University of Southern California Law School

Transcription

1 University of Southern California Law School Leal Studies Workin Paer Series Year 2011 Paer 80 The Selection of Thirteenth-Century Disutes for Litiation Daniel M. Klerman USC Law School, This workin aer is hosted by The Berkeley Electronic Press (beress) and may not be commercially reroduced without the ermission of the coyriht holder. htt://law.beress.com/usclws-lss/art80 Coyriht c 2011 by the author.

2 The Selection of Thirteenth-Century Disutes for Litiation Daniel M. Klerman Abstract Priest and Klein s seminal 1984 article arued that litiated cases differ systematically and redictably from settled cases. This article tests the Priest-Klein selection model usin a data set of thirteenth-century Enlish cases. These cases are esecially informative because juries rendered verdicts even in settled cases, so one can directly comare verdicts in settled and litiated cases. The results are consistent with the redictions of the Priest-Klein article, as well as with the asymmetric-information selection models develoed by Hylton and Shavell.

3 The Selection of Thirteenth-Century Disutes for Litiation Daniel Klerman USC Law School I thank Jennifer Arlen, Omri Ben-Shahar, Lisa Bernstein, Steven Choi, Mary Dudziak, Thomas Gallanis, Nuno Garoua, Joshua Getzler, Tom Griffith, Alon Harel, Bruce Hay, R.H. Helmholz, Keith Hylton, Gre Keatin, Kevin Kordana, Timur Kuran, Tom Lyon, Ed McCaffery, Evan Osborne, Eric Posner, Richard Posner, Richard Ross, Mark Ramseyer, Yoram Shachar, Robert Sitkoff, Kathy Sier, Matt Sitzer, Nomi Stolzenber, Eric Talley, and articiants in the Yale Law Economics and Oranization Worksho, Harvard Olin Conference on the Economics of Courts, USC Law School Faculty Worksho, Interdiscilinary Center (Herzliya) Faculty Worksho, Caltech Social Science History Seminar, Georetown Olin Worksho, Tel Aviv Law and Economics Worksho, Hebrew University Law and Economics Worksho, American Society for Leal History 2000 Annual Meetin, 15 th British Leal History Conference, 2001 Medieval Academy of America Annual Meetin, and American Law and Economics Association 2000 Annual Meetin for their helful comments and encouraement. I also thank Fred Boehmke for his assistance with the simulations in Section 3. This research was suorted by the National Science Foundation (Law and Social Science Proram, rant # SBR ), the Social Science Research Council, a Fulbriht Fellowshi, and three USC Law School Summer Research rants. Daniel Klerman, Charles L. and Ramona I. Hilliard Professor of Law and History, USC Law School, University Park MC-0071, 699 Exosition Blvd, Los Aneles, CA dklerman@law.usc.edu. Phone: Fax: SelrCT.doc Hosted by The Berkeley Electronic Press

4 Abstract Priest and Klein's seminal 1984 article arued that litiated cases differ systematically and redictably from settled cases. This article tests the Priest-Klein selection model usin a data set of thirteenth-century Enlish cases. These cases are esecially informative because juries rendered verdicts even in settled cases, so one can directly comare verdicts in settled and litiated cases. The results are consistent with the redictions of the Priest-Klein article, as well as with the asymmetric-information selection models develoed by Hylton and Shavell. 2 htt://law.beress.com/usclws-lss/art80

5 1. Introduction Geore Priest and Benjamin Klein s 1984 article, The Selection of Disutes for Litiation, 1 has roven to be one of the most influential articles in leal scholarshi enerally, and it has been esecially influential in emirical work. 2 In the nearly three decades since its ublication, hundreds of researchers have relied uon and aumented its rincial hyothesis, that litiated cases differ systematically and redictably from settled cases. 3 Priest and Klein s analysis has been esecially imortant to emirical leal studies, because it imlies that quantitative analysis is incomlete and robably misleadin if it does not take into account disutes that never reached final judment. The intellectual ower of the Priest-Klein hyothesis has, in some resects, outaced its emirical verification. While leal scholars have mounted a sustained camain to test the hyothesis in a number of leal settins, ranin from torts to tax, 1 Geore Priest and Benjamin Klein, The Selection of Disutes for Litiation, 13 J. Leal Stud. 1 (1984). 2 Fred R. Shairo, "The Most-Cited Law Review Articles Revisited," 71 Chicao-Kent L. Rev. 751, 771 (1996) (rankin it 99 th of all law articles); James E. Krier & Stewart J. Schwab, "The Cathedral at Twenty-Five: Citations and Imressions," 106 Yale L. J. 2121, 2145 (1997) (rankin it 81 st of all law articles); William M. Landes & Richard A. Posner, "Heavily Cited Articles in Law," 71 Chi-Kent L. Rev 825, 838 (1996) (rankin it 28 th of all articles in "redicted 'lifetime' citations"). 3 See, e.., Keith Hylton, Asymmetric Information and the Selection of Disutes for Litiation, 22 J. Leal Stud. 187 (1993); Steven Shavell, Any Frequency of Plaintiff Victory at Trial is Possible, 25 J. Leal Stud. 493 (1996); Joel Waldfoel, "The Selection Hyothesis and the Relationshi between Trial and Plaintiff Victory, 103 J. Pol. Econ. 229 (1995); Theodore Eisenber, "Testin the Selection Effect: A New Theoretical Framework with Emirical Tests," 19 J. Leal Stud. 337 (1990); Theodore Eisenber & Henry S. Farber, "The Litiious Plaintiff Hyothesis: Case Selection and Resolution," 28 RAND J. Econ. S92 (1997); Samuel R. Gross & Kent Syverud, "Gettin to No: A Study of Settlement Neotiations and the Selection of Cases for Trial, 90 Mich. L. Rev 319 (1991); J. Mark Ramseyer and Minoru Nakazato, The Rational Litiant: Settlement Amounts and Verdict Rates and Jaan, 15 J. Leal Stud. 263 (1989); Joel Waldfoel, "Reconcilin Asymmetric Information and Diverent Exectations Theories of Litiation," 41 J. Law & Econ. 451 (1998); Kuo-Chan Huan, How Leal Reresentation Affects Case Outcomes: An Emirical Persective from Taiwan, 5 J. Emirical Leal Stud. 197 (2008). 3 Hosted by The Berkeley Electronic Press

6 the results have been mixed. 4 This article endeavors to test the Priest-Klein hyothesis usin a data set of rivate rosecutions of crime from thirteenth-century Enland. While the Priest-Klein analysis is usually alied only to civil cases, it is aroriate to test it on these criminal cases, because, unlike modern criminal cases, they were rivately rosecuted by the victim or a relative, who could settle the case and retain the entirety of the roceeds. I find that the data are larely consistent with the theoretical redictions. My results are articularly notable for at least two reasons. First, this is the only existin study to evaluate the Priest-Klein selection theory usin data from before Perhas more imortantly, the data set of medieval rivate rosecutions facilitates a uniquely owerful test of selection theories, because courts at the time elicited jury verdicts even after settlement, thereby allowin for direct comarison of litiated and settled cases. As discussed more fully in section 2, judes in mid-thirteenth-century Enland decided to disreard settlements in criminal cases and to send settlin defendants to trial. This ractice was made ossible by the fact that juries at this time were selfinformin and so did not need the rosecutor s cooeration at trial to learn about the case. 6 Because modern data never include ost-settlement verdicts, revious emirical 4 For a review of the literature throuh 1995, see Daniel Kessler, Thomas Meites, and Geoffrey Miller, "Exlainin Deviations from the Fifty-Percent Rule: A Multimodal Aroach to the Selection of Cases for Litiation," 25 J. Leal Stud. 233 (1996), See also the articles by Waldfoel, Eisenber, Gross & Syverud, and Ramseyer & Nakazato cited in the revious footnote, the two articles cited in the next footnote, and Peter Sieelman & John J. Donohue III, "The Selection of Emloyment Discrimination Disutes for Litiation: Usin Business Cycle Effects To Test The Priest-Klein Hyothesis," 24 J. Leal Stud. 427, 429 (1995); Peter Sieelman & Joel Waldfoel, "Toward a Taxonomy of Disutes: New Evidence throuh the Prism of the Priest/Klein Model," 28 J. Leal Stud. 101 (1999). 5 In fact, with one excetions, all emirical studies that I am aware of use data from after The excetion is Stanton Wheeler et al., "Do the 'Haves" Come Out Ahead? Winnin and Losin in State Sureme Courts, ," 21 Law & Soc. Rev. 403 (1987). 6 Daniel Klerman, Was the Jury Ever Self-Informin? 77 S. Cal. L. Rev. 123 (2003). 4 htt://law.beress.com/usclws-lss/art80

7 tests of selection theories could test only redictions reardin litiated cases. 7 Such tests have been inconclusive, in art, because the redictions about litiated cases necessitate assumtions about the underlyin distribution of all cases (i.e. both settled and litiated cases). As such, a critic of selection theories could always arue that rior emirical tests were flawed, because the settled cases, if they could be observed, miht have characteristics identical to the litiated cases. Because the medieval data studied here include jury verdicts in both settled and litiated cases, they can rovide a more conclusive test of the Priest-Klein hyothesis. Priest and Klein arued that litiated cases are a biased samle of all leal disutes, because most cases settle and because settled cases are likely to differ from litiated cases. In eneral, they arued that litiation is more likely when the case is close and both sides have a rouhly equal chance of revailin at trial. Thus, their most heralded rediction is that litiated cases will tend, on averae, to result half in verdicts for the laintiff and half in verdicts for the defendant. They are careful to acknowlede, however, that a number of factors will cause deviations from this rediction, includin the underlyin distribution of all cases (settled and litiated), the roortion of cases which settle, and most imortantly for this article, asymmetric stakes. When defendants have more at stake in the litiation (erhas because they fear bad ublicity or an adverse recedent), they will be more cautious, settlin even close cases and litiatin only when 7 There are, however, two notable excetions. In their article, Linda Stanley and Don Coursey used an exerimental desin, which ave them information on the settled cases. Linda R. Stanley & Don L. Coursey, "Emirical Evidence on the Selection Hyothesis and the Decision to Litiate or Settle," 19 J. Leal Stud. 145 (1990). In her research, Leandra Lederman comared characteristics of settled and litiated tax cases, includin the size of the stakes and the decade the jude was aointed to the bench, but not (for the obvious reason) the verdict or judicial decision. Leandra Lederman, "Which Cases Go to Trial: An Emirical Study of Predictions of Failure to Settle, 49 Case W. Res. L. Rev. 315 (1999). It should also be noted that Waldfoel's articles, cited in the revious two footnotes, examine not only the ercentae of laintiff victories in litiated cases, but also the ercentae of cases that were litiated. 5 Hosted by The Berkeley Electronic Press

8 they are confident they will revail. As a result, when defendants have more at stake, litiated cases will contain more ro-defendant verdicts. Conversely, if laintiffs have more at stake, litiated cases will contain more ro-laintiff verdicts. The imlications of asymmetric stakes are imortant for the analysis of my data because it is doubtful that laintiffs (i.e. rosecutors) and defendants valued conviction equally. Althouh society may have had a lare stake in conviction, the rosecutors and defendants valuations are what matter here. Because rosecution was rivate, only the rosecutor s and defendant s ersonal stakes affected the decision to settle. I arue below in Section 2 that defendants in thirteenth-century rivate rosecutions had more at stake than rosecutors, because uilty verdicts usually resulted in a substantial fine levied aainst the defendant, which was aid exclusively to the kin. In contrast, the rosecutor obtained more attenuated benefits from the conviction, consistin of the avoidance of a small fine for false rosecution and, erhas, reutational benefits and retributive satisfaction. Under such circumstances, the Priest-Klein model redicts that litiated will contain a hiher roortion of victories for the defendant. The data larely conform to the theoretical redictions. Defendants were found uilty as chared more than twice as often in settled cases (82%) than in litiated cases (37%). While this difference would seem remarkably consistent with the Priest-Klein hyothesis, the analysis is comlicated by artial uilt verdicts and missin data. Partial uilt verdicts are verdicts in which the defendant was acquitted of the accusation made by the rosecutor, but still found uilty of a less severe, related offense. For examle, defendants accused of mayhem (causin serious bodily injury) were often found uilty of simle battery, and defendants accused of homicide were sometimes found uilty only of 6 htt://law.beress.com/usclws-lss/art80

9 bein an accessory. I also classify as artial uilt verdicts situations where the rosecutor accused the defendant of multile crimes (e.. beatin and theft), but the jury found the defendant uilty of only one (e.. just the theft). Existin theoretical work on the selection of disutes for litiation assumes only two ossible verdicts (liable or not liable) and does not address situations where three verdicts (innocent, artially uilty, uilty-aschared) are ossible. Thus, in order to conduct a enuine test of the Priest-Klein hyothesis usin the medieval data, it will be necessary to eneralize their theory to incororate the ossibility of artial verdicts. After makin such modifications, I still find that the Priest-Klein framework enerates redictions that are consistent with the medieval data. This modification miht also have some relevance to modern civil cases involvin multi-count comlaints. For examle, if a comlaint alleed both fraud and nelience, a jury verdict findin liability for nelience but not fraud could be analyzed as a artial verdict within the framework develoed here. While the ractice of renderin verdicts even in settled cases makes the thirteenthcentury data esecially valuable for testin selection theories, such verdicts were not rendered and recorded in every case. Moreover, for an overlain set of cases, it is unclear whether a case actually settled or was merely droed by the rosecutor for other reasons. The number of cases lackin information on either or both of these dimensions (uilt and/or settlement) is lare enouh that it could, under some assumtions, erase the differences between settled and litiated cases. For this to occur, it would be necessary for the missin data to be drawn from a oulation of disutes that divered sharly and imlausibly from the recorded data. Such a diverence is ossible. Indeed, a key insiht of the Priest-Klein analysis is the necessity of exlorin the lausibility of such 7 Hosted by The Berkeley Electronic Press

10 diverences between observed and unobserved cases. Nevertheless, as will be discussed in section 5.A, under the most lausible assumtions about the characteristics of the missin data, the difference between settled and litiated cases remains lare and statistically sinificant. The data set of thirteenth-century rivate rosecutions is also unique in containin jury verdicts in cases which the rosecutor initiated but then droed before trial without settlement. As relatively simle economic models of litiation would redict, juries reorted that defendants were innocent in the overwhelmin majority (74%) of such cases. Prosecutors rationally decided that it was not worth their time to litiate these cases, which were likely to lead neither to a neotiated settlement nor to the retributive satisfaction of a trial conviction. Section 2 of this article rovides eneral backround on thirteenth-century rivate criminal rosecutions and arues that the stakes in these cases were robably larer, on averae, for the defendant than for the rosecutor. Section 3 reviews the Priest-Klein theory of the selection of disutes for litiation, modifies it to take into account the ossibility of artial-uilt verdicts and of trials after settlement, and enerates emirical redictions. Section 4 describes the data, tests the emirical redictions, and discusses the results. Section 5 addresses some limitations of the data, includin the imlications of missin data and the ossibility that jurors inferred uilt from settlement. Section 5 also investiates asymmetric information, differences between crimes, and alternative exlanations for differences between settled and litiated cases. 8 htt://law.beress.com/usclws-lss/art80

11 2. Thirteenth-Century Private Prosecutions of Crime 8 Durin the middle aes, and indeed until the twentieth century, much if not most crime in Enland was rosecuted rivately by victims or their relatives. 9 In the thirteenth century, such rosecutions were usually called aeals. Such aeals should not be confused with modern aeals, because they had nothin to do with the correction of errors by a suerior court. To aeal someone meant simly to rosecute him for crime. Aeals could be brouht for a wide array of offenses, most tyically for assault (includin beatin, woundin and mayhem), homicide, theft (includin simle larceny, robbery and burlary), and rae. Lawyers seldom reresented either arty. The best way to understand a thirteenth-century rivate rosecution is to read a case. Some of the more imortant arts are underlined. John son of Benedict aealed Ivo Quarel, Osbert Cokel and Henry Wyncard in county court of [breach of the] kin's eace, wounds and imrisonment etc. And he [John] now comes and does not want to rosecute them. Therefore let him be committed to jail and his sureties, Ayltro Balliol and Walter son of Odo, are in mercy [fined]. And Ivo and the others come [to court]. And the jurors testify that they [John, Ivo, Osbert and Henry] have settled and they [the jurors] say that, in truth, the aforesaid Ivo and the others came to the roerty of Matthew of Leyham in Barford and fished there without Matthew's ermission and contrary to his wishes. The aforesaid John came alon and asked them for a lede, 10 and the aforesaid Ivo would not ive him one, but instead [Ivo] struck the aforesaid John in the head with a hatchet and made two wounds each three 8 Nearly all of the information in this section is discussed in reater deth in Daniel Klerman, Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1, (2001). 9 Private rosecutions could take two forms: rosecution by indictment and rosecution by aeal. In the thirteenth century, rosecution by indictment was not yet a sinificant source of rivate rosecution, so rosecution by aeal was the dominant form of rivate rosecution. On the transition from aeal to indictment, see Daniel Klerman, Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1, 5-8 (2001). 10 John robably asked Ivo and the others for a lede that they would show u at court if they were sued for fishin without ermission. 9 Hosted by The Berkeley Electronic Press

12 inches lon down to the crest of the head. And they [Ivo and the others] beat him badly. And afterwards they took him and bound him and ut him in a boat and took him from this county [Bedfordshire] to the county of Huntindonshire to Ivo's house at Buckden. There they draed him with a roe to a window of Ivo's solarium and forced him to break the window with an ax. And they ainted the wall near the window with the blood flowin from the wounds the aforesaid Ivo had iven the aforesaid John, and they draed him throuh the window and set uon him a blanket and some linen sayin that he had stolen them. 11 And they raised the hue [and cry] and caused the men who resonded to the hue [and cry] to understand that eihteen thieves had come to his house, and that all excet the aforesaid John had otten away. So they ut the blanket and the linen on him and took him to Huntindon and [they] ave him to the sheriff to be incarcerated. And he remained in rison until his tithin delivered him. 12 Therefore let the aforesaid Ivo and the others be taken into custody. 13 Later Ivo Quarel came and made fine for forty marks 14 [i.e. romised to ay the kin forty marks to be released from custody] by sureties Ralf Ridel [and eleven others]. 15 In this case, John accused Ivo and others of woundin and false imrisonment. Like nearly all rosecutors, John initiated the case in county court. In fact, otential rosecutors were required to sue at the first county court meetin after the commission of the alleed offense. County court met every four weeks, so rosecutors had a month or less to initiate their cases. Trial was ostoned until royal justices arrived in the county, which usually meant a delay of several years. Durin that interval, John, Ivo and the 11 When a thief was cauht in the act, the arehenders often tied the stolen oods to the thief before brinin him to local authorities. See R.C. van Caeneem, Enlish Lawsuits from William I to Richard I, Selden Society vol. 107, (1990). 12 The tithin robably secured his release, endin trial, uon a romise that they would ensure his resence at trial. Every adult male was required to be in a tithin, a rou whose most imortant function was roducin its members' attendance in court when necessary. 13 The justices had no intention of keein Ivo in rison. Imrisonment, or the threat of it, was used not as unishment, but to induce convicts to ay fines. 14 A mark was a unit of currency, equivalent to two-thirds of a ound. 15 Public Record Office, London, JUST 1/4, m. 30 (Bedfordshire 1247). This case exists only in Latin in manuscrit form. Transcrition and translation by the author. Dr. Paul Brand heled with two difficult-to-read words. 10 htt://law.beress.com/usclws-lss/art80

13 other defendants settled the case for undisclosed consideration. As a result, when the case came to trial, John told the judes he did not want to rosecute. Althouh it was not officially condoned, settlement was common. Judicial treatment of settled cases varied over the thirteenth century. Durin most of the early art of the century, settlements were tolerated. When judes were aware of settlement, they fined both arties. Nevertheless, such fines were uncommon, because judes seldom knew when a case a settled. They knew when a case was not rosecuted, but they did not know whether non-rosecution reflected settlement, rosecutorial realization that the case was weak, or some other factor. As a result, defendants who settled were, de facto, acquitted. Startin in 1234, however, judes bean to ask jurors routinely whether the arties had settled, and thus bean more often to fine those who had settled. The case quoted above reflects this new olicy towards settlement. Althouh the jude s inquiry reardin settlement is not recorded, the jurors resonse They have settled is. Because, as discussed below, the jury was local and "self-informin," it would have known whether the case had settled. In addition to inquirin about settlement, durin the eriod and after 1239, judes took an interest in the non-rosecuted cases and usually sent the accused to jury trial in site of the fact that these cases were often settled. As exlained in the next ararah, these cases were tried without a rosecutor. Or, as one thirteenth-century commentator ut it metahorically: in such cases, the kin acted as rosecutor. 16 If convicted, the defendant would be fined or haned for the offense, just as he would have 16 Bracton on the Laws and Customs of Enland, ed. Geore E. Woodbine, trans. Samuel. E. Thorne ( ), 2:402, f. 142b (When an aellor defaults, the "the kin may roceed ex officio."). Of course, the kin would not himself actually rosecute, and there were no rofessional or official rosecutors to rosecute on his behalf in ordinary cases. 11 Hosted by The Berkeley Electronic Press

14 been if he had not settled. This, indeed, is what haened in the case quoted above. In site of the settlement, Ivo and the others were tried. The jurors reorted that Ivo both wounded John and caused him to be falsely imrisoned. The jury thus rendered a uiltyas-chared verdict, which made Ivo liable for a forty mark fine. Forty marks was a very lare amount, about $40,000 in 2011 dollars. Because this was the eriod of the self-informin jury, the rosecutor s articiation at trial was not required for conviction. Little or no evidence was resented at trial. Instead, the jury, which was comosed exclusively of local men, includin men from the defendant's villae, was exected either to have investiated the disute before trial or to decide the case based on reutation and rumor. Trial was an oortunity for the jury to reort what it knew to the judes, not for jurors and judes to hear evidence from witnesses and the rosecutor. 17 The record of the case quoted above reflects this ractice. There is no indication that John testified. Indeed, no evidence at all seems to have been resented at trial. Instead, immediately after statement of the accusation and ascertainment of the resence or absence of the arties, the jurors testify to what haened, thus renderin a uilty verdict. Guilty verdicts were common in settled cases. As discussed in section 4, defendants were convicted of all chares in eihty-two ercent of settled cases, even thouh the rosecutor usually contributed nothin at trial to these cases, and may not even have shown u. The ossibility that juries inferred uilt from settlement is discussed in section 5.2. As a result of the two inquiries described above (inquiry into whether nonrosecuted defendants had settled and inquiry into whether they were uilty), after 1239, it is ossible to ascertain both how far the rosecutor went with the case and the uilt of 12 htt://law.beress.com/usclws-lss/art80

15 the defendant. For examle, in the case above we know that the rosecutor settled and that the defendants were uilty. In other cases, rosecutors roceeded to trial or droed their cases. Droed means initiated rosecution in county court, but neither settled nor rosecuted to trial. Fiure 1 illustrates the key decisions made by each actor rosecutor, defendant, and jude and the sequence of their actions 18 : Jude sends settled case to trial Prosecutor and defendant settle Jude resects settlement. Case does not o to trial Prosecutor continues rosecution Prosecutor and defendant do not settle. Case oes to trial Prosecutor sues Prosecutor dros case Prosecutor does not sue Fiure 1. Sequence of Actions The fact that defendants would be sent to jury trial even if they settled, of course, severely undercut their incentive to settle. 19 Why settle if one is to be tried and unished anyway? The fact that any defendants settled durin eriods when settlin defendants 17 Daniel Klerman, Was the Jury Ever Self-Informin? 77 S. Cal. L. Rev. 123 (2003). 18 Note, however, that althouh the rosecutor s decision to dro or continue the case is deicted as takin lace rior to the decision to settle, the rosecutor miht dro the case after the arties failed to settle. 19 In addition, as fully exlored in Daniel Klerman, Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1 (2001), the dilution of the defendants 13 Hosted by The Berkeley Electronic Press

16 were tried thus requires some exlanation. There are three rincial reasons defendants miht still settle. First, sendin settled defendants to trial was a new and unannounced olicy, so some defendants settled with the exectation that their settlements would be resected and then were surrised to be sent to trial. This is an esecially lausible exlanation for the eriod , when the olicy was relatively new. Because royal judes visited the countryside to hear criminal cases only every few years, it would have taken about a decade for the inhabitants of every county to become aware of the new olicy. Second, until the 1260 s, judes did not aly the new olicy with absolute rior. They obtained jury verdicts in only about sixty ercent of non-rosecuted cases. As a result, settlement could still be rational even for well-informed defendants, because there was a sinificant robability it would rotect them from trial and unishment. The data analyzed in this article are restricted to the eriod for which these two exlanations are most lausible, that is from 1239 to the early 1260's. Third, the fact that a defendant miht be sent to trial even if he settled increases the rosecutor s incentive settle. Because the rosecutor could obtain the benefits of both settlement and a conviction, the rosect of trial after settlement lowered the rosecutor s minimum settlement demand. This, in turn, made it ossible for the defendant to settle even if he made a low offer. This henomenon is discussed in reater deth and is modeled at the end of section 3. Two other reasons may also hel exlain why defendants continued to settle: unishments imosed on settlin defendants may have been less severe than on those who contested their cases, and settlement may have rovided non-leal benefits to the defendant, such as artial reair of the reutational damae conviction miht imose. incentive to settle also reduced victims incentive to rosecute, and thus sinificantly reduced the number of rivate rosecutions brouht. 14 htt://law.beress.com/usclws-lss/art80

17 While most verdicts were simly uilty or not uilty, jurors could also ive more nuanced verdicts indicatin that the defendant was artially uilty. For examle, a jury could reort that a defendant accused of beatin and theft was uilty of the beatin but not the theft, or a jury could reort that a defendant accused of mayhem (maimin bodily injury) was uilty only of battery (simle, non-maimin beatin). The sanctions imosed when juries returned these artial verdicts tended to be much smaller than the sanctions when the defendant was found uilty as chared. The defendant was never haned, and the fines were, on averae, only a third as lare as the fines imosed when the defendant was found uilty as chared. As in the suit between Ivo and John, uilty defendants were usually fined, with the fines aid to the royal treasury. Those convicted of homicide and theft were sometimes haned, and, when haned, forfeited their lands and chattels. Nevertheless, a rosecutor who secured a conviction received neither damaes from the defendant nor a bounty from the state. An unsuccessful rosecution, however, would result in imosition of a small fine for havin made a false accusation. Droed rosecutions also resulted in a small fine, and settled rosecutions miht result in a somewhat larer fine. These fines were usually much smaller than the fines aid by uilty defendants. For examle, althouh the fine is not recorded in the case quoted above, John robably aid a fine of one mark or less. In contrast, Ivo aid forty marks. As the revious ararah suests, the stakes in thirteenth-century rivate rosecutions were robably asymmetric and larer for defendants, because convicted defendants usually aid a fine to the royal treasury, while revailin rosecutors received no monetary benefit from conviction, excet avoidance of the small fine which would 15 Hosted by The Berkeley Electronic Press

18 otherwise be imosed for makin a false accusation. Because rosecution was rivate, only the rosecutor s and defendant s stakes mattered for settlement. Althouh society s stake in a conviction miht have been lare, there was no institutional mechanism which forced the arties to take it into account in their settlement neotiations. 20 That defendants stakes were larer, at least on averae, is crucial to testin the Priest-Klein theory, because, as discussed in the next section, the Priest-Klein theory yields radically different redictions deendin on whether the stakes are symmetric or if rosecutors have larer stakes. If one considers only rosecutors monetary stakes, the conclusion that defendants stakes were larer would be nearly unassailable. Nevertheless, the analysis is comlicated by the fact that rosecutors robably received nonmonetary benefits from conviction. A conviction miht, for examle, deter others from committin offenses aainst the rosecutor in the future. 21 A rae victim miht also vindicate her reutation throuh a successful rosecution. In addition, rosecutors miht derive satisfaction from retribution, that is, from the unishment of the defendant. The manitude of these nonmonetary benefits is difficult to measure. It is ossible that they were so lare as to make rosecutors stakes equal to or larer than defendants. Nevertheless, as exlained in the next ararah, it seems most lausible to infer that the nonmonetary benefits were sinificant but small comared to the sanctions a convicted defendant would bear. 20 Of course, the fact that judes bean to ut defendants to trial in site of settlement reflected society s stake in the case. 21 See David Friedman, "Makin Sense of Enlish Law Enforcement in the Eihteenth Century," 2 U. Chi. Roundtable 475 (1995). 16 htt://law.beress.com/usclws-lss/art80

19 A rouh sense of the imortance of the nonmonetary benefits can be arnered by examinin how otential rosecutors chaned their behavior when settlement olicy chaned. As noted above, sometimes settlements were resected, while at other times they were not. When settlement was resected, victims could be motivated to rosecute by the rosect of a monetary (or in-kind) settlement, as well as the nonmonetary benefits mentioned above (deterrence, reutation, and retribution). When judes took jury verdicts in settled cases, however, defendants became reluctant to settle, because settlement did not rovide rotection aainst conviction and unishment. 22 Thus, when judes took jury verdicts in settled cases, rosecutors who continued to rosecute were motivated almost exclusively by the nonmonetary benefits of rosecution. When judes took jury verdicts in settled cases, otential rosecutors brouht only one-third as many rosecutions. 23 This suests that the nonmonetary benefits of rosecution were relatively small comared to the benefits of settlement, but still lare enouh to motivate some rosecutions. The fact that defendants entered into settlements voluntarily suests that the benefits of settlement were smaller than the sanctions uon conviction. Thus, the nonmonetary benefits of successful rosecution to the rosecutor were robably, on averae, smaller than the sanctions imosed on the defendant uon conviction. Since the rosecutor's stakes were rincially the nonmonetary benefits of rosecution and the defendant s stakes were rincially the sanctions, it is reasonable to conclude that the stakes were, on averae, asymmetric and larer for the defendant. 22 Daniel Klerman, Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1 (2001). 23 Id. 17 Hosted by The Berkeley Electronic Press

20 It should also be noted that, althouh the discussion so far has focused on the nonmonetary asect of the rosecutor s stakes, the defendant s stakes also had a nonmonetary comonent, even when the formal sanction was a fine. Conviction almost certainly injured the defendant s reutation. This reutational injury would increase the defendant s stakes at trial and thus lend further lausibility to the idea that the defendant s stakes were larer than the laintiff s. 3. Theory and Predictions In their seminal 1984 article, Geore Priest and Benjamin Klein analyzed the decision to settle or litiate and showed that litiated cases are not a random samle of all disutes. They arued that the cases which are most likely to be litiated are those where liability is hihly uncertain. When liability is certain, both laintiff and defendant will find it advantaeous to settle in order to avoid litiation costs. Similarly, when the defendant is clearly not liable, the laintiff will not even bother to sue, or, if she does sue, she will be willin to settle for a small amount. Thus, only the hard cases, in which laintiff and defendant can come to very different assessments of the outcome, are likely to be litiated. Because only the close cases will be litiated, Priest and Klein arue that, under certain, secial circumstances, laintiffs and defendants should each revail in about fifty ercent of the litiated cases. Priest and Klein s 1984 article also analyzed several factors which miht cause deviations from the fifty-ercent tendency, most imortantly asymmetric stakes. When the stakes for the defendant are larer, Priest and Klein showed that defendants should revail more often in litiated cases. In such situations, litiation is still more likely in 18 htt://law.beress.com/usclws-lss/art80

21 close cases, but defendants have stron incentives to settle cases in which they are more likely to be found liable and to litiate only those in which they are more likely to be found not liable. Because their losses from an adverse verdict are larer than laintiffs' ains from a favorable verdict, defendants are willin and able to make settlement offers that laintiffs will find attractive, even when defendants think a laintiff victory hihly uncertain. Conversely, only when the facts are clearly in the defendant s favor, will a defendant find the risk of an adverse judment more attractive than settlement. As a result, Priest and Klein redict that, when the stakes are larer for the defendant, defendants should revail more often than laintiffs in litiated cases. Conversely, if laintiffs have more at stake, litiated cases will contain more ro-laintiff verdicts. Fiure 2 illustrates the selection effect when the defendant has larer stakes. Number of Disutes Litiated Pro-defendant Verdicts Litiated Pro-laintiff Verdicts Pro-defendant Verdicts Y * Pro-laintiff verdicts Y Fault of the Defendant Fiure 2. Priest & Klein Model 19 Hosted by The Berkeley Electronic Press

22 Y measures the deree of the defendant s fault, and Y * is the decision standard. If the defendant s fault exceeds Y *, the defendant is found liable. If the defendant s fault is less than Y *, the defendant is found not liable. Cases resultin in litiation are shaded. Litiated cases resultin in verdicts of liability are shaded heavily, while those resultin in no liability are shaded lihtly. The unshaded area under the curve reresents settled cases. The fiure illustrates two imortant asects of the Priest-Klein model. First, litiation is most likely in close, hard cases. Thus, the shaded areas flank the decision standard, Y *. Second, because the stakes are asymmetric and larer for the defendant, defendants are more likely to settle cases which they exect to lose, and conversely more likely to litiate only those cases that they are likely to win. Thus, the shaded area to the left of the decision standard is larer than the shaded area to the riht. Of course, this rah is a simlification. Sometimes there will be litiation even when the defendant s fault is far from Y *, and sometimes the arties will settle even when the defendant s fault is close to Y *. Nevertheless, the fiure hels to illustrate the eneral tendency in the selection of disutes for litiation. The analysis in the recedin ararah is comlicated by the ossibility that rosecutor and defendant could disaree not only about whether the defendant was uilty, but also about the deree of the defendant s uilt. As noted above in section 2, juries did not always simly reort that the defendant was uilty as chared or comletely innocent. Rather, jurors sometimes returned verdicts of artial uilt, declarin the defendant uilty of a less severe offense than chared or of some but not all chared offenses. Priest and Klein do not discuss the ossibility of such artial verdicts, but their model can be 20 htt://law.beress.com/usclws-lss/art80

23 adated to take them into account. This modification miht also have some relevance to modern civil cases involvin multi-count comlaints. For examle, if the comlaint alleed both fraud and nelience, a jury verdict findin liability for nelience but not fraud could be analyzed as a artial verdict within the framework develoed here. The biest chane is that one needs two decision standards, Y and Y, rather than just one, Y *. Y reresents the deree of culability just sufficient to lead a jury to render a artial uilty verdict, while Y reresents the deree of culability just sufficient to lead a jury to render a uilty-as-chared verdict. Fiure 3 illustrates the Priest-Klein model modified to take into account three ossible verdicts. Litiated Innocent Verdicts Litiated Partial Guilt Verdicts Litiated Guilty-As-Chared Verdicts Number of Disutes Innocent Verdicts Y Partial Guilt Verdicts Y Culability of the Defendant Guilty-as-chared Verdicts Y Fiure 3. Modified Priest & Klein Model 21 Hosted by The Berkeley Electronic Press

24 As in Fiure 2, litiated cases are shaded, while unshaded areas under the curve reresent settled cases. As in the oriinal Priest-Klein model, disutes are more likely near the decision standards, that is, near Y and Y. Since the stakes are asymmetric, one exects more artial uilt verdicts in disutes neary and more innocent verdicts in disutes near Y. In Fiure 3, Y and Y have been laced so that the roortion of innocent, artially uilty, and uilty-as-chared verdicts aroximates the roortion in the dataset, as described in the next section. 24 That is, Y and Y are close toether, because there are relatively few artial-uilt verdicts in the dataset, and Y is rather far to the left, because there are fewer innocent than uilty-as-chared verdicts in the dataset. It is thus ossible to make some rouh redictions from the fiure about the settled and litiated cases. Guilty-as-chared verdicts will be relatively uncommon amon the litiated cases, because there are many uilty-as-chared cases in the full data set, but disutes near Y will result in relatively few such verdicts. Conversely, artial uilt verdicts will be disrorortionately common amon the litiated cases, because such verdicts are uncommon in the full data set, but disutes near Y will result in many such verdicts, while disutes near Y will contribute a few more. Finally, there will be more not uilty verdicts amon the litiated cases, because disutes near Y will result in lare numbers of innocent verdicts, while such verdicts are somewhat under-reresented in the full data set. 24 For this ararah, the droed cases have been omitted from the analysis. As discussed below, they require an additional modification of the Priest-Klein framework. 22 htt://law.beress.com/usclws-lss/art80

25 While the discussion above hels to develo intuitions, it is also useful to analyze the roblem more formally. The formal analysis tracks Priest and Klein s article, modifyin the mathematics to take into account artial verdicts. 25 In each disute, the defendant s true culability is Y. Prior to trial, each otential litiant forms an estimate of Y. Let Yˆ π be the rosecutor s estimate of the defendant s culability, and let Yˆ be the defendant s estimate of the defendant s culability. 26 Thus: and Yˆ = Y + π ε π Yˆ ε, = Y + where ε π and ε are indeendent random variables with zero exectation and identical standard errors, σ ε, where σ ε is know to the arties. Given σ ε and their estimates of Y, each arty can estimate the robability of a artial uilt verdict and a uilty-as-chared verdict. The rosecutor s and defendant s estimates of the robability of a artial verdict are: and Pˆ π = P( Y Y < Y Y π Pˆ = P( Y Y < Y Yˆ ). ˆ ) Similarly, the arties estimates of the robability of uilty-as-chared verdicts are: and Pˆ π = P( Y Y Y π Pˆ = P( Y Y Yˆ ). ˆ ) 25 Those unfamiliar with the Priest-Klein model may find it helful to review aes 6-12 and of The Selection of Disutes for Litiation, 13 J. Leal Stud. 1 (1985). 26 I use the Greek letters π and, rather than and d, to denote laintiff and defendant to avoid confusion reardin, which already denotes a artial uilt verdict. 23 Hosted by The Berkeley Electronic Press

26 Let verdict, and let J π and J π and J reresent the value to rosecutor and defendant of a artial uilty J reresent the value of a uilty-as-chared verdict. If, as I maintain, the stakes were asymmetric and reater for the defendant, then J π < J and J π < J. For simlicity, I will exress the rosecutor s valuation of artial uilty and uilty-as-chared verdicts as roortional to the defendant s valuations, so J π = βj and J π = βj, where β reresents the asymmetry of the stakes, 0 < β < 1. In addition, since the sanction for a artial verdict was less than the sanction for a uilty-as-chared verdict, J and J π and can be rewritten as α J and αβ J, resectively, where α reresents the deree to which artial verdicts resulted in lesser sanctions than uilty-as chared verdicts, 0 < α < 1. The arties will settle when the rosecutor s minimum settlement demand (A) is less than or equal to the defendant s maximum settlement offer (B). 27 A and B can be reresented as follows: and ˆ ˆ A = Pπ Jπ + Pπ J π C + S = αβpπ J + βpπ J C + B = Pˆ J + Pˆ J + C S = α Pˆ J + Pˆ J + C S, ˆ ˆ S where C π and C are the rosecutor s and defendant s litiation costs, and S π and S are their resective settlement costs. Since the arties settle when A B, the conditions for litiation can be simlified to: β Pˆ Pˆ + α( βpˆ Pˆ ) > ( C S) / J (1) π π 27 This is a stron and somewhat unrealistic assumtion, as it assumes that barainin never breaks down. Nevertheless, I have made the assumtion, because Priest and Klein make it. I briefly discuss the imlications of relaxin this assumtion in the last ararah of section htt://law.beress.com/usclws-lss/art80

27 where C = Cπ + C, = S + S S π, and J = J. It is difficult to make analytic redictions about what ercentae of litiated cases will result in innocent verdicts, artial verdicts, or uilty-as-chared verdicts, because these ercentaes will vary with the underlyin distribution of all cases (both settled and litiated) and with the values of α, β, σ ε, and (C-S)/ J. Some redictions can be made about the limit case, as σ ε, the standard deviation of the arties errors, oes to zero. 28 Nevertheless, since arties were robably unable to redict outcomes with recision, it is most helful to consider a rane of values of σ ε. Simulations are most aroriate for this task. Table 1 shows some simulation results under various assumtions. 28 See the next three footnotes. 25 Hosted by The Berkeley Electronic Press

28 Table 1. Simulation Results Assumtions Results Litiated cases Settled cases % litiated % innocent % artially % uilty as % innocent % artially % uilty as (C-S)/J β σ ε uilty chared uilty chared * * * * * * * * * & * * * * * * * * * All results were rounded to the nearest inteer and are based on simulations usin 50,000 observations (disutes). For all results, α = 0.33, as discussed in Section 2. In addition, it was assumed that culability was distributed normally amon the oulation of litiated and settled (i.e. non-droed) cases, and that the oulation of litiated and settled cases contained 25% innocent defendants, 23% artially uilty defendants, 52% uilty-as-chared defendants. These ercentaes were derived from the actual oulation of litiated and settled cases, as described in Table 2 below. 26 htt://law.beress.com/usclws-lss/art80

29 Althouh the table shows that many results chane dramatically as the arameters are chaned, a few results are robust across most secifications: 1) The ercent uilty-aschared is almost always hiher amon settled cases than amon litiated cases. 29 2) The ercent innocent is usually hiher amon litiated cases than amon settled cases. 30 3) The ercent artially uilty is almost always hiher amon litiated cases than amon settled cases. 31 The above three results ive rise to three redictions about the data: Prediction 1. There should be a reater roortion of innocent verdicts amon the litiated cases than amon the settled cases. Prediction 2. There should be a reater roortion of artial uilt verdicts amon the litiated cases than amon the settled cases. Prediction 3. There should be a reater roortion of uilty-as-chared verdicts amon the settled cases than amon the litiated cases. These redictions are consistent with those derived informally above from analysis of Fiure 3. The ararahs before and after Fiure 3 suest intuitions suortin the redictions. Of course, while these redictions hold true almost reardless of the value of the various arameters, many results chane dramatically with the arameters. For examle, the ercent litiated varies dramatically with the value of (C-S)/J. When (C- 29 It can be roven that, in the limit, as σ ε oes to zero, this result will hold as lon as the roortion of uilty-as-chared verdicts in the entire oulation of disutes is reater than (1-α)/2. Since the data suest that α = 0.33 and that the roortion of uilty-as-chared verdicts was 52%, this rediction holds in the limit. 30 This result can be roven, in the limit, as lon as β is not very close to 1. Note that four of the five excetions in the simulations occur when β is hih (β = 0.75). Note that results for the arameters (C- S)/J=0.2, β=0.5, σ ε =0.5, aear to be an excetion only because of roundin. Without roundin, the ercent innocent would be hiher amon the litiated than amon the settled cases (25.09 versus 24.74), althouh the difference is not statistically sinificant. 31 This result holds in the limit, as lon as the roortion of artial uilt verdicts in the entire oulation is less than half. 27 Hosted by The Berkeley Electronic Press

30 S)/J is ositive, litiation is very unlikely. This is similar to the simle Priest-Klein result with asymmetric stakes and reflects the fact that when litiation is more exensive than settlement (as nearly all modern scholarshi assumes), settlement is mutually advantaeous unless the arties estimates of the robability of revailin are extremely far aart. On the other hand, when (C-S)/J is neative, litiation is very common. A neative value of (C-S)/J means that settlement is more costly than litiation. While this is imlausible in modern litiation, it may have been true for medieval rivate criminal rosecutions, because litiation involved few monetary costs (because there were no lawyers) but settlement could result in the imosition of sinificant fines. Of course, litiation had some costs, such as the cost of travel and the cost of the litiants time, but it is at least ossible that the settlement fine was larer than these costs. Priest and Klein make no redictions about droed cases, that is, about cases which the rosecutor initiated in county court but then neither settled nor ursued to trial. Priest and Klein seem to assume that all cases will either be litiated or settled, and thus leave no room for such cases. This oversiht is artially exlained by the fact that the theoretical art of their article was written with emirical testin in mind, and few modern data sets distinuish between settled and droed cases. It also reflects Priest s view that, as a matter of theory, laintiffs will be able to extract some settlement (albeit a small one) even from weak cases, excet when litiatin aainst defendants who have 28 htt://law.beress.com/usclws-lss/art80

31 adoted a stratey of contestin every filed action to a verdict. 32 This view is idiosyncratic and in no way essential to the broader theory. 33 A more lausible aroach to droed cases builds on the models of litiation develoed by Gould, Landes, Posner, and Shavell, uon which Priest and Klein also relied for most of their analysis. Under these models, as lon as litiation is costly, if the rosecutor estimates that the exected benefit of litiation is low, then the rosecutor will not even initiate rosecution. If however, the rosecutor must initiate the rosecution before he has sufficient information, and if he later learns that the exected benefit is low (most lausibly, because he learns that the robability of conviction is low), then there will be cases which the rosecutor initiates but then dros. Such cases will consist redominately of weak cases. The fact that otential medieval rosecutors had to initiate their cases in county court within four weeks suests that they would often have had to initiate rosecution before they had sufficient information. Litiation was costly, not so much in monetary terms, but in time. Prosecutors were seldom reresented by counsel, so they did not have to worry about lawyer s fees, but litiation did require travel, time in court, and erhas time lobbyin jurors. As a result, one would redict that rosecutors would dro sinificant numbers of cases. This analysis leads to a fourth rediction: Prediction 4. Innocent verdicts should redominate amon the droed cases. So far, the analysis has assumed that settlement rotected the defendant from trial and unishment. Nevertheless, as discussed in Section 2, this assumtion became 32 Priest, Reexaminin the Selection Hyothesis: Learnin from Wittman s Mistakes, 14 J. Leal Stud. 215, 225 (1985). 33 Priest concedes that this assumtion contradicts the model of litiation develoed by Shavell and others, which was standard at the time. Id at 222, 224. Priest's osition, however, has ained some 29 Hosted by The Berkeley Electronic Press

32 increasinly false. Judes bean to send defendants to trial in site of settlement. By the 1250 s, they did so in sixty ercent of the cases. 34 While at first this new olicy robably took litiants by surrise, toward the end of the eriod studied here it is likely that the arties bean to anticiate that there was a sinificant chance that defendants would be tried and unished in site of settlement. It is thus imortant to adjust the model to take this into account. Let λ be the robability that the defendant will be tried in site of settlement. The fact that the defendant may be tried in site of settlement makes settlement less attractive to the defendant and lowers his maximum settlement offer (B) by the exected disutility of trial, ˆ [ ˆ λ P J P J ]. 35 On the other hand, the fact that + the defendant may be tried (and unished) in site of settlement makes settlement more attractive to the rosecutor and lowers his minimum settlement demand (A) by ˆ [ ˆ λ P J + P J ]. As a result, A and B can be rewritten as follows: π π π π and ˆ ˆ A = ( 1 λ )[ Pπ Jπ + Pπ Jπ ] C + S = (1 λ)[ αβpπ J + βpπ J ] C + B = ( 1 λ)[ Pˆ J + Pˆ J ] + C S = (1 λ)[ αpˆ J + Pˆ J ] + C S ˆ ˆ S, Since the arties settle when A B, the conditions for litiation can be simlified to: βpˆ π C S Pˆ Pˆ Pˆ + α( β π ) > (2) (1 λ) J suort from Lucian Arye Bebchuk, "A New Theory Concernin the Credibility and Success of Threats to Sue, " 25 J. Leal Stud. 1 (1996). 34 See Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1, 39 (2001). 35 Note that I am assumin that the defendant incurred no litiation costs from trial after settlement. That is, the defendant s maximum settlement offer is not further decreased by λ C. This is historically accurate, because trials after settlement often occurred without the resence of either arty, a fact made ossible by the self-informin jury. For similar reasons, I am also assumin the rosecutor incurred no litiation costs from trial after settlement. If one assumed that the arties incurred litiation costs from trial after settlement, the results would be only slihtly different. See next footnote. 30 htt://law.beress.com/usclws-lss/art80

33 where C = Cπ + C, = S + S S π, and J = J. The only difference between this condition which takes into account the ossibility of trial after settlement and condition (1) above which does not is that the riht hand side is divided by ( 1 λ). Since the effect of asymmetric stakes comes throuh the left hand side, Predictions 1 throuh 3 remain valid. The rincile effect of dividin the riht hand side by the ( 1 λ) is to make settlement more likely when C S is ositive (i.e. when litiation is more exensive than settlement), and to make litiation more likely when C S is neative (i.e. when settlement is more exensive than litiation). The intuition behind these effects is easiest to ras in the extreme case, where trial after settlement is a certainty (i.e. λ = 1). If litiation costs are hiher than settlement costs, then the arties should always settle, because settlement avoids litiation costs 36 but otherwise leads to the same result (i.e. trial) as settlin. For the same reason, when settlement costs are hiher than litiation costs, the arties should always litiate. For moderate robabilities of trial after settlement, the effects are corresondinly moderate: an increased tendency to settle when litiation costs are hiher than settlement costs and an increased tendency to litiate when settlement costs are hiher than litiation costs. These effects are confirmed by the simulations. Durin the 1250 s, the robability ( λ ) that the defendant would be tried in site of settlement was sixty ercent. Thus, takin into account the ossibility of trial after settlement means considerin values of (C-S)/J that are two and a 36 Note here that I am assumin that the arties incurred no litiation costs from trial after settlement. See revious footnote. If one assumed that the arties incurred litiation costs from trial after C [ S /(1 λ)] settlement, the riht hand side of condition (2) would be. This would have the effect of J makin settlement always less attractive, irresective of whether settlement costs were larer or smaller than settlement. Since, as discussed in section 4, it is most lausible to assume that settlement costs were reater than litiation costs, the redictions for the data are the same under either assumtion. 31 Hosted by The Berkeley Electronic Press

34 half (1/(1-0.6)) times hiher. By dislayin results for (C-S)/J equal to ± 0. 5as well as ± 0.2, Table 1 already illustrates the effect of multilyin (C-S)/J by 2.5. Predictions 1 throuh 3 hold equally well for ± 0. 5 as for ± Nevertheless, litiation is less likely for 0.5 than for 0.2, and more likely for 0.5 than for This analysis leads to a fifth rediction: Prediction 5. If litiation costs are hiher than settlement costs (i.e. if (C-S)/J>0), then the roortion of litiated cases should o down as arties learn that settlement may be disresected. Conversely, if litiation costs are lower than settlement costs (i.e. if (C-S)/J<0), then the roortion of litiated cases should o u as arties learn that settlement may be disresected. The fact that judes miht send a defendant to trial in site of settlement should have little direct effect on a rosecutor's decision to dro a case. Perhas the increased chance of settlement (when litiation is more exensive than settlement) miht ersuade some rosecutors to continue rosecution of cases they miht otherwise have droed. Conversely, the increased chance of litiation (when settlement is more exensive than litiation), miht induce some rosecutors to dro some cases they miht otherwise have continued. More imortantly, however, at the same time that judes bean sendin settled cases to trial, they also bean sendin droed cases to trial. This could have made droin a case more attractive, as a rosecutor could still et the satisfaction of conviction without the trouble of continued rosecution. On the other hand, droin a case involved foroin the ossibility of settlement and thus would have been an unattractive otion for all but the weakest cases. As a result, Prediction 4 that innocent 37 When (C-S)/J ets closer to one or neative one, the redictions bein to chane, althouh I do not yet fully understand why. Nevertheless, because there were few or no lawyers in these cases, both litiation costs and settlement costs were relatively low, so it is not necessary to consider such values of (C-S)/J. 32 htt://law.beress.com/usclws-lss/art80

35 verdicts should redominate amon the droed cases should still hold even when rosecutors anticiate that judes will send settlin defendants to trial. 4. Emirical Results The data used in this article were oriinally athered in order to calculate and exlain the chanin frequency with which aeals were brouht in the eriod The data set contains over a thousand aeals from select districts in fourteen Enlish counties, ranin from Kent in the south to Yorkshire in the north. The districts were chosen because a larer ercentae of their records survive. Some of the cases have been rinted and translated, while others exist only in Latin, archment manuscrits stored in the Public Record Office in London. Assault (includin beatin, woundin and mayhem) was the most common crime, reresentin about thirty-nine ercent of all aeals, but homicide (27%), theft (12%, includin simle larceny, burlary and robbery), and rae (10 %) were also common. In twelve ercent, the crime was rare (such as arson) or unrecorded. Most rosecutors were male, but a sinificant minority (36%) were female. 39 The cases from the eriod are the most useful to test selection theories, because, as mentioned above, they involve defendants cauht by surrise by the new treatment of settled cases and defendants who settled before the new treatment of settled cases became routine. The recise endin oint of 1263 was chosen, because that is the 38 Daniel Klerman, Settlement and the Decline of Private Prosecution in Thirteenth-Century Enland, 19 Law & Hist. Rev. 1 (2001). This article contains a detailed discussion of the data and summary statistics. 39 The lare number of female rosecutors is discussed in Daniel Klerman, Women Prosecutors in Thirteenth-Century Enland, 14 Yale J. L. & Hum. 271 (2002). 33 Hosted by The Berkeley Electronic Press

36 year an "eyre" (judicial circuit of the countryside) ended. The revious one had ended in 1258 and the next one did not end until Only cases from after 1239 are analyzed, because only such cases usually contain jury verdicts even in non-litiated cases. Nevertheless, even after 1239, there are many settled cases without jury verdicts. In addition, there are a number of non-rosecuted cases for which jurors did not reort whether the case had been settled or droed. The imlications of the incomlete data are discussed in section 5.A. The table below summarizes the jury verdicts in those cases for which there are data both on the defendant s uilt and on whether non-rosecuted cases were droed or settled: The table is consistent with the redictions of the revious section. On the most basic level, the data stronly suest that there is a selection rocess at work. The litiated cases are very different from both the settled and the droed cases in ercent uilty as chared, ercent artially uilty, and ercent innocent. The table is also 34 htt://law.beress.com/usclws-lss/art80