Benefit-Cost Analysis As A Source Of Information About Welfare

National Research Council (US) Steering Committee on Valuing Health Risks, Costs, and Benefits for Environmental Decisions; P. Brett Hammond; Rob Coppock

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Steering Committee on Valuing Health Risks, Costs, and Benefits for Environmental Decisions; Hammond PB, Coppock R, editors. Valuing Health Risks, Costs, and Benefits for Environmental Decision Making: Report of a Conference. Washington (DC): National Academies Press (US); 1990.

Cover of Valuing Health Risks, Costs, and Benefits for Environmental Decision Making

Valuing Health Risks, Costs, and Benefits for Environmental Decision Making: Report of a Conference.

Show details

< Prev Next >

4Benefit-Cost Analysis As A Source Of Information About Welfare

Peter Railton

"Benefit-Cost Analysis: Threat or Menace?" is the theme of many philosophical discussions, but it is not the theme of this one. It does not seem to me that the very idea of benefit-cost analysis is a moral outrage or a conceptual absurdity. On the contrary: I am prepared to think that good benefit-cost analysis could do much to improve the reasonableness of policy making. At the same time, however, I am quite uncertain as to whether the actual application of benefit-cost analysis has generally had a salutary effect on environmental policy making, in part because I have some doubts about the justifiability of certain practices that appear common in actual applications of the process. My aim in this paper is to make some of these doubts clear and to indicate roughly how benefit-cost analysis might be carried out in a way that would substantially mitigate them.

My strategy in discussing these doubts is inspired by a question put to me by someone involved in the practice of benefit-cost analysis: "What do I have to believe in order to do this?" I will consider various actual and possible practices in benefit-cost analysis as it is applied to governmental regulation of risk and discuss what one might have to believe in order to defend such practices. I will then ask whether such beliefs seem tenable. When they do not, I will suggest what one might more credibly believe about the matter at issue and how these changes in belief might affect the practice of benefit-cost analysis.

The first part of this paper considers several questions about the scope and comprehensiveness of benefit-cost analysis in policy evaluation. In particular it is concerned with the compatibility of benefit-cost analysis with conceptions of social justice in which distribution, desert, or entitlement play a role. The second part of the paper deals with the notion of utility that might lie behind the economist's appeal to preferences in social policy assessment, and the relation of this underlying notion to those of price and willingness to pay. Of special interest is how diminishing marginal utility affects the use of prices to measure willingness to pay. The third part of the paper looks at the question of measuring costs and benefits that will arise in the future. This section argues the need for a distinction between, on the one hand, discounting future prices or willingness-to-pay indicators, a procedure that may be necessary for the accurate measurement of future costs and benefits, and, on the other hand, discounting future utilities themselves, a procedure that may lead to systematic mismeasurement. By way of conclusion, the paper suggests a role for benefit-cost analysis in highlighting future utilities that would otherwise lack adequate representation in the present. In contrast to its reputation as a threat to intergenerational justice, benefit-cost analysis could—in virtue of its commitment to accuracy in measurement—provide information that would make it uncomfortable for policy makers to ignore intergenerational inequities.

I have been asked to contribute a philosopher's perspective to the assessment of benefit-cost analysis. My hope is that this paper engages constructively with the practitioner's concerns in raising the question, "What do I have to believe?" I should, however, emphasize at the outset that my suggestions come from the standpoint of an outsider to the theory and practice of benefit-cost analysis, without the advantages—and disadvantages—of close involvement in the state of the art.

I should emphasize, too, that I am an outsider to the politics of environmental regulation. In particular, I am unable to address the question of whether benefit-cost analysis is a Trojan horse for a particular political tendency in environmental policy. If that currently is where the real issue over the utilization of benefit-cost analysis lies, then this paper is an exercise in naivete and I might be compared to a loyal but benighted citizen of Troy who offers the Greeks advice on how best to paint their gift.

Finally, I should also emphasize that I make no attempt to survey the full range of moral and philosophical issues that surround the justification of benefit-cost analysis, but instead focus almost entirely upon certain questions related to the measurement of benefits and costs. I have chosen this focus for a particular reason: if benefit-cost analysis fails in its job of measurement, then its greatest claim to the attention of those involved in environmental regulation—its capacity to provide information about the relative magnitude of a wide range of disparate benefits and costs, some of which might otherwise receive little notice or quite arbitrary treatment—will become too weak to justify consideration of its use. As I hope to indicate, however, benefit-cost analysis may be able to do a reasonably good job of measurement if those applying it are explicit about the difficulties involved and if they adjust their claims about the decisiveness, accuracy, and comprehensiveness of its assessments to the reality of its limited measurement techniques.

Scope And Comprehensiveness

To carry out benefit-cost analysis, what must an individual believe about its scope? The short answer to this question is, ''Very little, if one puts it only to very little use." In contrast, if a person thinks benefit-cost analysis can be a powerful tool of policy assessment, then he or she must have fairly powerful beliefs about what can come within its scope.

Let us consider, for example, an extreme position that is probably held by very few people: benefit-cost analysis affords the appropriate normative criterion for social choice. To believe this, one would have to believe that all the benefits and costs relevant to the assessment of policies can be satisfactorily accommodated within benefit-cost analysis.

How hard would it be to believe that? Benefit-cost analysis is founded on the idea of a potential Pareto improvement, that is, a change in which the gainers benefit sufficiently to be able to compensate the losers and still come out ahead. How might this standard—which is usually seen as capturing an idea of efficiency—provide the appropriate normative criterion for social choice? The clearest rationale would probably run as follows. Changes meeting this standard can be thought of as enlarging the total pool of benefits that are socially available. If certain complications are ignored and if population remains constant, and moreover if it is assumed that over time the benefits and burdens of social policies tend to be distributed randomly among individuals, then, when policies are implemented that all satisfy a standard of potential Pareto improvement, the expected value of all social positions will tend to rise. This rationale comes close to capturing a classical utilitarian standard of social choice (because overall utility increases if the expected value of all social positions rises). More surprisingly, it comes close to capturing a hypothetical contract standard of social choice because rational individuals who are concerned about advancing their own well-being will tend to prefer social arrangements with higher expected utility per position. Because hypothetical contract theory and utilitarianism are the two dominant philosophical trends in the theory of social justice, it might be argued that reliance on a potential Pareto improvement standard would be consistent with promoting justice in well-recognized senses.

Yet this argument does not really make it easy to believe that benefit-cost analysis gives an exhaustive account of the values that are relevant to normative social choice—even the single value of justice. In the first place, this approach assumes that benefits and burdens are randomly distributed; in the absence of further argument, however, it is hard to believe that this condition will be met so reliably that distribution in particular cases or in cumulative patterns need not be considered. After all, the standard of potential Pareto improvement says nothing about existing inequalities, and there is no guarantee that changes that would meet the standard would not bring about arbitrarily large increases in inequality. Serious normative worries might therefore arise for distribution-sensitive conceptions of justice.¹ In the second place, the standard says nothing about whether the present distribution is in any sense deserved, and there is no guarantee that changes that would meet the standard would not allow gains or losses to go to individuals who did not earn or otherwise deserve them. Serious normative worries might therefore arise for desert-based conceptions of justice. Even if one were to suppose that many potential Pareto improvements would be compatible with desert or equity, some would not be. Thus, on the basis of most conceptions of justice, the benefit-cost standard could not be an adequate criterion of normative social choice.

It seems to me that most theorists and practitioners of benefit-cost analysis recognize these limitations and do not advocate the use of benefit-cost analysis as a sufficient condition for policy approval. It is not uncommon, however, to hear benefit-cost analysis put forward as a necessary condition for approval, a preliminary "test" that policies must meet before further tests (of equity, desert, and so on) can be applied. Yet reasonable demands of equity or desert may sometimes require changes of a kind that would not pass a potential Pareto improvement test. For example, redistribution designed to remedy past injustices might not yield more for the gainers than it costs the losers; still, the gainers might deserve the change, and equity might be served by it. In a society in which past injustices call for such remedies, to restrict social choice to policies that represent potential Pareto improvements would be to decide, in effect, against these demands of desert or equity.

A defender of benefit-cost analysis might respond to these considerations by suggesting that the scope of benefit-cost analysis be enlarged to encompass all normatively relevant features in social choice. One way to accomplish this would be to identify people's willingness to pay for equity or desert. Yet surely, the question policy makers face is not simply "How much do our constituents care about whether things are just by their own lights?" but also "What is justice—what does it require?" However useful it may be to know the answer to the former question in order to answer the latter, the former cannot entirely replace the latter. In many cases, considerations of justice appear to function not as factors or weights within benefit-cost analysis but rather as constraints on the proposals that are prepared for such analysis. A benefit-cost analysis of the regulation of a toxin does not even consider such cost-cutting options as massive government deception about health effects or refusing to offer health care to individuals of certain races. A utilitarian might argue that such alternatives need not be analyzed because people are already convinced that they are contrary to maximizing social benefit. A Kantian, on the other hand, might insist that, regardless of whether social benefit is maximized, the government should not deceive or discriminate in these ways. These viewpoints constitute two poles of a substantive debate about the foundations and requirements of justice. Fortunately, it is not necessary to resolve this debate, or even enter into it, in order to carry out benefit-cost analysis—that is, as long as a benefit-cost standard is not treated as the appropriate normative criterion of social choice.²

If a benefit-cost standard is not seen as a necessary or sufficient condition of social choice, how, then, should benefit-cost analysis be viewed? It might be used as an information-yielding device, a way of generating and bringing together within a quantified scheme a great deal of data about how people are likely to be affected by alternative policy choices. Both utilitarians and contractarians consider it important to know how policies compare with regard to the sorts of benefits that benefit-cost analysis is best equipped to measure. For example, if a policy appears to be consistent with the constraints of justice and yet is found to be inefficient, or not cost-effective, or suboptimal with respect to aggregate expected value, that information will be viewed as normatively relevant by utilitarians and contractarians alike. One need not take sides in a controversy over the theory of justice to assign considerable importance to the information yielded by benefit-cost analysis in social decision making.

Viewing benefit-cost analysis as an information-yielding rather than a decision making device has implications for how such analyses should be presented. For example, a benefit-cost analysis will be more informative (although less decisive) if it reports disaggregated as well as aggregated effects—for instance, by telling policy makers which elements within the population will receive which costs and which benefits. If benefit-cost analysis is to help policy makers to choose justly, and if justice is distribution-or desert-sensitive, then disaggregated information may at times be more crucial than overall net results. Similarly, a benefit-cost analysis will be more informative (although less conclusive) if it refrains from collapsing all uncertainties and reports error bands or sensitivities rather than point measures. There are risks to justice that are every bit as significant as risks to health; it is surely a matter of justice how much of such risk governments should permit. It may be tempting to practitioners of benefit-cost analysis to see their real goal as a grand sum; it may also be tempting to policy makers to say to analysts, "Look, just tell me: what's the bottom line?"—perhaps in the hope that this figure will somehow decide difficult questions of justice for them. Such temptations should be resisted, however, if benefit-cost analysis is to play its appropriate role in policy deliberations.

Thus far, the discussion has centered on possible justice-based constraints on benefit-cost analysis without singling out any particular conception of justice. Such generality seems appropriate: because there is genuine controversy over the nature of justice and what it requires, it would be inappropriate to consign to an analytic technique the task of deciding which social outcomes are to be sought in this country.³

Surprisingly, the range of conceptions of justice that would accord a significant role to the sort of information produced by benefit-cost analysis may be even broader than has thus far been suggested. There may be such a role even within the conception of justice that seems most opposed to social aggregation and balancing: Lockean natural rights theories (e.g., Nozick, 1974). According to the more extreme forms of such theories, individual rights in person and property are natural rights of exclusion existing apart from social arrangements and constraining permissible social policy by setting up normative boundaries that cannot be crossed without the consent of the owner. Thus, for example, if the government were to allow dust and diesel fumes from the building and operation of a new intercity raft line to drift, unbidden, onto private property, then the integrity of that property would have been violated. It would be no defense, in this view, for the government to argue that the aggregate benefit realized from the new rail line would exceed the aggregate cost it would impose. A Lockean natural rights conception of justice, is disaggregative, since each individual is entitled to say what can and cannot cross his or her property line. In the example, if a given individual does not want the benefits or burdens of the new line, then the government is not free to send its dust and fumes onto his or her property, however beneficial the line might be to society as a whole. If control of all dust and fumes would be prohibitively expensive, then private property rights may stand in the way of building the line at all.

Of course, were the government to secure the voluntary consent of all those whose property boundaries would be at risk of violation, construction could proceed. Given the nastiness of dust and diesel fumes, individuals presumably would demand some compensation in return for their permission. With person and property at stake, individuals would be entitled, on a natural rights theory, to hold out for whatever they could get—even to refuse to consent at any price. As a result, a single individual could wield veto power over a scheme of wide-ranging social benefit, and this would give rise to perverse incentives. Individuals could wait until the government obtained near-unanimous consent to some beneficial scheme, and then hold out for compensation much in excess of any harm they might themselves bear.

Yet the constraints of individual veto power and its perverse incentives might be removed if it were possible to go ahead with the beneficial scheme and then provide after-the-fact compensation to all whose property had been adversely affected. This way of proceeding would not allow individuals to set their own levels of after-the-fact compensation, lest they demand arbitrarily large sums and in that way continue to exercise an effective veto. Instead, some fairly objective measurement of the extent of harm to individuals would be established so that commensurate compensation could be determined. The test of potential Pareto improvement addresses just this sort of measurement problem, although its application within a Lockean scheme would have to involve actual rather than merely possible compensation. This requirement in turn would mean including transaction costs in the assessment. Once that were done, however, a policy or activity that appeared to be a potential Pareto improvement could be pursued under this violate-and-compensate gambit.⁴ Thus, a government—or, for that matter, individuals—would be greatly interested in receiving the sort of information that benefit-cost analysis could provide using a framework of actual compensation.

From a philosophical point of view, it is not clear whether those whose fundamental sympathies are Lockean should be happy with the violate-and-compensate gambit, for it may be incompatible with full respect for individual property and consent. Yet a Lockean scheme without this gambit might result in so many inflexibilities as to disqualify it for any actual application. In the modern interconnected world, it is simply not possible to secure actual consent to all impositions of risk or to refrain from all risk-imposing activities. Indeed, even with the gambit, the Lockean scheme would be extraordinarily cumbersome, owing to the problem of identifying and appropriately compensating all actual victims.⁵ For now, however, it is sufficient to note that Lockeans who are concerned with real-world applications are bound to seek something like this sort of flexibility and that for such purposes it would be important to have the sort of information about relative magnitudes of harm and good that benefit-cost analysis may afford.

In sum, individuals with various perspectives on justice may find the information produced by benefit-cost analysis to be greatly valuable. Once benefit-cost analysis is understood as a process meant to yield information rather than to make decisions, practitioners of benefit-cost analysis need not take sides in controversies over the nature of justice. At the same time, it would be quite controversial—indeed, it would almost certainly amount to taking sides against widely held deontological conceptions of justice—to attempt to assimilate the theory of fights into a benefit-cost framework by assigning monetized costs to violations of rights and then entering such costs into an aggregative, balancing scheme. In a range of cases, it would seem that rights are better understood either as (a) constraints upon the possible projects whose costs and benefits decision makers are prepared to assess by analytic means or, more weakly, as (b) markers of areas of special social concern in which policy makers are uncomfortable with a straightforward balancing of costs and benefits. Either way, decision makers will want to give the issue of rights a distinctive role in their deliberations and not blend such considerations into a homogeneous mixture of costs and benefits.⁶

Measuring Costs And Benefits At Particular Points In Time

What does an individual have to believe he or she is measuring when performing benefit-cost analysis? I have been pursuing the idea that benefit-cost analysis should be seen as an information-yielding process. What, then, does benefit-cost analysis provide information about, and how accurate can it be?

One view on this matter has already been called into question, namely, the view that benefit-cost analysis provides an assessment of the relative normative weight of all considerations, including all matters of justice or rights. It seems to me much less implausible to believe that benefit-cost analysis finds its proper subject matter in the realm of welfare. This view might seem to run counter to the orthodox conception that benefit-cost analysis is based solely on the notion of preference. Yet why, when social policy is being chosen, do decision makers think that costs and benefits based on individual preference have any special relevance? Why not look to some less "subjective" measure? The usual answer involves the principle of consumer sovereignty: other things being equal, policy makers should respect the individual's judgments about what matters to him or her and how much it matters. This principle is not a piece of democratic theory, to the effect that "the people shall judge." For example, the principle of consumer sovereignty is applied to preferences but not to beliefs. Benefit-cost analysts seek out expert opinion on the likely effects of risky activities on the environment or on human health; they feel no compulsion to base their estimates of the probability of outcomes on some average drawn from popular opinion. Why, then, do they seek to base their estimates of the utilities of outcomes on individual judgments?

The answer presumably is that, although the public believes that scientists are more knowledgeable than the average American about natural phenomena, it does not believe that, in general, a group of experts would be more reliable than any given individual in ascertaining the extent or nature of that individual's welfare—not, at least, once that individual was informed about the options he or she faced. Scientists have the most sustained and most detailed experience of the natural world, but individuals have the most sustained and most detailed experience of how choice and outcomes affect their particular well-being. This answer would also explain why the principle of consumer sovereignty is applied to the preferences of adults rather than children. If individuals learn about the nature of their well-being through experience, and especially the experience of choice, and if moreover adults in general have better information about the options they face than do children, then the opinions of adults are more likely to be accurate.

By contrast, if one were simply an outright skeptic about the notion of welfare or about the possibility of learning about one's own good, there would be no clear rationale for giving more credence to the opinion of a given person than to the opinion of some randomly selected third party about the effects of a given social policy on that individual's well-being. In addition, there would be no clear rationale for granting informed preferences more authority than uninformed ones. The most plausible way to provide such rationales, and thereby to underwrite the theory and practice of benefit-cost analysis, is to believe that there is such a thing as individual welfare and that individuals do a better job of recognizing it in their own case than do third parties, especially when individuals are well informed.⁷

It may be somewhat unfashionable to speak in such terms. At times, it seems that, some contemporary economists wish to see no more in the notion of preference than the mere fact of consumer choice among bundles of goods. Yet unless one believes that behind choice lies an effort by individuals actually to achieve their goals, it is unclear why choice behavior even warrants the term preference. It certainly becomes problematic to determine how choice behavior could be a basis for deciding whether and to what extent individuals have received benefits, borne costs, or undergone compensation. What relevance would benefit-cost analysis have in normative policy making unless it was believed that preference and willingness to pay tended to reflect real gains or losses to the quality of individual lives?⁸

Perhaps the concept of well-being has been out of favor in economics circles because well-being, unlike behavior, is not publicly observable. Yet there is nothing inherently unscientific in positing the existence of unobservables—whether electrons or viruses or utilities—to organize and explain observation, as long as one is also responsive to evidence. Indeed, there is something especially odd about calling well-being "unobservable"; unlike an electron, it is something of which each person has had the most intimate experience. Furthermore, given the similarities among all human beings (e.g., they are all made of flesh and blood), it seems idle to imagine vast differences in the range and character of the inner lives of those whose environment and behavior are broadly similar. Indeed, this worry is really closer to metaphysical skepticism about other minds than to the methodological concerns of scientific psychology.⁹

corresponds to something less stipulative, I will gladly concede the point—at least so long as it is recognized that the objection arises with equal force against revealed preference.

Let us say, then, that the best case to be made on behalf of the relevance of benefit-cost analysis to normative social choice is that it provides systematic information concerning the extent to which a person's well-being will be affected by alternative policies. In principle, benefit-cost analysis is able to provide this sort of information because it attaches monetized values to the benefits and costs experienced by different individuals and therefore permits comparison and aggregation. In this respect, it resembles a Benthamic felicific calculus. The mechanism for the assignment and comparison of values differs from Bentham's, however, because the test of a potential Pareto improvement appears to involve no appeal to interpersonal comparisons of cardinal utility.¹⁰ Let us suppose, for example, that Mr. Smith could receive benefit B only if risk R were imposed on Mr. Jones. Yet suppose as well that Jones is indifferent regarding the choice between continuing the status quo and bearing additional risk R while receiving compensation $C; suppose, too, that Smith is indifferent regarding the choice between continuing the status quo and receiving benefit B at a cost to him of $D. If $D is greater than $C, imposing risk R to bring about benefit B passes the potential Pareto improvement test. Smith gains enough to compensate Jones—although it is not assumed that he actually does so—and still come out ahead.

It should be noted, however, that the choice behavior just mentioned was described in the language of indifference, benefit, cost, and compensation. Were one to discover that the behavior of Jones, which underlies the imputation of indifference, was the result of real or imagined intimidation by Smith or, alternatively, of a simple misunderstanding of what was being asked of him, then his behavior would not support an interpretation in such terms (indifference, compensation, etc.). Moreover, if these terms did not fit, it would be difficult to believe that the test revealed anything interesting about gains or losses in well-being. To take the test of potential Pareto improvement seriously for purposes of social choice, it is necessary to believe that the sorts of choice behavior on which the assessments are based can be plausibly interpreted as reflecting the choosers' views about how their well-being is likely to be affected.

Now, the real question of this section can be posed: Under what circumstances will a test of potential Pareto improvement reliably be informative about net effects on well-being? Obviously, it is necessary to ensure as far as possible that the parties to the choice behavior in question are neither in reality nor in their imagination being coerced and that they are not making simple mistakes of perception. Five other sources of possible inaccuracy may be of special relevance to the case of environmental risk assessment.

Diminishing Marginal Utility In The Intrapersonal Case

I have been suggesting that behind the use of choice behavior and the imputation of preferences lies the view that there really is such a thing as an individual's welfare and that, with experience, individuals tend to acquire some knowledge about their welfare and to act accordingly. One thing that can be said with considerable confidence about individual welfare is that most benefits exhibit diminishing marginal utility, a phenomenon people experience directly in their own lives.

Recognizing this phenomenon has an important effect on the interpretation of tests of potential Pareto improvement. Indicators of willingness to pay that are based on prices or related choice behavior will reflect the marginal utility of "the last unit." For example, an analyst might use the amount I am willing to pay as an entry fee into a wilderness area or for a fishing permit to measure the cost to me of converting some forest area to commercial use or of the environmental degradation caused by acid rain. This measure, however, will only indicate the marginal utility to me of "the last unit" of wilderness consumption, and the policy in question might have as one possible effect a dramatic alteration of the amount of wilderness or fishing available to me. Whenever such nonincremental changes are possible, the proper measure must take into account the higher marginal utility of the "first" unit of consumption, the "second" unit of consumption, and so on. That is, it is necessary to take consumer surplus into account.¹¹ The issue is not whether dramatic changes are the likely result of any given policy but whether they are a possible result. If so, then the proper way to calculate the expected value of nonincremental outcomes is to take the product of their probability and the sum of ''price" and any consumer surplus.

The discrepancy between this procedure and one based on a marginal measure can be quite large. One need only reflect upon the difference between the amount one would be willing to pay to achieve a 1 percent increase in physical mobility and the amount one would be willing to pay to avoid utter physical immobility—would the former be as much as one one-hundredth of the latter? Yet among the consequences of risky policies are not only marginal effects—the exposure of a population to an incremental amount of unwanted risk—but also, typically, the nonmarginal effect of actual injury to individuals. Here, too, the proper way to calculate the expected value of such nonmarginal outcomes—that is, to reflect the full difference between normal health and significant illness or disability—is to take the product of the probability of the various nonincremental harms and the sum of their "price" plus any consumer surplus.

Diminishing Marginal Utility In The Interpersonal Case

Diminishing marginal utility, I have suggested, is something each person experiences directly. It is, moreover, something each individual exhibits in his or her behavior through changing marginal rates of substitution. One might further say that, in the absence of compelling evidence to the contrary, it is reasonable to assume that people who have similar observable physical and social characteristics and who exhibit similar choice behavior will also be broadly similar in the amount of utility they derive from specific benefits. In particular, it is reasonable to assume that those who are otherwise similar but who stand far apart on a scale of income or wealth will derive different marginal utility from money. For example, certain tasks (e.g., some forms of dirty, difficult, nonself-directed labor) are almost universally disliked. The amount of such activity that can be obtained from individuals simply by holding out the inducement of, let us say, $10, is almost always less when their wealth is great and almost always diminishes as their wealth increases.

Returning to the potential Pareto improvement test as applied to Jones and Smith, let us suppose that Jones is very much like Smith in most respects but is considerably less affluent. For Jones, the marginal utility of a dollar can be expected to be significantly greater than it is for Smith. In the earlier example, Smith would pay $D to receive benefit B and Jones would accept compensation $C to run associated risk R. Yet it does not follow from the fact that $D is greater than $C that the gain to Smith is of sufficient magnitude to offset the loss to Jones because the dollars Smith would pay are paid at a lower marginal utility than the dollars Jones would receive. Of course, if amount $C were actually to be transferred from Smith to Jones as compensation, then no problem would arise because the marginal utility of these dollars would then be that of Jones. It is not part of the potential Pareto improvement test that such a transfer takes place, however. Because there is no actual compensation, the test involves what is in effect an interpersonal utility comparison of a $C-valued-by-Jones loss with a $D-valued-by-Smith gain. Owing to diminishing marginal utility, one would expect that, if such a comparison were based solely on the dollar amounts involved, the measure would be inaccurate in a systematically regressive way, exaggerating gains or losses to the more affluent in comparison with those to the less affluent.

Still, it might rightly be asked whether this type of comparison will lead to systematic inaccuracies in large-scale policy assessment Under certain conditions, inaccuracies will be reduced: when willingness to pay is assessed using fairly broadly based behavioral indicators (e.g., prices) to obtain average values; and when the benefits and burdens of large-scale policies are rather uniformly dispersed over varied populations. But when such conditions are not met, the test will tend to yield errors. This might not be worrisome if the net errors that remain within assessments of individual policies themselves varied randomly from the assessment of one policy to the next, so that within sequences of policies, the effects of assessment errors would tend to cancel each other out over time. I know of no general reason to expect that this sort of balancing will occur; on the other hand, there is a general argument—based upon diminishing marginal utility—showing a systematic tendency toward nonrandom, regressive error. As a result, it is difficult to believe that policy assessment can legitimately ignore questions about inaccuracies arising from effects of uneven distribution.¹²

For example, errors can creep in if one compares a willingness-to-pay indicator drawn from a population in which the marginal utility of money is relatively high (e.g., a risk premium in the wage of miners) with a willingness-to-pay indicator drawn from a population in which the marginal utility of money is relatively low (e.g., an aesthetic premium in the price of property in settings of unusual natural beauty).¹³ A second example involves errors that can arise in comparisons of the benefits of controlling job-site pollutants liable to produce nonincremental health effects in the small exposed population of producers with the cost of incrementally increasing prices to the large population of consumers. A third example combines the intrapersonal and interpersonal effects to magnify the possible error. In this case, a marginal willingness-to-pay indicator is used to measure an incremental cost to one relatively affluent population (e.g., an increase in the price of strawberries); this figure is then set alongside a marginal indicator used to measure a nonincremental harm to a less affluent population (e.g., an increase in disability among migrant laborers who work in strawberry fields in which a short-lived but toxic fungicide has been applied).¹⁴ In such cases, some and perhaps considerable adjustment of monetized values would seem to be required to improve the capacity of willingness-to-pay indicators to represent utilities.

However ordinalist its official theory, benefit-cost analysis cannot plausibly dismiss interpersonal comparison and diminishing marginal utility. Even the idea that a potential Pareto test offers a way of measuring increases in the total social pie requires the belief that, for example, Smith's $D-valued-by-Smith gain can be compared with Jones's $C-valued-by-Jones loss. Too much is known about what money, or any other good, means to an individual's well-being to convince anyone that this comparison can be settled simply by asking whether $D is greater than $C. Too much is known, that is, to be persuaded that an uncorrected potential Pareto improvement test captures as fully as possible the welfare effects of social choice.¹⁵

Preferences Involving Poor Information Or Other Cognitive Defects

To the extent that one is prepared to use market prices or choice behavior as indicators of willingness to pay and, ultimately, of welfare, one must believe that the people involved are rational, reasonably well informed, confronted with an appropriate array of options, and so on. Prices, for example, directly measure not willingness to pay but the tendency to pay, a mixture of willingness and real or imagined constraints, of knowledge and ignorance.

In choices in which outcomes (a) depend on complex causal sequences, (b) show great latency, or (c) involve options of markedly different salience, there is good reason to expect everyday choice behavior to embody serious informational and deliberative defects. These conditions notoriously obtain in many areas of choice behavior that are relevant to environmental policy making: labor markets for jobs involving risks, the market for insurance, the market for safety devices, and so on. The wages of coal miners tell something about risk premiums, but they also speak volumes about restricted options, historical bargaining positions, and the lack of information.¹⁶ If prices are to be used to estimate willingness to pay, one must either believe that these distortions are not significantly present in the particular markets of concern, or one must attempt to correct prices to reduce the influence of such distortions.

Price correction need not run counter to the fundamental idea underlying consumer sovereignty if the adjustments are based on efforts to create real-life choice situations in which agents are given good and usable information, a wide range of options within which fine discrimination is possible, and ample resources from which to choose. Experiments of this kind, which might bear some resemblance to the famous negative income tax experiments, could be carried out in representative populations on an appropriately large scale. The preferences expressed in such experiments could then be used to develop more general projections concerning how individuals' preferences would evolve were they well informed, well placed to choose, and so on. Consumer sovereignty could thus be upheld, although at the cost of expensive experimentation. It seems to me that, by calling for the funds necessary to underwrite experiments to improve the quality of the information yielded by their work, advocates of benefit-cost analysis could simultaneously exhibit their commitment to empiricism and help mitigate the widespread suspicion that they favor technocracy over autonomy.

Here, too, another worry about benefit-cost analysis might be addressed. It is well known that individuals give markedly different answers to questions about how much they would pay to avoid a certain increase in risk, as opposed to how much they would demand to be paid to bear an increased risk of the same magnitude. The latter sum tends to be larger than the former, often by several orders of magnitude. Various cognitive mechanisms have been proposed to explain this discrepancy. Pending a better theory of what is going on in such judgments, it seems to me inappropriate—and bound to raise suspicion—that benefit-cost analysis tends to seize on the smaller figure when measuring willingness to pay. No doubt benefit-cost analysts feel that the very large sums demanded for compensation do not reflect the sort of realistic valuation of money that comes into play when individuals are asked what they would be prepared to pay out of pocket. By the same token, however, the rather small sums individuals say they would pay may reflect a lack of realistic understanding of the magnitude of the risks involved (money paid out of pocket being much more salient than probabilities of harm). Accuracy in measurement might be served by working with both figures as something like upper and lower bounds rather than accepting the lower figures as more credible. Both figures seem likely to reflect some departure from cognitively ideal deliberation.

Finally, it is important to take indirect welfare effects into account whenever cognitive defects are present. Let us suppose that a fear based on poor information is widespread and that fuller public information will reduce but not dispel it. A policy that appears to be optimal when its returns are calculated on the basis of informed preferences may nonetheless cause considerable anxiety in suspicious or information-resistant segments of the population. This anxiety and the resultant individual and social dysfunctions are a real welfare cost of the policy and should not be dismissed as unworthy of attention. How much weight to place on unreasonable fears is a substantive matter in the policy arena. The job of benefit-cost analysis, by contrast, is to let policy makers know, as fully as possible, the likely welfare effects of acting in the presence of such fears. In making their assessment, benefit-cost analysts will ensure, to the extent possible, that they have the best available information about the actual harm individuals are likely to experience rather than the harm that uninformed individuals expect to undergo. At the same time, it is important to note that actual harm includes not only direct health effects but also social, psychological, and medical problems arising from (warranted or unwarranted) anxiety.

Preferences Not Related To Welfare

By and large, the preferences that make themselves felt in prices and other indicators of willingness to pay can be expected to reflect individuals' explicit or implicit pursuit of their own well-being. Yet there are various areas in which strongly held values that are largely independent of personal well-being also influence choice behavior.¹⁷ Nowhere is this clearer than in environmental issues, an area in which concerns about posterity, about the kind of society Americans want and the sorts of things they think should be done to bring it about, inextricably enter into people's preferences. This problem may be especially large when political behavior—such as support for various sorts of environmental policies—is used evidence for a willingness to pay.

Typically, as in the case of cognitive defects, there will be indirect welfare effects of such preferences. For example, if I very much want a wildlife area to be protected, then I prefer not to hunt there, and I also prefer that no one else hunt there. Should I learn that people are hunting there, I would be disturbed in a way that affects my well-being. Such indirect welfare effects are a genuine part of any calculation of social welfare, even though their basis lies in other-directed preferences. This approach is to be distinguished from one that gives other-directed preferences a direct role in utility assessment (because arguably, apart from my agitation on learning of hunting in the preserve, my welfare is not affected adversely by the mere fact that hunting is occurring).

If such indirect welfare effects were always of the same magnitude as direct effects on preference fulfillment, it would be possible to ignore the question of whether a given willingness-to-pay indicator reflects self-or other-oriented preferences. It is impossible to believe that these effects will be that similar, however. For example, my willingness to pay for some benefit may reflect preferences I have about how things will go after my death. Once I am gone, however, the extent to which this preference is satisfied can be affected although my welfare cannot.

If the goal in such cases is to assess gains and losses in individual welfare, analysts will be misled to some degree by looking at individual preferences. And yet such a widened scope of interest does not create a fundamental tension with the doctrine of consumer sovereignty. The basis of the doctrine of consumer sovereignty is the individual's reliability to judge his or her own well-being; in the above instance, preferences regarding other matters are being considered. Thus, some adjustment in the interpretation of choice behavior may be called for whenever other-directed preferences can be expected to play a large role.¹⁸ This approach does not at all preclude the measurement of indirect welfare effects, although it requires some attention to the sorts of preferences that are consulted when assessing such effects. Careful analysis would avoid amalgamating other-directed and personal-welfare-based preferences.

The Absence Of Appropriate Markets

To believe that benefit-cost analysis can achieve a reasonable degree of accuracy in its information-yielding role, one must believe that the process is able to contend with those benefits that are important to welfare but that have no existing appropriate markets. The welfare of individuals is much affected by many features of the world—for example, the aesthetics of daily life, the degree of community enjoyed—for which it is difficult to produce direct willingness-to-pay measures.

Two approaches come to mind. First, an analyst might give quantitative measures only for those benefits for which an appropriate market exists and note explicitly the incompleteness of the analysis. One way to accomplish this is illustrated by Porter's benefit-cost analysis of mandatory deposits on beverage containers (Porter, 1978). A major benefit that could result from a container deposit law would be a reduction in litter, with attendant aesthetic gains, an increase in community pride, and so on. How is one to give a monetized value to such benefits, however? Porter wisely does not try; instead, he carries out a sensitivity analysis that indicates a range of willingness-to-pay values, within which one's willingness to pay for these difficult-to-quantify goods would have to fall in order to outweigh the estimated costs of the deposit system.

A second approach attempts to find or create some proxy market or other indirect way of assigning a quantitative measure. Per capita government expenditures on litter removal, for example, might be considered as an indicator of willingness to pay. The danger of this approach is its arbitrariness; the danger of the first approach is that nonquantifiable benefits may simply drop out of sight in favor of those that can be more readily measured.

One advantage of the second approach is that it may hold "soft" (difficult-to-quantify) benefits before the public eye and even stimulate debate over their magnitude and how best to measure them. In any event, the second approach will be needed even for the "hard" benefits that are segregated out by the first approach because actual markets involve externalities and therefore their prices may need adjustment The problem of externalities, it need hardly be said, is particularly acute in the environmental case. For example, Porter measures disposal costs for solid wastes simply by looking at what cities were paying per ton in the 1970s to have refuse hauled to the landfill and buried. A fuller accounting, however, would also consider what is now painfully evident: the limited space available for the purpose of solid waste disposal and the long-term costs of filling that space with nondegrading bottles and cans. Moreover, Porter's analysis looks at energy consumption in container production versus consumption in container recycling (because this measure is fairly readily gauged), but it ignores the resulting levels of pollution from production rather than recycling, the comparative environmental degradation in obtaining raw materials, and so on. These external factors are difficult to quantify and in any given case may be of little significance. The point is that they are easily overlooked even in very careful analyses, so that it is potentially misleading to summarize the results of such analyses as one-dimensional acceptance and rejection zones for policies. A legislator may believe he is looking at hard facts when he faces a $27 per year per citizen value of litter reduction necessary to make the deposit law a net benefit. In reality, he is looking at the easy facts—the ones most amenable to quantification. The reliability of an analysis should not be measured by how much its monetized values rely directly on market prices because the prices themselves may not reflect the true costs of a specific benefit.

A third approach may help remedy some of the problems of the first two: environmental policy might deliberately be used to reshape the boundaries of markets and to internalize external costs. This approach might involve a fairly elaborate governmental role in structuring some markets, but it would permit the use of market mechanisms to determine prices. Governmental intervention, then, is not always the enemy of a free market approach.

Within the realm of individual welfare, quantification of certain central benefits—for example, the aesthetic or social qualities of the environment within which one lives or works—may always elude analysts, even in cases in which market mechanisms have been modified to include externalities. Either the policy analyst must believe that these difficult-to-quantify contributors to welfare can be ignored—perhaps because they tend to cancel one another out, although this would be hard to believe in general—or he or she must be committed in advance to building many caveats into the final benefit-cost assessments and to bringing many more qualms and considerations directly to the attention of decision makers.

The five sources of possible inaccuracy that have been discussed—diminishing marginal utility in the intrapersonal case, diminishing marginal utility in the interpersonal case, preferences involving poor information or other cognitive defects, preferences not related to welfare, and the absence of appropriate markets—are all subject to some degree of correction, and this fact suggests something about how benefit-cost analysis might be improved as a measure of individual welfare. Of course, to take these difficulties seriously, it is necessary to take the notion of individual welfare seriously.

Economists sometimes air rather skeptical views about the concept of well-being, and one evident effect of such skepticism is to promote impatience with efforts to look closely at how adequate one measure of utility might be in comparison with another. I do not think that methodological or philosophical considerations require such skepticism, and as far as I can see the best case for defending the normative relevance of the enterprise of benefit-cost analysis depends in the end on rejecting it. It may be a deep-seated fact rather than a mere heuristic that microeconomics texts tend to begin with cardinal utility theory and to retain the vocabulary of utility even in the brave new world of ordinalism.

It would be possible, given the desire, for the government greatly to reduce the mystery about utility by greatly increasing the support given to experiments aimed at providing better information about well-being, as perceived through the lens of choice. Moreover, it would be possible, given the desire, for the government greatly to increase the support given to basic social and psychological research and to research into the toxicity or carcinogenicity of various chemicals, the behavior of substances in the atmosphere or the soil, the medical consequences of social dislocation, and so on. The United States justifies a vast military budget by pointing to the dangerous possibility of being caught unprepared. Yet even as I write, this nation is in danger of being caught unprepared in the realm of environmental policy making. At bottom the problem may not be the difficulty of doing good research on relating utility to willingness to pay but rather an unwillingness to pay for research with considerable utility. No doubt, this unwillingness is another example of a cognitive defect.

Measuring Costs And Benefits Over Time

What does one have to believe to discount future costs and benefits? One concern that must be addressed if preferences are to be corrected for informational or cognitive defects is the fear that such correction will in fact result in a particular portion of the total population—in this case, technocrats—substituting their preferences for those of the population at large. Yet the very same sort of concern arises from certain uses of intertemporal discounting. For example, the use of intertemporal discounting is sometimes justified by pointing to a "social time preference"—people, it is said, prefer to have a given benefit occur sooner rather than later. (This phenomenon is not entirely uncontroversial, either as a descriptive¹⁹ or as a normative matter, but let us assume it for now.) People alive today, then, will prefer benefits in the present. By symmetry, people of the future will prefer benefits in the future, which is their present. Thus, if existing tune preferences are used to discount the benefits and costs of a social policy with long-term consequences, benefit-cost analysis will, in effect, be assessing the overall value of a policy by substituting the preferences of a particular portion of the affected population—those who are alive in the present—for the preferences of the population at large.

The above is not in itself a complaint of intertemporal injustice, although it could easily become the basis for such a complaint. Instead, the complaint is, in the first instance, one of mismeasurement, of failing to take into account the preferences of many of those on whom the consequences of the policy will fall. It is as if the costs and benefits of a national ambient air standard could be assessed only by consulting the preferences of those in Kansas. Indeed, the mismeasurement is worse than that bemuse the time preferences of those alive in the present are not—unlike, say, those living in Kansas—even approximately representative of the time preferences of all those affected by the policy. People tend to show a fairly strong preference that benefits be received within their lifetimes, with costs postponed until sometime in the future. The comparison therefore should be with measuring the value of a national standard on acid rain by consulting the preferences only of those in the Ohio Valley.

Still, it might be argued that this comparison is misleading. If persons who are alive today are saving at an optimum rate, as they should be in an ideal market, then the welfare of future generations will be assured through economic growth. This situation is quite different from one in which people in the Ohio Valley benefit from cheap power while inflicting net losses on those downwind of them. Thus, the interests of future generations are not really being left out of account if people act on existing time preferences.

I would like to draw attention to two features of this rationale for following current time preferences.

When Saving Is Optimal

First, this rationale does not involve the discounting of future utilities. Under conditions of optimal saving, if people who are alive today were to ignore their own time preferences and force themselves to invest more and consume less, the total amount of consumption that everyone will enjoy over time would be reduced. This efficiency-based argument is temporally neutral; it functions to show that there is no real inconsistency in practice between, on the one hand, making decisions about consumption versus investment based on social time preference, and, on the other, counting all welfare effects equally, whenever they occur.

What role would discounting play in such a scheme?²⁰ Because future generations are assumed to be more prosperous than those alive today, they will be willing to pay more (in constant dollars) to receive a given benefit or avoid a given harm. Yet these higher, time-indexed, willingness-to-pay indicators will not reflect comparably large gains in utility, owing to the diminishing marginal utility of money. Discounting must be used to telescope these higher figures in a way that permits comparability with present willingness-to-pay indicators. This approach avoids the overestimation of future costs and benefits, which could result in mismeasurement of the relative value of competing policies that differ in terms of when their costs and benefits will obtain. From the standpoint of the present, a future monetized measure can be discounted to solve a problem of intertemporal comparison and obtain an accurate portrait of effects on utility over time. Once again, it is apparent that no discounting of future utilities—as opposed to prices—is involved.

Such discounting of monetized measures for the purpose of intertemporal comparison should presumably take as its rate not the real rate of economic growth but rather the rate at which the marginal utility of money is declining as such growth occurs. This rate can be estimated only roughly and, owing to disparities of wealth, will represent only an average within a population. Presumably, however, the upper bound is the real rate of economic growth because even at high finite levels of wealth, additional dollars will continue to have some positive utility. If growth is taking place and wealth differentials are not increasing, the lower bound will be zero. Thus, when discounting is used for the sake of intertemporal comparison, the usual rates employed in benefit-cost analysis, which often exceed the real rate of growth of the economy, are much too high and would be expected to yield systematic undermeasurement of future gains and losses.

Discounting for purposes of intertemporal comparison only makes sense if one is operating with time-indexed willingness-to-pay indicators. A similar effect could be achieved by using present willingness-to-pay indicators for future costs and benefits and then not discounting.²¹ Disaster occurs, however, when one mixes these two strategies and uses present willingness-to-pay indicators for future costs and benefits and then also applies discounting. This approach produces the astonishing result that, after a few years, the outcomes of policies, no matter how harmful or beneficial, have almost no measured welfare effect. Such a result is hardly credible, for it would imply that it will not matter to people in the not-too-distant future how life goes for them.

It is possible that legislators or administrators who ask for benefit-cost analysis want nothing more than a measure of effects on the welfare of people alive today. It is also possible that, if benefit-cost analysis is performed in a climate of optimal growth, no real underrepresentation of future utility will result. Still, the function of benefit-cost analysis, as understood here, is to give the fullest possible accounting of welfare effects. Such an accounting might simply confirm the view that, in practice, acting on existing time preferences will not lead to policy choices that would in effect underrepresent future utility. But one would want to see this confirmation actually carried out.

This point brings me to the second feature of the rationale for following existing-time preferences that I would like to discuss.

When Saving Is Not Optimal

Currently, there is a vigorous debate about optimal saving, and I am in no position to judge it. Yet, it does seem difficult to find people who believe that, once all externalities have been taken into account, present rates of saving are optimal. Indeed, reflecting on the environmental problems of toxic wastes, atmospheric pollution and warming, nonrenewable resources, ozone layer depletion, and so forth, raises the real possibility that future generations will be worse off than today's. It is simply not clear that the economic growth made possible by rapid exploitation of the environment will be great enough to enable future societies to make up the losses connected with resource exhaustion and environmental degradation. To effect widespread ecological damage is quickly and rather cheaply done, but even the richest of societies do not appear to be able to afford to repair this damage fully without severe economic burdens.

One might (were it not an understatement of the difficulty) invoke what I am tempted to call the law of styrofoam pellets, a special case of the second law of thermodynamics: on a windy day it is the work of a moment to send styrofoam lentils all over hill and dale—-I need only open a packing box and wave it aloft. To reassemble even nine tenths of the pellets in the original box, however, would give someone a week's work.

The wholesome and optimal expansion of the economy cannot be taken for granted. Without optimal saving and growth, the realm of operation becomes that of the second best, where one can no longer be certain that the effort to approximate an efficient solution in one area brings overall efficiency closer. So, most bets are off. In particular, because it is no longer safe to assume that people will be moving forever upward along curves of increasing utility, decision makers cannot ignore the problem of trying to gauge future utilities accurately in order to make comparative assessments of policies. To follow existing time preferences and employ a blanket discount function in the world of the second best is thus to adopt an attitude that carries a considerable possibility of underrepresenting actual welfare effects.

Competing environmental policies may differ greatly in regard to when they impose costs and how they affect the relative scarcity of certain goods. Although for practical purposes it may be necessary to limit how far into the future costs and benefits are traced, some remote consequences of present activity can even now be foreseen.²² I recall as a teenager being disturbed in a way I could not quite express by a photo in the middle of an encyclopedia article on radioactivity. It showed a sailor pushing concrete blocks off the stern of a ship, and each block was full of highly radioactive waste. ''The concrete blocks are expected to last 100 years," said the caption cheerily. Elsewhere in the same article, however, was the information that many of the radioactive isotopes present in such blocks have half-lives of thousands or millions of years. Policy makers can choose to make more or less sturdy concrete blocks, to make more or less expensive storage facilities, or to alter behavior so as to create more or less of the waste for disposal. The alternatives presented in these choices may mean cheap waste disposal in the present but an expensive or even irreversible problem in the future, or expensive waste disposal in the present and a cheaper and more reversible problem in the future. Benefit-cost analysis can help in determining the long-run expected value of such choices—but not by engaging in blanket discounting in a second-best world.

To be sure, discounting will be of use even so. One must, for example, be sensitive to the opportunity costs of capital, to worries about displacing investment, and to the compounding effects of taking benefits earlier rather than later. Yet in all such cases, future costs and benefits must be measured without actually discounting future utilities. Certain goods tend by natural (or "natural") mechanisms to multiply over time—rabbits and capital, for example. Consequently, if one wants to know the quantity of such goods that are needed on hand now in order to have some particular quantity of them on hand at a given time in the future, something like a present-value calculation based on the rate of increase can be carried out. This calculation, however, will not reveal how much of this good should be on hand at any particular future time, given present resources—that is a question one must try to answer by asking what will be needed or wanted, and how badly, at that time. This latter question supposes an evaluative stance of temporal neutrality, reflecting commitment to a full accounting of welfare effects. Such neutrality is consistent with discounting in regard to decisions about how best to deploy the "naturally increasing" instrumentalities of welfare.

This mention of natural increase through productivity gains brings the discussion to another wrong turn that may be taken in calculating opportunity costs. Practitioners of benefit-cost analysis have rightly rejected the idea that the value of a life is measured by its contribution to economic production and instead have developed measures based on willingness to pay. Such measures yield values that are seemingly more in line with intuitive views about the relative importance of life. Nevertheless, even though a particular sum—for example, $2 million—is set as the monetized value represented by a life, this value cannot be treated as if it were a $2 million investment Suppose decision makers are asked to choose between two policies, one of which emphasizes saving lives in the present while the other emphasizes achieving some other benefit 10 years in the future. If the monetized value of a life were a monetary asset, it should be possible to calculate the opportunity cost of losing a life as opposed to saving it by considering the return on $2 million. This kind of approach however, would be misleading about the effects of saving a life in comparison with other benefits. Monetization is a potentially useful way of valuing various benefits for the purpose of comparison; it is not, however, a device for turning all benefits into interest-beating assets that carry the present value of their monetized measure. It would be unfortunate if the hard-headed practicality behind monetization were to yield to mystification just at the point when its proper goal has been accomplished.

In sum, in this second-best world, decision makers cannot believe that they are entitled to carry out blanket discounting of future costs and benefits on the basis of an assumption that the welfare of future generations is being adequately protected by the underlying productivity of the economy. Policy assessment must endeavor to capture the actual expected long-term costs and benefits of today's choices as accurately as possible, and to weight them, using the best available estimates of the probabilities of outcomes rather than some uniform temporal discount. If measurement is to be consistent and accurate, it must use the same criteria for people alive in the future and their well-being that are used for those alive today. If consumers are sovereign, then they must be sovereign, whenever they come into being. One of the rationales for benefit-cost analysis is market failure; without optimal savings and growth, there can be no clearer example of market failure than the failure of the preferences of future generations to enter into the determination of prices today.

By providing a means by which to represent future preferences in present calculations, benefit-cost analysis can allow decision makers to confront the actual effects their choice will have upon welfare in the future.

Thus, the use of benefit-cost analysis not only is a matter of accurate measurement but could also involve the promotion of intergenerational justice by giving voice to temporally absent preferences. Benefit-cost analysts committed to a full accounting could become, in effect, representatives of otherwise disenfranchised future generations. It remains to be seen whether practitioners of benefit-cost analysis will seize on this—perhaps unexpected—opportunity to plant their analytic banner on the moral high ground.

Acknowledgments

I would like to thank Allan Gibbard, Edward Green, Roger Noll, Robert Solow, and Hal Varian for helpful criticisms of an earlier draft of this paper. I am especially grateful to Hal Varian for suggesting useful readings and for helping me to see how to frame some of the issues I wish to raise in terms that might be intelligible to economists. If I have failed in this task, it is surely not his fault. I am also indebted to Douglas MacLean for numerous helpful discussions and for bringing to my attention Page (1977) and appendix F of Parfit (1984), both of which I found most useful in writing this paper.

References

Ben-David, Shaul, Allen V. Kneese, and William D. Schulze 1979A Study of the Ethical Foundations of Cost-Benefit Analysis Techniques. Working paper. Washington, D.C.: Resources for the Future.
Conservation Foundation 1987. Risk Management Case Study. Unpublished manuscript. Washington, D.C.
Leonard, Herman B., and Richard J. Zeckhauser 1986. Coat-benefit analysis applied to risks: Its philosophy and legitimacy. Pp. 31-48 in Douglas MacLean, editor. , ed., Values at Risk. Totowa, N.J.: Rowman and Allanheld.
Lind, Robert C. 1982. Introduction. Pp. 1-94 in Robert C. Lind, editor. et al., Discounting for Time and Risk in Energy Policy. Washington, D.C.: Resources for the Future.
Nozick, Robert 1974. Anarchy, State, and Utopia. New York: Basic Books.
Page, Talbot 1977. Conservation and Economic Efficiency. Baltimore, Md.: Johns Hopkins University Press.
Parfit, Derek 1984. Reasons and Persons. Oxford: Clarendon Press.
Porter, Richard C. 1978. A social cost-benefit analysis of mandatory deposits on beverage containers. Journal of Environmental Economics and Management 5:351-375.
Railton, Peter 1986. a Facts and values. Philosophical Topics 14:5-31.
1986. b Locke, stock, and peril: Natural property rights, pollution, and risk. Pp. 89-123 in Mary Gibson, editor. , ed., To Breathe Freely: Risk, Consent, and Air. Totowa, N.J.: Rowman and Allanheld.
Rawls, John 1971. A Theory of Justice. Cambridge, Mass.: Harvard University Press.
Sen, Amartya 1985. Well-being, agency, and freedom: The Dewey Lectures 1984. The Journal of Philosophy 82:169-221.

Footnotes

1: Such conceptions include not only hypothetical social contract theory of the Rawlsian variety (see Rawls, 1971) but also classical utilitarianism, which, owing to diminishing marginal utility, is also distribution sensitive. This question is discussed in the second section of this paper.
2: Some approaches (e.g., Ben-David, Kneese, and Schulze, 1979) have sought to capture these competing conceptions of justice within something like a benefit-cost framework. These attempts come to grief, however, in their inability to achieve an adequate representation of a fundamentally deontological theory of justice. The agent-centered constraints of such theories may simply lack a non-agent-centered, or global, value-theoretic expression.
3: Even a utilitarian might justifiably believe that the proper role for benefit-cost analysis is in providing information rather than in decision making because a great many factors influencing direct or indirect utilitarian assessment are not well captured by the specific analytic techniques of benefit-cost analysis.
4: An example of this approach is eminent domain. It prevents individuals from exercising veto power over public projects or extracting exorbitant sums in exchange for giving their consent, and it involves compensation by some objective standard.
5: Nozick (1974) considered adding such a gambit to Lockean theory, although he also showed some preference for using a system of tort law to carry out compensation. For further discussion of risk and Lockean theory—with or without the gambit—see Railton (1986b).
6: I should perhaps emphasize that this claim does not depend on rejecting a utilitarian analysis of rights. It is an open question whether the system of rights considered optimal by a utilitarian would permit various sorts of balancing in public policy making. The optimal system of rights in the long run might be one that imposed some fairly strict limits on balancing.
7: In their defense of benefit-cost analysis, Leonard and Zeckhauser (1986:35) put forward "the basic tenet that people know what is good for them," based on "the assumption that agents are well informed." For a philosophical defense of a conception of individual good based on something similar to informed preference, see Railton (1986a).
8: The question of factors other than individual welfare that influence preferences (e.g., social values, posthumous goals) is considered later in this paper.
9: These remarks suggest a reply to die-hard positivists. Even classical behaviorism has availed itself of the notions of positive and negative reinforcement, aversion, satiation, expectation, and so forth, by giving them operational definitions. At least one strain of revealed preference theory seems to treat preference in a similarly operationalist way. Questions about whether a given choice really reveals an individual's preferences are blocked by pointing out that this definition simply follows from the stipulation of what "revealed preference" is to mean. Is there any reason, then, why the positivist should not consider operationalizing the notion of cardinal utility by tying it to specific patterns of behavior—perhaps along with physiological evidence concerning, say, galvanic skin response and pupil dilation? If the objection is raised that it is difficult to see the interest, for purposes of social choice, in working with such a concept of utility unless it
10: Whether, in a sense, the test actually involves a form of interpersonal comparison is discussed later.
11: I am grateful to Hal Varian for pointing out to me that the intrapersonal phenomenon I wish to describe here can be expressed using the economists' notion of consumer surplus.
12: One possible source of balancing is that certain benefits may actually have higher utility for the more affluent, because the enjoyment of such benefits requires the possession of other resources. More benefit from an Alaskan wilderness area may accrue to the more affluent because they, unlike others, are able to get there. It is difficult to believe, however, that this sort of balancing will be sufficient to offset regressive effects in the general run of environmental and health-related cases to which benefit-cost analysis is applied.
13: A much more important example of this phenomenon, and one with systematic import, concerns future generations—under the assumption that they will be substantially better off than people who are alive today. This question is discussed in the third part of this paper.
14: This example is not fanciful. It reflects procedures used in an unpublished risk management case study (Conservation Foundation, 1987) that is meant to be fairly sophisticated methodologically. I say "reflects" because the exact bases for the mortality and morbidity values this study uses are not stated, although the values seem consistent with familiar marginal willingness-to-pay indicators.
15: Corrections to benefit-cost measurement discussed in this section involve attempts to capture welfare and not the problem of capturing other values (e.g., process or equity). The concern with diminishing marginal utility, for example, reflects the need to remove certain distortions of the potential Pareto improvement test or willingness-to-pay indicators. As it happens, these distortions have something like regressive distributive implications. The argument for correcting them, however, does not depend on any assumption favoring the incorporation of some measure of distributive justice into benefit-cost analysis; rather, it depends only on two assumptions: that what is wanted is a measure of welfare effects that is as accurate as possible, and that marginal utility declines.
16: The wages of coal miners may also reveal something about social norms—for example, ideas about what one ought to do to live up to social expectations and to provide for one's family, what deference one owes to those with higher social standing, or what activities are socially valued. Such considerations may in any given case have a significant effect on willingness to pay, although they will greatly complicate the interpretation of choice behavior as a straightforward indication of individual evaluation of risk. A policeman or fisherman may demand less in compensation for a job-related risk that is deemed socially expected or gender appropriate, or that is connected with a socially valued activity, than he would demand for a risk with none of these features (e.g., passive exposure to an environmental toxin). For further discussion of preferences and values, see the next section.
17: Amartya Sen (1985) especially has drawn attention to such phenomena.
18: It may not seem quite so philistine to endeavor to exclude other-directed preferences from direct consideration in benefit-cost analysis once one recognizes the limited, "welfarist" purpose of benefit-cost analysis and once one notices that the values that other-directed preferences reflect quite properly play a significant role in actual policy choice. However, the charge of philistinism may stick to anyone who insists upon treating benefit-cost analysis as the decisive normative criterion of social choice.
19: For example, people continue to set aside savings even when, as has often been the case in recent decades, the real after-tax rate of interest on savings is indistinguishable from zero (Lind, 1982:84). However, since people save for various reasons, one should be somewhat skeptical of using saving behavior too directly as a measure of time preference.
20: I am indebted here to Allan Gibbard's comments on the original version of this paper.
21: The result may not be quite the same, however, bemuse it is possible that, on average, members of future generations, owing to their higher standard of living, will receive more utility than members of generations alive today from, for example, a year of life.
22: Actually, such limits may be necessary for theoretical purposes as well. In general, an infinite time horizon makes comparisons of costs and benefits impossible—both will simply grow indefinitely. Still, a distant but finite time horizon, along with some further constraints about foreseeable effects beyond that horizon (e.g., concerning something like the net rate of utility growth) would do a much better job of representing future utilities than straight discounting at any significant rate.
: Peter Railton is associate professor in the Department of Philosophy at the University of Michigan.

Bookshelf ID: NBK235531

Contents

< Prev Next >

PubReader
Print View
Cite this Page
National Research Council (US) Steering Committee on Valuing Health Risks, Costs, and Benefits for Environmental Decisions; Hammond PB, Coppock R, editors. Valuing Health Risks, Costs, and Benefits for Environmental Decision Making: Report of a Conference. Washington (DC): National Academies Press (US); 1990. 4, Benefit-Cost Analysis As A Source Of Information About Welfare.
PDF version of this title (1.7M)

In this Page

Scope And Comprehensiveness
Measuring Costs And Benefits At Particular Points In Time
Measuring Costs And Benefits Over Time
Acknowledgments
References

Recent Activity

Clear Turn Off Turn On

Benefit-Cost Analysis As A Source Of Information About Welfare - Valuing Health ...
Benefit-Cost Analysis As A Source Of Information About Welfare - Valuing Health Risks, Costs, and Benefits for Environmental Decision Making

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

Bookshelf