首页 | 农村土地改革 | 法经济学茶座 | 法经济学论坛
     网站首页≯ 侵权法≯
因果关系和可预见性

 CAUSATION AND FORSEEABILITY

Omri Ben-Shahar

Tel Aviv University

© Copyright 1999 Omri Ben-Shahar

 

Abstract

This chapter begins with a survey of the implicit role of causation in the

writings of the early, pathbreaking economic analysts of tort law. It then

clarifies the basic distinction between retrospective (ex post) causation and

prospective (ex ante) causation, a distinction that forms the core of many

subsequent economic discussions of causation. Next, the explicit role of

causation doctrines in inducing optimal care and activity levels is examined

under the strict liability and the negligence regimes. The analysis is then

extended to cover several complications often plaguing the determination of

causation: uncertainty over causation, joint actions among tortfeasors and

unforeseeability of harm.

JEL classification: K13

Keywords: Foreseeability, Retrospective and Prospective Causation, Optimal

Care, Strict Liability, Joint Actions

 

1. Introduction

 

The contribution of economic analysis to the clearer understanding of the

function of law is particularly evident in the law of causation. The vast juristic

literature deliberating the proper meaning of causation has left a trail of doubt

and uncertainty. Prominent traditional tort scholars have conceded that ‘there

is perhaps nothing in the entire field of law which has called for more

disagreement, or upon which the opinions are in such a welter of confusion’

(Keeton et al., 1984, p. 263), and that ‘both courts and textbook writers still fall

back when deciding issues in causal terminology’ (Hart and Honoré, 1985, p.

1). Economic analysis helps distinguish the different problems that are involved

and offers a unified approach to their resolution.

 

The attempts of traditional tort scholarship to make sense of the law of

causation have led to the classification of the debates into two separate

doctrines, cause-in-fact and proximate cause. The cause-in-fact doctrine

incorporates the law’s endeavor to define and to conceptualize the criteria that

would determine when an act is part of a causal chain that ends with the injury.

Here, the but-for test is the most common intuitive criterion for inferring such

a factual causality relation. Of all the acts that pass any of the cause-in-fact

tests, the law narrows down the responsibility to those satisfying additional,

‘legal’, tests, which are mostly embodied in the proximate cause doctrine.

Liability is imposed only upon a sub-set of the acts that are causally linked to

the injury, those that survive the scrutiny of a variety of normative judgments

regarding their proximity to the harmful event. As Cooter (1987) nicely labeled

it, the proximity doctrine portrays causation as a ‘decaying transitive relation’:

as the chain of causal inference extends (‘a caused b, b caused c, ...’), the

relationship between removed links weakens without being destroyed.

 

The economic analysis of the law of causation illuminates both the

cause-in-fact and the proximate cause doctrines. Economic analysis applies

positive tools from decision theory and statistics to clarify the definition of a

cause-in-fact, and to resolve some of the confusion regarding the relative

contribution of a given factor to the harmful consequence. Under the normative

economic analysis, the proximate cause doctrine’s designated role is to expand

or shrink the scope of liability, in order to achieve efficient deterrence.

 

This chapter is structured as follows: it begins with a survey of the implicit

role of causation in the writings of the early, pathbreaking economic analysts

of tort law. It then clarifies the basic distinction between retrospective (ex post)

causation and prospective (ex ante) causation, a distinction that forms the core

of many subsequent economic discussions of causation. Next, the explicit role

of causation doctrines in inducing optimal care and activity levels is examined

under the strict liability and the negligence regimes. The analysis is then

extended to cover several complications often plaguing the determination of

causation: uncertainty over causation, joint actions among tortfeasors and

unforeseeability of harm.

 

2. Causation in Early Economic Analysis of Law

 

The original economic theory of tort law deliberately rejected an explicit role

for a causation doctrine in determining liability. Coase’s (1960) view was

particularly resolute in its exclusion of a formal causation element. Coase

describes an injury as a result of mutual and symmetric interaction among

parties. Like particles that randomly collide with each other in space, actions

of individuals may conflict and cause one-sided or mutual harm. Thus, the

phrase ‘the injurer acted and, when coming across the victim, caused an injury’

is interchangeable with the phrase ‘the victim acted and, when coming across

the injurer, caused an injury’. Both passive and active factors are equally

necessary in making the harm occur.

 

Since liability cannot be placed solely on the basis of causation, as both the

injurer and the victim are necessary causes, it ought to be decided according to

a cost-benefit analysis, which will determine the identity of the party that can

alter its actions more cheaply and avoid the injury. As Calabresi (1970)

explained, for instrumental reasons the least-cost avoider should be singled out

as the cause of an injury. The forward looking social objective - minimization

of accidents’ costs - will be furthered if a party that can prevent an accident

with a lower cost than the harm arising from the accident will be regarded as

the sole legal cause of the accident and be held liable. Hence, under this view,

causation is not a preliminary condition for evaluating liability, but it is the

conclusion of the evaluation (see Cooter, 1987).

 

Landes and Posner (1983, 1987, pp. 228-255) have reinforced this view and

argued explicitly that causation has no role in determining liability. Inasmuch

as the purpose of tort law is to promote economic efficiency, the injurer should

be regarded as the cause of an injury when he is the lower-cost avoider of it,

and not otherwise. Therefore, they claimed, ‘the idea of causation can largely

be dispensed with in an economic analysis of torts’. When efficiency analysis

is conducted to determine liability, it can be fully pursued without reference to

causation. Inefficient behavior is synonymous with causing an expected harm.

 

The symmetry among the roles of the injurer and the victim, as well as the

absence of any independent requirement of causation, became well evident

when Brown (1973) formulated his rigorous model of accidents. This model -

the benchmark for subsequent economic analysis of tort law - assigned

symmetric roles to the injurer and the victim, by making the expected harm a

function of care levels taken by both. A party’s action can raise the probability

of harm and, thus, can only be a cause of an expected harm.

 

Thus, in early economic analysis of tort law, cause is reduced to efficient

prevention: the assignment of legal cause is dependent solely upon the

judgment about the economic efficiency of preventive measures. The inquiry

into causation carries no additional message once a cost-benefit analysis of the

care choices has been completed. This characterization of causation, which

prominent scholars have labeled ‘causal minimalism’ (see Hart and Honoré,

1985, pp. lxvii-lxxvii), has led authors to argue that causation serves goals

other than efficiency (Epstein, 1973, 1979, 1987; Borgo, 1979; Cooter, 1987)

or that it merely represents an older method of conducting efficiency analysis

(Grady, 1989; Miceli, 1996).

 

3. Prospective Causation

 

Building on the analytical framework of Calabresi (1970) and Brown (1973),

subsequent treatments of causation distinguished among two concepts of factual

linkage between acts and harms. Calabresi (1975) classified the empirical tests

of causality into two types, which he labeled causal link and but-for cause.

Both describe effects of actions on outcomes. An act is a but-for cause if,

without it, the injury would not have taken place. In contrast, an act has a

causal link to an injury if it increases the probability of its occurrence. As

Shavell (1980) later rephrased the distinction, causation can be either

retrospective or prospective. Retrospective causation exists if, all else held

fixed, but for the action the harmful consequence would not have occurred.

Prospective causation exists when an action raises the probability of the

harmful consequence. Thus, the distinguishing factor between the two types of

causation is the time perspective of the evaluation. Retrospective causation is

backward-looking, answering the counterfactual inquiry of whether the action

was a necessary condition for the outcome. Prospective causation, in contrast,

is forward-looking, answering the ex ante inquiry of whether the action

increased the likelihood of injury (see also Rizzo, 1981; Miceli, 1996).

 

This distinction, and in particular the development of analytical tools to

focus on prospective-probabilistic causation, has helped the economic literature

advance both in its normative and positive study of the law. In the normative

dimension, probabilistic causation became a building block of economic models

of tort law. As Shavell (1980, p. 475) has explicitly phrased it, ‘the first-best

level of care is determined by the cost of taking care and the degree to which

lack of care is a cause of expected losses’. For an action (‘low care’) to raise

the probability of a consequence (‘harm’) relative to another action (‘high

care’), there must be states of the world in which harm occurs only if that

action is taken, and not if the other action is taken.

 

The prospective causation concept has also advanced the positive analysis

of tort law. Perhaps the sharpest example of the contrast between retrospective

and prospective causation theories, and the clearest application of prospective

causation analysis, arises within the family of ‘coincidental’ accidents cases.

In the famous case of Berry v. Borough of Sugar Notch, 43 A. 240 (1899), an

excessively speeding streetcar happened to arrive at a point along its route just

when a tree fell above that point, and struck it. A strict retrospective causation

inquiry would have identified the action of speeding as a but-for cause, since

the accident would not have occurred had the streetcar travelled more slowly.

Applying the traditional retrospective approach, the court sensed the illogic of

assigning liability for such an arbitrary episode, thus had to resort to elusive

concepts such as ‘coincidental harm’ or ‘abnormal risk’ in order to screen such

results and derive a general principle that will restrict the scope of liability. In

contrast, under prospective causation inquiry the action of speeding is

recognized to have not affected the likelihood of harm of the type that occurred.

Ex ante, a tree can fall at any point along the route, and the speed at which the

vehicle is moving does not increase its probability of being hit. The result that

the court reached can be easily aligned with the logic of prospective causation.

(See Honoré, 1983, pp. 50-55, for early applications of the prospective

causation concept.)

 

4. Causation and Socially Optimal Care

 

The causation requirement, although not an explicit element in the ordinary

economic model of tort law, can be isolated and characterized in economic

terms. The basic proposition made by Shavell (1980, 1987, pp. 105-126), and

reiterated by Rizzo (1981), Landes and Posner (1983, 1987) and Cooter (1987),

claims that the desirability of any precautionary action should be determined

only with reference to states of the world in which failure to take the action

would lead to greater expected losses. In determining the level of care that is

optimal, the benefits of care should be balanced against its costs. But whereas

the costs of care accrue before the ensuing state of the world materializes and

regardless of the actual state of the world that will materialize, the benefits of

care arise only in those states in which taking care reduces harm. For example,

in the case of City of Piqua v. Morris, 120 N.E. 300 (1918) the defendant failed

to take sufficient measures against floods. However, a particularly severe flood

occurred, one that even appropriate precautions would not have withstood.

Thus, in evaluating the desirability of anti-flood measures, only the chance for

moderate floods should be counted.

 

The idea that care can be cost-justified only with reference to states of the

world in which it can reduce the harmful consequence was formulated by

Shavell (1987, pp. 118-121) in causation terminology. Shavell defined care (or

lack thereof) to be a necessary cause of harm if, given some state of the world,

a different level of care would have led to a different level of expected harm. He

then proceeded to show that the socially optimal level of care depends only on

states of the world in which the injurer’s care would be the necessary cause of

any losses that occur.

 

In order to examine the extent to which liability rules can implement

optimal care, a causation restriction was formally introduced to the structure of

liability rules. Shavell’s (1980) concept of the scope of liability incorporates the

causation restriction. The scope of liability is defined as the set of states of the

world under which liability can be applied. The scope of liability is said to be

restricted if, given a harmful consequence, there are some states of the world

in which the injurer is not held liable. The scope of liability will be unrestricted

if, anytime there is a harmful consequence, and no matter what state of the

world surrounded it, the injurer will be held liable. The design of a liability

regime includes, in addition to the determination of due care (in negligence)

and the magnitude of liability (both in strict liability and in negligence), the

determination of the scope of liability. If an act is not a necessary cause of the

injury, the injury may be left outside the actor’s scope of liability.

 

Adding the determination of a scope of liability into the analysis nicely

extends the early economic models, by capturing the effects of conditions

beyond the control of the human actors. When the probability or magnitude of

injuries depend upon external conditions, analyzing such conditions within the

formal structure of liability rules is necessary.

 

5. Causation Under Strict Liability

 

Under strict liability, courts have to determine the magnitude of liability and

its scope. Assuming the magnitude of liability equals the victim’s actual harm,

what remains to be determined is whether the accident is to be included within

the scope of liability. Two principal propositions can be made regarding the

incentive effects of the determination of the scope of liability:

 

5.1 The Effect of the Scope of Liability on the Level of Care

The injurer will have optimal incentives to take care as long as the scope of

liability includes at least all the states of the world in which the injurer’s care

is a necessary cause of the harm. If the scope of liability is too restricted, and

does not include all the states in which the injurer could alter the harmful

consequence with its care, then the injurer will have insufficient incentives to

take care. In this case, the injurer will ignore some of the social benefits of its

care - the reduction in expected harm occurring in states of the world outside

the scope of liability - and will underinvest in care. If, in contrast, the scope of

liability is optimally restricted, and includes only states of the world in which

the injurer’s care is a necessary cause, the injurer will bear only the increment

in expected losses due to its actions, and will have optimal incentives to take

care. Similarly, if the scope of liability is unrestricted, so that whenever harm

occurs, and regardless of the state of the world, the injurer is held liable, the

injurer will engage in optimal care. Notice that an unrestricted scope of liability

does not, in itself, distort the injurer’s incentives to take care. Even if the

injurer is liable for harms which its care could not have prevented, it will not

exercise excessive care. Taking more care will not prevent the harm in the

states of the world in which care is not a necessary cause, and thus will not

reduce its expected liability. Hence, the injurer’s incentives to take care can be

distorted only by an overly-restricted scope of liability, not by an unrestricted

one (see Shavell, 1980, 1987, pp. 105-110; Landes and Posner, 1987, p. 236).

 

5.2 The Effect of the Scope of Liability on the Level of Activity

If the scope of liability is too restricted, and does not include all the states in

which the injurer’s care is a necessary cause, it was already established above

that underinvestment in care will arise. This underinvestment can also lead to

excessive incentives to engage in the activity, as the injurer will not bear the

full ‘externality’ of its activity. The cost of engaging in the activity is reduced

by the incremental reduction in the investment in care and by the incremental

reduction in the expected liability and, thus, an injurer may engage in an

activity even when it is undesirable from a social point of view. Similarly, if the

scope of liability is too broad or unrestricted, it may discourage an injurer from

engaging in a socially desirable activity. Although the injurer who faces an

unrestricted scope of liability will not take excessive care, the injurer will face

an inflated expected liability. As Shavell (1980, 1987, p. 108) has termed it, the

injurer may find the unrestricted scope of liability to be ‘crushing’. An activity

that is worthwhile may be deterred by imposing upon the actor costs of losses

that would have been occasioned regardless of this activity. For example, if a

car manufacturer is held liable for accidents arising from bad conditions of

roads, such that cannot be avoided by extra prevention devices in the car’s

design, it may be led to reduce the volume of production. Hence, for injurers to

engage in optimal levels of activity, courts have to restrict the scope of liability

appropriately, which, according to some (for example, Burrows, 1984), may

demand too much sophistication from the legal system, and, according to others

(for example, Wright, 1985) does not reflect prevalent causation doctrines.

 

6. Causation Under the Negligence Rule

 

Under the negligence rule, courts have to determine the level of due care, the

magnitude of liability and the scope of liability. Assuming that the magnitude

of liability equals the victim’s actual harm, what remains to be determined is

which harms should be factored into the determination of the standard of due

care, and under what states of the world the accident is to be included within

the scope of liability. Shavell (1980, 1987, pp. 105-121) has made the following

propositions concerning the incentive effects of causal determinations:

 

6.1 The Determination of the Optimal Standard of Care

The due level of care should equal the optimal level of care, as determined by

considering the effect of care only in circumstances in which care is a necessary

cause - that is, only in states of the world in which taking care would reduce

harm. Care that has no bearing on the occurrence of harm should be excluded

from the negligence standard.

 

6.2 The Effect of the Scope of Liability on the Actual Level of Care

Once a standard of due care is set, the scope of liability has only limited

incentive effects. Whether the scope of liability is optimally restricted (to

include only states of the world in which the injurer is the necessary cause), or

whether the scope of liability is too broad or unrestricted, the injurer will take

the due level of care (assumed to be set optimally). Further, unlike the

activity-crushing effect of strict liability, under the negligence rule an

unrestricted scope of liability does not necessarily deter the injurer from

engaging in the activity. The injurer is induced to take due care and thereby

avoid liability, and thus becomes indifferent as to the actual scope of liability

(Landes and Posner, 1983, 1987, p. 236). As long as the exaggerated scope of

liability does not boost the level of due care, it has no adverse incentive effects

per se. In contrast, if the scope of liability is too restricted, and does not include

all the states in which the injurer’s care could have reduced harm, the injurer

may (but not necessarily) be led to take too little care. The injurer will compare

the cost of due care to the cost of liability in its inefficiently restricted scope. If

the cost of liability is smaller, the injurer’s incentives to take due care will be

distorted.

 

6.3 The Scope of Liability in an Imperfectly Operating Negligence System

Inasmuch as the application of the negligence rule is plagued with error and

uncertainties, it contains an element of strict liability (the injurer may bear

liability even if he were not negligent). In this case, the unrestricted scope of

liability can have the crushing effect that is associated with the operation of a

strict liability rule (Shavell, 1980, 1987, p. 108), Landes and Posner, 1983,

1987, p. 236).

 

Even after Shavell’s (1980) analysis of the optimal scope of liability,

showing that liability should be restricted only to accidents that would not have

occurred had the injurer employed due care, most economic models of

negligence, including part of Shavell’s (1987) book, continued to implicitly

assume that the scope of liability is unrestricted, and that liability turns solely

upon the injurer’s negligence. That is, if the injurer were negligent, no matter

how slight its deviation from due care, it is liable for any accident that arises,

including accidents that additional care would not have prevented. Grady

(1983, 1984, 1989), Kahan (1989) and Marks (1994) have shown that the

analytical framework which assumes unrestricted scope of liability is not in line

with either tort doctrine or optimal incentive design. To adhere to Shavell’s

analysis and restrict the scope of liability so that it includes only accidents that

were caused by the injurer’s negligence, would imply that an injurer who takes

less than due care is not liable for every harm that arises, but only for those

harms which would not have arisen had the injurer taken due care. Thus, if the

injurer takes slightly less than due care, the proper scope of its liability would

include only the (slight) incremental harm that occurred due to this deviation,

and will not include all harms that would have occurred anyway. In the

ordinary case in which care reduces the probability of an accident but not its

magnitude, if the accident occurs the negligent injurer will have to pay

damages with a probability less than one.

 

Using Kahan’s (1989) illustration, suppose the proper height of a fence

surrounding a stadium is 10 feet, and the field owner erects a fence of 9 feet.

If a ball flies over the fence and causes harm, the scope of liability should be

 (and, as a matter of common law, is) restricted to those accidents caused by

balls flying over the fence at heights between 9 and 10 feet. Only those

accidents are caused by the field owner’s negligence. Making the field owner

liable for all harms caused by flying balls, including those that fly at heights

exceeding 10 feet, would mean imposing an unrestricted scope of liability.

 

Until Kahan (1989) exposed it, most economic models managed to conceal

their incorrect characterization of the scope of liability. The reason these

models successfully overlooked this restriction relates to the discussion above,

which suggested that in the case of a perfectly operating negligence system an

exaggerated scope of liability does not have a distorting effect. Since the

perfect-information models of negligence find that the injurer will have the

proper incentives to take optimal care even under the exaggerated scope of

liability regimes, and since there is no crushing of activity effect, they suppress

the significance of the exaggerated scope of liability. But, as Kahan clearly

demonstrated, an unrestricted scope of liability will have different incentive

effects from an optimally restricted scope of liability in cases in which the

application of the negligence rule is plagued with information imperfections.

 

Grady and Kahan’s analyses also suggest that the proper characterization

of causation should eliminate what is otherwise considered a prominent feature

of the negligence rule: the discontinuity of the injurer’s cost function at the

point of due care. This feature of discontinuity plays an important role in

models analyzing injurers’ behavior under uncertainty (see, for example,

Craswell and Calfee, 1986; Shavell, 1992; Ben-Shahar, 1995). If the injurers’

cost function is continuous, as Grady (1989), Kahan (1989), and Cooter

(1989b) have demonstrated it to be, the incentive to deliberately engage in

excessive care, to ensure compliance with the uncertain legal standard, is

significantly diminished.

 

7. Uncertainty over Causation

 

When an injury occurs, its origin may be ambiguous. Several reasons may

account for the uncertainty. First, it may be that separate factors created similar

risks simultaneously, and the actual injury cannot be clearly traced to any one

of them. Second, it may be that the injury manifested itself a long period after

the risk was created or the accident occurred, in which case the cause is

difficult to identify. The principal question that needs to be addressed in the

face of causal uncertainty is under what conditions should a party be liable for

injuries that are uncertain to have been caused by its actions?

 

Two basic approaches to liability in the face of uncertainty over causation

can be proposed. The first approach applies an all-or-nothing criterion to

determining liability. An all-or-nothing criterion holds that either there is no

liability or, if liability is imposed, then it equals the full losses of the victim.

The most common all-or-nothing criterion is the threshold probability rule,

under which full liability is imposed upon the defendant if the probability that

it caused the accident exceeds a threshold level. Potentially, any threshold can

be set, including one that would require proof of causation exceeding any

reasonable doubt. However, the prevalent doctrine applying the threshold

probability rule is the ‘preponderance of the evidence’ standard of many

common law jurisdictions, which incorporates a threshold probability of 50

percent. (In some cases the law reverses the burden of proof and presumes that

the defendant is the cause of the injury. Then, the defendant needs to satisfy the

50 percent threshold in proving that he is not the cause of the injury.)

 

The second approach to resolving uncertainty over causation incorporates

a proportional liability criterion. Under this approach, whenever there is a

positive probability that the defendant caused the injury, liability will be

imposed, but its magnitude will be reduced proportionally to account for the

uncertainty. The most common proportional rule, known as the ‘market share’

approach, sets the defendant’s liability equal to the actual harm multiplied by

the probability that the defendant caused the injury.

 

Traditionally, the law of torts has been governed by the first approach of

all-or-nothing. Full liability is assigned to a party whose acts are assessed to be

a substantial factor in bringing about the harm. A ‘preponderance of

probabilities’ - a threshold of 50 percent - is required for imposition of liability,

and without it no liability is inflicted. However, beginning in the 1980s, and

coming as a response to the onslaught of mass exposure or catastrophic injury

torts, American courts have been more willing to apply the second approach.

In the case of Sindell v. Abbott Laboratories 607 P.2d 924 (1980), which

involved the mass disaster of the DES drug, the court determined each

manufacturer to be liable for a fraction of every victim’s harm, with liability

determined in proportion to the manufacturer’s market share. The debate over

the market share doctrine has since occupied many branches of tort scholarship,

including law and economics. The next two sections examine the economic

justifications for the two approaches.

 

8. The Case for Threshold Probability Rules

 

The first systematic analysis applying economic methods to compare the two

liability approaches was presented by Kaye (1982), who proposed to show the

superiority of the 50 percent threshold probability rule over any other threshold

probability rule as well as over the proportional approach. Kaye’s argument is

based on the assumption that in situations of uncertainty over causation the

social objective of tort adjudication is to minimize the ex post costs of erroneous

liability decisions. Ignoring any ex ante incentive effects that the rules may

have, and assuming that the two types of potential errors courts could make -

false positives and false negatives - are just as costly, Kaye shows that the 50

percent threshold rule is the error-minimizing one. To illustrate the essence of

Kaye’s argument, consider a case in which the harm is $100 and the probability

that the defendant caused it is 40 percent. Under the ‘preponderance of the

evidence’ rule, the defendant will not be liable, and the expected error costs will

equal $40 (there is a 40 percent chance that the defendant is truly the tortfeasor,

in which case it underpays by $100, for an expected error cost of 0.4×$100 =

$40). In contrast, if the court applies the proportional liability rule and sets the

defendant’s liability at $40, the expected error costs will be $48 (there is a 40

percent chance that the defendant is truly the tortfeasor, in which case it

underpays by $60, for an expected error cost of 0.4×$60 = $24; and there is a

60 percent chance that the defendant is not the tortfeasor, in which case it

overpays by $40, for an expected error cost of 0.6×$40 = $24; the sum of the

expected costs of the two possible errors is $24 + $24 = $48).

The all-or-nothing feature embodied in the threshold liability rule has an

additional potential advantage, suggested by Levmore (1990), of reducing the

degree of uncertainty. If uncertainty is assumed to be endogenous, and to vary

according to the incentives of the parties to bring evidence to court, then the

liability approaches can be compared with respect to the degree of uncertainty

they generate. Here, Levmore claims, a high threshold probability will produce

the most evidence and lead to the least uncertainty. When uncertainty is great

and tortfeasors are difficult to identify, plaintiffs face a complete denial of

recovery under a threshold rule that sets a sufficiently high threshold

probability. This will induce plaintiffs to invest more in developing evidence

and identifying the true injurers. In contrast, under a market share regime

plaintiffs need not invest in identifying the true injurers, since they are fully

compensated regardless of the degree of uncertainty. (Of course, if the

defendant, rather than the plaintiff, were assumed to be the party that can

develop superior evidence, then a market share rule will give the defendant the

greatest incentives to bring evidence.) Again, ignoring the ex ante incentive

effects of the rules and focusing on the ex post characteristics of the

adjudicatory regime, a case is made for the threshold probability rule.

 

Apart from minimizing uncertainty and the ex post costs of uncertainty, the

threshold probability approach may offer the additional advantage of reducing

administrative costs. As Shavell (1987, p. 117) suggests, there are three distinct

reasons that the administrative costs will likely be higher under the

proportional liability approach. First, the volume of cases is likely to be higher

under the proportional liability approach, because actions in which the

probability of causation is less than the threshold could be brought. Second,

more defendants could be sued in a typical case under the proportional

approach, raising the costs of litigating the case. And third, litigation under the

proportional approach requires the added judicial determination of the precise

probability of the defendant being the cause of the injury, whereas under the

threshold probability approach the only thing that matters is whether the

probability of causation exceeds the threshold. Some of these excess costs may

be diminished under the enforcement regime that Rosenberg (1984) has

proposed, which borrows features from ‘public law’, such as class actions,

damage scheduling, and insurance fund judgments.

 

9. The Case for the Proportional Liability Rule

 

An economic analysis purporting to show the potential superiority of the

proportional rule was presented by Shavell (1985, 1987, pp. 115-118, 123-126).

Shavell evaluated the effects of the two approaches with respect to a different

social objective from the ex post error-costs minimization criterion. Adopting

an ex ante focus, Shavell took the objective of liability regimes to create optimal

deterrence, that is, the minimization of accident losses. In this perspective,

legal errors do not involve a welfare cost per se, and their minimization could

perhaps be taken as a measure of fairness but not as a proxy for optimal

deterrence (see also Kaplow, 1994). Once embracing the social objective of

optimal deterrence, Shavell has managed to demonstrate that the proportional

liability approach is the superior method of inducing socially desirable

behavior.

 

The main propositions made by Shavell are as follows. First, the threshold

probability criterion distorts the incentives of parties to take care. If the

probability of causation is systematically below the threshold probability, the

party will face too little liability and will take less than optimal care. This

problem of underdeterrence under the conventional threshold probability rule

was labeled by Levmore (1990) and Farber (1990) as the problem of recurring

misses. For example, if the probability of a party being the cause of a typical

injury is systematically 40 percent (as in the case of a manufacturer holding a

40 percent share of the market for a harm-causing product), the party will

always escape liability under the 50 percent threshold rule. The net of liability

will miss this party repeatedly. Thus, the party will have no incentives to take

care. The underinvestment result arises both under the negligence rule and

under strict liability (see also Landes and Posner, 1987, pp. 263-269; Robinson,

1985).

 

Similarly, if the probability of causation is systematically assessed above the

threshold, the injurer may have excessive incentives to take care. This

distortion will arise under a strict liability regime, since the injurer will pay for

all losses, more than it actually causes. The injurer may take excessive care for

a subtle reason. Since the injurer pays for all losses only if he is determined to

by the likely cause, the injurer will have the incentive to reduce the chance that

this determination would be made. By taking excessive care, the injurer may

be able to reduce the posterior probability that he will be designated as the

likely cause of the injury. That is, extra care may shift the preponderance of

probabilities, and clear the injurer from liability altogether. Notice that this

overinvestment result will not arise under a negligence regime, since the injurer

will take due care and will avoid the excessive liability (unless, of course, the

level of due care is ill-defined). This is another illustration of the general

proposition discussed in Section 6 above, which claimed that an unrestricted

scope of liability - that is, liability for consequences that a party did not cause -

does not in itself lead to distorted incentives, and only enhances the motive to

take due care under a negligence regime.

 

In contrast to the distorted outcome under the threshold probability rule, the

proportional liability approach leads to socially optimal levels of care (see

Delgado, 1982; Rosenberg, 1984; Shavell, 1985, 1987, pp. 115-118; Levmore,

1990). The injurer faces expected liability equal to the expected loss associated

with its behavior, and will behave as it would in the absence of uncertainty over

causation. For example, a manufacturer who causes 40 percent of the harms of

a particular type will pay for losses in every case, including the 60 percent of

the cases which it did not cause. But in every case its liability will equal 40

percent of the individual harm, thus it ends up bearing liability of 40 percent

of the total harm, equal to the fraction it caused.

 

Another important proposition made by Shavell concerns the incentives of

parties to engage in the activity that produces the harm. Again, the threshold

probability rule distorts ex ante incentives. When the probability of causation

is systematically above the threshold, the injurer will be overdeterred from

engaging in activity under the strict liability regime. (Under the negligence rule

the injurer takes due care and escapes liability, thus engages in the same level

of excessive activity as it does in the absence of uncertainty over causation, see

Shavell, 1987, pp. 21-32). Likewise, when the probability of causation is

systematically below the threshold, the injurer escapes liability and, as a result,

engages in excessive levels of activity, both under negligence and under strict

liability. In contrast, under the proportional liability rule, the injurer’s

incentives to engage in the activity are the same as they would be in the absence

of any uncertainty over causation.

 

Several authors have argued that the market share approach may lead a

free-rider problem which will cause underinvestment in care. Marino (1991)

demonstrated that care practiced by one firm produces benefits to other firms.

By reducing the probability of harm associated with its products, a firm

produces a positive externality captured by the other firms in the form of

reduced expected liability. That is, each firm will underinvest in care because

it bears the cost of care in full but can appropriate only a share of its benefits.

The magnitude of this underinvestment will diminish as the firm’s market

share increases, and the underinvestment problem will disappear if the firm is

a monopoly. In similar spirit, other authors have argued that proportional

liability will not generate optimal incentives for safety research. Delgado (1982)

and Rose-Ackerman (1990) have pointed out the public good characteristics of

safety improvements, and speculated that the infliction of full liability on an

injurer who has the greatest opportunity to conduct safety research may be

superior to the division of liability according to proportional causation.

 

The dichotomy between the proportional liability approach and the

all-or-nothing threshold probability approach reflects the tension between ex

ante and ex post goals of the tort system. A framework that seeks to unite the

two approaches has been offered recently by Porat and Stein (1997). Under this

framework, a liability rule should be designed to give incentives to parties who

are the cheapest evidence providers, to reduce the ex post uncertainty in

assessing liability. The ingenious mechanism these authors examine is titled

‘liability for uncertainty’ - imposing liability for an injury whose cause is

uncertain on the party that created or had the best opportunity to prevent that

uncertainty. This will lead the parties to invest optimally in removing

uncertainties, and when ex post uncertainty is eliminated, the ordinary liability

mechanisms can operate to generate optimal incentives to reduce the primary

damage. Thus, for example, DES manufacturers can either eliminate the

uncertainty over causation (thus avoid liability for uncertainty), in which case

the all-or-nothing approach will apply and will generate optimal care

incentives, or the manufacturers can choose not to eliminate the uncertainty

over causation, in which case they will be liable for the injuries based on their

proportional contribution to the creation of uncertainty.

 

10. Risk-Based Liability and Safety Regulation

 

The market share approach is a doctrinal step away from the strict

fundamentals of causation. But it is not the most radical step. With the growth

of mass exposure torts, and due to the large degree of uncertainty over

causation in these torts, authors have advocated an even more unorthodox legal

mechanism which will practically abandon any causality requirement between

the injurer’s action and actual harm. The idea is to structure a liability regime

based solely on ‘probabilistic causation’. Under this regime, liability is

proportional not to the harm itself, but rather to the risk of harm which the

actor produces; it is applied regardless of whether this risk actually

materializes. For example, an individual who uses a product and later discovers

that she is under a particular risk, that may or may not develop into actual

harm, may recover damages equal to her expected harm. Thus, liability is

assigned strictly on the basis of the creation of unreasonable risk, independent

of any injury. Contrary to the dominant role that the causation requirement was

given in other influential theories (as in Epstein, 1973), here the causation

element is essentially eliminated.

 

Robinson (1985) and Landes and Posner (1984, 1987, pp. 263-269) have

argued that awarding its expected losses to each potential victim exposed to the

risk of harm will create the proper incentives for injurers to take care and to

select the correct activity levels. In the context of mass exposure accidents, and

in light of the severity of risk-spreading and bankruptcy concerns, this view has

triggered serious attention (see, for example, Roe, 1984; Celli, 1990).

 

Viewed ex post, this approach gives many potential victims a windfall, as

they are going to be compensated without actually bearing any risk; at the same

time, actual victims will be undercompensated. But viewed ex ante, this

approach can provide superior incentives for care relative to other approaches

that have to await the actual, oftentimes lagged, harm and, thus, dilute the

deterrent force of liability. Obviously, a troubling aspect of a risk-based liability

regime is its administrative costs. Litigation need not be conditional on

occurrence of harm and thus could be more frequent, let alone more

complicated (see Celli, 1990). At the same time, each victim is awarded only

a fraction of the actual harm, which may reduce the incentives to take legal

action and, consequently, will lead to underdeterrence.

 

A different approach to monitoring incentives in cases that pose inherent

difficulties in ascertaining causation is a centralized approach, relying on

administrative regulation to enforce optimal risk reduction. Several authors

(Shavell, 1984, 1993; Cooter and Ulen, 1989, p. 420) have examined the

advantages of relying on regulatory authorities to monitor and deter risk

creation in the period before harm manifests itself. These authors have

suggested that regulation of safety may perform better than a risk-based liability

system in preventing mass torts. The main justifications for the superiority of

a regulatory regime are: (1) the government may be a better

information-gatherer than the injurer; (2) injurers may be judgment-proof in

catastrophic harms, thus liability will not generate sufficient deterrence; (3)

suits may not be brought in all cases, due to their costs and to victims’ lack of

information, thereby diluting the deterrent effect of liability; (4) administrative

costs of decentralized liability regimes may be higher.

 

11. Causal Apportionment among Joint Tortfeasors

 

Uncertainty over causation may involve another dimension. Apart from the

difficulty of identifying the actual cause-in-fact - the party whose act caused the

particular accident - courts may face the difficulty of determining the relative

causal share of each of several tortfeasors. There may be information as to the

probabilistic contribution of each act - what ex ante risk each act imposes - and

how the risks change when acts occur simultaneously. However, when two acts

operate simultaneously to cause harm, the contribution of each act to the

combined risk has to be determined before courts can apply either the threshold

probability approach or the proportional liability approach. This determination

involves a ‘disentanglement’ of the harm-production process, a logical exercise

which has proven to be problematic.

 

To illustrate the problem, imagine two fires that are set simultaneously and

independently and combine to destroy a field. It is estimated that, ex ante, the

first fire alone had a 10 percent chance of destroying the field, the second fire

alone had a 20 percent chance, and together they had a 50 percent chance of

destruction. That is, their joint operation creates a synergistic effect. If both

fires were set and the destruction occurred, how should liability be divided

across the two ‘causes’? Or, suppose a particular illness can be contracted

either by use of a product (1 percent) or, independently, by smoking (4 percent).

However, if an individual both uses the product and smokes, the risk increases

to 15 percent. Again, a significant synergistic effect exists. What fraction of the

harm can a smoker that has used the product recover from the manufacturer?

In both examples, how should the synergistic effect arising from the multiple

causes joined together be divided across the causal contributors?

 

The problem of allocating the shares of liability in accidents that have

multiple causes is said to have ‘generated a bewildering variety of legal rules

and nomenclature’ (Kaye and Aickin, 1984), and have perplexed scholars who

expressed ‘doubts that there exists a single factotum satisfactory formula for

dividing damages’ (Kruskal, 1986). The first systematic treatment of the causal

apportionment problem was offered by Rizzo and Arnold (1980) (see also Rizzo

and Arnold, 1986). These authors proposed an apportionment scheme that

assigns to each act a share of liability that consists of two elements. The first

element is proportional to the act’s ‘marginal product’, which Rizzo and

Arnold defined to be the probability of harm given this act operating alone. The

second element is a fraction of the synergistic effect, which Rizzo and Arnold

arbitrarily selected to be one half.

 

In the literature that followed, Rizzo and Arnold’s allocation scheme was

generally rejected. Kaye and Aickin (1984) found flaws with the statistical

framework applied by Rizzo and Arnold, as well as with the material

justifications for the apportionment scheme. In particular, Kaye and Aickin

suggested that Rizzo and Arnold’s definition of an act’s marginal product - the

increase in the probability of harm, given this act operating alone - is no more

appropriate than many other possible definitions, such as the incremental

increase in the probability of harm, given that the other act is also operating.

Kaye and Aickin argued also that there is no one logical way to divide the

synergistic effect across the acts, and thus logic alone cannot provide a

guideline for resolving the causal apportionment problem. Hence, any

apportionment scheme should be evaluated mainly according to the incentive

effects it generates.

 

Further criticism of Rizzo and Arnold was voiced by Kruskal (1986),

Kelman (1987) and Rose-Ackerman (1990), who again claimed that there is no

one principled way to apportion damages that will not be arbitrary and that will

make economic sense. An alternative to the Rizzo and Arnold method of

defining an act’s marginal product can be derived from the cooperative

game-theory concept of the Shapley value. This method offers a more

structured way to define an act’s marginal product, based on its expected

marginal contribution to the probability of harm, averaged over all possible

combinations of acts (see Ben-Shahar, 1996). While this approach enjoys some

intuitive appeal that the previous method did not have, it has the similar

shortcoming in its reluctance to consider the ex ante effects of the

apportionment rule on the incentives for care among multiple tortfeasors.

 

When multiparties are responsible for an injury, there may not exist an

apportionment rule that leads to efficient incentives and keeps total liability

equal to harm. To provide the right incentives to all parties, damages exceeding

full harm may need to be assigned. For example, Landes and Posner (1980,

1987, pp. 193-201) have focused on the effects of liability apportionment on

incentives for care under the negligence regime. They have argued that joint

tortfeasors can be led to take due care under a no-contribution rule - that is, a

rule that makes each party liable for the entire damage and allows the victim

to determine each tortfeasor’s liability share. If the total cost of care is less than

the expected harm, at least one of the injurers will have the incentives to take

due care (his cost of care is less than half the expected harm), thereby placing

the entire liability on the other and leading the other to take due care as well

(see also Wittman, 1981, for a related argument in a joint but sequential care

setting). Thus, in the celebrated case of Summers v. Tice, 199 P.2d 1 (1948),

where two hunters independently and simultaneously fired in the direction of

a victim but only one - unidentified - hit, joint liability with no contribution will

lead both to take optimal care. Notice that a doctrinal justification for making

each hunter fully liable can be obtained through Porat and Stein’s (1997) idea

of liability for uncertainty. Each hunter is liable since, but for his action, the

apportionment difficulty would not have existed: either the fatal bullet was shot

by him, in which case but for his action there would have been no injury, or the

fatal bullet was shot by the other, in which case but for his action there would

have been no uncertainty.

 

The problem with the negligence-based no contribution rule is that it may

lead injurers to engage in excessive levels of activity. Miceli and Segerson

(1991) study a different apportionment rule, one that would potentially lead

joint tortfeasors to take efficient care and make efficient activity decisions. They

propose a ‘decoupled’ strict liability regime, under which each tortfeasor pays

a sum equal to the difference between actual damages and the damages that

would have resulted were he inactive. That is, all tortfeasors are held strictly

liable simultaneously, but each receives a ‘credit’ for the expected damages that

would have occurred in his absence. Since this rule may lead to payment

exceeding actual harm, the excess can be paid as a fine to the state.

 

12. Unforeseeability

 

Whether the objective probability of an accident is low or high should not in

itself affect its inclusion or exclusion from the scope of liability. As Shavell

(1980) explained, even if the probability of the harmful consequence is very

small, the act that is the cause of the increase in probability should carry

liability. The effects on incentives to take care and on the level of activity will

be correspondingly small, as they ought to be. In addition, the added expected

administrative costs of adjudicating a low-probability event are small, since

these costs will be incurred only with a small probability. Thus, the benchmark

claim is that whenever an act is a necessary cause of the harm, liability should

follow (see also Rizzo, 1981).

 

However, to determine the incentive effects that any scope of liability

generates, it is not the objective probability of harm that matters, but the

subjective probability - the ex ante assessment of the possible consequences as

it is made by the injurer. Calabresi (1975) was the first to state explicitly that

there is no sense in trying to deter an act which is a necessary cause of the

injury by threatening to impose liability on an injurer who assigns a subjective

probability of zero to the injury. An injurer who does not foresee a harmful

consequence cannot be meaningfully labeled the least-cost avoider. As Shavell

(1980) clarified and generalized, whenever the subjective probability is very

low in absolute terms, and lower than the objective probability in relative terms,

liability will not produce sufficient ex ante behavioral effects to justify the

increase in the costs of dispute resolutions. Thus, under the doctrine of

unforeseeability, accidents whose probabilities are likely to be underestimated

by injurers should be excluded from the scope of liability. (But see Rizzo, 1981,

for an alternative view, advocating the use of objective probabilities in

determining the abnormality of events.)

 

It may be that an injurer does not foresee some specific low-probability

consequences that subsequently materialize. However, the same injurer may

still be in a position to associate an activity with unforeseen risks. The injurer

may recognize a large variance of outcomes even if it does not recognize the

nature of each outcome in the distribution. In this case, assigning liability for

unforeseen harms will have the desirable effect of reducing the level of an

activity that is known, ex ante, to cluster many unforeseen risks (see Shavell,

1980; Landes and Posner, 1987, p. 250). For example, handing a loaded gun

to a child leads to many unforeseeable risks (apart from the obvious ones), and

can be deterred by imposing liability on this action. Of course, when injurers

systematically fail to recognize the unforeseeable consequences of their actions,

other forms of deterrence may be required, such as criminalizing the actions

(see Calabresi, 1975).

 

The scope of liability for low probability events has another important

effect, in determining the incentives of potential injurers to investigate and

contemplate the potential consequences of their actions. That is, the amount of

information individuals have regarding risk and risk avoidance can be thought

of as endogenous, influenced in part by liability rules. The effects of liability

rules on the incentives to acquire accurate information ex ante have been

studied in more general contexts by several authors (see, for example, Kaplow

and Shavell, 1992, 1996; Shavell; 1992; Ben-Shahar, 1995). Specifically, as the

scope of liability for low-probability harms expands, individuals will have

greater incentives to learn and anticipate these harms, and to take proper

measures to avoid them (or liability for them). Thus, the unforeseeability

doctrine should be replaced by a doctrine of ‘expensive foreseeability’: only

risks that are too costly to anticipate and foresee will be excluded from the

scope of liability. Therefore, in operating the hand formula to determine

whether lack of care should be considered negligent, courts have to account not

only for the direct costs of care, but also for the costs of foresight (Calabresi,

1975; Grady, 1984; Landes and Posner, 1987, pp. 239-247).

 

The concept of unforeseeability in tort law differs from the concept of

unforeseeable damage in contract law. The tort doctrine excludes only ‘freak’

accidents from the scope of liability and serves to avoid unnecessary dispute

resolution costs. The contract doctrine serves only to limit damages to the

average, reasonable, level. However, in economic terms, both tort and contract

doctrines can be justified by their effects on ex ante information acquisition.

The limited scope of liability in tort should apply only when the risk is too

costly for the injurer to anticipate, thus it creates incentives to foresee. The

limited scope of liability in contract has the important effect of inducing

high-risk contractual parties to reveal their idiosyncratic traits ex ante, so as to

secure full compensation and to reduce the chance of breach (see Bebchuk and

Shavell, 1991).

 

Two prominent tort doctrines can further illustrate the role of foreseeability

in monitoring incentives. The first doctrine distinguishes between categories

of harms and can be illustrated by the well-known case of Palsgraff v. Long

Island R.R. (162 N.E. 99 (1928)). In that case, as the result of a railroad

employee’s negligence, a parcel containing fireworks fell from the train,

exploded and caused the crowd to panic and to knock down scales that were

standing on the other platform, in a manner that injured a passenger. Since the

employees did not know of the parcel’s content, the court found the harm to be

unforeseeable and outside the scope of negligence. This result is found by most

economic writers to be justified (see Calabresi, 1975; Shavell, 1980; Landes

and Posner, 1983, 1987, pp. 246-247). Since the injurer discounted the

probability of that type of accident, liability would not have generated better

incentives and would not have led to prevention of the accident. The injurer

may be the least-cost avoider with respect to losses from injuries to passengers

and, at the same time, not the least-cost avoider with respect to non-passengers.

Making the railroad liable for injuries to non-passengers from acts that were

negligent due to the risk they posed to passengers will either have no effect in

reducing risks to non-passengers, or may lead the railroad to engage in socially

excessive investments to identify freak accidents.

 

Another well-known tort law doctrine is the eggshell skull doctrine,

according to which courts impose liability for bodily harm equal to the full

severity of the injury, even if the extent of the injury was unforeseeable due to

a pre-existing condition of the victim. This may seem to contradict the basic

economic insight, which established that when the probability of a high loss is

systematically underestimated, holding the injurer liable for the total loss does

not increase the incentive to take care (Shavell, 1980). However, as Landes and

Posner (1983, 1987, pp. 249-250) have explained, there is economic sense in

holding liability for an unforeseeable extent of injury. First, allowing only

average damages could potentially generate optimal care incentives, but only

if carried out in all cases, including in cases in which victims have ‘thick

skulls’ (that is, awarding these victims average damages despite the fact such

damages are known to overcompensate them). Put differently, if at the high end

of the distribution of harms liability is capped, but at the low end of the

distribution liability equals actual harm, average liability will fall short of

average harm and the incentives for care will be diluted. As long as victims

with low harms get their actual damages, victims with eggshell skulls should

also get their full damages, to maintain the correct level of average liability ex

ante (see also Kaplow, 1994; Kaplow and Shavell, 1996). In addition, liability

for an unlimited extent of injury may have a desirable activity-reducing effect,

resulting from the injurer’s de facto strict liability.

 

13. Causation and Litigation Costs

 

The restriction on the scope of liability that the causation requirement embodies

has, in itself, an ambiguous effect on the administrative costs of the legal

system. A restricted scope leads to a lower volume of suits that are filed, saving

the litigation costs in cases that are clearly outside the scope of liability. On the

other hand, if the scope is unrestricted there may be less harms (through a

reduction in the level of activity) and thus less suits, and each suit that is filed

may be less costly to litigate, as there is no causation issue to resolve (see

Shavell, 1980, 1987, p. 109; Landes and Posner, 1983).

 

Informational imperfections and their legal treatment have important effects

on the costs of resolving disputes. When courts are uncertain about causation,

a significant portion of the trial effort may be devoted to disentangling the

causation process. Applying simple standards such as the ‘preponderance of the

evidence’ rule may reduce administrative costs, sufficiently so to overshadow

its inferior incentive effects. Similarly, when the court’s ex post assessment of

the probability of harm exceeds the injurer’s ex ante assessment, administrative

costs of determining liability may tip the scale towards categorizing the harm

as unforeseeable. Lastly, when causation is difficult to verify ex post, but

probabilitic linkage is known to exist ex ante, the costs of decentralized dispute

resolutions may exceed the costs of a centralized regulatory scheme, outlining

the proper bounds of the civil liability system.

 


中国法律经济学网登载此文出于学术研究之目的,绝不意味着中国法律经济学网赞同其观点或证实其描述。以上内容仅供研究者学习与交流,无意侵犯版权。如有侵犯您的利益,请告知。我们将尽快删除。

加入日期:2007/1/11 14:48:34浏览次数:2186
发表评论
名号:
内容:
验证: 5278
法律经济学网
联系站长: 柯华庆 lawgame@263.net 京ICP备09028584号
北京市昌平区中国政法大学法学院(102249)
本网站由卡卡鱼网提供技术支持 网站总访问量:1996356