Legal liability for accidents (a branch of tort law) is a means by which society can reduce the risk of harm by threatening potential injurers with having to pay for the harms they cause. Liability is also frequently viewed as a device for compensating victims of harm, but we will emphasize that insurance can provide compensation more cheaply than the liability system. Thus,

we will view the primary social function of the liability system as the provision of incentives to prevent harm. There are two basic rules of liability. Under strict liability, an injurer must always pay for harm due to an accident that he causes.

Under the negligence rule, an injurer must pay for harm caused only when he is found negligent, that is, only when his level of care was less than a standard of care chosen by the courts, often referred to as due care. (There are various versions of these rules that depend on whether victims’ care was insufficient, as we will discuss below.) In fact, the negligence rule is the dominant form of liability; strict liability is reserved mainly for certain especially dangerous activities (such as the use of explosives). Our discussion of liability begins by examining how liability rules create incentives to reduce risk. The allocation of risk and insurance will then be considered, and following that, the factor of

administrative costs. Then we take up a number of important topics bearing on liability: the magnitude of liability (damages), causation, and the judgment proof problem (assets insufficient to pay for harm). Finally, we consider the subjects of product liabilIty and intentional torts.3

2.1. Incentives

In order to focus on liability and incentives to reduce risk, we assume in this section that parties are risk neutral. Further, we suppose that there are two classes of parties, injurers and victims, and that they are strangers to one another, or at least are not in a contractual relationship.

3A comprehensive economic treatment of accident law is contained in Shavell (1987a), which this section largely follows. See also Landes and Posner (1987a) and Calabresi (1970), an early, innovative, informal economic analysis of

liability. Under the negligence rule, suppose that the due care level x^ is set equal to x*, meaning that an injurer will have to pay h if x < x* but will not have to pay anything if x x*. Then it can be shown that the injurer will choose x*: clearly, the injurer will not choose x greater than x*, for that will cost him more and he will escape liability by choosing merely x*; and he will not choose x < x*, for then he will be liable (in which case the analysis of strict liability shows that he would not choose x < x*).

For example, injurers might be drivers and victims pedestrians, or injurers might be polluting firms and victims affected residents. To begin with, we assume that accidents are unilateral in nature: only injurers can influence risks. Then we consider bilateral accidents, in which victims as well as injurers affect risks. We

also examine two types of action that parties can take that alter risk: first we consider their level of care (such as driving speed) and then their level of activity (number of miles driven).

2.1.1. Unilateral accidents and the level of care. Here we suppose that injurers alone can reduce risk by choosing a level of care. Let x be expenditures on care (or the money value of effort devoted to it) and p(x) be the probability of an accident that causes harm h, where p is declining in x. Assume that the social objective is to minimize total expected costs, x + p(x)h, and let x* denote the optimal x. Under strict liability, injurers pay damages equal to h whenever an accident occurs, and they naturally bear the cost of care x. Thus, they minimize x + p(x)h; accordingly, they choose x*. Thus, under both forms of liability, injurers are led to take optimal care. But note that under the negligence rule, courts need to be able to calculate optimal care x* and to be able to observe

actual care x, in addition to observing harm. In contrast, under strict liability courts do not need to do the former two; they only need to observe harm.4

It should also be noted that, under the negligence rule with due care x^ equal to x*, negligence would never actually be found, because injurers are induced to choose x* and thus would be exonerated if they were sued after causing an accident. Findings of negligence may occur, however, under a variety of modifications of our assumptions. Courts might make errors in observing injurers’ actual level of care so that an injurer whose true x is at least x* might

mistakenly be found negligent because his observed level of care is below x*. Similarly, courts might err in calculating x* and thus might set due care x^ above x*. If so, an injurer who chooses x* would be found negligent (even though care is accurately observed) because x^ exceeds x*. As emphasized by Craswell and Calfee (1986), the possibility of errors in the negligence   determination leads injurers to choose incorrect levels of care; one possibility is that they would take excessive care in order to reduce the risk of being found negligent by mistake.5 There exist other explanations for findings of negligence as well, including the possibility that individuals may not know x* and thus take too little care, the judgment-proof problem, which may lead an actor to choose to be negligent (see section 2.6), and the inability of a person to control his behavior perfectly at every moment or of a firm to control its employees.

4Compare the discussion of corrective taxes versus regulation in section 3.6.2.

5This might explain the phenomenon of “defensive medicine,” on which see Danzon (1985) and, for empirical

evidence, Kessler and McClellan (1996). Whether there is a tendency toward excessive care depends upon the degree of

legal error and on whether injurers who are found negligent are held responsible for all harm caused or only the incremental

harm attributable to their negligence. On the latter, see Grady (1983) and Kahan (1989).

2.1.2. Bilateral accidents and levels of care. We now assume that victims also choose a level of care y, that the probability of an accident is p(x,y) and is declining in both variables, that the social goal is to minimize x + y + p(x,y)h, and that the optimal levels of care x* and y* are positive.6 Under strict liability, injurers’ incentives are optimal conditional on victims’ level of care, but victims have no incentive to take care because they are fully compensated for their losses. However, the usual strict liability rule that applies in bilateral situations is strict liability with a defense of contributory negligence, meaning that an injurer is liable for harm only if the victim’s level of care was not negligent, that is, his level of care was at least his due care level y^. If victims’ due care level is set by the courts to equal y*, then it is a unique equilibrium for both injurers and victims to act optimally: victims can be shown to choose y* in order to avoid having to bear their losses, and injurers will choose x* since they will in fact be liable, as victims will not be negligent.7

Under the negligence rule, optimal behavior, x* and y*, is also the unique equilibrium. Injurers can be shown to choose x* to avoid being liable, and since victims will therefore bear their losses, they will choose y*.8 Two other variants of the negligence rule are negligence with the defense of contributory negligence (under which a negligent injurer is liable only if the victim is not negligent) and the comparative negligence rule (under which a negligent injurer is only partially liable if the victim is also negligent). These rules are also readily shown to induce optimal behavior in equilibrium. Thus, all of the negligence rules, and strict liability with the defense of contributory

negligence, support optimal levels of care x* and y* in equilibrium, assuming that due care levels are chosen optimally. Courts need to be able to calculate optimal care levels for at least one party under any of the rules, and in general this requires knowledge of the function p(x,y). The main conclusions of this and the last section were first proved by Brown (1973).9

6In some early, less formal literature on accidents, for example, Calabresi (1970), reference is made to the notion of the “least-cost avoider,” the party — injurer or victim — who can avoid an accident at the lower cost. The idea of a leastcost

avoider relies on the assumption that each party can undertake a discrete amount of care, independently sufficient to

prevent an accident.

7That this equilibrium is unique follows from three observations: (1) Victims never have an incentive to take care y

exceeding y* (for once they take due care they will be compensated for their losses). (2) Victims will not choose y less than

y*, for if they do so, they will bear their own losses, injurers will take no care, and victims thus will minimize y + p(0,y)h.

But y + p(0,y)h = 0 + y + p(0,y)h > x* + y* + p(x*,y*)h > y*, implying that victims must be better off choosing due care y*

than any y < y*. (3) Because in equilibrium victims thus take due care, injurers choose x to minimize x + p(x,y*)h, which is

minimized at x*

.8Uniqueness is demonstrated by the following: (1) Injurers will not take excessive care in equilibrium. (2) If injurers take inadequate care, victims will take no care, so injurers will minimize x + p(x,0)h, which exceeds x* + y* +  p(x*,y*)h, which exceeds x*. Thus, injurers are better off taking care of x*. (3) Since injurers must choose x* in equilibrium, victims will choose y*.

9Diamond (1974) proved closely related results shortly afterward. See also Green (1976), who focuses on the case of heterogeneous injurers and victims. The failure of the negligence rule to control the level of activity arises because negligence is defined here (and for the most part in reality) in terms of care alone

2.1.3. Unilateral accidents, level of care, and level of activity Now let us reconsider unilateral accidents, allowing for injurers to choose their level of activity z, which is interpreted as the number of times they engage in their activity (or if injurers are firms, the scale of their output). Let b(z) be the benefit (or profit) from the activity, and assume the social object is to maximize b(z) – z(x + p(x)h); here x + p(x)h is assumed to be the cost of care and expected harm each time an injurer engages in his activity. Let x* and z* be optimal values. Note that x* minimizes x + p(x)h, so x* is as described above in section 2.1.1, and z* is determined by b(z) = x* + p(x*)h, which is to say, the marginal benefit from the activity equals the marginal social cost, comprising the sum of the cost of optimal care and expected accident losses (given optimal care).

Under strict liability, an injurer will choose both the level of care and the level of activity optimally, as his object will be the same as the social objective, to maximize b(z) – z(x + p(x)h), because damage payments equal h whenever harm occurs. Under the negligence rule, an injurer will choose optimal care x* as before, but his level of activity z will be socially excessive. In particular, because an injurer will escape liability by taking care of x*, he will choose z to maximize b(z) – zx*, so that z will satisfy b(z) = x*. The injurer’s cost of raising his level of activity is only his cost of care x*, which is less than the social cost, as it also includes p(x*)h. The excessive level of activity under the negligence rule will be more important the larger is expected harm p(x*)h from the activity.

2.1.4. Bilateral accidents, levels of care, and levels of activity. If we consider levels of care and of activity for both injurers and victims, then none of the liability rules that we have considered leads to full optimality (assuming that activity levels are unobservable). As just explained, the negligence rule induces injurers to engage excessively in their activity. Similarly, strict liability with a defense of contributory negligence leads victims to engage excessively in their

activity (the number of times they expose themselves to risk), as they do not bear their losses given that they take due care. The reason that full optimality cannot be achieved is in essence that injurers must bear accident losses to induce them to choose the right level of their activity, but this means that victims will not choose the optimal level of their activity, and conversely.10 The distinction between levels of care and levels of activity was first emphasized in Shavell (1980c), where the results of this and the last section were shown.

2.1.5. Empirical evidence on the effect of liability on safety. Only a modest amount of empirical work has been undertaken on the effect of liability on accident risks. See Dewees, Duff, and Trebilcock (1996) for a general survey of the literature that exists, and, among others, Devlin (1990), Landes (1982), and Sloan (1994) on liability and auto accidents, Danzon (1985) and Kessler and McClellan (1996) on liability and adverse medical outcomes, and Higgins (1978), Priest (1988), and Viscusi (1991) on liability and product safety.

2.2. Risk-bearing and Insurance

We consider next the implications of risk aversion and the role of insurance in the liability system, on which see Shavell (1982a). Several general points may be made.

First, the socially optimal resolution of the accident problem obviously now involves not only the reduction of losses from accidents, but also the protection of risk-averse parties against risk. Note that risk bearing is relevant for two reasons: not only because potential victims may face the risk of accident losses, but also because potential injurers may face the risk of liability. The former risk can be mitigated through first-party insurance, and the latter through liability insurance.

Second, because risk-averse individuals will tend to purchase insurance, the incentives associated with liability do not function in the direct way discussed in the last section, but instead  are mediated by the terms of insurance policies. To illustrate, consider strict liability in the unilateral accident model with care alone allowed to vary, and assume that insurance is sold at actuarially fair rates. If injurers are risk averse and liability insurers can observe their levels of care, injurers will purchase full liability insurance coverage and their premiums will depend on their level of care; their premiums will equal p(x)h. Thus, injurers will want to minimize their costs of care plus premiums, or x + p(x)h, so they will choose the optimal level of care x*. In this instance, liability insurance eliminates risk for injurers, and the situation reduces to the previously analyzed risk-neutral case. 10However, there exist ways to induce fully optimal behavior using tools other than conventional liability rules.

For example, if injurers have to pay the state for harm and victims bear their own losses, both victims and injurers will choose levels of care and of activity optimally. On the possibility of such decoupling of what injurers pay and what victims receive, see note 102.

If, however, liability insurers cannot observe levels of care, ownership of full coverage could create severe moral hazard, so would not be purchased. Instead, as we know from the theory of insurance, the typical amount of coverage purchased will be partial, for that leaves injurers an incentive to reduce risk. In this case, therefore, the liability rule results in some direct incentive to

take care because injurers are left bearing some risk after their purchase of liability insurance. But injurers’ level of care will still tend to be less than first best.

This last situation in which liability insurance dilutes incentives leads to our third point, concerning the question whether the sale of liability insurance is socially desirable. (We note that because of fears about incentives, the sale of liability insurance was delayed for decades in many countries and that it was not allowed in the former Soviet Union; further, in this country liability insurance is sometimes forbidden against certain types of liability, such as against punitive damages.) The answer to the question is that sale of liability insurance is socially desirable, at least in basic models of accidents and some variations of them. In the case just considered, the reason is evident. Injurers are made better off by the presence of liability insurance, as they choose to purchase it. Victims are indifferent to its purchase by injurers because victims are fully

compensated under strict liability for any losses they sustain. In particular, it does not matter to victims that the likelihood of accident may rise due to the sale of liability insurance. This argument must be modified in other cases, such as when the damages injurers pay are less than harm. In that circumstance, the sale of liability insurance may not be socially desirable. See section 2.6.

Fourth, consider how the comparison between strict liability and the negligence rule is affected by considerations of risk-bearing. It is true that the immediate effect of strict liability is to shift the risk of loss from victims to injurers, whereas the immediate effect of the negligence rule is to leave the risk on victims (injurers will tend to act non-negligently). However, the presence of insurance means that victims and injurers can substantially shield themselves from

risk. Of course, as was just discussed, insurance coverage may be incomplete due to moral hazard; this makes risk-bearing of some relevance to the comparison of liability rules, but which rule becomes more favorable is not obvious. Finally, as we stated at the outset of section 2, the presence of insurance implies that the liability system cannot be justified primarily as a means of compensating risk-averse victims against loss. Rather, the justification for the liability system must lie in significant part in the incentives that it creates to reduce risk. To amplify, although both the liability system and the

insurance system can compensate victims, the liability system is much more expensive than the insurance system (see the next section).11 Accordingly, were there no social need to create incentives to reduce risk, it would be best to dispense with the liability system and to rely on insurance to accomplish compensation.12

11Also, victim compensation through liability generally implies that possibly risk-averse injurers bear risk. In some contexts, such as auto accidents, one supposes that injurers are not substantially less risk-averse than victims.

12Some jurisdictions have implemented “no-fault” (essentially, first-party insurance) regimes for automobile accidents. See Dewees, Duff, and Trebilcock (1996). Also, there are intermediate schemes, like workers’ compensation, that provide compensation and charge experience-rated premiums to injurers to instill incentives to reduce risk. See Moore and Viscusi (1990)

2.3. Administrative Costs

2.3.1. Administrative costs of the liability system. The administrative costs of the liability system are the legal and other costs (notably the time of litigants) involved in bringing suit and resolving it through settlement or trial. These costs are substantial; a number of estimates suggest that on average, administrative costs of a dollar or more are incurred for every dollar that a victim receives through the liability system. In contrast, the administrative cost of receiving a dollar through the insurance system is often below fifteen cents.13The factor of administrative costs affects the comparison between the forms of liability. On one hand, we would expect the volume of cases — and thus administrative costs — to be higher under strict liability than under the negligence rule. This is because, under strict liability, a victim can collect whether or not the injurer was at fault, whereas under the negligence rule fault must be established, so that in many cases of accident there will be no suit or, if there is a suit, it will be dropped after little has been spent.14 On the other hand, given that there is a case, we would anticipate administrative costs to be higher under the negligence rule than under strict liability, because under the negligence rule due care will be at issue.

2.3.2. Administrative costs and the social desirability of the liability system. The existence of administrative costs and their significant magnitude raises rather sharply the question whether it is worthwhile for society to bear them to gain the benefits of the liability system — the incentives to reduce risk. Unfortunately, it is quite possible for suits to be attractive for private parties to bring even if the social benefits of the liability system are small and make it socially undesirable.

2.4. Magnitude of Liability: Damages

The magnitude of the payment a liable party must make is known as damages, because it is normally set equal to the harm the victim has sustained. In this section, we discuss various issues relating to damages.

2.4.1. Basic theory. As a general matter, damages should equal harm under strict liability for incentives to be optimal in the unilateral model of accidents. Clearly, for injurers to be led to choose optimal levels of care, their expected liability must equal expected harm p(x)h, meaning that damages d should equal h. Likewise, for their levels of activity to be optimal, the same must be true.15

We should add that this point essentially carries over to the situation, not yet considered, where the magnitude of harm is stochastic.

14Farber and White (1991) provide evidence that many medical malpractice cases are dropped after discovery, when plaintiffs learn that the defendant probably was not negligent. Relatedly, Ordover (1978) analyzes a model in which victims

are uncertain about injurers’ negligence; the result is that some victims of negligence may not sue and others who are not victims of negligence might sue.

15In the bilateral model, damages equal to harm would also be optimal under a rule of strict liability with a defense of contributory negligence if victims’ activity level is not variable. If their activity level is variable, then optimal damages may well be less than harm, for this will induce victims to moderate their level of activity.

Under the negligence rule, analysis of the optimal magnitude of damages is somewhat different. Recall that if damages equal harm h, injurers will be induced to take care of x* (assuming that due care x^ = x*). It is also the case that damages higher than h would induce

injurers to take care of x*: this will increase the incentive to be non-negligent, to choose x*, but it will not lead injurers to take excessive care because they can escape liability merely by taking care of x*. Moreover, it can be shown that damages somewhat below h will also induce due care because, by taking due care rather than slightly less care under the negligence rule, injurers do not

just reduce liability slightly but avoid liability altogether.16 Thus, optimal damages are not unique but range from a level somewhat below h to any greater level. When, however, one introduces the possibility of uncertainty in the negligence determination (see section 2.1.1), the situation becomes more complicated. For example, we noted that error in the negligence determination

might lead injurers to take excessive care to reduce the risk of being found negligent by mistake. If so, a level of damages exceeding h would only exacerbate this problem, and it might be beneficial for d to be lower than h.17

To sum up, we can say that in simple cases damages should equal harm under strict liability and under the negligence rule, although there are complications, such as that concerning uncertainty in the negligence determination. In fact, the law generally does impose damages equal nto harm, but subject to some exceptions (which we will note in sections 2.4.3 and 2.4.5).

2.4.2. Nonpecuniary elements of loss. Accidents often involve nonpecuniary losses, such as pain and suffering. To provide injurers with proper incentives to reduce accidents, they should pay for all nonpecuniary harms that they cause. However, it may be better for the state to receive these payments than for victims to receive them. Victims would often not elect to insure against

nonpecuniary losses because these losses would not create a need for money, that is, raise their marginal utility of wealth.18 Parents usually would not insure against the death of a child, for example, as this frequently would not generate a need for money, however devastating the loss would be for the parents. Thus, as initially proposed by Spence (1977), liability for pecuniary losses accompanied by an appropriate fine for nonpecuniary losses may be socially desirable.19

16 This point depends upon the particular formulation of the negligence rule (whether a person who takes less than due care is responsible for all harm caused or only the increment to harm resulting from x falling below x^). See Kahan (1989).

17An additional issue is that erroneous findings of liability tend to remedy the problem of excessive levels of activity under the negligence rule, raising the possibility that setting damages above h would be desirable.

18 For empirical evidence, see Viscusi and Evans (1990).

19 Alternatively, victims might enter into contracts under which insurers would receive pain and suffering recoveries in exchange for a reduction in premiums on other coverage.

2.4.3. Punitive damages. When an injurer’s behavior departs substantially from what is appropriate, damages in excess of harm, so-called punitive damages, may be imposed. If imposition of such damages causes expected liability to exceed expected harm, injurers will be induced to take excessive precautions, at least under strict liability, and they will also reduce their levels of activity undesirably.20

higher should be damages when the party is found liable. Accordingly, a firm that dumps toxic wastes at night, or an individual who tries to conceal a bad act, should have to pay punitive damages, but not an injurer who in a noticeable way causes harm. On these points and others, see, for example, Cooter (1989), Diamond (1997), and Polinsky and Shavell (1998a).

2.4.4. Accuracy of damages. Much expense is incurred in litigation about the magnitude of a victim’s harm, which raises the question of what the social value of greater accuracy is and whether the private value of accuracy is different from the social value. As stressed in Kaplow and Shavell (1996b), the private value of accuracy about harm generally exceeds the social value.

To explain, there is social value in establishing harm accurately if and only if injurers know the harm they might cause when they choose their level of care. For example, if an injurer anticipates that the atypically large harm he might cause will be accurately measured, he will exercise an appropriately high degree of care, as is socially desirable.21 However, injurers often lack (and could not reasonably obtain) considerable information about the harm they might cause when they decide on their precautions. Drivers, for example, know relatively little about how much harm a potential victim would suffer in an accident (the seriousness of injuries, the magnitude of lost earnings). Thus, drivers’ incentives to avoid accidents would be largely the same if, instead of using precise measurements of harm, courts employed rough averages (based, perhaps, upon

abbreviated litigation over damages or upon figures from a table). Nevertheless, victims and injurers have very strong incentives to spend to establish damages accurately in court. A victim will always be willing to spend up to a dollar to prove that harm is a dollar higher, and an injurer will always be willing to spend up to a dollar to prove that harm is a dollar lower.

20 Under a perfectly operating negligence rule, punitive damages would not affect injurers’ behavior, as explained in

section 2.4.1. But if there is uncertainty in the negligence determination, the problem of excessive precautions may be

exacerbated by punitive damages; also, punitive damages may reduce injurers’ activity levels (although this effect may be


21Relatedly, a prospective injurer’s incentive to acquire information about the harm he may cause (whether it will be atypically large) will be greater when he knows that harm will be accurately determined.

2.4.5. Components of loss that are difficult to estimate.

Some components of loss are hard to estimate, for example, the decline in profits caused by a fire at a store (as opposed to the cost of repairing the store) or certain non npecuniary harms, and the law sometimes excludes such difficult-to-measure elements of loss from damages. This legal policy might be justified when the cost of ascertaining a component of loss outweighs the value of the improvement in incentives that its inclusion would accomplish. However, the cost of estimating a component of loss would be low if rough estimates were used (and the analysis of the last section suggests that this often would not much compromise incentives to reduce risk). Therefore, the policy of excluding components of loss that are hard to evaluate may be unwarranted.

2.5. Causation

2.5.1. Basic requirement of causation. A fundamental principle of liability law is that a party cannot held liable unless he was the cause of losses. For example, if cancer occurs in an area where a firm has polluted, the firm will in principle be liable only for the cancer that it caused, not for cancer due to other carcinogens.

This principle is clearly necessary to achieve social efficiency under strict liability, because otherwise incentives would be distorted. Socially desirable production might be rendered unprofitable if the firm were held responsible for all cases of cancer.

Under the negligence rule, restricting liability to accidents caused by an actor may be less important than under strict liability: if negligent actors were held liable for harms they did not cause, they would only have greater reason to act non-negligently, but would not take excessive precautions if there were no uncertainty surrounding the negligence determination. In the

presence of such uncertainty, however, relaxation of the causation requirement might adversely affect incentives. Further, under both liability rules, absence of the causation requirement might raise the volume of litigation and thus administrative costs. On the basic causation requirement and incentives, see originally Calabresi (1975) and Shavell (1980a).22

22On the causation requirement under the negligence rule, see also Grady (1983) and Kahan (1989), who study restriction of liability to losses that are in excess of the possibly positive losses that the actor would have caused had he not been negligent. The possibility that a party would not be said to be the proximate cause of losses on account of coincidence (as opposed to the freak character of losses) is illustrated by the following case: a

2.5.2. Uncertainty over causation. In many situations there is uncertainty about causation. For example, it may not be known which manufacturer out of many sold the product that resulted in injury, or whether harm was due to the defendant firm or to background factors (was cancer attributable to a firm’s pollutant or to unknown environmental carcinogens?). The traditional approach of the law is to hold a defendant liable if and only if the probability that the defendant was the cause of losses exceeds 50%. This approach can lead either to inadequate or to excessive incentives to reduce risk. For example, a firm that supplies only 20% of the market demand will escape liability for any harm caused by its product (assuming that harm cannot be traced to particular firms). Consequently, the firm will have no liability-related incentive to take precautions. If, however, a firm’s market share exceeds 50%, the firm will be liable for all harms due to the product that it and other firms sell, for it will always be correctly said to be more likely than not the cause of harm. Thus, the firm’s liability burden will be socially excessive (under strict liability). These potential problems of inadequate and of excessive incentives may arise under any liability criterion based on a threshold probability of causation; they are not unique to a 50% threshold. Essentially this point has been made frequently, and it is formally developed in Shavell (1985b).

The legal system has recently adopted (in limited settings) the approach of imposing liability in proportion to the likelihood of causation. Under this approach, a firm supplying 20% of the market would be liable for 20% of harm in every case. Note, therefore, that the firm’s liability bill would be the same under this regime as it would be if it paid for all the harm in the 20% of cases it truly caused — implying that its incentives would be socially appropriate. That the proportional liability principle engenders optimal incentives (without there being a need to establish causation in particular cases) is an advantage of the principle relative to the traditional threshold probability criterion. See Rosenberg (1984) and Shavell (1985b).

2.5.3. Proximate causation. Even if a party is a cause of losses, he may still escape liability under tort law because he was not the “proximate cause” of losses, where proximately-caused losses are, mainly, those that came about in an ordinary manner and that were not the product of coincidence. Liability often is not found for freak accidents, such as where a dog imbibes nitroglycerin left at a mining site and then explodes, injuring nearby persons. Allowing parties to

escape liability for such unusual accidents is sometimes thought not to undermine incentives, on the ground that no one could have foreseen such accidents. This argument, however, is subject to the criticism that courts may find it difficult to discriminate between accidents that can and cannot

be foreseen. Moreover, the argument leads to the reductio ad absurdum that there should never be liability: any accident may be viewed as extraordinarily unlikely (of essentially zero probability) if it is described in sufficient detail. speeding bus happened to be at just the “right” point on its route to be struck by a falling tree; the bus company escaped liability for the injuries to passengers even though they would not have occurred but for the excessive speed of the bus. Allowing parties to escape liability for such coincidental accidents might not affect precautions, however. One presumes that the probability of a bus being struck by a falling tree is independent of its speed, so that imposing liability would not affect the speed at which buses are driven. On proximate causation, see Calabresi (1975) and Shavell (1980a).

2.6. Judgment-Proof Problem

The possibility that injurers may not be able to pay in full for the harm they cause is known as the judgment-proof problem and is of substantial importance, for individuals and firms often pose risks significantly exceeding their assets (a person of modest means could cause a devastating fire; a small firm’s product could cause many deaths).

First, if there is another party who has some control over the behavior of

the party whose assets are limited, then the former party can be held vicariously liable for the losses caused by the latter. Thus, holding a large contractor liable for the accidents caused by a small subcontractor or an employer for accidents caused by its employees will induce the former to control the risks posed by the latter.23 See Kornhauser (1982) and Sykes (1981).

Second, parties with assets less than a specified amount could in some contexts be prevented from engaging in an activity. However, such minimum asset requirements are a somewhat blunt instrument for alleviating the incentive problems under consideration. A third response to inadequate incentives, one closely related to asset requirements, is regulation of liability insurance. See Shavell (forthcoming). One form of insurance regulation would mandate purchase of (perhaps full) coverage.24 This approach would be especially appealing when insurers can observe the precautions taken by injurers. An opposite form of insurance regulation would prohibit purchase of liability insurance. This could improve incentives to take care if insurers cannot observe injurers’ precautions, because in that case insurance coverage would dilute incentives to take care when these incentives are inadequate to begin with.

Imposing liability on corporations for behavior of their judgment-proof subsidiaries or requiring that liability of shareholders be unlimited (at least with respect to third-party tort victims) might serve a similar function. See Hansmann and Kraakman (1991). Also, liability might be imposed on parties who supply services to potentially judgment-proof entities and are in a position to monitor them, such as accountants, lawyers, and lenders. See Kraakman (1986) and Pitchford (1995).

Many jurisdictions require liability insurance of those who drive, although the required amount is usually small perhaps $20,000); the result is that there is an excessive incentive to drive (which may be greatest among the young, who also impose the largest risks). Fourth, the use of Pigouvian taxes equal to expected harm may help to alleviate the judgment-proof problem. When harm will be caused with a low probability, the expected harm will be much less than actual harm; hence, parties with limited assets may be able to pay the appropriate tax on risk-creating behavior even though they could not pay for the harm itself.

A fifth way of correcting for dilution of incentives is for the state to regulate parties’ behavior directly, such as with traffic laws or by insisting that food and drugs meet certain safety requirements. Regulation, however, may involve inefficiency because regulators’ limited knowledge of risk and the cost and ability to reduce it. (We discuss regulation further in the section 3.6.2.)

A final way of mitigating dilution of incentives is resort to criminal liability. A party who would not take care if only his assets were at stake might be induced to do so for fear of imprisonment.

2.7. Product Liability

We have not yet considered accidents where the victims are customers of the injurer (or more generally where victims are in some contractual relationship with injurers). In this case, the role of liability in providing incentives may be attenuated or even nonexistent. The reason, obviously, is that firms producing risky products may be unable to sell them or may have to accept a reduction in price commensurate with the risk of loss attaching to the products.

If customer knowledge of product risk is perfect, then firms’ incentives to reduce risk will be optimal even in the absence of liability. For example, if the expected losses caused by a product risk are $100, the firm will have to accept a $100 lower price than otherwise, so it will be willing to spend up to $100 to eliminate the risk. Therefore, liability is not needed to generate incentives toward safety.

If, however, customer knowledge of risk is imperfect, liability is potentially useful in reducing risk.25 In the absence of liability, firms that increase safety generally will be unable to obtain an increase in the price fully reflecting the reduction in risk.26 (Indeed, in the extreme case where customers cannot observe anything about the true risk, firms would have no incentive to reduce

it.) Therefore, the prospect of liability for product-caused harms will increase incentives to reduce risk. Also, imposing liability will result in prices that reflect the full costs of products, leading to more efficient purchasing decisions.

A question concerning liability is whether court-determined liability or customer-selected liability, namely, warranties, is likely to be better.27 The answer depends on the nature of customers’ information or lack thereof and on other factors.

25See, for example, Goldberg (1974). Another potentially useful policy is supplying information about risk to customers, on which see Magat and Viscusi (1992) and Viscusi and Magat (1987).

26If a sufficient fraction of customers are informed about risk, producers may be induced to offer better products to all customers. See, for example, Schwartz and Wilde (1979).

27See generally Priest (1981) and Rubin (1993).

Suppose instead that customers misperceive risk. Then their selection of warranties may be skewed, as emphasized by Spence (1977). For example, if customers believe the risk of a product failure causing a loss of $10,000 to be 1% when it is really 5%, then a warranty would not be purchased: a seller of a full warranty would have to charge $500 for it, but the perceived expected value of it would be only $100. In this circumstance, it might be better for the courts to impose liability because that would create incentives to reduce risk.

producer liability might dilute their incentives to do so (assuming that defenses such as contributory negligence are unsuccessful because of difficulties in observing customers’ behavior). Also, the administrative costs of liability are high. Thus, whether imposition of liability will improve social welfare, given customers’ ability to purchase warranties, involves a complicated weighing of considerations, and courts’ ability to do this is not clear.

2.8. Liability versus Other Means of Controlling Accidents

Liability is only one means of controlling harm-causing behavior; safety regulation and Pigouvian taxes are among the alternatives, as we indicated in section 2.6. For a general comparison of methods of controlling harm, see our discussion of regulating externalities in section 3.6.2.

2.9. Intentional Torts

To this point, we have examined liability for accidents, but we have not dealt explicitly with so-called intentional torts, such as assaulting someone or stealing his property (which also are crimes).28 See Landes and Posner (1981). An intentional tort may be defined as a harm than an injurer causes in which either of two things are true: the injurer acted in a manner that caused harm to occur with a very high probability, or the injurer obtained utility from the victim’s suffering itself. It would be possible to apply the foregoing analysis of accidents to intentional torts without modification. The conclusions reached did not depend on the magnitude of the probability of harm or on the source of benefits to injurers. However, both of these aspects of intentional harms suggest changes in assumptions that could alter our analysis and conclusions.

First, in situations where harm would be very likely to occur, bargaining between injurers and victims would often be possible.29 If so, it may be desirable to forbid injurers from harming victims unless they obtain consent in advance, presumably in exchange for payment.

Second, where injurers derive utility directly from the fact that harm is suffered by victims, some analysts suggest that injurers’ utility should not count in assessing social welfare. If so, deterrence becomes more valuable. Also, it may be optimal to deter some harmful acts even when the injurer’s benefit exceeds the victim’s loss, which calls for damages greater than harm, or for supplemental sanctions, notably imprisonment. (One suspects, however, that with most such intentional torts, injurers’ benefits rarely exceed victims’ losses.)