In this section, we consider civil litigation, that is, the bringing of lawsuits by private actors to enforce their rights in the areas of civil law that we have just  discussed. Until now, we have largely assumed that the operation of the legal system is frictionless, in the sense that the bringing and adjudication of lawsuits is without cost. We now analyze the implications of the expense involved in the operation of the legal system.

We begin with what may be called the basic theory of litigation: the choice of a party who has suffered a loss whether to sue; if suit is brought, the choice of the litigants whether to settle with each other or instead go to trial; and the choice of litigants, before or during trial, of how much to expend on litigation. Then we discuss various extensions to the basic theory of litigation, including nuisance suits, the shifting of legal fees, lawyers and agency problems in litigation, and legal discovery. We subsequently consider the provision of legal advice, the appeals process, alternative dispute resolution, and the formulation of legal rules.

Private incentive to sue

As a general matter, the plaintiff will sue when the cost of suit cP is less than the expected benefits from suit. The expected benefits from suit incorporate potential settlements and trial outcomes, but in this section we usually assume for simplicity that if suit is brought, the plaintiff obtains as a judgment a certain amount h equal to harm suffered. Thus the plaintiff will sue if and only if his litigation cost, cP, is less than h. (Obviously, if there is only a probability p of winning this amount, the plaintiff, if risk neutral, would sue if and only if cP < ph; and if the plaintiff is risk averse, he would be less likely to sue.) The effect on the private incentive to sue of many variations in the legal environment is straightforward to identify, as we will note below.

Socially optimal suit versus the private incentive to sue. The private incentive to bring suit is fundamentally misaligned with the socially optimal incentive to do so, given the social costs and social benefits of suit. The deviation between the privately-motivated and socially appropriate level of suit could be in either direction. The general reasons for these conclusions may be understood as follows.

On one hand, there is a divergence between social and private costs that can lead to socially excessive suit. Specifically, when a plaintiff contemplates bringing suit, he bears only his own costs; he does not take into account the defendant’s costs or the state’s costs that his suit will engender.

of suit, its deterrent effect on the behavior of injurers. But the plaintiff does consider his private benefit, the gain he would obtain from prevailing. This private gain is not a social benefit but instead is a transfer from the defendant; it could be either larger or smaller than the social benefit. The contrast between the socially optimal and private incentive to sue is initially examined in Shave.

It should be clear from out discussion that the point that the private and social incentives to bring suit may diverge is robust. On one hand, it will always be the case that the private cost of use of the system will be less than the social. And, on the other hand, the private benefits from suit will be what the plaintiff will win from suit, usually money, whereas the social benefits from suit will ordinarily be different: they will always include deterrence benefits and may also include compensation of victims (if insurance is unavailable) and the setting of precedent. These benefits litigants either will not take into account or will tend to weigh differently from their social importance.

Implications of the social and private divergence. The main implication of the social and private divergence is that state intervention may be desirable, either to correct a problem of excessive suit — notably, by taxing suit or barring it in some domain — or a problem of inadequate suit — by subsidizing suit in some way. For the state to determine optimal policy, however, requires it to determine the effects of suit on injurer behavior and weigh them against the social costs of suit. It is thus not correct for the state to base policy on some simple, even though superficially appealing, criterion, and notably, whether the plaintiff’s expected gains from suit would have exceeded the aggregate litigation costs.

Whereas this section was concerned with the implications of litigation costs for the frequency of suit, another litigation-cost related issue has to do with the level of precautions taken by injurers. The optimal level of precautions should reflect not only the direct harm that would be caused by an accident, but also litigation costs. To induce this higher level of

precautions, injurers who are sued should pay not only the harm, but also (perhaps as a penalty) litigation costs borne by the plaintiff and by the state.

Another response to the problem of inadequate suit on account of small stakes, when there are many victims so that aggregate stakes are substantial, is the use of class actions. See Dam (1975). Yet another approach is to encourage suits

with pro-plaintiff fee-shifting (see section 5.4.2) or damage multipliers (see Kaplow 1993), or to use “decoupling” — that is, a system in which damages are raised (to enhance deterrence) and plaintiffs’ recoveries are raised (if suit must be encouraged, but not necessarily by the amount of damages) or are lowered (if there otherwise would be excessive suits), on which see Polinsky and Che (1991) and Shavell (1997, 1999).

It should also be emphasized that the importance of the private-social divergence in incentives to sue may be substantial. This is suggested by the fact that the costs of use of the legal system are high; indeed, legal costs may on average actually equal the amounts received by those who sue.103 Hence, the incentives created by the legal system must be significant to justify its use.

Settlement versus Trial Assuming that suit has been brought, we now take up the question whether parties will reach a settlement or go to trial. A settlement is a legally enforceable agreement, usually involving a payment from the defendant to the plaintiff, in which the plaintiff agrees not to pursue his claim further. If the parties do not reach a settlement, we assume that they go to trial, that is, that some tribunal determines the outcome of their case. In fact, the vast majority of cases settle.105 We discuss here two different models describing whether settlement occurs and then consider the socially optimal versus the private decision whether to settle.

Exogenous beliefs model. One model of settlement versus trial presumes that parties have each somehow come to a belief about the probability of the trial outcome; let pP represent the probability of the plaintiff prevailing in his opinion, and pD be that same probability in the defendant’s opinion. Let w be the amount that would be won (for simplicity assume that they agree about w). Assume also that the parties are risk neutral. The plaintiff’s expected gain from trial, net of litigation costs, is pPw – cP. This is the minimum amount he would accept as a settlement, rather than go to trial. The defendant’s expected loss from trial, including his litigation costs, is pDw + cD; this is the maximum amount he would pay in settlement rather than go to trial. Hence, a settlement is possible if and only if pPw – cP pDw + cD, in which case the settlement amount will be in the settlement range [pPw – cP, pDw + cD]. Note that if the parties agree on the probability p, the settlement range will be positive and

cP + cD in length. A settlement range does not exist, and trial will occur, when pPw – pDw > cP + cD. This means that the expected award in the plaintiff’s opinion exceeds the expected award in the defendant’s opinion by more than the sum of litigation costs. Thus, trial will tend to occur when the plaintiff is sufficiently more optimistic about winning than the defendant believes he should be.

civil cases were resolved without trial; see Administrative Office (1995). These figures, however, overstate the settlement rate because many of the cases not tried were dismissed by a court rather than being settled. On the other hand, many disputes are settled before any complaint is filed.

Lowenstein et al. (1993) and Mnookin (1993) suggest that litigant overoptimism is plausible.

The model under discussion originated with Friedman (1969), Gould (1973), Landes (1971), and Posner (1973), and was further elaborated in Shavell (1982c). It has the virtue of clarifying several basic intuitions: that settlement is fostered by litigation cost savings and by risk aversion, and that trial might result when plaintiffs expect to gain more than defendants expect to lose.

The model also helps to explain the striking predominance of settlement in actuality. First, lawyers, who are experts on the law, are typically advising both litigants, and much information is acquired and comes to be shared by the opposing sides; we should thus expect beliefs of the two sides to be similar. Second, the costs of trial tend to be substantial.

These observations suggest that a settlement range typically exists and thus that settlement would be likely to occur.107 The model has the additional virtue of being simple and easy to manipulate, because it focuses on the calculation of the settlement range.108 The model is unsatisfying, however, in two respects. It does not explain the origin of parties’ beliefs.

And it does not include a description of rational bargaining between the parties; thus, it does not explain whether there will be a settlement when there is a positive settlement range or the amount of any settlement within the range.

The observations also raise interesting questions about the timing of settlement — will it occur early or late? (In fact, many cases settle early, but many also settle late, on the eve of trial or even during trial.) A reason for settlement to be delayed is that at the outset of settlement negotiations, information may be disparate; but, as noted, over time, as information is acquired and shared, the parties’ beliefs tend to converge. A reason for settlement to occur early, however, is that this maximizes the parties’ savings in litigation costs. To express the point differently, as time passes, more litigation costs are sunk, meaning that the savings from settlement are lowered, tending to decrease the chances of settlement. For analysis of the timing of settlement (in a model of asymmetric information), see Spier (1992a).

For example, the larger the possible judgment amount w, the greater the chance of trial, for a larger judgment magnifies the effect of differences of opinion in the likelihood of trial outcomes (pPw – pDw, when positive, is increasing .

However, larger judgments tend to reduce the likelihood of trial if litigants are risk averse.

Asymmetric information model

A second type of model of settlement versus litigation presumes that there is asymmetry of information between litigants and includes an explicit account of bargaining. The simplest of such models is that of Be bchuk (1984), in which there is one-sided asymmetry of information and bargaining consists of a single take-it-or-leave-it settlement offer made by the party without private information.109 Suppose, for example, that the defendant has private information about the probability p that the plaintiff will win at trial (perhaps the defendant possesses private information bearing on whether he will be found negligent).

The plaintiff makes a settlement offer x, knowing that low p defendants will reject his offer and high p defendants will accept; specifically, if pw + cD < x, the defendant will reject and the plaintiff will therefore obtain only pw – cP, but if pw + cD x, the defendant will accept and pay x. The plaintiff, who knows the probability distribution over p, chooses x to maximize his expected payoff from settlement or trial.111 The higher his offer x, the more he will obtain if his offer is accepted, but the greater is the likelihood of rejection and thus of his bearing trial costs.

At the optimal offer for the plaintiff, there will generally be a positive probability of trial and of settlement. Furthermore, it can be shown that the higher are litigation costs, the more likely is settlement, and that risk aversion increases the likelihood of settlement. This model, note, is roughly consistent with the previous one of section 5.2.1 in the sense that trial results due to disparate beliefs (arising out of the asymmetry of information). In particular, the plaintiff’s opinion of the probability of winning is the mean probability E(p) over the distribution of defendants, and trial will occur if the defendant’s p is sufficiently low in the distribution. In addition, the comparative statics of the present model are similar to that of the previous one (for instance, as just noted, higher litigation costs make settlement more likely). The primary virtues of asymmetric information models are twofold.

Asymmetric information models of trial versus settlement have been refined and extended in various ways. See, for example, Daughety and Reinganum (1994) (in which asymmetry of information is two-sided), Hay (1995) (in which

unobservable case preparation contributes to asymmetry of information), Reinganum and Wilde (1986) (in which the informed party makes the offer, and the uninformed party makes an inference from it), Schweizer (1989) (in which asymmetry of information is two-sided), and Spier (1992a) (in which there are multiple rounds of bargaining and, as discussed in note the focus is on the timing of settlement). For a useful survey of asymmetric information models of litigation, see Farmer and Pecorino (1996), and for a general survey of asymmetric information models of bargaining, see Kennan and Wilson (1993). For an empirical investigation of litigation that emphasizes asymmetric information, see Farber and White

Asymmetric information could also concern the magnitude of the judgment or factors independent of the trial itself, such as parties’ degree of risk aversion, their short-run need for funds, their tastes for litigation, and, as already noted, unobservable aspects of case preparation. Specifically, the plaintiff’s expected payoff as a function of x is 0z(pw – cP)f(p)dp + (1 – F(z))x, where

z =(x – cD)/w), f is the density of p, and F is the cumulative distribution of p.

First, they include an explicit account of bargaining and thus of the probability of settlement and the magnitude of the settlement offer. (But the ability to predict the probability of settlement and the magnitude of the

settlement offer is to some extent specious. Under the bargaining models studied, essentially arbitrary modeling choices are made over such matters as who makes the offer, the informed or the uninformed party; these choices substantially influence the probability of settlement and the settlement offers.)112 Second, the models explain differences of opinion that give rise to trial in terms of differences in possession of information. (However, the models do not explain why there should be such differences in information, given the incentives for sharing of information and its orced disclosure through legal discovery.

Socially optimal versus privately-determined settlement. The private and the social incentive to settle may diverge for reasons related to those explaining the difference between the private and the social incentive to sue. First, because the parties involved in litigation do not bear all the costs of a trial — the salaries of judges and ancillary personnel, the forgone value of juror time, implicit rent on court buildings — the parties save less by settling than society does, which suggests that the private incentive to settle is socially inadequate.114 Second, when there is asymmetric information, parties will fail to settle — and thus litigation costs will be incurred — when their demands turn out to have been too aggressive.

But their desire to obtain from each other a greater share of their litigation cost savings does not itself translate into any social benefit. Third, the prospect of settlement may reduce deterrence because defendants gain from settlement. This need not, however, be socially undesirable because settlement lowers the real total social cost of harmful acts, making less deterrence appropriate.

Also, the division of surplus in settlement may affect deterrence. Fourth, the prospect of settlement may increase deterrence because it lowers plaintiffs’ expected litigation costs and thus increases the chance of suit. These latter factors are not, of course, taken into account by the parties to settlement negotiations.

112For an attempt to address this problem, see Daughety and Reinganum (1993).

113The normative question concerning the social versus the private value of settlement has received little attention

relative to the positive question of when parties will settle. On the normative question, see Polinsky and Rubinfeld (1988b),

Shavell (1997, 1999), and Spier (1997).

In addition, the parties generally do not properly take into account each others’ costs when there is asymmetric information. For example, the adverse consequence of rejecting an offer and going to trial involves incurring one’s own trial

costs but not the other side’s.

Indeed, settlement reduces ex post costs by the sum of the plaintiff’s, defendant’s, and court’s costs, but deterrence is reduced only by the fraction of the savings in the plaintiff’s and defendant’s costs that is captured by the

defendant in settlement bargaining.

Finally, by averting trial, settlement may have other effects on social welfare. For example, trials may reveal socially valuable information (such as about product hazards that consumers could guard against) or lead to new precedents. These are also factors that parties may ignore or treat inappropriately (a firm might have a socially perverse incentive to avoid trial to conceal information about product hazards). The state can act to correct a divergence between private and social incentives to settle. A factor that should be stressed in considering optimal social policy is that if settlements were to reduce deterrence undesirably, this does not imply that trial should be fostered; deterrence could be enhanced by raising damages to induce settlements for greater amounts or by imposing a tax on defendants (regardless of whether they settle). Trial is desirable only when there is no less costly way to raise social welfare, and a conjecture is that the usual social problem is that there are too many trials, not too few.116

Litigation Expenditures

Private incentives to spend on litigation. Here we focus on litigant expenditures given that suit has been brought. (We should note that litigation expenditures are made prior to trial as well as during trial; indeed, most are incurred in cases that settle.) Suppose that each litigant’s expenditures are made noncooperatively, as in Braeutigam, Owen, and Panzar (1984), Katz (1987, 1988), and Posner (1973).117 Under this assumption, a plaintiff will make litigation expenditures as long as this raises his expected return from settlement or trial (net of litigation costs), and a defendant will make such expenditures as long as this lowers his expected total outlays. The effects of each litigant’s expenditures will generally depend on what the other does; indeed, the two will often be spending to rebut one another.118

5.3.2. Social versus private incentives to make litigation expenditures. There are several sources of divergence between social and private incentives to spend during litigation Such expenditures have a negative social value. Third, even if expenditures do improve the accuracy of outcomes, they may not be socially optimal in magnitude. By analogy to what we stressed in section 5.1.2, the parties decide on their expenditures based on how they influence the litigation outcome, without regard to the influence (if any) on incentives. This could lead to expenditures that are too great or too small, relative to what is socially correct.

In fact, courts attempt to promote settlement in a variety of ways.

We are abstracting from the possibility that the parties might be able to enter into agreements to limit litigation expenditures.

One may contrast a system in which a single authority, perhaps the tribunal, makes decisions about litigation

expenditures. This is done to an extent in many European countries for criminal proceedings. In the United States, federal trial court judges occasionally use special masters or court-appointed experts to perform similar functions.

119See generally Kaplow (1994a)

An important instance of the possibility that expenditures could be socially excessive concerns the assessment of damages. See Kaplow and Shavell (1996b). Suppose that the presently estimated harm deviates from the truth by $100. Then one of the litigants will be willing to spend up to $100 to prove the correct amount (it will be the defendant if the estimate exceeds the correct level, and the plaintiff if the estimate is too low). It can be shown that the social value

of the more accurate estimate tends, however, generally to be lower than $100, because the social value of accuracy is based on its effects on incentives. Indeed, there will sometimes be no beneficial incentive effect from more accurate assessment of harm, such as when errors (in the absence of additional expenditures) are unbiased and not predictable ex ante by potential injurers.

In particular, potential injurers, at the time they choose their precautions, will often know only a probability distribution of possible harm, so litigation expenditures ex post that provide a precise assessment of a particular victim’s actual harm would not affect incentives.120 Because private and social incentives to spend on litigation may diverge, it may be beneficial for expenditures to be either controlled or encouraged.

Expenditures on determining whether a party is liable (as opposed to the magnitude of damages) could also be socially excessive or inadequate.121 To illustrate the latter possibility, suppose that the cost of establishing that a defendant was negligent exceeded the amount of harm suffered. Plaintiffs would not have an incentive to make the necessary expenditure, with the result that negligence might not be discouraged. But if the deterrent effect of liability were significant, that result would be undesirable. (Suppose that deterrence would eliminate most negligently caused harm, so that ex post litigation costs would not often have to be incurred.

Extensions of the Basic Theory

We consider here various extensions to the basic theory discussed above; for the most part, these extensions are concerned with the description of litigation rather than with its normative analysis.

If compensation of risk-averse victims were important due to the unavailability of insurance (see section 2.2), more accurate compensation will have social value, although parties’ incentives to make litigation expenditures would still tend to be excessive because the insurance benefit of avoiding a $1 error in compensation is less than the maximum of $1 that a party would be willing to expend to correct the error. See Kaplow (1994b).

The determination of liability in the context of law enforcement (see section 6) is analyzed in Kaplow and Shavell (1994a).

In the light of our analysis in section 5.1.2, it should be clear that cases opposite to those illustrated in the text — that is, inadequate incentives to prove damages or excessive incentives to establish liability — can also arise.

123For empirical studies that bear on the theory of litigation, see, for example, Hughes and Snyder (1995), Ramseyer and Nakazato (1989), and Viscusi (1986a, 1988).

Nuisance suits

A nuisance suit is often defined as a suit that the plaintiff brings even though he would not actually pursue his case to trial, because the expected award he would obtain is less than the trial cost; in this sense a nuisance suit is a negative expected value suit. We should first point out that we cannot infer that nuisance suits should not be brought: as we stressed in section 5.1.2, it is quite possible that the social deterrence benefits of a type of suit make it desirable to bring even though litigation costs exceed the expected judgment.124 (Nor, as we emphasized, can it be assumed that non-nuisance suits — positive expected value suits — ought to be brought.)

A major question about nuisance suits, and the one to which primary attention has been given, is why they are brought in view of their negative expected value. One important explanation concerns asymmetric information: that plaintiffs who are not willing to go to trial are not identifiable to defendants and ride on the coattails of plaintiffs who would be willing to go to trial. As a consequence, the plaintiffs who are unwilling to go to trial are able to settle for a positive amount with defendants; see Bebchuk (1988) and Katz (1990a). Another possibility, not premised on asymmetric information, is that a plaintiff can initiate a suit at low cost and, although he would lose if the defendant undertook substantial litigation effort, he would prevail if the defendant did not. In this case, the defendant might prefer to settle to avoid paying defense costs; see Rosenberg and Shavell (1985).

An additional reason concerns the point that, as plaintiffs spend continuously on litigation, their willingness to go to trial increases (because the amount that they would then save by not going forward diminishes); see Bebchuk (1996).

This is obviously not to deny that some types of nuisance suits are undesirable. For example, where plaintiffs would not prevail because their cases are fictitious, their bringing of suits would tend to distort incentives as well as waste resources on litigation.

125There is also a literature on how nuisance suits (or, relatedly, suits with a low probability of success) might be discouraged. See Bebchuk and Chang (1996), Katz (1990a), and Polinsky and Rubinfeld (1993, 1996).

Shifting of legal fees

Thus far, we have assumed that parties bear their own legal costs, a regime referred to as the American rule. By contrast, under the English rule, the loser pays the legal costs of both sides. Fee-shifting may also be one-way, favoring the plaintiff (that is, shifted only to the defendant, if the plaintiff wins) or favoring the defendant (shifted only to the plaintiff, if the defendant wins).126 Fee-shifting has clear implications for the incentive to sue; for example, under the English rule, suit is encouraged, relative to a regime of no fee-shifting, if the plaintiff’s probability of winning is sufficiently high, because then his expected costs of trial fall.

Fee-shifting may increase the chance of trial, essentially because it accentuates differences in litigant estimates of the expected gains and losses from trial; see Posner (1977a) and Shavell (1982c). Under the English rule (the effects under one-way fee shifting are similar), if the plaintiff and the defendant are each optimistic about winning, then each will be optimistic about passing on his legal expenses to the other, which tends to reduce the settlement range and increase the chances of trial.127 However, fee-shifting tends to raise the amounts the parties will spend at trial, as a party’s expenditure will only be a cost to him with a probability rather than with certainty.128 This attenuates the rise in the chance of trial. (The increase in litigation costs is, of course, significant in itself.129) Also, fee-shifting makes trial riskier, so that if parties are risk averse, it can reduce the chance of trial.

A variant of simple fee-shifting is an offer-of-settlement scheme, according to which fees are shifted only if a settlement proposal is rejected and the amount actually awarded differs in a specified way from the rejected proposal. For instance, if a defendant rejects a plaintiff’s offer and the actual trial award exceeds that offer, fees might be shifted to the defendant. The effects of such schemes on settlement are complex and not readily summarized. See Bebchuk and Chang (1997), Miller (1986), and Spier (1994).

For a description of the use of fee shifting, see Derfner and Wolf (1995). Fee-shifting favoring only plaintiffs is used to stimulate suits where the private incentive is thought to be inadequate, whereas fee-shifting favoring only defendants is usually proposed as a means to discourage frivolous litigation.

The same conclusion holds, for closely related reasons, in the asymmetric information model of Bebchuk (1984). 128See Braeutigam, Owen, and Panzar (1984), Hause (1989), and Katz (1987). 129In Katz’s (1987) simulation, the English rule increases costs by 125 percent.

5.4.3. Additional elements of trial outcomes. We have assumed that the only outcome of a trial is a judgment paid by the defendant and received by the plaintiff, but there are other possibilities. First, a trial outcome may have implications for a litigant beyond the immediate judgment. For example, a firm may believe that a loss at trial would invite a string of future law suits; thus, a loss would be more costly for it than the judgment.130 This would tend to make

settlement more likely, as it would raise the amount the defendant firm would be willing to pay in settlement. Second, a litigant may care whether a trial is held per se: a plaintiff might, say, wish the defendant to be exposed to public scrutiny. This would make trial more likely. Or a party might want to avoid a trial because it would result in the airing of embarrassing facts or the disclosure of valuable business information, which would tend to make trial less likely. Third, in whether cases that go to trial are representative of the underlying population of cases, and notably, whether the likelihood of plaintiff victory at trial or the amounts won are typical of the cases that settled. This question is important because, often, the most readily available data is on cases that go to trial, whereas the great majority of cases settle. As Priest and Klein (1984) first emphasized, the cases that go to trial may be quite different from settled cases.

For example, if in 99% of cases defendants would be found liable for a certain amount, but in 1% of cases defendants would prevail, then, if plaintiffs cannot distinguish the two groups, they will likely insist on a settlement amount that the former defendants would pay and the latter would reject. Hence, defendants would win all cases that go to trial, which would be wholly unrepresentative of the cases that settled. In general, cases that go to trial are not representative of the underlying population of cases, and the proper manner of making inferences is complex.132

5.4.5. Lawyers as agents of litigants. Because clients and their lawyers are in a principal and agent relationship, the general problems of principals and agents are relevant. Consequently, to the degree that clients cannot observe lawyers’ effort levels and lack legal expertise, a fee arrangement linked to lawyers’ performance might have joint value to them, but it would impose risk on lawyers (although it would simultaneously reduce clients’ risk, and many clients — particularly individuals or small entities — may be risk averse).133 130See, for example, Che and Yi (1993).

See Shavell (1993b). For instance, suppose that for each parent in a custody dispute, the value of custody is equivalent to $1,000,000, each parent believes custody would be awarded with probability 50%, and the cost of trial for each is $10,000. Then to induce either parent to settle and give up the opportunity of custody, an offer of at least $490,000 would have to be made, yet neither parent may have assets nearly equal to that amount.

132Priest and Klein (1984) suggested that cases that go to trial would be won by plaintiffs approximately 50% of the time, regardless of the underlying population of cases. This somewhat surprising conclusion of theirs is correct given their assumptions; but it is not borne out in fact, and does not hold under general assumptions about the population of cases and bargaining over settlement and trial. See Eisenberg (1990), Eisenberg and Farber (1997), Hylton (1993), Shavell (1996), Waldfogel (1995b), and Wittman (1985).

133Another problem in the agency relationship is that, at the time of contracting, the client may not know the lawyer’s quality, and there may also be asymmetric information regarding the strength of the client’s case.

It frequently are paid a fraction of the amount they obtain for their clients under a so-called on tingent fee agreement.134 In addition, lawyers are implicitly rewarded on the basis of performance in the sense that they (and their firms) acquire reputations, so that their future business depends on performance. Lawyers’ conduct is also controlled to some extent by the threat of suit by clients for malpractice, by court-mandated penalties, and by bar association

discipline. See Wilkins (1992). Principal-agent problems that are specific to the legal context arise in the decisions to sue and to settle versus go to trial. See Miller (1987). For example, when lawyers are paid on a contingent fee basis, they might have perverse incentives to favor not bringing suits or to settle,

because their own gain would be only a fraction of the total gain from winning. See Danzon (1983) and Hay (1996).135 When lawyers are paid on an hourly basis, it is often said that they have an excessive incentive to sue and to reject settlement offers in favor of trial. (This claim, however, assumes that their hourly rate exceeds their opportunity costs; if, for example, additional, more profitable work comes into the office after the hourly rate is set, then hourlycompensated lawyers may have an excessive incentive to settle.)

5.4.6. Insurers as agents of litigants. Insurers often play a role in litigation. In accident suits, for example, plaintiffs may own medical or disability insurance policies with (subrogation) clauses giving their insurers the right to bring suit and conduct litigation, and defendants frequently hold liability insurance policies that give insurers a role in litigation. Conflicts may arise between litigants and their insurers as their agents in litigation when the coverage ceiling is less than the amount at stake in litigation. See Meurer (1992) and Sykes (1994). To illustrate, suppose that there is a 20% chance that trial would result in a finding of liability and losses are $500,000; also assume that the defendant’s liability coverage ceiling is $150,000. The liability insurer would prefer to reject a settlement offer of $75,000, even though the offer falls below the expected judgment of $100,000. (If the settlement offer is accepted, the insurer pays $75,000 for sure, whereas if there is a trial the insurer makes a payment of $150,000 only 20% of the time, which has an expected cost of $30,000.) By similar reasoning, a plaintiff’s insurer would tend to want to settle for less than plaintiffs would like, which would increase the chance of settlement. Note, however, that reputational interests of insurers as well as the possibility of renegotiation between insurers and insureds serve to mitigate their conflicts of interest.

134See generally Rubinfeld and Scotchmer (1998) on contingent fees. We note that payment arrangements that are contingent upon outcomes may be common because lawyers who nominally charge hourly rates may submit higher bills when successful and may trim their bills when they lose.

135Hay (1997) discusses how bifurcated contingent fees (which pay a higher rate if there is no settlement) can help to address this problem. Additional principal-agent problems arise in the context of class actions, where many plaintiffs are joined in a class and there is a free-rider difficulty with regard to supervision of attorneys. See Coffee (1986) and Macey and Miller (1991).

5.4.7. Voluntary sharing of information. In the discussion of settlement versus litigation in section 5.2, we assumed that the information of parties was somehow exogenously determined: either information was in the background and formed parties’ perhaps disparate beliefs, or else information was explicitly presumed to be asymmetric. However, litigants in general have strong motives to share information. See Shavell (1989a). Most obviously, parties will want to share favorable information in order to foster settlement and to improve its terms. A plaintiff, for example, would want to show the defendant information establishing that his losses were in fact higher than the defendant otherwise believes; in this way, the plaintiff can induce the defendant to pay more in settlement and perhaps avoid an impasse leading to trial. Likewise, a defendant would want to show the plaintiff evidence pointing toward his lack of responsibility, in order to  convince the plaintiff to accept a lower settlement offer. In addition, parties will want to reveal information to avoid negative inferences that would be made from their silence. If a plaintiff says nothing about the magnitude of his losses, the defendant will be likely to infer that the plaintiff is withholding information that his losses are lower than average, and if this inference is made, the defendant will not be willing to make an average offer. Both this incentive to avoid negative inferences and the incentive to reveal favorable information tend to produce significant voluntary disclosure and help to explain the high rate of settlement.136

Nevertheless, some information will not be shared, and this helps to explain why some cases do not settle. First, a party may decide against disclosing information because revealed information can often be countered at trial if the opposing side has foreknowledge of it. Second, information may be difficult to share, even though a party wants to do that. For instance, a plaintiff might know that his business losses from a breach of contract will be high, but not be able

to demonstrate this during settlement negotiations (because, say, experts will have to be hired for trial to verify the losses). Another difficulty faced by a party who wants to reveal favorable information is that it may consist of the absence of unfavorable information. (For example, if the defendant was not drinking before a traffic accident, his favorable information may be the nonexistence of anyone who saw him drinking, and he may have no way to demonstrate this.137) Third, information may not be shared because it is unfavorable and the negative inference drawn from silence is not too strong. Note that the negative inference from silence will be weakened to the extent that some parties do not disclose favorable information for the first two reasons just given.

136Farber and White (1991) find that many malpractice cases settle after plaintiffs obtain information from defendants.

137If the case does not settle, the plaintiff may ultimately be able to verify the defendant’s claim implicitly: investigations may fail to locate any person who saw the defendant drinking (whereas if there really is a witness, there is some probability that the witness would be located)

5.4.8. Required disclosure of information — legal discovery. The courts may require that a litigant disclose certain information to the other side; this practice is known as discovery. It is commonly believed that discovery significantly increases the likelihood of settlement because it reduces differences in parties’ information. But, as just emphasized, there may well be substantial voluntary sharing of information, so the influence of compulsory disclosure will not be so great and is in fact nonexistent in a natural model of disclosure. See generally Shavell (1989a), and see also Hay (1994).138

Discovery will, nevertheless, tend to increase the rate of settlement and also will affect the terms of settlements. First, when parties would otherwise withhold favorable information to disable the opponent from countering it at trial, discovery will force disclosure, which in turn will

make settlement more likely. Second, when parties would otherwise withhold unfavorable information (because the negative inference from so doing would not be too strong), discovery will mandate disclosure and lead to settlement on less favorable terms. It should be noted, however, that such parties with unfavorable information would have settled in the absence of discovery.139 Settlement will increase overall because, when those with unfavorable information are required to disclose it, more generous offers will be made to those who remain silent in the face of discovery (perhaps those with favorable information who cannot verify the strength of their cases). Third, the prospect of legal sanctions for false statements may make more credible parties’ insistence that they lack certain unfavorable information (such as the assertion that there is no witness who could testify to the party having been drinking before an accident); this would encourage settlement of such cases.140

Discovery may also be used strategically. Obeying discovery requests is often expensive because significant time and resources may be needed to produce the desired information.

138Other models of discovery are Sobel (1989) and Mnookin and Wilson (1998); because these articles do not compare outcomes when there is discovery with outcomes with voluntary sharing of information, they are hard to interpret. See also Cooter and Rubinfeld (1994), which considers discovery, but in a model without an explicit treatment of asymmetric information.

139Those who withhold unfavorable information when there is no compulsory discovery seek to mimic others with favorable information who remain silent (whether because they strategically withhold information or because they cannot credibly verify their favorable situation). Accordingly, they receive settlement offers that reflect the average characteristics of the silent group; being those in the group with the least favorable cases, they will be the ones who settle, on terms that are better than they can expect if they were to disclose their unfavorable information. See Shavell (1989a).

140There may, however, be limitations on the feasibility of enforcement of discovery obligations. If a side fails to divulge unfavorable information, often this will not come to light (because the case may settle beforehand or because, even if there is a trial, the other side may never learn the truth in any event). Accordingly, very high sanctions for misrepresentations and possibly selective investigation (perhaps by the state) of the veracity of discovery responses may be necessary, although the present system does not follow either course.

5.4.9. Criminal adjudication. The analysis of suit and settlement for criminal adjudication (see, for example, Landes (1971) and the literature cited in section 6.3.8 on plea bargaining) is in some respects similar to that for civil adjudication, but there are differences in parties’ incentives that are worthy of note. First, in criminal cases the complaining party is a public prosecutor.

Accordingly, litigation decisions will not be based on a simple comparison of litigation costs and the expected gain because the prosecutor neither directly bears these costs nor benefits monetarily from winning (costs are borne by the state and there is no actual recovery). Instead, a prosecutor’s decisions will be dictated by the complex of factors determining his salary and his professional future. Second, a criminal defendant is often impecunious and will have been assigned a public defender. Not having to pay for his defense, such a defendant will not save legal expenses by settling, making him less willing to settle than otherwise. But those who serve as public defenders or are appointed to represent indigent defendants will often have limited budgets and receive low compensation, so they may exert less effort than what defendants would demand were they not liquidity constrained. Also, criminal defendants, and especially first-time defendants, may not care so much about the magnitude of punishment as about the fact of a criminal conviction or about having to spend some time incarcerated.

5.4.10. Additional aspects of legal procedure. There are many aspects of legal procedure that merit study but which we do not examine here, due mainly to their having received only limited treatment in the literature. Topics include the burden of proof,141 rules of evidence (and tribunals’ making inferences from evidence),142 the use of juries,143 the behavior of judges,144 summary adjudication,145 class actions,146 sequential versus joint adjudication of multiple issues in a single case,147 the sharing of liability among multiple defendants,148 and the advantages of the adversarial system of adjudication (in which each side substantially controls its litigation activity) versus the European inquisitorial model (in which the tribunal controls much litigation activity).149

141See Davis (1994), Hay and Spier (1997), Kaplow (1994a), Posner (1973), Rubinfeld and Sappington (1987), and Sobel (1985).

142See Daughety and Reinganum (1995), Froeb and Kobayashi (1996), Lewis and Poitevin (1997), Schrag and Scotchmer (1994), and Shavell (1989b).

143See Klevorick, Rothschild, and Winship (1984) (jury deliberations) and Schwartz and Schwartz (1996) (peremptory challenges).

144See Cohen (1992), Elder (1987), Higgins and Rubin (1980), Kimenyi et al. (1993), Kornhauser (1992a, 1992b), Posner (1993a), Ramseyer (1998), and Rasmusen (1994).

145See Posner (1986) on summary jury trials.

146See Che (1996), Dam (1975), and Miller (1998).

147See Landes (1993).

148See Easterbrook et al. (1980), Klerman (1996), Kornhauser and Revesz (1994), and Polinsky and Shavell (1981).

149See, for example, Langbein (1985) on the German system

5.5. Legal Advice

Because legal advice is costly, individuals must make decisions whether or not to obtain it, and questions about its social desirability also arise. In discussing the topic of legal advice, it is useful to consider separately ex ante legal advice — obtained when a party is contemplating an action — and ex post legal advice — secured after a party has acted or someone has been harmed, which is to say, at the stage of possible or actual litigation.

A notable difference between the types of advice is that ex ante advice can channel behavior directly in conformity with law, whereas ex post advice comes too late to accomplish that (although it has indirect effects on behavior). Ex ante legal advice was first studied from an economic perspective in Shavell (1988) and Kaplow and Shavell (1992); ex post legal advice was initially investigated from this standpoint in Kaplow and Shavell (1989, 1990).150

5.5.1. Ex ante legal advice: when acts are contemplated.

Advice has private value to a party who is considering taking some action with a possible legal consequence if the advice might lead him to alter his decision. The private value of legal advice is just an instance of the conventional definition of the expected value of information to a decisionmaker, as presented for instance in Raiffa (1968).

The social, as opposed to the private, value of ex ante legal advice inheres in the social desirability of advice-induced changes in parties’ behavior. In general, advice has positive social value because it promotes adherence to legal rules. The specific nature of the comparison between the social and the private values of legal advice depends on the form of liability. When liability is strict, the private value of legal advice is the same as its social value.

This basic

conclusion follows essentially because a party’s liability burden equals the harm he causes. When, however, liability is based on negligence, the private value of legal advice can be shown to exceed its social value. The explanation is in part that if a person avoids negligence because of advice, his liability saving will generally be larger than the reduction in expected harm he accomplishes, for he will escape liability entirely even though his non-negligent behavior might still cause harm.

150Legal advice was further studied in Bundy and Elhauge (1991, 1993) and Fischel (1998).

5.5.2. Ex post legal advice: at the stage of litigation. The private value of ex post legal advice resides in the possibility that the advice will lead a party to change his decisions about suit, settlement, and trial.

In considering the social value of ex post legal advice, observe first that because such advice is, by its nature, imparted to parties only after they have acted, it cannot have aided them initially in conforming with the law. A firm that does not know whether discharging a chemical waste into a river will violate an antipollution statute obviously cannot be led to behave appropriately by

learning what the law is after it decides about discharging the chemical. This simple but fundamental observation means that ex post advice does not raise social welfare in the direct way that ex ante advice does. Nonetheless, ex post advice certainly may influence behavior and social welfare. Ex post advice that defendants obtain in the course of a lawsuit may affect social welfare by lowering sanctions for those who knowingly violate the law, that is, ex post advice may dilute deterrence of undesirable conduct. Lawyers may lower expected sanctions by advantageous use of legal strategy and, importantly, by counseling defendants on the selection of evidence to present and to suppress. Given that individuals anticipate that their expected sanctions for causing harm will be reduced due to the subsequent availability of legal advice, fewer individuals will be deterred from engaging in undesirable behavior. Thus, legal advice may have negative social value, a point that was early emphasized by Bentham (1827). This reasoning, however, is incomplete, in part because the state may be able to raise overall sanctions to offset the dilution of deterrence due to advice.

5.5.3. Other aspects of legal advice. Subversion of the law. One issue that we have not mentioned is that advice may directly subvert the law. Lawyers may lower the effective magnitude of sanctions by helping clients to hide assets, and lawyers may also decrease the likelihood of sanctions if they have knowledge of enforcement strategies (such as how the tax authorities choose whom to audit). Of course, lawyers are not supposed to thwart law enforcement, but they have an economic incentive to do so and can fairly easily avoid punishment for it (lawyers give advice in private and can phrase their advice in hypothetical but readily understood terms). From the social perspective, legal advice that frustrates law enforcement I obviously undesirable.

Confidentiality of legal advice. The legal system protects the confidentiality of communications between lawyers and their clients under wide circumstances. Confidentiality of legal advice will benefit clients when there is a positive probability that disclosure of advice would lower its value. This would usually be true of advice about the selection of evidence to present in litigation: such advice generally would be robbed of effectiveness if it were disclosed to the opposing side and the court. Confidentiality is also of obvious importance to those obtaining

advice subversive of the law. By contrast, confidentiality often should not matter to parties obtaining advice about the legality of an act or about magnitude or likelihood of sanctions, because disclosure of such advice will usually not disadvantage them. (This is because they seek advice with the intention of following the law.151) Still, whatever is the character of strictly legal advice, maintaining the confidentiality of much business or personal information about clients themselves will frequently be of importance to the clients.

Because protection of confidentiality can benefit clients (and never is a disadvantage to them), it encourages clients to consult with and reveal information to their lawyers. This in itself is sometimes thought to imply that confidentiality is socially desirable. That reasoning, however, is mistaken: confidentiality is socially desirable only if the legal advice that confidentiality encourages is socially desirable, and as has been explained above, that may not be the case. Confidentiality of legal work product. The legal system also protects the confidentiality of legal work product (documents and other records of lawyers’ effort) that they generate on behalf  of clients in connection with litigation. The protection of work product is accomplished principally by denying opposing litigants the right to legal discovery of it.

As Easterbrook (1981) stressed, protection of work product encourages lawyers to engage in research on their clients’ cases, for much of the value of the research would be lost if it became immediately known to the other side. But whether protection of work product is socially desirable is not evident a priori; for it depends on whether or not the legal advice that the work product supports is socially desirable.152 A further complication is that, even when the advice is socially desirable, the private value of advice, and thus the amount of work product, may be socially excessive.

151See Shavell (1988).

152To illustrate, investigation may be necessary to determine or document facts that will improve accuracy, but the social value of greater accuracy may or may not exceed the cost of investigation. Also, with work product protection, two

parties may engage in duplicative efforts.

The appeals process — the process whereby a litigant disappointed with the decision of a first-order tribunal can seek reconsideration before a higher tribunal — is a widely observed feature of adjudication; in virtually all legal systems today, there exists a fairly general right of appeal of trial court decisions. An important social justification for the appeals process concerns correction of error. See Shavell (1995b). Suppose that litigants possess information about the occurrence of error and that appeals courts can frequently verify it. Then litigants may be induced to bring appeals when errors are likely to have been made but not otherwise, because of the costs of appeals. This

outcome may be fostered by imposition of fees for bringing appeals, so as to discourage appeal

when decisions were likely to have been correct.153 In other words, if there is an appropriate price for pursuing appeals, the appeals process can harness the information that litigants have about the occurrence of error and tend to remedy it. When this process functions well, appeals not only result in error correction, they also do so cheaply, for the legal system is burdened with reconsidering only the subset of cases in which errors were more probably made. This may render society’s investment in the appeals process inexpensive in comparison to the alternative it has of improving the accuracy of the trial process (by investing in the length and quality of trial court adjudication). Under that alternative

approach, extra expenditure would be required in all cases rather than only in the subset of cases that are appealed. The appeals process, in other words, may be an economical way of correcting error by taking advantage of litigants’ information that it has occurred.

5.7. Alternative Dispute Resolution

When parties need to resolve a dispute, they may turn not only to the state-sanctioned method of dispute resolution, namely, trial before a court, but also to arbitration and other forms of alternative dispute resolution (ADR).154 In examining ADR, it is helpful to distinguish between ex ante agreements to employ ADR — arrangements made before disputes arise — and ex post resort to ADR — use of ADR after disputes have arisen. See Shavell (1995a).155

5.7.1. Ex ante ADR agreements. Ex ante ADR agreements may be adopted because they are to the mutual benefit of the parties to a contract.156 In particular, ADR may lower the cost of resolving disputes or reduce risk.

153In fact, however, public fees for appeal are nominal, although private costs may be nontrivial.

154We focus on binding ADR; nonbinding ADR, such as mediation, is often used to foster settlement.

155On other issues raised by ADR, see Landes and Posner (1979). It should also be mentioned that there is a literature considering arbitration alone, not arbitration as an alternative to state-authorized litigation. See, for example, Ashenfelter (1989, 1992), Ashenfelter and Bloom (1984), and Farber (1980).

156We observe that to obtain many of the benefits noted in the text, the agreement to use ADR must be made ex ante; if the parties wait until a dispute arises, it will often be in the interest of one of the parties to refuse t accept ADR.

Second, ADR may engender superior incentives for the parties through greater accuracy of results. Suppose, for instance, that substandard performance of a contract would be correctly assessed by expert arbitrators under ADR but not by courts. Then the parties to the contract might well prefer to adopt ADR because it would induce good performance, thereby raising the willingness of the promisee to pay for the contract. Third, ADR may result in improved incentives to engage in disputes or to refrain from that. For example, it may be that the number of disputes brought under the legal process would be excessive, dissipating substantial resources of the parties without instigating mutually desirable changes in behavior; thus an ADR agreement that would serve to limit the number of disputes would be advantageous.

Because ex ante ADR agreements made by knowledgeable parties raise their well-being, it seems that ex ante ADR agreements should ordinarily be enforced by the legal system, as they are in fact. It is sometimes suggested that society should go further and subsidize ADR. A subsidy might be justified on second-best grounds, because the state already subsidizes ordinary litigation by not charging litigants for its full costs. It would seem, however, that the optimal solution is to remove the latter subsidy, unless it is justified on the ground of inadequate private incentives to sue.

5.7.2. Ex post ADR agreements. Parties will tend to make ex post ADR agreements in order to reduce dispute resolution costs and risk. On this account, ex post ADR would also tend to be socially desirable. A full evaluation of ex post ADR, however, must recognize other effects, notably, how the prospect that parties would adopt ADR ex post would affect their ex ante behavior. The proper analysis is similar to that bearing on the private versus the social value of settlements, in section

5.8. Formulation of Legal Rules

Economic analysis of the operation of the legal system often takes the legal rules being enforced as given. The formulation of legal rules itself, however, raises interesting economic issues.158 One issue concerns the optimal level of detail of rules. On one hand, greater detail allows better tailored control of behavior. On the other hand, greater detail involves higher compliance and litigation costs. Moreover, it cannot be assumed that parties will become informed of the precise content of more detailed rules. See Diver (1983), Ehrlich and Posner (1974), and Kaplow (1995a).159

157Indeed, parties’ adoption of ADR can be seen as a form of out-of-court settlement because use of ADR means that there will be no trial and instead the parties will be bound by the alternative they have chosen.

158For a survey of the literature, see Kaplow (1998).

159The subject of legal complexity has received particular attention in the context of the income tax. See, for example, Blumenthal and Slemrod (1992), and Kaplow (1996).

Another issue is whether rules should be formulated fully ex ante, or instead should be incompletely specified initially and fully articulated only ex post, during adjudication of particular disputes. Fuller ex ante specification is more costly for the state, but may provide greater predictability for parties and hence induce better behavior, and it also may reduce adjudication costs. See Diver (1983), Ehrlich and Posner (1974), and Kaplow (1992c). Full ex ante specification of legal rules tends to be advantageous when the governed behavior is frequent and has common characteristics, essentially because of economies of scale (the rule is formulated only once). A closely related subject is the issuance of precedents by courts; for example, major disagreements about issuing precedents concern the degree to which details of rulings beyond those necessary to decide the case before a court should be specified and when courts should take the opportunity to announce new legal rules or modify existing ones.160 Additional issues are presented by the frequent need to modify legal rules. New rules, if fully and immediately applicable, will typically affect the returns to previous investments. The prospect of such application of new rules imposes risk on actors and also affects their investment decisions, but the latter effect tends to be efficient when the legal reform reflects certain economically relevant information or changed circumstances.161 See generally Kaplow (1986a, 1992a).

5.9. Relevance to General Incentive Schemes

In closing, we suggest that the topic of the operation of the legal system should in substantial respects be viewed as a basic one in the theory of incentives. This is because incentive schemes often require that parties come before authorities who apply rules — that is, the incentive schemes — and this adjudication process is costly. (Even if the adjudication is informal, it will involve expense.) Therefore, many of the issues that we addressed are of relevance. Notably, questions arise concerning private versus system-appropriate motives to come before authorities who apply rules, for individuals will not take into account total adjudication costs nor the incentive effects of adjudication (such as whether the bringing of employee complaints will induce better behavior in a firm). Also, many of the more particular issues that we considered about litigation and the design of legal procedure — including settlement, discovery, and appeals — have general analogs in other incentive systems.