, compared to values that are not absolute in this way, have various predicted properties (Baron
1.2 Effects of interest: quantity, omission, completeness, context
On the basis of the theoretical approach just described, we propose to examine the determinants of four effects: insensitivity to quantity, omission bias, preference for complete solutions, and the impact of context on preference.
Hypothetical decisions may include quantitative information. In these cases, we can think of the decision as having two dimensions: quantity (scope) and problem type. The problem type is more "prominent" in the sense of Tversky, Sattath, and Slovic (1988). It determines the expression of attitudes, which we hypothesize to be governed by prominence. And it would determine choice responses. This kind of insensitivity to quantity may be reduced when subjects evaluate different quantities of programs (of the same type) side by side. This would make quantity more ëvaluable." Hsee (1996) has found that evaluability affects the weight of an attribute in choice and judgment. However, increased sensitivity to quantity does not necessarily imply that quantity is appropriately weighted. Indeed, Dekay and McClelland (1996) as well as Kahneman, Ritov, and Schkade (1998) both found that when different degrees of threat to animal species were presented in the same context (either as probability of survival, or as extent of damage), ratings of interventions were determined by an additive combination of the species evaluation and the scope of the problem.
The scope insensitivity found with PVs seems to be of a different type than that found as a result of evaluability. In other cases, making some quantity evaluable increases the attention it gets, as found by Hsee (1988). But subjects in PV studies explicitly say that quantity should not matter. It is a moral commitment rather than an attentional neglect.
Another effect we have explored (Ritov and Baron, 1990, 1992, 1995; Baron and Ritov, 1993, 1994) is "omission bias." People favor bad events caused by their omissions over the same bad events caused by their action. Recent work has found that this applies even when the events are different. We have also found (Ritov and Baron, 1999) that omission bias is related to PVs. If a person has a PV for (e.g.) destroying species, this value seems to apply to action rather than inaction. The person might be unwilling to take an action that would cause the extinction of one species, in order to save five, even though someone else who cared about species as a consequence (but not as a PV) would be willing to do this, or more. While the values people held protected varied considerably, the basic finding of greater omission bias for protected values, held across a wide array of issues, ranging from endangered species, to withdrawal from occupied territories (for Israeli respondents). This is consistent with the view that PVs result from rules of action rather than values for consequences. Baron and Leshner (2000) found that expressions of PVs could be reduced by asking subjects to think of counterexamples, i.e., cases in which they would sacrifice the value in question. Such instructions concerning PVs also reduced omission bias that involved the same PVs. A common counterexample was of the form, Ï would sacrifice this value to prevent greater harm to the same value." This is, of course, exactly what is tested in omission bias.
The third effect of interest is the preference for complete elimination of a risk or problem. In a way, this is an extension of the idea of quantity insensitivity. If what matters is whether a risk is present or not, the magnitude of a risk will not matter, and people may prefer complete elimination of one problem to incomplete reduction of others, even if the total benefit is greater with incomplete reductions. Ritov, Baron, and Hershey (1993) examined preferences for reductions of multiple risks. They found that, when complete elimination of one risk is possible, subjects tended to prefer this option over other alternatives, even if the total risk reduction was greater with the alternatives. Preference for complete elimination of one risk was present both in decisions concerning individual consumption of risk-carrying foods, and in decisions concerning willingness to pay for public goods, such as clean air and pure water.
Finally, we shall explore the role of the two processes, spontaneous emotional response and PVs, in accounting for context effects. The effect of context on preferences has been studied with different tasks, and in a variety of domains. A special case of a context effect occurs when the relative positions of two objects on a judgment scale vary when they are evaluated together or separately. Such context shifts have been demonstrated in the domain of public goods as well as in other domains. Recent research (Kahneman and Ritov, 1994; Kahneman et al., under review) proposed to account for the shift by assuming that in separate evaluation each object is judged in relation to a distinctive context that it evokes, whereas joint evaluation imposes a shared context for both objects. Returning to the example of a ban on ivory trade discussed earlier, direct comparison of the elephant problem with the health risks to farm workers yields a shift in the relative positions of these two problems: the relative preference for addressing the health problem increases. Protected values may play a role in the context shift as well. This potential role will be discussed below.
1.3 Response measures
We shall examine these effects, when possible, in five different response measures, pricing, attitude measures, tradeoff measures, rating and ranking of programs . Pricing responses are of the sort used in contingent valuation surveys. These responses are of interest because such surveys are still in widespread use.
Attitude responses were used in previous research on evaluation of public goods by Kahneman and Knetsch (1992), and Kahneman and Ritov (1994). They include questions about importance of an issue, how the respondent would feel about contributing to the solution of a problem, how much concern a problem arouses, and how upset would the respondent expect to feel upon thinking about the issue. More recently, attitude responses have been used in a study of punitive awards (Kahneman, Schkade and Sunstein, 1998), in which monetary awards were predicted from respondents rating of the degree of outrage they felt about different cases.
Tradeoff measures ask the respondent to match the quantity of one good to that of another. In principle pricing responses fit in this category, but we want to treat separately those cases in which both goods are public. Protected values are cases in which subjects say they do not want to make tradeoffs. All our tradeoff measures will allow this as a response option.
The act-omission distinction is tested by a particular kind of tradeoff measure in which the two goods are identical, except that harm from action is traded off against harm from omission. A typical question is "How many acres of forest would you destroy in order to save 100 acres?"
2. Objectives, hypotheses, and expected significance
The research seeks to discover the role the two modes of evaluation, spontaneous reaction to the target and endorsement of absolute values, in evaluation of public policies. Specifically we explore four properties of judgments about public policies:
1. preference for complete solutions;
2. insensitivity to quantity
3. omission bias.
4. context effects.
Each of these properties violates the economic model. When they are used as the basis of policy, the policy is inefficient. These properties can result from protected values. PVs are already shown to affect insensitivity to quantity. They may also affect the preference for complete solutions. In particular, if people are insensitive to quantity, they could think that a small amount of a bad thing is just as bad as a large amount, especially if that thing resulted from action in which they played a role (e.g., as a citizen of the country that did it), and particularly when they don't consider that bad thing in comparison with other kinds of bad things.
We shall employ the five different measures discussed earlier: WTP, ranking, rating, attitude, tradeoff, and PV. Two measures are seen as potential mediators: attitude or emotional response and PVs.
The manipulations we propose are:
1. Varying context (including separate vs. joint evaluation)
2. Evaluating complete/incomplete solutions
3. Varying quantity [magnitude]
4. Varying the cause of outcomes: act vs. omission
5. Judging central vs. peripheral category members
6. Evaluating identified vs. statistical targets.
These manipulations and the hypotheses we propose to test are briefly described below.
1. Context.
Previous research has suggested that people compare programs to whatever set of contrasting alternatives is available. If people evaluate a single program, or a set of programs of the same type, they compare each program to others of the type. If they evaluate a broader set of programs, then their evaluations can change considerably. We hypothesize that evaluation context will affect PVs and omission bias. In particular, we hypothesize that expressions of PVs will be affected by the same factors that affect expressions of attitudes. The more general hypothesis at issue here is that PVs are extreme expressions of attitudes. For example, PVs may be expressed for saving species more readily in the context of other environmental goods than in the context of measures that save human lives. The same effects have been found in expressions of attitudes.
2. Magnitude of effects
One of the properties of protected values is that people often take quantity to be irrelevant. Those with a protected value for preservation of ancient archeological sites, for example, say that the quantity of harm is irrelevant. Yet, sonme evidence (Baron and Leshner, 2000) suggests that people are less likely to endorse protected values when quantity is small, e.g., when the issue is destroying a few square meters of a massive site. It is possible that protected values arise when the subject thinks of the magnitude of harm as larger. Thus, in so far as possible, we will manipulate magnitude in all our experiments. We will this use magnitude and endorsed PVs as independent predictors of other effects. The hypothesis of interest is that both magnitude and PVs have independent effects.
3. Completeness of solution
As indicated above, people often take quantity to be irrelevant. On the other hand, they may think of the of quantity of a specified action in a simplistic way, by asking whether or not the proposed quantity is sufficient to completely eliminate the problem. Previous research offers mixed findings concerning preference for complete elimination. Some new results (described later) suggest that preference for complete solutions occurs more in some domains than others. We propose to study the determinants of this effect, by using a wide range of problems, while manipulating the extent to which proposed interventions offer complete (or partial) solutions. We will examine the correlations between preference for complete solution as reflected in WTP and independent attitude and PV measurement for the same problems. The general hypothesis here is that PVs contribute to preference for complete solution.
4. Act vs. Omission
Baron and Spranca (1997) proposed that PVs involve rules concerning actions (as opposed to values for outcome). This assertion implies that PVs should show a particularly large bias against harmful acts that undermine the value in question, as opposed to harmful omission (omission bias). Our recent findings (Ritov and Baron, 1999) supported this assertion. We propose to extend this research by exploring the role of both processes, spontaneous emotional evaluation and PVs as determinants of the omission bias.
5. Central vs. peripheral category membership.
When people say that they want to protect endangered species, they may be thinking splendid birds, majestic trees, elephants, or dolphins. Yet, most of the loss of biodiversity is in insects and plants that we might call weeds. People may assign value to categories, such as species, on the basis of members of the category that come to mind when they think of it in a particular context, such as protection of species from extinction. Their general attitudes toward the category might not apply to members seen as more peripheral in the context, such as insects. By eliciting evaluations of general categories as well as central and peripheral members of the very same categories (in varying orders) we propose to test the hypothesis that above assertion applies not only to spontaneous emotional response, but to PVs as well.
6. Identified vs. statistical.
People use probabilities as an excuse for action or inaction, whichever they favor. For example, the two sides of the global warming debate both point to uncertainty as a reason for whichever option they support. So comparison of probabilistic cases to specified numbers might affect all three of our dependent measures. Note that we can manipulate identified vs. statistical outcomes even while manipulating the quantity of outcomes. For example, we can say, "forests A, B, and C" vs. "three forests out of the ten". We hypothesize that preference for complete solutions and insensitivity to quantity will be greater for identified cases than for statistical ones.
3. Comprehensive description of the methodology and plan of operation.
We describe here in more detail the manipulations we will use, and how they are related to the effects we propose to study.
A. Evaluation context
Ritov and Kahneman (1997) have proposed that evaluations involve an implicit norm. When people are asked to evaluate saving elm trees, they compare the value of elm trees to other trees. They might assign a high value to such a project. When asked about finding a cure for viral meningitis, they might assign a lower value because they would compare this to other diseases. The rank order of objects within category is assumed to be relatively stable, in two respects. First, different response measures yield similar ordering, provided the objects are judged one at a time. Second, adding other objects from the same natural category should not affect the rating of each object.
The implicit norm can also be changed by including other items in the same task. So, if the task included both elm trees and viral meningitis, they would be comparing people and plants. Each would serve as a norm for the other. In this light, subjects, expressing their spontaneous evaluation, might value the disease more highly than the tree, although they evaluated elm trees higher than viral meningitis, when the two problems were presented separately. Again, if judgement is based on protected values, this reversal may not occur. We shall attempt to extend this result by using items that fall into two different categories, such as malaria vaccine development as an example of a health program or an example of fighting poverty. For problems such as this, whose categorization is ambiguous, simply naming the category (one of the relevant ones) may change the evaluation.
These effects of adding items within or between categories may not apply when preferences are based on protected values. It is as yet unclear whether a PV for ßaving animals from extinction" applies to termites as well as pandas. Moreover, it is unclear whether the expression of a PV is labile, influenced by context, or not. The answer to this question bears on how seriously we should take expressions of PVs. The more seriously we must take them, the more of a problem they create for practical measurement of public values. We shall thus compare the effects of same-category vs. different-category context on PVs and non-PVs, where the PVs are assessed in the two context conditions, both for each category and for its individual measures.
Another, related question, concerns the very tendency to base a decision on protected values. Bazerman et al. (1998) suggest that in assessing a single alternative, the way to make a decision is less apparent than in comparison, or choice. In the former case, the decision maker must develop his or her own intuitive decision model. One approach Bazerman and his associates (Bazerman et al., 1998) suggest, invokes the distinction between what an individual wants and what he thinks he should do. They argue that in evaluating a single alternative, people may be more inclined towards what the "want", as opposed to what they think they ßhould" do. The same assumption has been used to explain preference reversals between WTP for private and public goods (Irwin et al., 1993). However, it is possible that in the struggle to make sense of a single alternative, a protected value may provide the ëasy way out". This hypothesis suggests single object evaluations, may yield higher endorsement of protected values than joint evaluations, particularly when the alternatives evaluated jointly are from different categories. Thus, we shall compare PV expressions (as defined earlier) for within-category vs. single-item tasks. Another possibility, mentioned earlier, is that PVs are best understood as extreme expressions of attitudes. In this case, PVs will be expressed differently for single objects only if attitudes are expressed differently, and this result has not been found.
We propose to test these ideas, in a set of studies, by using three conditions: separate evaluation; joint evaluation of members of the same category; and joint evaluation of members of different categories. Moreover, these tasks will help pull apart the determinants of four different kinds of evaluation responses: pricing (as in contingent valuation); relative importance judgments; omission bias; and PVs. In the pricing response, subjects will indicate willingness to pay for a program to reduce some harm, such as the decline of elm trees or the incidents of bacterial meningitis. The amount of the good will be specified and varied either across widely separated cases or across cases presented side by side. In the tradeoff response, the subject will rate the importance of the program in comparison to some specific other program (one that is rated highly). In the omission bias response, subjects will indicate how many units of the bad event they will tolerate resulting from their action, in order to prevent a fixed number of units (e.g., a million cases or trees) from happening through their omission. In the PV manipulation, subjects will answer questions about the bad event in question, in particular, whether they think it should be allowed in return for some benefit if that benefit is great enough.
B. Central vs. peripheral category membership
We hypothesize that general descriptions of categories will make subjects think of members most central for the particular context. A question about protection of endangered species in general will evoke emotional responses as strong as a question about dolphins (a central member) and much stronger than questions about mosquitoes. We also hypothesize that protected values will follow the same pattern as, and be mediated by, these emotional responses. We have already carried out a pilot study (but without completing data analysis) using categories of this type.
In another study, we will prime categories by presenting either a central or peripheral member before the category itself. We expect fewer PVs for the categories, and less emotion, when the prime is peripheral. (Subjects will also make judgments about the prime.) Finally, we shall also manipulate whether the action being considered is banning or discouraging the unwanted harm, e.g., cutting pristine forests. Banning is sure to stop the central abuses. Discouraging might not. Thus, banning should be more sensitive to PVs. If this happens, it will suggest that those with PVs would support more categorical laws, which cover some entire category, if that category includes what is of primary concern to them. For example, they might want to ban genetic modification of humans if their primary concern is the use of genetic modification to create a "master race." But this would also ban the treatment of genetic defects.
C. Complete vs. partial elimination
In judging possible improvements in the state of the world, is the complete elimination of a problem valued over and above other improvements, possibly alleviating problems without eliminating them? Previous research offers mixed results. First, consumers of risky products, for example, were shown to be willing to pay a disproportional price for the elimination of one type of risk, as compared to their willingness to pay for a comparable overall risk reduction which does not involve elimination of any particular risk (Viscusi, Magat, and Huber, 1987). In this case the risk stemmed from the use of an individually purchased product, and hence the disproportional value of complete elimination of problems might have been limited to private goods.
As described earlier, willingness to pay for environmental goods may be largely insensitive to relevant quantitative information. WTP's failure to increase with the size of the proposed environmental good often extends to the special case of partial versus complete elimination of the relevant problem (Kahneman, 1986; McFadden and Leonard, 1993). Kahneman and Ritov (1994) varied, between subjects, both the scope of the problem and the completeness of proposed solutions. They constructed three versions of each of three problems. For example, one version of a problem pertaining to the shrinking of rain forests mentioned the threat to the rain forest in both Brazil and Ecuador, and proposed an intervention in both countries. Another version mentioned the threat in both countries but proposed an intervention for Ecuador alone. Finally, the third version restricted the stated problem to Ecuador, and matched the proposed intervention to the problem. The other two problems involved species of birds threatened by pollution, and the spread of AIDS in some African countries. Across the three problems, Kahneman and Ritov found no significant effect either of the problem's scope or of the intervention's completeness. It is worth noting, however, that the only problem for which the partial solution was valued significantly less than the complete solution was the spread of AIDS in Africa.
Ritov and Zohar (in preparation) examined two factors which may account for the mixed findings regarding preference for complete solutions: a distinction between problems which involve threats to human life and problems which do not pose such threats, and a discriminative evaluation of problems directly affecting the respondent. They examined those two factors by varying each independently, and found that while willingness to pay was lower for non risky ecological problems, and for problems not directly affecting the respondents, decrease in WTP due to incompleteness of the proposed solution occurred only for life threatening problems.
The scope of problems examined by Ritov and Zohar was limited, and does not allow for a definite conclusion. It seems, however, that the preference for complete solutions occurs more in some domains than others. The factors determining which domains remain to be investigated. We will examine this question in a greater variety of domains.
The two evaluation processes suggest different hypotheses concerning the interaction of problem domain and preference for complete solution. The attitude model of public goods valuation (Kahneman and Ritov, 1994; Ritov and Kahneman, 1997), asserts that concern about the problem is a component, or an expression of the spontaneously evoked attitude (Kahneman, Ritov, and Schkade, 1998). Recent research underlies the importance of affective response in risk assessment studies (Slovic, 1997). Although there is no evidence that concern increases sensitivity to quantity in general, it is plausible that complete elimination of problems arousing great concern will be relatively over valued. Problems involving human mortality, are likely be more disturbing than other public problems. The extra value attached to complete solution of problems involving risk to human lives, may, thus, result from the increased concern about these problems.
Baron, Hershey, and Kunreuther (2000) found worry to be a strong predictor of desire for risk reduction. Problems involving human mortality, are likely be more disturbing than other public problems. The extra value attached to complete solution of problems involving risk to human lives, may, thus, result from the increased concern about these problems.
A different explanation of the Ritov and Zohar findings is based on the concept of protected values. As Baron and Spranca (1997) show, protected values stem from rules about actions, which people hold themselves morally obliged not to violate, regardless of consequences. One such rule could be never to put any human life in danger. If subjects espouse this rule, they should regard complete elimination of a threat to human life as infinitely better than a partial reduction, regardless of the reduction's quantitative aspects. The extra value assigned to complete elimination of health risks could stem from the fact that the preservation of human life is shared as a PV by may people.
The two possible sources of the complete solution effect have different implications. If concern is the primary reason for preferring a complete solution, then different degrees of concern should yield corresponding variation in preference for complete solution. If, on the other hand, the effect is mostly due to PVs, its occurrence should depend on the degree to which the relevant value is protected. In a set of studies we plan to explore the complete solution effect, by expanding the scope of outcomes to include other health risks, and by asking about concern and absoluteness (PV endorsement), in addition to WTP. For example, in one experiment we shall employ a between subject factorial design, with environmental problems varying in whether or not a risk to human life is involved, whether or not they occurred in the area in which the respondent lives, and whether the proposed intervention would partially solve or completely eliminate the problem. Subjects will provide their WTP for the proposed intervention, rate their concern about the problem, and indicate whether or not they consider the pertinent value to be protected.
In another experiment the preference for complete elimination of risk to human life will be examined in a within-subject design, varying two factors: whether or not the problem affects the area in which the respondent lives, and whether the proposed intervention would partially solve or completely eliminate the problem. As personal relevance is likely to affect the individual's concern more than his or her endorsement of protected values, this study will enable to more clearly differentiate concern from absoluteness as determinants of preference for complete solutions.
D. The action target: identified or statistical
A public problem sometimes takes the form of helping identified individuals, as in the case of a child who was critically wounded in a rare accident and is in need of a costly emergency treatment. People are often willing to expend greater resources to help such individuals, than they are in other cases, in which the problem's target is not identified. This so called ïdentified victim effect" has been the subject of a study by Jenni and Loewenstein (1997), who examined possible sources of the effect. Their research suggested that the most important cause of the disparity in treatment of identifiable and statistical lives involves the proportion of the lives at risk that would be saved. Their finding is consistent with the more general phenomenon that people are concerned about the distribution of risk among populations (Slovic, Fischhoff, and Lichtenstein, 1982), and, in particular the concern about relative rather than absolute risk reductions (Baron, 1997b; Fetherstonhaugh et al., 1997).
Jenni and Loewenstein (1997) supported their argument by demonstrating that, in the case of traffic accidents, varying the proportion the lives saved affected the rating of importance assigned to reduction of risk. It is possible, however, that relative risk reduction affects evaluation differently, depending on whether the lives at risk are statistical or identifiable. This hypothesis will be tested in the present research.
The present research further proposed that beyond the proportion of lives saved, protected values may also be related to the ïdentified victim effect". It is plausible to assume that protected valued may apply more in the case of an identified victim than in the case of a statistical one. It is easier to regard not saving statistical lives as an omission, than not saving an identified person. As protected values reflect deontological rules about actions, holding the saving of human life a protected value would dominate evaluation of the former case more than of the latter. We shall briefly describe the results of a pilot study testing the above hypothesis.
Pilot study
Method
Seventy five students at Ben-Gurion university participated in this study. The questionnaires were run in class sessions. Each subject was randomly assigned to one of the versions.
The first part of the questionnaire posed a single question. A 2X2 between subject design was used, with identifiable vs. statistical victims as one factor, and the number of saved victims (1 out of 5 vs. 3 out of 5) as the other factor. All versions began with the sentences Ïmagine you are working for the ministry of health, and are in charge of an area with a population of one million people. The budget at your disposal is very limited."
The identified victims versions then continued:
"Five toddlers from a single daycare are brought to the hospital with an acute poisoning caused by a mushroom which develops in puddles and small water reservoirs. This type of mushroom is extremely rare. The previous known case of poisoning by the mushroom occurred ten years ago. All five toddlers are in critical condition. The doctors suggest a new experimental treatment. This treatment is extremely costly, and its results are uncertain. If the treatment is given to all five victims, it is expected that 3 (1) of them will be saved. You decide not to authorize the treatment, and all five toddlers die".
The statistical victims version delineated, instead, the following situation:
Äbout once every ten years a rare type of mushroom, which develops in puddles and small water reservoirs, causes poisoning in children. During the next ten years it is expected that five toddlers will die of poisoning by this mushroom. If the sensitivity of a child to this mushroom is known in time, it is possible to give her immediate treatment, which will reduce the risk of death. A relevant test is available, but in order to have the results in time, it needs to be taken before exposure to the mushroom. If this test is given to all children in the area you are in charge of, it is expected that the number of deaths due to this cause will decrease from 5 in ten years to 2 (4) in ten years. Giving the test to all children will be costly, in light of your limited budget. You decide not to give the test. In a few years a new case of the mushroom poisoning occurs, and five toddlers die as a result."
Subjects in all versions were asked to rate the degree to which they would feel personally responsible for the death of the five children. The rating scale ranged from 1 (not responsible at all) to 7 (most responsible).
After completing the first part of the questionnaire all subjects completed an identical second part, in which protectedness of the relevant value was assessed. The list was made up of different possible actions (or inactions) by government. The two target statements for the present study were: "Not taking action to avoid potentially fatal diseases" and "withholding treatment from patients with life threatening diseases, because of budget constraints". For each statement the subjects selected one of the following three responses, which best described their position: Ï do not object to this", "This is acceptable if the benefits are sufficiently significant", and "this is unacceptable, no matter how beneficial the outcome may be". Subjects were classified as having a protected value only if they selected the ünacceptable" option for both target statements.
Results
Table 1 presents the means of responsibility rating, by type of victim, proportion of saved victims, and value protectedness. The ratings were subjected to an analysis of variance by the above three factors. The main result of interest is that protectedness interacted with type of victim (F=5.21, p=.02): the difference between the means of subjects who hold a protected value, and those who do not, was greater for identified victims than for statistical ones. These results indicate that a decision concerning identified victims is concerning statistical victims. perceived as more closely linked to the relevant (protected) value than a decision
The distinction between identified and statistical victims may, in some cases, be a framing effect. Some evidence exists for such a framing effect. Keller and Sarin (1988) gave subjects hypothetical options like the following:
Option 1:
Option 2:
Option 3:
People generally prefer option 3 to option 2, and option 2 to option 1. That is, people are concerned with equity before the resolution of uncertainty as well as after. The interesting thing about options 2 vs. 1, though, is that "person 1" and "person 2" are arbitrary. They are identified only by number. So the preference here for ïdentified individuals" may be subject to a framing effect, especially if we make clear that the numbers are assigned at random.
A further hypothesis involves the link between protected values and omission bias. Ritov and Baron (1999) showed that subjects holding protected values exhibited more omission bias than subjects not holding PV's. If PV's are more clearly applicable in the case of identified victims, then omission bias due to protected values should also be more notable in that case, as compared to the case of statistical victims.
More generally, in a new set of studies, we ask whether identified outcomes differ from statistical ones in their effect on preference for complete solutions, PVs, sensitivity to quantity, and omission bias. For example, one experiment will replicate the design of the pilot study, with the modification that subjects will be asked to make a choice, rather than just rate the responsibility of the agent. Another experiment will examine the effect of omission vs. commission, in the context of victim type (identified vs. statistical) and protected values. A 2 (victim type) X 2 (omission/commission) within subject design will be used, in a task similar to that of pilot study. If our hypothesis is correct, we expect to find that omission bias due to protected values (the difference between omission and commission in subjects with PV, compared to the same difference in subjects without PV) will be greater for identified victims than for statistical ones.
Overview of expected results and their significance
The over-arching hypothesis is that people have two modes of evaluating public polices, one quantitative and consistent with standard economic views and one more intuitive. Political action - including individual action with effects on public outcomes - is often based on little thought, because the effects are mostly on others and the individual his little self-interested incentive to think quantitatively. Our proposal concerns the determinants and the effects of intuitive thinking. We are interested in two main determinants: primary emotional responses to descriptions of issues; and protected values, that is, values seen as absolute. These may be related.
In particular, we hypothesize that the intuitive mode will lead to, or increase, the following effects:
1. preference for complete solutions
2. insensitivity to quantity
3. omission bias
4. context effects
5. the identified victim effect
We shall both measure the two causal variables of interest - emotional attitudes and protected values - and attempt to manipulate them. The manipulations will include the context in which the object is embedded (e.g., environment vs. environment plus health) and the relation of the object to a category (e.g., general description of the category, description of a central member likely to evoke a strong response, description of a peripheral member.)
In principle, these two manipulations and five dependent measures can lead to ten series of experiments. However, many of the measures can be combined, such as act/omission, identifiable victims, and complete/incomplete solutions. We shall also explore different types of issues: health, environment, rights, etc. Our manipulation of context will involve combining these or keeping them separate.
From a theoretical point of view, our results could elucidate the nature of the biases in question. Of particular interest are interactions among the effects. We shall seek such
interactions systematically, extending research we have already done. For example, an interaction between the identified-victim effect and omission bias would suggest that the former effect was the result of a rule against hurting known people through actions. This would suggest, in turn, a hypothesis of the origin of the act/omission distinction in thinking, namely, that it arises from such a rule, which might easily be learned by children who are punished for specific transgressions. We cannot here anticipate all possible interactions, but the search for them is a major focus of the research.
The practical significance of this project is to clarify the nature of thinking about public issues. Such clarification can lead people to understand better why policy decisions are often sub-optimal. They can lead, in principle, to institutional ways of improving decisions, such as ways of getting public input about risk analysis. But, more generally, as the psychology of judgment extends its reach, it becomes part of the public's own understanding of itself, and - if the understanding is accurate - this can only help.
General methods for all studies
We shall carry out several series of studies using the kinds of public issues that we have used in previous research. These include reduction of environmental harm from pollution, prevention of environmental harm, reduction and prevention of risks to human life and health, and also other goods such as cultures, languages, historic buildings, and works of art. We shall also include items concerned with the violation of rights, such as arbitrary imprisonment and prevention of free speech. These are all policy issues that evoke PVs in some people. They can all be quantified, and we can present hypothetical cases in which harms are identified or statistical.
We shall carry out our research both with paper questionnaires (mostly in Israel) and on the World Wide Web. The use of the Web for research is of interest in its own right. The NetLab Workshop (NSF, 1997) recommended encouragement of just this sort of effort, and Baron's present research is on the cutting edge, or one of many such edges. In particular, the use of JavaScript allows programming of a kind of interview, in which later questions are contingent on earlier answers, as well as checking of responses as they are made, feedback about practice items, etc. Current experiments are available at http://www.psych.upenn.edu/ baron/qs.html. This method has several advantages over pencil-and-paper questionnaires:
1. The subjects are more varied, and special efforts can be made to solicit particular groups of subjects. This is more true each year, as more people use the web. For example, current research involves getting responses from patients who have had chemotherapy.
2. It is easier to make sure that subjects answer all the questions, give answers in the appropriate range, and meet minimal conditions on ordering.
3. Interactive questioning can be programmed so that it can be described clearly and replicated. In this regard, the method is similar to computer assisted telephone interviewing, but it is much less expensive. Included here are error checking, challenges to past responses, and use of practice trials with feedback about them.
4. Some forms of ``cheating,'' such as completing the same study twice, are more difficult than in other methods. Subjects must provide their name and address (and, for Americans, their social security number) in order to be paid. Other forms of cheating, such as submitting the same form under different names, are impossible with current methods.
5. Although random sampling of a population is impossible (as it is in other methods based on convenience samples), the sample is more varied than those that use solely students. Up to 50respondents (depending on the study) are non-students, and more than half are female (despite fears that the web is male dominated). More than 10variety.
6. Costs of data entry are eliminated.
7. Several studies have found that the quality of data collected on the web is as high (or sometimes higher) than that of data collected in the laboratory (Birnbaum, 2000).
The general experimental procedure (which is continuing to evolve) is to use JavaScript to randomize the stimuli separately for each subject. Each session contains 20-100 judgments, depending on their difficulty, and pay ranges from 3 to 6, again, depending on the difficulty and time required. In some cases, stimuli are blocked, and the blocks are randomized too. Typically, each item is presented on a screen, so that the subject cannot back up.
The research is continuous, an extension of current studies, which involve simultaneous data collection, analysis and reporting of earlier studies. Each experiment typically involves 50 subjects and the data are collected in a few weeks. Thus, dozens of experiments can be done in a year. The results are monitored as they come in, and studies that seem to be misunderstood can be redesigned.
Technical details are discussed in Baron
4. Detailed account of available resources.
The studies proposed do not require great resources. The paper and pencil questionnaires will be prepared using the available copying machines. They will be filled out mostly by students fulfilling partial requirements in introductory courses. Personal computers of the PI's will be used to enter and analyze the data.
5. Progress report on recent BSF-supported project
Project title: "The cognitive psychology of social issues"
Principle investigators: Ilana Ritov and Daniel Kahneman.
Period of support: 1995 - 1998
The supported research sought to explore the determinants of subjective evaluation of public goods, particularly when those are presented as a single, context free alternatives. Our research gradually led us to focus on the attitude model as the basic scheme for understanding the way people evaluate public problems.
The core of an attitude is a spontaneously evoked valuation, which assigns to the entity an affective value that can range from extremely positive to extremely negative. An attitude object considered on its own is implicitly compared to a set of objects of the same general kind. The alternatives that an object brings to mind tend to share its essential attributes, and are drawn from the category to which it would naturally be assigned. This within-natural category comparison process diminishes the impact of the more prominent attributes, as these are shared by all objects under comparison. Thus, for example, saving a threatened animal species may be rated higher than reducing lead poisoning incidence, because when each problem is encountered in isolation it is spontaneously compared to other problems of the same category.
We further explored the precise nature of the object, which invokes the attitude, subsequently expressed as WTP, satisfaction, or a political stand . When a respondent reads, for example, that "The population of the Peregrine Falcon in a coastal preserve has declined by 25be the 25birds of pray in general, or a particular image of a single Peregrine Falcon in distress. We argued, as a general psychological principle, that the object of evaluation is more likely to be the latter. Indeed, we obtained similar evaluations of problems, when asking subjects to consider a particular instance, or a more generally informative description of each problem. The attitude model provides an integrated framework for understanding systematic violations of economic rationality. In particular, it yields the following important implications: (a) In as much as they reflect the core emotional evaluation, different measures of attitude should yield similar responses. (b) Evaluation by exemplar accounts for insensitivity to quantitative aspect of the problem, when it is judged on its own. (c) When the context of judgment is different from the spontaneously evoked context, relative preferences may reverse.
The research is summarized the following papers:
Ritov, I. and Kahneman, D. (1997). How people value the environment: Attitudes versus economic values. In Bazerman, M., Messick, D., Tenbrunsel, A., and Wade-Benzoni, K. (Eds.). Psychological Perspectives to Environmental and Ethical Issues. The New Lexington Press.
Kahneman, D., Ritov, I., and Schkade, D. (1998). Economic preferences or psychological expressions: an analysis of dollar responses to public issues. Journal of Risk and Uncertainty, 19, 203-235.
Kahneman, D., Schkade, D., Ritov, I., and Sunstein, C. (under review). The relativity of attitudes: adjectives, dollars, and judgment reversals.
6. Relevant bibliography on research area
Bazerman, M. H., Moore, D., Tenbrunsel, A. E., Wade-Benzoni, K., and Blount, S. (1998). Explaining Joint Versus Separate Preference Reversals. Manuscript
Baron, J. (1997). Biases in the quantitative measurement of values for public decisions. Psychological Bulletin, 122, 72-88.
Baron, J. (1997b). Confusion of relative and absolute risk in valuation. Journal of Risk and Uncertainty, 14, 301-309.
Baron, J., Hershey, J., and Kunreuther, H. (2000). "Determinants of priority for risk reduction: the role of worry. Risk Analysis, 20, 413- 427.
Baron, J.,
Baron, J. and Ritov, I. (1993) Intuitions about penalties and compensation in the context of tort law. Journal of Risk and Uncertainty, 7, 17-33. Reprinted also in: Camerer C. and Kunreuther, H. (eds.). Making Decisions About Liability and Insurance. Kluwer Academic Publishers (1993).
Baron, J. and Ritov, I. (1994) Reference points and omission bias. Organizational Behavior and Human Decision Processes, 59, 475-498.
Baron, J.,
Baron, J. and Spranca, M. (1997). Protected values. Organizational Behavior and Human Decision Processes, 70, 1-16.
Birnbaum M. H. (Ed.) (2000). Psychological Experiments on the Internet. New York: Academic Press.
DeKay, M. L., and McClelland, G. H. (1996). Probability and utility components of endangered species preservation programs. Journal of Experimental Psychology: Applied, 2, 60-83.
Fetherstonhaugh, D., Slovic, P., Johnson, S., and Friedrich, J. (1997). Insensitivity to the value of human life: A study of psychophysical numbing. Journal of Risk and Uncertainty, 14, 238-300.
Hsee, C. K. (1996). The evaluability hypothesis: An explanation for preference reversals between joint and separate evaluations of alternatives. Organizational Behavior and Human Decision Processes, 67, 242-257,
Irwin, J. R., Slovic, P., Lichtenstein, S., and McClelland, G. H. (1993). Preference reversals and the measurement of environmental values. Journal of Risk and Uncertainty, 6, 5-18.
Jenni, K. E. and Loewenstein, G. (1997). Explaining the ïdentifiable victim effect". Journal of Risk and Uncertainty, 14,
Kahneman, D. (1986). Valuing environmental goods: An assessment of the contingent valuation method. In R. Cummings, D. Brookshire, and W. Schultze (Eds.) Valuing environmental goods: An assessment of the contingent valuation method. Totowa, NJ
Kahneman, D., and Knetsch, J. (1992). Valuing public goods: the purchase of moral satisfaction. Journal of Environmental Economics and Management, 22, 57-70.
Kahneman, D. and Ritov, I. (1994). Determinants of stated willingness to pay for public goods: A study in the headline method. Journal of Risk and Uncertainty, 9, 5-38.
Kahneman, D., Ritov, I., and Schkade, D. (1998). Economic preferences or psychological attitudes: an analysis of dollar responses to public issues. Journal of Risk and Uncertainty, in press.
Kahneman, D., Schkade, D. A., and Sunstein, C. R. (1998). Shared outrage and erratic awards: The psychology of punitive damages. Journal of Risk and Uncertainty, 16, 49-86.
Keller, L. R.,
Lim, C. S.,
McFadden, D. and Leonard, G. K. (1993). Issues in the contingent valuation of environmental goods: Methodologies for data collection and analysis. In Hausman (Ed.) Contingent Valuation. A Critical Assessment. North-Holland
NSF (1997). NetLab Workshop Report. http://www.uiowa.edu/ grpproc/netlab.htm
Ritov, I. and Baron, J. (1990) Reluctance to vaccinate: ommission bias and ambiguity. Journal of Behavioral Decision Making, 3, 263-277.
Ritov, I. and Baron, J. (1992) Status-quo and omission biases. Journal of Risk and Uncertainty, 5, 49-61.
Ritov, I. and Baron, J. (1995). Outcome knowledge, regret, and omission bias. Organizational Behavior and Human Decision Processes, 64, 119-127.
Ritov, I. and Baron, J. (1999). Protected values and omission bias. Organizational Behavior and Human Decision Processes, 97, 79-94.
Ritov, I., Baron, J. and Hershey, J.C. (1993) Framing effects in the evaluation of multiple risk reduction. Journal of Risk and Uncertainty, 6, 145-159.
Ritov, I. and Kahneman, D. (1997). How people value the environment: Attitudes versus economic values. In Bazerman, M., Messick, D., Tenbrunsel, A., and Wade-Benzoni, K. (Eds.). Psychological Perspectives to Environmental and Ethical Issues. The New Lexington Press.
Slovic, P. (1997). Trust, emotion, sex, politics, and science. Surveying the risk-assessment battlefield. In Bazerman, M., Messick, D., Tenbrunsel, A., and Wade-Benzoni, K. (Eds.). Psychological Perspectives to Environmental and Ethical Issues. The New Lexington Press.
Slovic, P., Fischhoff, B., and Lichtenstein, S. (1982). Response mode, framing, and information processing effects in risk assessment. In R. M. Hogarth (ed.) New directions for methodology of social and behavioral science: The framing of questions and the consistency of response. San Francisco: Jossey-Bass.
Tversky, A., Sattath, S., and Slovic, P. (1988). Contingent weighting in judgment and choice. Psychological Review, 95, 371-384.
Viscusi, W. K., Magat, W. A., and Huber, J. C. (1987). An investigation of the rationality of consumer valuation of multiple health risks. Rand Journal of Economics, 18, 465-479.