Harvard Law Review Harvard Law Review Harvard Law Review

Harvard Law Review Forum

The Effect of Settlement in Kaplow’s Multistage Adjudication

Responding to Louis Kaplow, Multistage Adjudication, 126 Harv. L. Rev. 1179 (2013)

Professor Louis Kaplow’s article, Multistage Adjudication, presents an extremely useful framework for thinking about how the legal system should make decisions in preliminary stages of adjudication — an area of study that has received far too little attention in the academic literature.1 As such, Kaplow’s article is a very valuable starting place for integrating the existence of multiple stages of decisionmaking into normative analysis of legal procedure. It is in that context that I offer this comment.

As is typical, and desirable, when starting a new area of study, Kaplow makes a number of simplifying assumptions in order to make his framework and analysis clearer. One natural way to build on Kaplow’s analysis is to think about the effects of relaxing some of these assumptions. In this Response, I do just that. Kaplow’s analysis generally assumes that all cases are resolved, at one stage or another, by a court. And in the few paragraphs in which Kaplow does acknowledge the possibility of private settlement (section III.D.2 of his article), he seems to assume that the parties each have access to the same information.2 In reality, of course, the vast majority of cases settle, and they do so under conditions of asymmetric information.3 This comment will explore how the relaxation of these two assumptions (final resolution by the court and symmetric information) may influence Kaplow’s framework for thinking about optimal multistage adjudication.

In most cases, at least prior to extensive and costly discovery, there is likely to be a great deal of asymmetric information. Since saving on discovery costs is a primary motivation for settlement,4 we should expect to see much settlement prior to discovery, when there are still significant asymmetries in information. In general, one would expect the defendant to know much more than the plaintiff about the defendant’s likelihood of being found liable because he (I will use the male pronoun for the defendant and the female pronoun for the plaintiff) knows the type of act he committed (harmful or benign in Kaplow’s model). In such cases, multistage adjudication has a surprising effect on settlement dynamics and this effect, in turn, has important implications for optimal system design.5

Consider the following two-stage model. The defendant, because he knows the type of act he committed, has a precise estimate of his expected sanction. The plaintiff does not know with certainty the type of act the defendant committed, so she is aware only of probabilities that the sanction will take on various values. Prior to stage one, the uninformed plaintiff makes a settlement demand to the informed defendant.6 If the demand is accepted, the dispute is resolved.7 If the defendant rejects the demand, then both sides incur some costs8 and the court decides whether to continue or dismiss the suit. If the court continues the suit, then prior to stage two (trial) the plaintiff again makes a settlement demand which the defendant accepts or rejects. If he accepts, the dispute is resolved. If the defendant rejects the plaintiff’s demand, then the parties incur more costs and the court finds the defendant liable or not.

The settlement game prior to trial (stage two) is precisely the screening model first analyzed by Professor Lucian Bebchuk almost thirty years ago.9 These models are called “screening models” because the settlement demand screens the defendant by type. A defendant who knows he is very likely to be found liable (because his act was harmful) will accept the demand, while a defendant who knows he is very likely not to be found liable (because his act was benign) will reject it. Notice that the plaintiff goes to trial in exactly the cases in which she is most likely to lose. Also, note that the plaintiff’s demand at this stage will be lower the more likely she thinks it is that a defendant with a strong defense would reach this stage.

Consider how this situation affects the pre-stage one settlement. When facing any given settlement demand prior to the first stage of adjudication, the defendant compares this demand both to his expected liability at trial and to the settlement demand he expects the plaintiff will make should the case be continued. It is still the case that any demand is more attractive to a defendant with a weak defense (since the case is more likely to be continued and to result in a plaintiff victory) than to one with a stronger defense. As a result, any demand is more likely to be accepted by a weak defendant than a strong one. Furthermore, the lower the demand, the stronger the defendant must be to reject it. Thus, the rejection of a lower demand prior to stage one makes the plaintiff believe that she is up against a stronger defendant at stage two (since she will only face a defendant at stage two if he rejected the pre-stage one demand). Thus, lower pre—stage one demands induce lower pre-stage two demands. As a result, ceteris paribus, making a lower demand at stage one does not necessarily increase the probability that the defendant will accept it since, upon seeing this demand, the defendant can expect a lower demand at stage two as well.

Because the pre-stage two settlement demand (and thus the pre-stage one settlement demand as well) will depend on the plaintiff’s beliefs about the strength of the defendant’s defense, a more generous continuation rule has conflicting effects on the level of the pre-stage one settlement demand. On the one hand, the plaintiff can expect the defendant to accept a higher demand because the defendant is less likely to get the case dismissed (this is the effect Kaplow describes). On the other hand, because defendants will face a higher bar for getting their cases dismissed, plaintiffs at stage two will face defendants that are stronger on average than they would be if a less generous continuation rule were in place.10 This increase in average defendant strength means that the plaintiff will make a lower pre-stage two settlement demand. As a result, the plaintiff also has to reduce the pre-stage one settlement demand so that the defendant will not reject it and wait for the pre-stage two demand. In principle, these conflicting effects could lead to either higher or lower settlement demands. It is possible, however, to construct examples in which this second effect dominates and more generous continuation rules lead to lower settlement demands both before and after the first stage and, as a result, a lower probability of trial.11

If these lower settlement demands occur, then more generous continuation rules can lead to effects that are very different from the effects without settlement. As Kaplow demonstrates, without settlement more generous continuation rules clearly lead to more adjudication costs per case that enters the legal system but lead to more deterrence and more chilling, which reduces the number of cases entering the legal system. Introducing settlement under circumstances of asymmetric information, however, we see that more generous continuation rules can lead to lower settlement demands and more settlement. This lower probability of trial can reduce the adjudication costs per case that enters the legal system.

It is worth noting that, although the probability of settlement increases due to more generous continuation rules, in this model the reason for that change is not that it increases expected adjudication costs as Kaplow suggests.12 That plausible-sounding effect will occur in a model where there is some exogenous chance of a settlement failure at stage two. If settlement occurs solely because of the strategic interaction of the parties, however, then more generous continuation rules only increase the expected adjudication costs of cases for a defendant that is not going to settle anyway because he knows he is unlikely to be found liable at trial. Such a defendant will reject the plaintiff’s settlement demands because he knows the plaintiff has overestimated her chances of winning. A defendant whose private information is less favorable to him will plan to settle after losing at the first stage. Thus, he will not expect any increase in adjudication costs.13

On the other hand, the lower settlement demands lead to less deterrence for harmful acts (those that are most likely to lead to liability and thus most likely to settle). The greater continuation probability, however, leads to greater expected liability for benign acts (those that will not lead to early settlement).14 This greater expected liability certainly leads to greater chilling for benign acts and can lead to greater deterrence for harmful acts that are not the most harmful.15 Thus, there could be more or fewer cases entering the legal system. The net effect of more generous continuation rules on total adjudication costs therefore remains ambiguous but for nearly the opposite reasons as in the no-settlement model.

Notice that more generous continuation rules can, if they lead to more settlement, worsen the chilling/deterrence tradeoff by reducing the difference in expected liability faced by really harmful acts (that benefit from more generous continuation) and less harmful acts (which suffer from more generous continuation). This effect is reminiscent of the harmful effects that settlement can have on ex ante incentives found in A Model of Welfare-Reducing Settlement16 and Chilling, Settlement, and the Accuracy of the Legal Process.17

Up to this point, I have implicitly assumed that the plaintiff always has a credible threat to go to trial if the defendant rejects her settlement demand. As Professor Barry Nalebuff has pointed out in the context of settlement in single-stage adjudication, the plaintiff’s need to maintain the credibility of her threat to litigate can impede settlement.18 The plaintiff cannot make too low a settlement demand because if she does, then she must draw a very negative inference regarding her probability of winning if the defendant rejects her demand (she will infer that the defendant must know he is very likely to win to reject such a low settlement demand). This negative inference makes it more likely she will drop the case if her demand is rejected, which makes the defendant less likely to accept the demand, eliminating the primary benefit of a low demand in the first place.

This problem can be remedied by strict continuation rules in multistage adjudication. A strict continuation rule can benefit both the plaintiff and the defendant because once a case has passed the first stage with a strict continuation rule, the plaintiff has a stronger reason to believe that her case is worth litigating. Passing a higher initial hurdle sends a stronger signal about the legal strength of the plaintiff’s case. This signal will enable the plaintiff to maintain a credible threat to litigate even after the defendant rejects a low settlement demand. Furthermore, by enabling the plaintiff to make a lower second stage settlement demand, the plaintiff must also then reduce her first stage settlement demand. Thus, when the credibility of the plaintiff’s threat to litigate is an important consideration, the effect of more or less generous continuation rules can be exactly the opposite of what it is when this credibility constraint is absent.19

The above analysis demonstrates that taking into account the possibility of settlement will have important effects on the analysis of multistage adjudication and optimal system design. Under the simplest models of settlement, in which settlement simply enables the parties to reach the likely outcome from litigation at less cost, abstracting from settlement is quite innocuous; this Response shows, however, that if there are important informational asymmetries, then the strategic incentives of the parties in settlement bargaining can have surprising effects. If the plaintiff always has a credible threat to continue her litigation, then more generous continuation rules can (in some circumstances) lead to more settlements, which would reduce (not increase) the adjudication costs per case. On the other hand, more generous continuation rules can also lead to lower settlement demands that worsen the deterrence/chilling tradeoff. That said, the analysis in this Comment has necessarily been informal and abbreviated. It demonstrates that considering settlement will have nontrivial effects on optimal multistage adjudication, but a more rigorous and complete treatment of settlement and multistage adjudication is necessary to provide more definitive results on exactly what those effects will be in various circumstances. Such an analysis is one of many extensions that should follow Kaplow’s pathbreaking article.

* Bernard J. Ward Professor of Law at The University of Texas School of Law and Florence Rogatz Visiting Professor of Law at Yale Law School. I thank Professor Louis Kaplow for helpful comments.

1. Louis Kaplow, Multistage Adjudication, 126 Harv. L. Rev. 1179 (2013).

2. Kaplow’s discussion of settlement in section III.D.2 relates the settlement amount to the expected sanction. Kaplow, supra note 1, at 1248-50. Therefore, Kaplow must assume settlement under symmetric information because only if parties are symmetrically informed is there only one expected sanction. Under conditions of asymmetric information, the uninformed plaintiff would face a distribution of possible sanctions while the informed defendant would know the precise value of the expected sanction. If parties are symmetrically informed, however, one would expect all cases to settle immediately, eliminating almost any role for adjudication costs in the analysis. Thus, the analysis of optimal multistage adjudication is much less interesting if parties settle under conditions of symmetric information.

3. See, e.g., Theodore Eisenberg & Charlotte Lanvers, What Is the Settlement Rate and Why Should We Care?, 6 J. Empirical Legal Stud. 111, 133 tbl.6 (2009), (finding that nearly eighty-two percent of tort cases filed in two district courts were settled prior to trial). Other data show that trial rates among all cases are under five percent, though some of the cases that do not make it to trial are dismissed or resolved via summary judgment rather than settled. See Abraham L. Wickelgren, Law and Economics of Settlement, in Research Handbook on the Economic Analysis of Torts (Edward Elgar ed., forthcoming 2013) (providing an overview of data on the rates of settlement).

4. See generally Lawyers for Civil Justice et al., Litigation Cost Survey of Major Companies (2010), available at http://www.uscourts.gov/uscourts/RulesAndPolicies/rules/Duke%20 Materials/Library/Litigation%20Cost%20Survey%20of%20Major%

5. I do not discuss the implications for optimal system design in this Response both due to space constraints and because these implications largely follow from Kaplow’s analysis after taking into account the different effects that settlement has on total system costs, deterrence, and chilling.

6. One could also consider an alternative settlement bargaining model in which the informed defendant makes the offer. In such a setting, there might be an equilibrium in which the defendant’s offer signals his type. That situation would require a different analysis. Such an analysis is worth pursuing, but space constraints prevent me from doing so here.

7. The defendant will accept a given demand only if it is less than his expected liability at trial plus his expected adjudication costs. As discussed in more detail below, however, the defendant might reject such a demand prior to stage one in hopes of receiving a better offer prior to stage two. It is also worth noting that since the plaintiff does not know the defendant’s expected liability at trial, her demand could exceed the maximum the particular defendant might accept. Lastly, this analysis omits any reputational reasons for settlement and assumes that both parties make settlement decisions to maximize their payoff from the case at hand.

8. For simplicity, I will assume there are no costs to negotiating a settlement. Obviously, this is not strictly the case, but given that such costs are likely to be very small relative to the costs of discovery and trial, it should not affect my conclusions very much.

9. Lucian Arye Bebchuk, Litigation and Settlement Under Imperfect Information, 15 Rand J. Econ. 404 (1984).

10. This conclusion assumes that a more generous continuation rule reveals less information. A defendant is less likely to reveal as much in a preliminary hearing that is less likely to affect the outcome; the latter seems plausible when continuation rules are more generous. Therefore it seems likely that if the continuation rule were more generous, then the plaintiff would think it was more likely that she was facing a defendant with a strong defense who did not reveal much of this strength in stage one, leading her to make a lower settlement demand before stage two.

11. These examples are available from the author upon request.

12. See Kaplow, supra note 1, at 1249.

13. It is not obvious that thinking about the plaintiff’s expected adjudication costs in this setting makes sense given that her actual costs in each stage have not changed and her probability of paying those costs is partly endogenous (determined by her own settlement demands).

14. In the numerical examples that I have worked out, the defendant’s expected liability is lower for all cases that settle early under the more generous continuation rule and for the most harmful cases among those that do not settle early.

15. This finding implicitly assumes a greater range of types of acts than in Kaplow’s model. He assumes there are only harmful and benign acts. In contrast, this analysis assumes that there is a continuous distribution of the external harm from the act. The reason for the change is that a settlement model with only two possible types generates misleading results since then the settlement demand can perfectly screen between the possible types of defendants. Admittedly, this expanded range makes the comparison between the settlement and the no-settlement analysis less clean. This distortion is not a significant problem, however, as long as Kaplow’s analysis is fairly robust to including a greater range of possible types, which I believe it is.

16. Abraham L. Wickelgren, A Model of Welfare-Reducing Settlement, Contributions to Econ. Analysis & Pol’y, May 2004.

17. Ezra Friedman & Abraham L. Wickelgren, Chilling, Settlement, and the Accuracy of the Legal Process, 26 J.L. Econ. & Org. 144 (2010). For an informal treatment of these effects, see Ezra Friedman & Abraham L. Wickelgren, No Free Lunch: How Settlement Can Reduce the Legal System’s Ability to Induce Efficient Behavior, 61 SMU L. Rev. 1355 (2008).

18. See Barry Nalebuff, Credible Pretrial Negotiation, 18 Rand J. Econ. 198, 198 (1987).

19. Nalebuff finds a similar effect in basic models of settlement. See id.