ewi003 57..75 Ex-Post Egalitarianism and Legal Justice Alon Harel Hebrew University of Jerusalem Zvi Safra Tel Aviv University Uzi Segal Boston College In any legal system, one finds numerous rules, practices, and constitutional pro- visions that are incompatible with utilitarian considerations. It is not merely util- itarianism that fails to explain a diverse range of rules and practices. Other theories that, like utilitarianism, involve ex ante considerations cannot explain them as well. There are two possible primary explanations for the prevalence of these nonutilitarian rules and practices: Kantian (deontological) explanations and a view we label ex post egalitarianism, which requires that the state decides on its action in an egalitarian manner ex post. Our approach allows for compar- isons among different societies by giving meaning to statements like ‘‘Society A is more egalitarian than society B.’’ Furthermore, we show that the more egal- itarian societies should also employ less extreme criminal law rules and should be more sensitive to various kinds of injustice, whether it is caused by individual wrongful behavior or by criminal law rules. 1. Introduction In any legal system, one finds numerous rules and practices as well as consti- tutional provisions that are incompatible with utilitarian considerations (i.e., with maximizing the sum of individual utilities). These rules and practices often grant benefits to an individual whose well-being is at risk; yet the costs of these benefits, in terms of utilities, to other individuals may outweigh these benefits. Thus, for instance, despite the persistent belief of economists that efficiency requires to impose the harshest possible sanctions, legal systems often impose light sanctions and consequently have to bear the high costs of increasing the probability of detection. It is not merely utilitarianism that fails to explain a diverse range of rules and practices. Other theories that, like utilitarianism, involve ex ante We thank Ariel Rubinstein for helpful discussions. Zvi Safra thanks the Israel Institute of Busi- ness research for its financial support. Uzi Segal thanks the National Science Foundation (grant 0111541) for financial aid. The Journal of Law, Economics, & Organization, Vol. 21, No. 1, doi:10.1093/jleo/ewi003 � The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oupjournals.org JLEO, V21 N1 57 considerations cannot explain them as well. An ex ante analysis ranks distribu- tionsofexanteutilitiesbeforepeopleknowtheactualoutcometheyreceive. 1 Ex post analysis, on the other hand, relates to the final distribution of utilities. This analysis, too, is done before information concerning the distribution of final out- comesisrevealed,butrelatestotheevaluationofdistributionsoffinaloutcomes. There are two possible primary explanations for the prevalence of these non- utilitarian rules and practices: Kantian (deontological) explanations and a view we label ex post egalitarianism. Kantian explanations are based on the con- viction that there are nonconsequentialist obligations. The state is sometimes obliged to act or not to act in certain ways even if acting differently is con- ducive to utility, or to the ex ante interests of individuals. Typically those explanations rely on autonomy-based considerations and are based on the belief that respecting the dignity of individuals constrains the promotion of their well-being. For contemporary sophisticated deontological explanations of various rules and practices, see Kamm (1996:143–204). Ex post egalitarianism, on the other hand, requires that the State decide on its actioninanegalitarianmannerexpost.ItdiffersfromtheKantianapproachsince it is founded on a comparison between the well-being of the people who are affected by a certain rule and practice rather than presupposes the existence of moral constraints, which are independent of consequentialist considerations. It differs from utilitarianism since it is concerned with equality, and not only with the maximization of the sum of utilities. Finally, it differs from ex ante analysis, as itconsidersthe well-being of individuals after they have the knowledge of the prevailing social outcomes. The purpose of this article is to demonstrate that some legal rules and practices that are traditionally justified in terms of Kantian explanations can alternatively be explained in terms of ex post egalitarianism. Two impor- tant qualifications should be stated. First, we do not reject Kantian consider- ations. It is not claimed that the rules and practices examined here cannot be explained on the basis of Kantian considerations. Instead, we simply claim that they can also be explained in terms of ex post egalitarianism. Second, it is not claimed here that ex post egalitarianism is an absolute value, which overrides any conflicting values. The rules and practices examined here demonstrate that ex post egalitarianism competes with other considerations such as ex ante anal- ysis and utilitarian considerations. To demonstrate our approach, consider the issue of sentencing practices. As was argued by Becker (1968), sentencing practices pose a great challenge for the utilitarian approach: If criminals react to the expected sanction, then greater deterrence can be achieved by either increasing the probability of de- tection or by increasing the size of the sanction. However, increasing the prob- ability of detection is much more costly than increasing the size of the sanction. Hence the most efficient way to deter criminals is to impose the harshest 1. Note that when we use the term ‘‘ex ante’’ we do not discuss the extreme situation of ex ante decision process—the situation in which the decision is made behind the ‘‘veil of ignorance,’’ before the individual is even aware of his personal characteristics. 58 The Journal of Law, Economics, & Organization, V21 N1 sanction possible and to reduce accordingly the probability of detection. In reality, however, sanctions do not conform to Becker’s recommendations and they are usually proportional to the degree of culpability and wrongfulness of the crimes committed. Numerous explanations were provided to the puzzle raised by Becker. First, if criminals are risk averse, an increase in the sanction is not a costless transfer payment. Increasing the sanction and reducing the probability increases the risk faced by individuals. Second, the system proposed by Becker eliminates marginal deterrence—the incentive to substitute less for more serious crimes. We suggest below the following explanation. Sanctions generate inequality because those who are subjected to the sanctions are worse off relative to those who are not. The harsher the sanction is, the larger is the inequality between two classes of individuals, namely those offenders who are subjected to the sanction and nonoffenders who are subjected to the sanction by mistake, and offenders and nonoffenders who are not subjected to the sanction. Harsh sanctions are indeed, as Becker observed, required by consideration of aggre- gate utilities. Yet social interests in harsh sanctions must be balanced against social concerns for equality. Therefore it may well be the case that what dic- tates limits on the size of sanctions is ex post egalitarianism and not retributive justice considerations. Our approach allows for comparisons among different societies by giving meaning to statements like ‘‘Society A is more egalitarian than society B.’’ Furthermore, we show that the more egalitarian societies should also employ less extreme criminal law rules and should be more sensitive to various kinds of injustice, whether it is caused by individual wrongful behavior or by the society’s criminal law rules. As we elaborate in Section 8, such societies would use less harsh sentencing practices; furthermore, they would try to avoid wrong- ful convictions by raising the required burden of proof and, at the same time, they would try not to sacrifice minorities for the sake of the whole society. In Section 2 we present social policies and discuss the ex post and ex ante approaches. We define the notion of being an ex post egalitarian society and provide a way for comparing levels of egalitarianism among societies. In Sec- tion 3 we present our results, and we present our results and offer some appli- cations in Section 4. Formal analysis is deferred to the appendix. 2. Social Policies and Their Evaluation Consider a situation where individuals are facing some uncertainty regarding outcomes that are controlled by society, for example, road accidents or some health-related issues. There are many individuals and we assume that the risks they face are independent of each other. If society is sufficiently large, we can assume, for practical measures, that the proportion of those who receive a cer- tain outcome equals the probability of receiving this outcome. For example, if each member of society is facing an independent chance of 2% to fall victim to a car theft, then in a large society the probability that the true proportion of victims will differ from 2% by more than e is negligible. Social policies are Ex-Post Egalitarianism 59 often aimed at controlling these proportions, but they rarely try (or are even able) to determine the recipients of each outcome. For example, lower speed limits or wider road shoulders reduce the probability of fatal car accidents, but they do not tell who will be involved in accidents. There are two extreme ways in which society can view such issues: ex ante and ex post. Suppose that society is facing a choice between two economic policies. One implies an annual increase of 5% in everybody’s utility, the other provides each person with an independent risk, where there is a 50% chance of a 10% increase and a 50% chance of no change in utility. From an ex ante per- spective, both policies are equally attractive, as all expected utility maximizers are indifferent between the two. But the two policies may differ from an ex post perspective. The first implies a 5% increase in the utility of all. As individual risks are independent, the second policy implies (for a sufficiently large so- ciety) a 10% increase in the utility of half of the population and no change in the utility of the rest. If initially everybody’s utility is 100, now half of the population will have 110 and half will get only 100. Ex post, the two pol- icies are not the same and a nonutilitarian society may not be indifferent between them. Which is the correct analysis, the ex ante or the ex post? Consider another situation, where one unit of an indivisible good needs to be given to one of two individuals. Both receive utility one from receiving it and zero otherwise. All allocation procedures lead to an uneven allocation of ex post utility, one person receives one, the other zero. Diamond (1967:765) suggested randomization over the two individuals as a tool for improving social well-being, thus cre- ating ex ante egalitarianism. There is an extensive discussion in the literature about whether social welfare is really improved by such randomizations. Broome (1984a,b) argues in favor of ex post analysis, claiming that equality of expected utilities is not a real equality. Harsanyi (1977) also argues against attributing ex ante egalitarianism any significant value. Epstein and Segal (1992) focus attention on ex ante analysis, while Karni and Safra (2000) and, more explicitly, Ben Porath, Gilboa, and Schmeidler (1997) try to com- bine both ex post and ex ante consideration into one evaluation function. This article offers an ex post analysis, while assuming that society is not necessarily utilitarian (i.e., evaluations of social policies do not depend solely on the sum of individual utilities). We restrict attention to situations where each social policy leads to a known ex post distribution of utilities. As argued above, even though we cannot predict individual outcomes, we nevertheless know, for each policy, what proportion of the population will receive each possible utility level. We evaluate such distributions using a social welfare function W, which, not being utilitarian, is not necessarily linear. A utilitarian society is interested in maximizing the sum of individual util- ities and does not pay attention to the diversity of the distribution of utilities. Preferences for equality imply that society will be willing to reduce the av- erage utility in order to make the distribution more concentrated around its mean. Such concerns can be represented by quasi-concave functions, where the social evaluation of the average of two equally attractive distributions 60 The Journal of Law, Economics, & Organization, V21 N1 is better than both. 2 Such functions can be monotonic in individual utilities, but are also sensitive to utility differences between individuals, and admit some trade-off between the sum of utilities and their disparity. We now present an outline of our model; a more formal analysis appears in the appendix. Given two distributions of utility x and y, the mixture ax þ (1 � a)y gives each individual a times his allocation under x plus 1 � a times his allocation under y. For example, if x ¼ (3, 1) and y ¼ (1, 5), 1 2 x þ 1 2 y ¼ ð2; 3Þ. Our first assumption is that mixtures of equally attractive ex post utility distributions improve social welfare. Quasi-Concavity. Let x and y be two distributions of utility. If W(x) ¼ W(y), then for all a 2 (0, 1), W(ax þ (1 � a)y) c W(x). Strict quasi-concavity requires that for all x 6¼ y, W(x) ¼ W(y) implies W(ax þ (1 � a)y) > W(x).3 As mentioned above, we are interested here in utility distributions and do not care for the identity of the individual recipients of these utilities. Distri- bution x of utilities can therefore be represented as a cumulative distribution, where Fx(u) is the proportion of individuals whose utility given x does not exceed u. Two distributions that lead to the same cumulative distribution should therefore be equally attractive. Symmetry. If the two utility distributions x and y satisfy Fx ¼ Fy, then W(x) ¼ W(y). The major concept of the article is that of egalitarian social welfare functions: Definition 1. We say that a social welfare function is ex-post egalitarian if it is strictly quasi concave and symmetric. For a meaningful comparative statics analysis, we need to be able to com- pare different societies. For this, we will need the following concept: Definition 2. A social welfare function W# is more egalitarian than another social welfare function W if, whenever the society with the social welfare func- tion W is willing to sacrifice the difference between the averages of the two distributions in order to improve equality, so does the society with the social welfare function W#. 2. Recall that from an ex ante perspective, such functions imply preferences for randomizations. 3. Note the relation of this definition to Diamond’s (1967:765) argument in favor of ex ante evaluation—choosing the utility distribution x with probability p and y with probability 1 � p yields each individual expected utility that is equal to p times his utility under x plus 1 � p times his utility under y. Preferences for randomization imply quasi-concave preferences over social lotteries. Assuming that all individuals are expected utility maximizers, such preferences imply that the social welfare function is quasi-concave in ex ante utilities (Epstein and Segal, 1992). Ex-Post Egalitarianism 61 (see the Appendix for a more precise definition). Figure 1 depicts this concept for the case of two individuals. In this picture, W# is more egalitarian than W and distributions between x and E[x] (the average of the distribution x) display more equality. In the region between x and the main diagonal, the indifference curve through x of W# is lower than the corresponding indifference curve of W, hence W# is showing more egalitarianism than W. In the literature, considerations of equality that are based on cumulative dis- tributions lead to the concept of aversion to mean-preserving spreads. How- ever, by Dekel (1986), if W is symmetric and quasi-concave, then W must also represent such aversion, hence this concept of equality is related to the concept of quasi-concavity discussed above. 3. More Egalitarian Societies Suppose that society has control over a decision variable a 2 [0, 1] that leads to a utility distribution c(a). We assume that a > b implies that there is s such that the distribution of c(b) þ s is a mean-preserving spread of the distribution of c(a). In other words, as a moves from zero to one, c(a) is becoming more and more concentrated (even with respect to a changing average). The curve c(a) is called a track. The optimal value of a depends on social preferences, and our aim here is to compare these optimal values for societies with different degrees of ex post egalitarianism. The main result of the article is that the more egalitarian the society, the further it is willing to sacrifice average utility in order to reduce the variation of the optimal utility distribution. This result is stated in the following theorem. Figure 1. W# is more egalitarian than W. 62 The Journal of Law, Economics, & Organization, V21 N1 Theorem 1. If W# is more egalitarian than W, then along a track c where there is a substitution between average utility and spread, the optimal point for the more egalitarian function has a weakly lower sum of utilities and more concentration than the less egalitarian function. We present a formal statement of this theorem and provide a proof in the appendix. Moreover, we provide three conditions for a strict result (Theorem 2). Another result, implied by both theorems, is that an egalitarian society chooses a utility distribution that is more concentrated than the one chosen by a utilitarian society. This result is stated in the following corollary (and is proved in the appendix). Corollary 1. Let W be an ex-post egalitarian social welfare function and let c be a differentiable track. The optimal utility distribution of W is more con- centrated (and has a lower sum of utilities) than the optimal distribution of a utilitarian society. 4. Applications In this section we discuss some applications of our main result, namely, that a more egalitarian society will go further in the direction of reducing utility spread, even at the cost of total utility. 4.1 Victims of Crime versus Victims of the Criminal Law System Under the most fundamental principles of evidence law, facts constitutive of the defendant’s guilt have to be proven beyond a reasonable doubt. 4 This rule is perceived by practitioners as well as scholars to be grounded in justice-based considerations; it is often described as a right of defendants against the state. The rhetoric of rights and justice used in the justification of the strict standard of proof is often hostile to utilitarian calculations. This is no accident; utili- tarianism cannot in general justify the heightened burdens of proof associated with this principle. To see it, assume that a legal system can adopt either a strict or a lenient rule of proof. Under the strict rule, guilt has to be proven beyond a reasonable doubt. Under the lenient rule, guilt has to be proven in a ‘‘satisfactory man- ner.’’ Each one of these alternative rules involves costs and benefits, and there 4. This is an old principle of common law. An important articulation of it can be found in the English landmark case of Woolmington v. DPP [1935] A.C. 462. The court stated there unambig- uously that ‘‘No matter what the charge or where the trial, the principle that the prosecution must prove the guilt of the prisoner is part of the Common Law of England and no attempt to whittle it down can be entertained’’ (see Woolmington v. DPP [1935] A.C. 481). In the United States the same principle is regarded as required by the due process clause. The Supreme Court stated that this clause ‘‘protects the accused against conviction except upon proof beyond reasonable doubt of every fact necessary to constitute the crime with which he is charged.’’ See In Winship, 397 U.S. 358, 364 (1970). The rule however had much longer and deeper roots which can be traced to Roman law (Williams, 1963:186–190). Ex-Post Egalitarianism 63 is some trade-off between convicting the innocent and acquitting the guilty (Williams, 1963:188, Wertheimer, 1977:51–52). It is reasonable to believe that a legal system that adopts the lenient rule fulfills better its task of protecting its citizens from crime than a system that adopts the strict rule and the costs of the strict principle in terms of deterrence may outweigh the benefits the strict principle provides to the innocent people who are better protected under it from wrongful conviction. In that case, util- itarianism dictates a rejection of the strict rule. In their efforts to justify the strict rule, utilitarians often develop justifica- tions that depend on some restrictive assumptions. Bentham (1825:197), for example, supported the claim that there ought to be a presumption in favor of the accused, writing: ‘‘Generally speaking, a too easy acquittal excites regret and uneasiness only among men of reflection; while the condemnation of an accused, who turns out to have been innocent, spreads general dismay; all se- curity appears to be destroyed; no defence can any longer be found, when even innocence is insufficient.’’ Yet moral theorists usually reject utilitarian justi- fications for the principle that guilt should be proven beyond a reasonable doubt (Dworkin, 1986:72,81–84). Some of their arguments rely on the intu- ition that wrongful conviction seems to be a direct interference in the lives of an innocent person. In contrast, the failure to prevent a crime is merely an omission on the part of the state. Expostegalitarianismisanin-betweenapproach.LikeBenthamandlikeutil- itarianism, it does not ignore the individual costs and benefits of different bur- dens of proof. However, it also considers social norms and notions of justice, whicharerepresentedbythequasi-concavityofthesocialwelfarefunction.Such functions pay more attention to the well-being of those members of society who areworseoffcomparedtotherest.Toillustrate,considerthefollowingstructure. Society is composed of innocents and criminals. Innocent people face the risk of wrong conviction and the risk of criminal victimhood. Criminals may or may not be caught and convicted (for simplicity, we ignore minor groups such as criminals who are not convicted but are victims of crime). For a given level of burden of proof a, we get an ex post utility distribution of the form cðaÞ ¼ ðui; piÞ5i¼1; where u1 is the utility of convicted innocents, u2 is the utility of convicted criminals, u3 is the utility of innocent victims of crime, u4 is the utility of innocents, and u5 is the utility of the unconvicted criminals. We as- sume that crime benefits the criminal, that punishment is more severe than the consequences of the criminal act, and that wrongful conviction is worse than just conviction to the bearer of the punishment. Therefore we assume u1 < . . . < u5 (see distribution s1 —the continuous line—in Figure 2). As explained in Section 2, pi is the proportion in the population of those who receive utility ui (and is also the probability of getting this utility level). Society now raises the burden of proof while simultaneously raising the probability of detection so that the level of crime in society remains untouched. The reason for this simultaneous act is the desire that the cost of the increase in crime level be shared by the whole society and not only by the additional vic- tims. Criminals are interested in the probability of being punished, which is the 64 The Journal of Law, Economics, & Organization, V21 N1 product of the probability of detection and the conditional probability of con- viction given detection. As crime does not change, this combined probability too does not change. Moreover, assuming one crime per criminal, the number of criminals is unchanged. Therefore the number of those who receive utility levels u2 and u5 is unchanged. Also, the size of the u3 part of society is not changed—these being the people who suffer from crime. Obviously the combination of higher burden of proof and higher rate of de- tection that does not change the crime rate will reduce the number of convicted innocents, hence the size of the u1 level will go down, while the size of the u4 group will increase. Better detection is of course not cost free. Suppose that society finances the cost by taxing those who are not convicted and did not suffer from crime, that is, groups 4 and 5. 5 The new distribution is depicted by curve s2 (the dotted line). Obviously the second distribution intersects the first one only once, and from below. If the first distribution is optimal, then the new distribution must have a lower expected value, otherwise, as it is less spread than before, a higher expectation would have prevented distribution s1 from being optimal, even for a utilitarian society, and certainly for an egalitarian one. We strengthen this statement by assuming that the expected value is actually monotonically decreasing as we move from the first to the second distribution. The conditions of Theorem 1 are thus satisfied (observe that the set of dis- tributions society can choose from is a track) (section 3). Even when the second distribution is shifted to the right by the whole increased cost of detection, there will be only one crossing point, and the expected value of the shifted distribution will be more than the expected value of distribution s1 ). We there- fore find out that the more egalitarian society will seek a higher level of burden of proof. Our analysis does not imply that the optimal social policy is to set the burden of proof at its highest possible level (‘‘beyond any reasonable doubt’’). We Figure 2. Burden of proof. 5. Arguably, one should tax group 3 as well (recall that groups 1 and 2 cannot be taxed, as they are serving jail terms). However, we will ignore this tax in order to satisfy the conditions of Theorem 1. For infinitesimal changes this tax is indeed negligible, as the number of those who suffer from (serious) crime is small compared to the number of those who are not affected by it, and the tax is small compared to the sanction against criminals. Ex-Post Egalitarianism 65 recognize the validity of two claims on the social ruling. The obligation to increase social well-being, but also the obligation to promote equality. Unlike utilitarianism, and unlike some of the above justifications for the strict rule, we refuse to give one claim lexicographic dominance over the other. Ex post egal- itarian social welfare functions do indeed take both factors into consideration. 4.2 Torture and Other Forms of Cruel Punishment One of the most puzzling concerns of the legal economist is the ceiling on the size of the criminal sanction. Since Becker (1968), legal economists tried to explain the reasons for the widely held intuition that the criminal law system should not impose the maximally possible sanctions. Becker’s argument is simple and yet compelling. An expected punishment of $1000 can be imposed by different combinations of fines and probabilities of apprehension. If the costs of collecting fines are assumed to be zero, regard- less of the size of the fine, but the costs of apprehending and convicting crim- inals rise with the probability of apprehension, 6 then the most efficient combination is a probability close to zero and a fine arbitrarily close to infinity (see Becker, 1968; see also Posner, 2003:219–227). Under this argument, the only optimal sanction for every crime is the most extreme—possibly torture and death. Such a sanction enables one to obtain that even very low probability of apprehension and conviction would be sufficient to guarantee a sufficiently large expected sanction. Yet our legal system does not conform to the Becker model. Instead, our system imposes a ceiling on criminal sanctions. Torture is prohibited altogether in modern legal systems and most modern legal systems do not impose capital punishment. There are several primary explanations for the ceilings on criminal sanc- tions. Some legal economists dispute some of the presuppositions of Becker’s analysis. They point out that criminals may be risk averse or risk preferring, and if criminals are not risk neutral, Becker’s analysis does not yield the same outcome (Polinsky and Shavell, 1979). Moreover, economists point out the importance of marginal deterrence, that is, the incentive to substitute less for more serious crimes as a reason for differentiating between the sanctions imposed for different crimes (Shavell, 1983:1245–1246; Posner, 2003:222). Traditional criminal law provides, Kantian justification to the ceiling on sanc- tions. The wrongfulness and the culpability of criminals dictate what they de- serve and criminal law should not impose any sanctions that exceed what is being deserved. For a use of this Kantian insight, see Nozick, (1981/2000:363– 397) and Fletcher (2000:454–491). Ex post egalitarian considerations could explain the very same phenome- non. Criminal sanctions yield benefits that are provided to the public at large at the cost of the particular individual upon whom they are imposed. The im- position of criminal sanctions contributes to the well-being of others by 6. These costs are increasing with the probability because higher probabilities imply more po- lice, prosecutors, judges, defense attorneys, etc. 66 The Journal of Law, Economics, & Organization, V21 N1 producing deterrence. Yet, if too large sanctions are imposed on a small group of individuals, the disparity between the well-being of those who bear the costs of deterrence and those who benefit from it is too large. The ceiling on the criminal sanction is aimed at constraining this disparity. As before, society is composed of innocents and criminals. The first group consists of victims of crime (with the utility u2) and of nonvictims (with the utility u3), while the second group consists of those who are punished (u1) and of those who are not (u4). 7 Here too we assume that crime benefits the criminals and that punishment is more severe than the consequences of the criminal act. Therefore we assume u1 < . . . < u4, see distribution s1 in Figure 3. Society now reduces the severity of punishment while simultaneously rais- ing the probability of detection so that the number of convicted criminals is not higher than before. It follows that the number of criminals is thus decreasing (as before, we assume one crime and one victim per criminal). This, of course, is not cost free (otherwise, society would not have been at an optimal point), and we assume, as before, that the financial burden falls on the shoulders of those who are not convicted and did not suffer from crime (see footnote 5). The new distribution is depicted by curve s2. In this distribution, the utility of convicted criminals is higher than before (as the punishment is less severe), and there are not more of them than before. The utility of unconvicted crim- inals is less than before (they have to pay for better detection), and since the overall crime rate is down and the probability of detection is higher, the size of this group diminishes. As there are fewer crimes, there are fewer victims of crime, but their utility is the same as before. Finally, less crime means more people who are neither criminals nor victims (the old u3 group), but since they Figure 3. Cruel punishment. 7. We ignore here the possible existence of innocent people who are wrongly convicted. For the sake of simplicity, we assume that all innocent people share the same utility. Relaxing this assump- tion would not affect the outcome of our analysis. For example, introducing an additional group of typical and nonactive potential criminals whose social state is inferior to innocent non-victims and superior to innocent victims would create a new utility level u# between u2 and u3. Since it is reasonable to assume that this group is small and less wealthy, and hence only little taxes are imposed on it, our analysis holds for this case as well. Ex-Post Egalitarianism 67 have to pay for better detection, their utility goes down. Obviously the new distribution intersects the first one only once, and from below. If the first distribution is optimal, then the new distribution must have a lower expected value, otherwise, as it is less spread than before, a higher expectation would have prevented distribution s1 from being optimal even for a utilitarian society. We thus assume that the expected value is monotonically decreasing as we move from the first to the second distribution. It is reasonable to assume that crime benefits criminals less than the harm it imposes on its victims (otherwise it would have been socially optimal to have more crime, at least for a utilitarian society). In other words, u4 � u3 < u3 � u2. Moreover, since the number of convicted criminals is reduced, the change in the number of victims must be higher than the change in the number of un- punished criminals. In other words, when the new distribution s1 is shifted to the right by the full amount of the increased cost of detection, its expected value will be higher than the original distribution, hence the social choice set is a track. The conditions of Theorem 1 are thus satisfied, and we therefore find out that the more egalitarian society will seek less severe punishment than the less egalitarian one. 4.3 Sacrificing Some for the Rest The problem of whether or not society should sacrifice some (hopefully few) individuals to save the rest is at least 2500 years old. In Iphigeneia at Aulis, Euripides describes the attempted sacrifice of Iphigeneia to appease Artemis’ wrath, thus obtaining fair wind to carry the Greek navy out of the bay of Aulis to sail for Troy (Hornblower and Spaworth, 1996:765). Similarly the rabbin- ical law deals with the situation of a group of people who are surrounded by enemies and are asked to surrender one of them to be killed. The law is that they should all be killed, but should not do it, unless the enemy specified the one they wanted (Tosefta, 1977:20). In both sources there is an unambiguous reluctance to accept the idea that social goals may be achieved by sharply re- ducing the well-being of some members of society. Figure 4. Sacrificing some for the rest. 68 The Journal of Law, Economics, & Organization, V21 N1 In this subsection we discuss a similar problem, where society can reduce the utility of a minority to improve the position of the majority. We assume a homogeneous society, where initially everyone has utility level u1. By re- ducing the utility of some members of society it is possible to increase the utility of the rest—two such possibilities are depicted in Figure 4 (distributions s2 and s3 distribution s1 represents the fully egalitarian situation). Such redis- tributions of utility can be obtained, for example, by a military draft, jury duty, and other such obligations. If the expected value of the third distribution is less than that of the second one, then there is no point in switching from s2 to s3 even in a utilitarian society. We therefore assume that the expected value in monotonically decreasing as society is moving from distribution s3 to s2. Shifting distribution s3 to the left to get the same expected value as that of distribution s2 will maintain the single, fewer crossing property of the two distributions, hence Theorem 1 may be applied. We obtain that the more egalitarian society will opt for fewer benefits for the privileged at the cost of the underprivileged. We would like to emphasize that our analysis does not imply that society should not recruit only part of the citizenry, as it may well be that the fully egalitarian distribution is suboptimal, for example, when a relatively modest reduction in utility for a small number of individuals can significantly increase the utility of the rest. But the ex post egalitarian society is certainly more sen- sitive than the utilitarian society to issues of utility distribution, and will there- fore ask for greater benefits for the privileged than the utilitarian society before it agrees to further sacrifices by the underprivileged. 5. Conclusion Our most fundamental rules and practices constrain the pursuit of maximiza- tion of utility. It is often natural to explain these constraints as grounded in deontological considerations founded on principles of dignity and inviolability of persons. This indeed has been the explanation traditionally given to many of the rules and principles constraining the pursuit of utility. This article provides an additional explanation—one that relies on principles of ex post egalitarianism. More specifically, it is argued that ex post egalitar- ianism can explain a broad array of rules and practices, including the require- ment that guilt be proven beyond reasonable doubt, the prohibition on torture and cruel and unjust punishment, and the prohibition of sacrificing some for the sake of saving others. These rules and practices constrain the pursuit of utility. At the same time, they reallocate utility in a fashion that promotes ex post egal- itarianism. More specifically, they reduce the costs which otherwise would be imposed on people whose well-being turns out ex post to be lower at the expense of those whose well-being turns out ex post to be higher. Unlike ex ante egalitarianism, which is often achieved through randomiza- tions over social members, ex post egalitarianism involves a real transfer of goods from some individuals to others. It is therefore important that these goods be divisible, as indeed are the goods in all of our examples. Ex-Post Egalitarianism 69 6. Appendix Let X be the set of real bounded random variables on the probability space (S, R, P), and let Fx denote the cumulative distribution function of x 2 X, where X is the set of all utility distributions. Generic elements of X are denoted x, y, z, while degenerate random variables are denoted r, s, t. Scalars are denoted a, b, c. A preference relation c is a binary relation on X that is complete, transitive, continuous, 8 and monotone with respect to the relation of first-order stochastic dominance (as usual, ; denotes indifference and _ denotes strict preference). The preference relation c is symmetric if for all x; y 2 X; Fx ¼ Fy 0 x ; y: It is quasi-concave if for all x; y 2 X; x ; y 0 "a 2 ð0; 1Þ; ax þ ð1 � aÞy c x (strict quasi-concavity is defined with strict preferences). Note that quasi concavity is defined with respect to outcomes, and not with respect to probabilities. By construction, quasi-concavity implies preferences for av- eraging, hence it is related to equity-seeking behavior. Let U denote the set of all symmetric and quasi-concave preference relations on X. The social welfare functions W and W# used in this article represent preferences in U. Example 1. Consider the set L � U of preference relations that satisfy positive linearity: for all a > 0 and for all r 2 v x c y5ax þ r c ay þ r Let c 2 L and pick a non-degenerate utility distribution x. The preference relation c can be fully reconstructed from the set fy 2 v : y ; x and E½y� ¼ E½x�g and from the unique degenerate distribution a that satisfy a ; x. The set L was characterized in Safra and Segal (1998). Definition 3. Let c, c# 2 U. The preference relation c# is more egali- tarian than c if for all x 2 v fy 2 v : y c x and E½y�E½x�g � fy 2 v : y#x and E½y�E½x�g The preference relation c# is strictly more egalitarian than c if these inclu- sions are strict for all non-degenerate x. Note that the relation ‘‘more egalitarian than’’ is a partial order on the set of preference relations U. Fact 1. Assume c# is more egalitarian than c and that they are both strictly quasi-concave. Then, for all x 2 X; fy 2 X : y ; #x and E½y� ¼ E½x�g ¼ fy 2 X : y ; x and E½y� ¼ E½x�g: That is, both induce the same preferences over subspaces in which the expected values are fixed. 8. We use the topology of weak convergence with the Lévy metric on it. This metric is defined by d(F, G) ¼ inffe > 0: For all x, G(x – e) – e � F(x) � G(x þ e) þ eg; see Huber (1977). 70 The Journal of Law, Economics, & Organization, V21 N1 Proof. Let x 2 X and denote Ix ¼ fy 2 X : y ; x and E½y� ¼ E½x�g and I#x ¼ fy 2 X : y ; #x and E½y� ¼ E½x�g: First consider y 2 Ix. By definition, y c# x. If y c# x, then x 2 fz 2 X : z c y and E½z� � E½y�gnfz 2 X : z c #y and E½z� � E½y�g; a contradiction to the assumption that c# is more egalitarian than c. Next consider y 2 I#x\Ix and let a denote the degenerate utility distribution in which all utilities are equal to E[x]. By strict quasi-concavity, a _ x and a _ y. If x _ y, then, by continuity, there exist k 2 (0, 1), such that z ¼ ky þ (1 � k)a satisfies z ; x. Since z 2 Ix, it follows by the above argument that z 2 I#x, hence z ; #y; a contradiction to the strict quasi-concavity of c#. Since y;Ix, it must be the case that y _ x. By the strict quasi-concavity of c# and the continuity of c, there exists z satisfying E[z] ¼ E[x], x _#z, and z _ x. Hence, z 2 fy 2 X : y c x and E½y� � E½x�gnfy 2 X : y c #x and E½y� � E½x�g; a contradiction to the assumption that c# is more egalitarian than c. n Let x8y denote that Fy is a mean-preserving spread of Fx while Fx 6¼ Fy. The following fact shows that if c# is (strictly) more egalitarian than c, then, whenever the addition of r > 0 suffices to compensate the preference relation c# for the lack of equity manifested by y, then r (more than) suffices for the preference relation c. This result demonstrates the close connection between our definition and the strong measure of risk aversion suggested by Ross (1981). Note that the analysis of Ross is restricted to a subset of preferences in U—preferences that correspond to expected utility preferences in the liter- ature on risk. Fact 2. If c# is more egalitarian than c, then, for all x; y; r 2 X such that x 8 y, y þ r c #x 0 y þ r c x: Similarly, if c# is strictly more equitable than c, then y þ r c #x 0 y þ r_x: Proof. Let x, y, and r satisfy x8y, y þ r c#x, and x _ y þ r. By mono- tonicity, there exists z ¼ y þ r# (r# > r) satisfying z ; x and z ;#x þ r$ _# x (r$ > 0). Hence x belongs to fw 2 X : w c z and E½w� � E½z�g; while it does not belong to fw 2 X : wc#z and E½w� � E½z�g; a contradiction to the assumption that c# is more egalitarian than c. n Definition 4. A track is a function c : [0,1] / v such that a > b 0 cðaÞ 8 cðbÞ þ s for some s 2 v.9 9. s is a function of a and b. Ex-Post Egalitarianism 71 Note that s can be either a positive or a nonpositive distribution. Also note that if c(a) is a degenerate distribution, than a¼1. Otherwise, if c(a) is a mean- preserving spread of c(1) þ s, then c(a)¼ c(1) þ s, while the definition of 8requires the two to be different from each other. Theorem 1. Let c, c# 2 U and assume that c# is more egalitarian than c. Let c be a track and let c (a) and c (a#) be the unique optimal distributions of c and c# along c, respectively. Then a# � a. Proof. Suppose that a > a# and let s satisfy c(a)8c(a#) þ s. Consider first the case s c 0. The quasi-concavity of c# implies that c(a)c#c(a#) þ s. By monotonicity, c(a#) þ s#c(a#), hence c(a)c#c(a#), a violation of the assump- tion that a# is the unique optimal point of c# along c. Suppose now that s < 0. By assumption, c(a#) c#c(a). Since c# is more egalitarian than c, it follows Fact 2 and from the fact that c(a)8c(a#) þ s that cða#Þ ¼ ½cða#Þ þ s� þ ð�sÞ_#cðaÞ 0 cða#Þ_cðaÞ: A contradiction to the assumption that a is the unique optimal point of c alongc. It thus follows that a# c a. n Figure 5. Proof of Theorem 2. 72 The Journal of Law, Economics, & Organization, V21 N1 Next we want to establish conditions under which a# is strictly greater than a. Define the positive tangent c#þ(a) to satisfy c#þðaÞ ¼ lim anYa ¼ cðanÞ � cðaÞkcðanÞ � cðaÞk ; where k�k is with respect to the Lévy metric (see footnote 8). Definition 5. The track c : [0,1] / v is equitably-differentiable if for all a the tangent c#þðaÞ is well defined and there exists �c > 0 such that for all 0 < c < �c; cðaÞ þ cc#þðaÞ8cðaÞ þ s for some s ¼ s (a, c) 2 v. The ray cðaÞ þ cc#þðaÞ is the linear continuation of the curve c from the point c(a). The condition of the last display equation says that small movements in this direction increase the level of equity. (Recall that all preferences in U are quasi-concave.) Definition 6. The preference relation c 2 U is smooth is for all non- degenerate x 2 v there exists a unique linear function gx : v / < such that y ; x implies gx(y) � gx(x), and moreover, gx is continuous in x. Since a hyperplane in X is an indifference set of a linear function, it follows that if a preference relation c 2 U is smooth, then for all x 2 X there is a unique hyperplane Hx supporting the indifference set at that point. Note that Fréchet differentiability may not imply this property (Machina, 2001; Safra and Segal, 2001). Let c, c# 2 U be strictly quasi-concave and smooth and assume that c# is strictly more egalitarian than c. Let Hx (similarly H#x) be the supporting hyperplane to the indifference set of c(c#) through the nondegenerate dis- tribution x. Let D denote the line generated by all degenerate distributions, D ¼ fr : r 2 Xg; and consider the two-dimensional plane T ¼ SpanfD, xg. By monotonicity, D is not contained in Hx, hence Hx \ T is a line. By monotonicity and quasi-concavity there exists t satisfying ftg ¼ (Hx \ T) \ D and, by symmetry and strict quasi-concavity, t < E[x] (see Figure 5). By construction, Hx \ T ¼ fxg þ Spanfx � tg. Let Jx (J#x) be the indifference curves of c (c#) in T through x. Let Gx ¼ fy 2 X : E½y� ¼ E½x�g: Clearly, Hx ¼ SpanfHx \ Gx, x � tg and, by Fact 1, H#x \ Gx ¼ Hx \ Gx. By definition, the curves Jx and J#x intersect each other at x such that the following holds: Jx lies above J#x between x and D and Jx lies below J#x other- wise. This, however, is not sufficient to imply that x þ (x – t) does not belong to H#x. To have this, we assume the following generic assumption: ð*ÞFor all nondegenerate x; HðxÞ 6¼ H#ðxÞ: Ex-Post Egalitarianism 73 Note that the set of points fx : Hx \ T ¼ H#x \ Tg is the complement of an open and dense set (by comparability, Hx \ T ¼ H#x \ T iff Hx ¼ H#x; by smoothness, the set of points where Hx 6¼ H#x is open; and the fact that c# is more strictly equitable than c implies that there is no open set on which Hx ¼ H#x). Theorem 2. Let c, c# 2 U be strictly quasi concave and smooth. Assume that c# is strictly more egalitarian than c, and that condition (*) is satisfied. Let c be an equitably-differentiable track and let c(a) and c(a#) be the unique optimal points of c and c# along c, respectively. If a < 1, the a# > a. Proof. Since a < 1, c(a) must be a non-degenerate distribution (see the discussion after Definition 2 above). By Theorem 1, it is sufficient to show that a# 6¼ a. Consider the tangent c#þ(a). By the optimality of x, x þ c#þ(a) 2 Hx. There- fore x þ c#þ(a) ¼ z þ d(x � t) for some d, where z 2 Hx \ Gx ( ¼ H#x \ Gx). We show that x þ c#þ(a) does not belong to Hx \ Gx. Assume the contrary. Since c is equitably differentiable, it follows that for sufficiently small c > 0, c(a) þ cc#þ(a)8c(a) (note that the s in the definition of an equitably differentiable track satisfies s ¼ 0). The smoothness of c implies that for sufficiently small c > 0, c(a þ c) c c(a); a contradiction. Therefore d 6¼ 0. It follows by condition (*) that x þ (x � t) does not belong to H#x. Now d 6¼ 0 implies that x þ c#þ(a) does not belong to H#x. Therefore the optimality of c(a#) for c# shows that c(a#) 6¼ x and a# 6¼ a. n Corollary 2. Let c, c# 2 U such that c# is strictly quasi concave and smooth and c is a utilitarian preference relation. Let c be an equitably- differentiable track and let c(a) and c(a#) be the unique optimal points of c and c#, respectively. If a < 1, then a# > a. Proof. We will use the notations of the previous proof. By Theorem 1, a# c a. The condition that c is a utilitarian preference relation implies that the supporting hyper plane to the indifference set of c through the nondegenerate distribution x satisfies Hx ¼ Gx (note that t ¼ E[x]) and the tangent vector x þ c#þ(a) belongs to Hx \ Gx. As in the former proof, the facts that c is equitable- differentiable and that c# is smooth, symmetric, and strictly quasi-concave imply that for sufficiently small c > 0, c(a þ c) c#c(a), hence a# > a. References Becker, G.S. 1968. ‘‘Crime and Punishment: An Economic Approach,’’ 76 Journal of Political Economy 169–217. Ben Porath, E., I. Gilboa, and D. Schmeidler. 1997. ‘‘On the Measurement of Inequality Under Uncertainty,’’ 75 Journal of Economic Theory 194–204. Bentham, J. 1825. A Treatise on Judicial Evidence. London: Paget. Broome, J. 1984a. ‘‘Uncertainty and Fairness,’’ 94 Economic Journal 624–32. ———. 1984b. ‘‘Selecting People Randomly,’’ 95 Ethics 38–55. Dekel, E. 1989. ‘‘Asset Demand Without the Independence Axiom,’’ 57 Econometrica 163–69. Diamond, P.A. 1967. ‘‘Cardinal Welfare, Individualistic Ethics, and Interpersonal Comparison of Utility: Comment,’’ 75 Journal of Political Economy 765–66. 74 The Journal of Law, Economics, & Organization, V21 N1 Dworkin, R. 1986. ‘‘Principle, Policy, Procedure,’’ in R. Dworkin: A Matter of Principle. Oxford: Oxford University Press. Epstein, L.G., and U. Segal. 1992. ‘‘Quadratic Social Welfare Functions,’’ 100 Journal of Political Economy 691–712. Fletcher, G. 2000. Rethinking Criminal Law. Oxford: Oxford University Press. Harsanyi, J.C. 1977. Rational Behavior and Bargaining Equilibrium in Games and Social Situa- tions. Cambridge: Cambridge University Press. Hornblower, S., and A. Spaworth. 1996. The Oxford Classical Dictionary. Oxford: Oxford Uni- versity Press. Huber, P.J. 1977. Robust Statistics Procedures. Society for Industrial and Applied Mathematics. Kamm, F.M. 1996: Morality, Mortality, vol. II. New York: Oxford University Press. Karni, E., and Z. Safra. 2002. ‘‘Individual Sense of Justice: A Utility Representation,’’ 70 Econ- ometrica 263–84. Machina, M. 2001. ‘‘Payoff Kinks in Preferences Over Lotteries,’’ 23 Journal of Risk and Un- certainty 207–60. Myerson, R. 1981. ‘‘Utilitarianism, Egalitarianism and the Timing Effect in Social Choice Prob- lems,’’ 49 Econometrica 883–97. Nozick, R. 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press. Polinsky, M., and S. Shavell. 1979. ‘‘The Optimal Tradeoff Between the Probability and Magni- tude of Fines,’’ 69 American Economic Review 880–91. Posner, R. 2003. Economic Analysis of Law, 6th ed. New York. Aspen Law & Business. Roemer, J.E. 1996. Theories of Distributive Justice. Cambridge MA: Harvard University Press. Ross, S.A. 1981. ‘‘Some Stronger Measures of Risk Aversion in the Small and in the Large with Applications,’’ 49 Econometrica 621–38. Safra, Z., and U. Segal. 1998. ‘‘Constant Risk Aversion,’’ 83 Jornal of Economic Theory 19–42. ———. 2002. ‘‘On the Ecomomic Meaning of Machina’s Fréchet Differentiability Assumption,’’ 104 Journal of Economic Theory 450–61. Shavell, S. 1985. ‘‘Criminal Law and the Optimal Use of Nonmonetary Sanctions as a Deterrent,’’ 85 Columbia Law Review 1232–62. Tosefta, 1977. Translated by J. Neusner. New York: Ktav. Wertheimer, A. 1977. ‘‘Punishing the Innocent Unintentionally,’’ 20 Inquiry 45. Williams, G. 1963. The Proof of Guilt: A Study of the English Criminal Trial, 3rd ed. London: Stevens. Ex-Post Egalitarianism 75