1

v: 3/12/2007 8:35 AM

STATE CIVIL APPEALS: AN EMPIRICAL PERSPECTIVE

Michael Heise and Theodore Eisenberg∗ Abstract Two findings dominate prior empirical studies of federal civil appeals. First, appeals courts are more likely to disrupt jury verdicts than bench verdicts. Second, trial court defendants fare better than plaintiffs on appeal. In this study we test the robustness of these two findings by extending our analysis of appeals from the federal to the state court context. In addition to complementing research on federal appeals, a greater understanding of state civil appeals is especially important insofar as the bulk of civil litigation—including appeals—takes place in state courts. Our study exploits the first large-scale database of state court civil appeals and our results largely confirm prior research. Specifically, state appellate reversal rates for jury trials and defendants exceed the reversal rates for bench trials and plaintiffs. Both descriptive analyses as well as more formal selection models point to appellate judges’ attitudes toward trial-level adjudicators as an important explanation for the outcome of civil appeals in state courts.

∗ Heise is Professor, Cornell Law School; Eisenberg is Henry Allen Mark Professor, Cornell Law School.

2

v: 3/12/2007 8:35 AM

State Civil Appeals: An Empirical Perspective

Theodore Eisenberg and Michael Heise∗ Abstract Two findings dominate prior empirical studies of federal civil appeals. First, appeals courts are more likely to disrupt jury verdicts than bench verdicts. Second, trial court defendants fare better than plaintiffs on appeal. In this study we test the robustness of these two findings by extending our analysis of appeals from the federal to the state court context. In addition to complementing research on federal appeals, a greater understanding of state civil appeals is especially important insofar as the bulk of civil litigation—including appeals—takes place in state courts. Our study exploits the first large-scale database of state court civil appeals and our results largely confirm prior research. Specifically, state appellate reversal rates for jury trials and defendants exceed the reversal rates for bench trials and plaintiffs. Both descriptive analyses as well as more formal selection models point to appellate judges’ attitudes toward trial-level adjudicators as an important explanation for the outcome of civil appeals in state courts. I. Introduction If we still do not know much about how our civil justice system behaves,1 we know even less about how the world of civil appeals behaves. And whatever we know about civil appeals, we know far less about state appeals than federal appeals even though the bulk of civil litigation—including appellate—takes place in state courts.2 Until recently, a paucity of helpful data impeded systematic analyses and a deeper, more textured understanding of civil appeals.

Limited appeals data have not dampened the formation of both opinions and conventional wisdom concerning the operation of and problems with our civil justice system in general and the appeals system in particular. Paradoxically, the absence of data can enhance the influence of opinions, anecdotes, and casual impressions. For example, in the appeals context advice to lawyers typically emphasizes the sacrosanctity of jury

∗ Heise is Professor, Cornell Law School; Eisenberg is Henry Allen Mark Professor, Cornell Law School. 1 See generally Michael J. Saks, Do We Really Know Anything About the Behavior of the Tort Litigation System--And Why Not?, 140 U. Pa. L. Rev. 1147 (1992). 2 Marc Galanter, The Vanishing Trial: An Examination of trials and Related Matters in Federal and State Courts, 1 J. Empirical Legal Studies 459, 506 (2004) (“The great preponderance of trials, both civil and criminal, take place in the state courts.”)

3

verdicts and warns that “appellate challenges to jury findings rarely succeed.”3 Not surprisingly, such advice informs and influences concrete decisions at the beginning of the litigation process when a lawyer’s decision about whether to select a jury trial pivots partly on popular perceptions about the comparative resilience of jury outcomes to appellate review.4 Holding conventional wisdoms and perceptions up to empirical light frequently uncovers surprises.5 One set of surprises about civil appeals emerged from studies of federal courts drawing on data from the U.S. Administrative Office (“AO”).6 Contrary to received wisdom, for example, prior studies have found that jury verdicts were more prone to reversal than bench verdicts.7 As well, defendants’ success with appeals systematically exceeded plaintiffs’ success.8 Moreover, these general findings presented with even greater force within specific case types.9

Efforts—such as ours—at extending existing research on federal civil appeals to state civil appeals accomplish several goals. Comparisons with state civil appeals provide an opportunity to assess the robustness of findings from prior work on federal appeals. Scholarly attention to state appeals is warranted in its own right as well because, as Professor Galanter notes, “the great preponderance of trials, both civil and criminal, take place in the state courts.”10 Finally, our study of state appeals benefits from a unique data set which, unlike the AO data set used in federal appeals studies, is designed expressly for systematic study of the appeals process11 and permits unprecedented analyses of the appellate processes in state courts. Although helpful descriptive work using the state appeals data set sketches the broad contours of the state appeals terrain,12 our study presents findings from the first systematic empirical analysis of state appellate activity that models critical aspects of the state civil appeals process.

3 See, e.g., George A. Somerville, Standards of Appellate Review, in APPELLATE PRACTICE MANUAL 25 (Pricilla Anne Schwab, ed.) (1992). 4 See, e.g., Roger Haydock & John Sonsteng, 81 TRIAL (1991). 5 See, e.g., Kevin M. Clermont & Theodore Eisenberg, Trial By Jury or Judge: Transcending Empiricism, 77 Cornell L. Rev. 1124 (1992)(finding that plaintiff win rates at trial before a judge and before a jury differ and in surprising directions). 6 See, e.g., Theodore Eisenberg, Appeal Rates and Outcomes in Tried and Nontried Cases: Further Exploration of Anti-Plaintiff Appellate Outcomes, 1 J. Empirical Legal Studies 659 (2004)(finding that reversal and appeal rates vary across case categories and by party status); Kevin M. Clermont & Theodore Eisenberg, Plaintiphobia in the Appellate Courts: Civil Rights Really Do Differ From Negotiable Instruments, 2002 U. Ill. L. Rev. 947 (2002)(exploring variation in party win rates); Kevin M. Clermont & Theodore Eisenberg, Appeal from Jury or Judge Trial: Defendants’ Advantage, 3 Amer. L. and Econ. Rev. 125 (2001)(finding a slight defendant “advantage” in the appellate courts). 7 See, e.g., Clermont & Eisenberg, Advantage, supra note 6, at 150-51, tbl.4. 8 Id. 9 See, e.g., Kevin M. Clermont, Theodore Eisenberg, & Stewart J. Schwab, How Employment-Discrimination Plaintiffs Fare in Federal Courts of Appeals, 7 Employee Rts. & Emp. Pol’y J. 547, 547-48 (2003) (describing how employment-discrimination plaintiffs “swim against the tide”). 10 Galanter, Vanishing, supra note 2, at 506. 11 Appeals studies using AO data, while important, must work with data sets not expressly designed for appeals studies. 12 See Bureau of of Justice Statistics Bulletin: Appeals from General Civil Trials in 46 Large Counties, 2001-2005 1 (July 2005)[hereinafter “BJS, Appeals”].

4

Findings from our study of state civil appeals largely comport with previous

findings from studies of federal appeals. In addition, our study supplies further analytic texture to comparisons between judges and juries as well as variations across case types within the comparatively understudied appeals context. For example, similar to civil trials in state courts,13 appeals are comparatively rare events. In our data set of 8,038 completed state court trials, only 12% stimulated an appeal. Of the 965 trials that continued to the appeal process, just over one-half culminated with a final appellate court decision. Finally, rarer still are cases that exhausted a state’s full appellate process, from trial court to the highest appellate tribunal. Of the 965 cases that initiated appeals, only 24 cases reached a state’s appellate court of last resort.14 For all practical purposes, trail court decisions resolve the bulk of legal disputes raised in state courts.

We also note that the state appeals playing field is uneven in two important ways. First, although state appellate courts affirmed trial court decisions in more than two-thirds of all cases, appellate courts were more likely to upset jury trials than bench trials. Second, defendants were far more successful than plaintiffs in securing a reversal. Not surprisingly, and as results from our selection model reinforce, a plaintiff victory in front of a jury was the most likely scenario to generate an appellate court reversal. Our results also suggest that appellate court attitudes and assumptions about trial courts are more influential on appeals’ outcomes than any selection effect. Section II describes our data, methodology, and research design. Section III presents descriptive results with a particular emphasis on the contours of the stream of cases from trial verdict through the appeals processes. We find that while appeals courts typically affirm trial court decisions, when reversals emerge they are more likely to favor defendants and bench trial decisions. Section IV reports regression results that confirm and refine the core results in Section III. Section V concludes. II. Data, Methodology, and Research Design Two distinct, though related, data sets supply the data necessary for this study. First, the “Civil Justice Survey of State Courts,” a project of the National Center for State Courts (NCSC) and the U.S. Bureau of Justice Statistics (BJS), gathers data directly from state court clerks’ offices on tort, contract, and property cases disposed of by trial during calendar year 2001. The 2001 data set covers state courts of general jurisdiction in a random sample of 46 of the nation’s 75 most populous counties. The 75 counties from which the sample of 46 was drawn include approximately 37 percent of the 1990 U.S. population and about one-half of all civil lawsuits filed in state courts.15 The 2001 data set includes information on 8,038 cases.16 13 Galanter, Vanishing, supra note 2, at 506. 14 The 24 cases that reached a state’s appellate court of last resort represent 2.5% of the cases appealed and 0.3% of cases litigated at trial. 15 For a general discussion, see Bureau of Justice Statistics Bulletin: Civil Justice Survey of State Courts, 2001: Civil Trial Cases and Verdicts in Large Counties, 2001 (Apr. 2004) [hereinafter “BJS, Trials 2001”]. For a more technical source and the data set codebook, see Bureau of Justice Statistics, U.S. Dep’t of Justice, Civil Justice Survey of State Courts, 2001, (Inter-Univ. Consortium for Pol. & Soc. Research, No.

5

The second data set used in this study enhanced the first. NCSC and BJS implemented a follow-up study that supplemented the 2001 trial study by tracking the 1,204 cases where the trial verdict or judgment, concluded during calendar 2001, was appealed to an intermediate appellate court or a state’s court of last resort by April 2005.17 Of this 1,204 appeals case universe, 47 appeals (3.9 percent) were excluded from many analyses because the appeals lacked critical information about which party prevailed at trial.

Because we are most interested in appeals from adverse trial court decisions, we excluded cases where it was not clear that the nature of the appeal was classically adverse. For example, if the trial court ruled for the plaintiff, most would expect that the defendant would be the appellant in any appeal. Conversely, where trial court ruled for the defendant one would expect the plaintiff to be the appellant. For the most part (in 83.4% of the appeals), these expectations were met. Discarding the small sub-group of non-adverse appeals generates a universe of 965 usable appeals and permits a more accurate picture of appeals by plaintiffs and defendants from trial court judgments entered against them.18 Consequently, the appeal rate is the percentage of trial court judgments for one party or the other that the losing party puts onto the appellate docket.19 After filtering appeals that did not involve traditionally adverse parties, computations of reversal and affirmance rates were straightforward. The reversal rate is the percentage of cases of those appeals that generate a formal legal conclusion that either reversed20 or remanded the trial court decision. The affirmance rate, by contrast, is the percentage of cases that generated an appellate court decision affirming the trial court.

Merging the two NCSC and BJS data sets generates a unique source of data: a longitudinal view of the universe of state appellate activity drawn from the most representative sample of state trial activity in the United States. Such a data set facilitates the systematic study of appeals of trials.21 With direct access to the state court clerk’s offices, as well as approximately 100 trained coders recording data, the data set avoids traditional limitations incident to relying upon litigants or third party to report. Self-

3957, June 2004) [hereinafter “ICPSR, Trials 2001”]. See also Eisenberg et al., Juries, Judges, and Punitive Damages: Empirical Analyses Using the Civil Justice Survey of State Courts 1992, 1996, and 2001 Data, 3 J. Empirical Legal Studies 263, 267-68 (2006)(describing the 1992, 1996, and 2001 data sets); Catherine M. Sharkey, Unintended Consequences of Medical Malpractice Damage Caps, 80 N.Y.U. L. Rev. 391, 446-450 (2005)(same). 16 Although the 2001 study’s final sample includes 8,311 cases (see ICPSR, Trials, 2001, supra note 15, at 4), the data set includes usable information on 8,038 cases. See id. at 5 (noting the final data set includes 8,038 cases); Sharkey, supra note 15, at 446 (“The 2001 dataset includes 8038 cases”). 17 For a general description of the appeals data set, see BJS, Appeals, supra note 12. Of the 1,204 cases that involved an appeal, 15 (or 1.5 percent) remained pending at the end of the April 2005 study period. Id. at 1. 18 Data on prevailing party at trial are missing for 3.9 percent of the appeals data set. 19 For a similar approach to defining the appeal rate, see Clermont & Eisenberg, Advantage, supra note 6, at 129. 20 By reversed we mean in whole or in part. 21 See, e.g., BJS, Appeals, supra note 12.

6

reports, common in many commercial verdict reporters, typically overstate plaintiff win rates and damage award levels.22

Despite the state civil appeals data set’s unique strengths, it is not without

limitations, however. Because our sample focuses on the nation’s largest counties and state courts of general jurisdiction, the data might not convey those aspects of the civil justice system, if any, peculiar to smaller counties or rural areas or to cases heard in special jurisdiction courts.23 Moreover, the focus on state courts renders unclear any implications for federal courts.24

Selection effects also influence our data set in various ways. As Table 1 illustrates, the stream of cases encounters an array of filters as it proceeds from the civil dispute stage through the trial and appeals processes. As only a small fraction of civil actions filed wind up reaching trial the pool of tried cases may systematically differ from the larger pool of civil disputes from which they emerged. The appeals process itself imposes additional filters. Only 12% of the tried civil cases initiated the appeals process. Fewer cases still (6.8%) pursued the appeals process to decision. Of the cases that began the appeals process just over one-half (56.9%) completed it.

Indeed, there are strong theoretical reasons to expect influences flowing from these various selection effects. Expectations theory predicts that objectively strong and weak civil cases will settle or reach conclusion prior to reaching a conclusion at trial.25 Litigants that take cases to trial do so partly because they can afford to pursue trial litigation and, presumably, because they sense some reasonable level of uncertainty—factual or legal—as to a case’s outcome.26 The sub-pool of cases that withstand settlement, motions for directed verdicts and summary judgments, and other factors is more likely comprised of cases whose totality of underlying merits resides somewhere in the “gray middle area.” A similar set of filters arise anew during post-trial appeals

22 See, e.g., Theodore Eisenberg et al., The Predictability of Punitive Damages, 26 J. Legal Stud. 623, 614, n.53 (1997)(finding bias in commercial verdict reporter samples); Deborah Jones Merritt & Kathryn A. Barry, Is the Tort System in Crisis? New Empirical Evidence, 60 Ohio St. L.J. 315, 324-26 (1999) (same); Erik K. Moller et al., Punitive Damages in Financial Injury Jury Verdicts, 28 J. Legal Studies 283, 335 (1999) (reporting reasonable levels of confidence in the jury verdict reporters but acknowledging some potential bias). 23 Despite the absence of a clear theoretical explanation, the influence of geography, and whatever that might entail, remains a consistent finding in the research literature. See, e.g., Eisenberg, Predictability, supra note 22, at 630-31 (finding geography relevant to punitive damages); Teresa A. Sullivan, et al., As We forgive Our Debtors: Bankruptcy and Consumer Credit in America 339-40 (1989)(geography influencing bankruptcy filings); Michael Heise, Justice Delayed?: An Empirical Analysis of Civil Case Disposition Time, 50 Case Western Res. L. Rev. 813, 836-38, 847-48 (2000)(noting a geographic influence on case disposition time). The influence of geography is not limited to American courts. See, e.g., Eisenberg, Predictability, supra note 22, at 631 n.26 (finding a geographic effect in courts in Japan, Sweden, and Finland). 24 For a discussion about how the financial stakes at issue might distinguish state and federal appeals see supra Subpart III.A. 25 Theodore Eisneberg, Litigation Models and Trial Outcomes In Civil Rights and Prisoner Cases, 77 Geo. L.J. 1567, 1571 (1989) 26 Professors Priest and Klein, among others, previously articulated and developed this point. See generally George L. Priest & Benjamin Klein, The Selection of Disputes for Litigation, 13 J. Legal Stud. 1 (1984).

7

processes.27 These various selection filters influence the case stream over time and in a manner that generates a distributionally skewed subset of appeals. III. Initial Observations

We now turn to those cases that initiated the appeals process, explore appeal rates, and assess how they vary across trial, party, and case types.

TABLE 1: STATE AND FEDERAL CIVIL TRIALS, APPEALS & OUTCOMES

State: Civil Trials

Appealed (%)

Appealed to Compl.

(%)

Reversal Rate (%)

Federal: Civil Trials

Appealed

(%)

Appealed to Compl.

(%)

Reversal Rate (%)

All trials 8,038 12.0 6.8 32.1 21,415 20.9 10.0 20.7 Judge trials 1,873 14.0 7.6 27.5 6,258 22.4 10.8 20.5 Jury trials 6,165 11.4 6.6 33.7 15,157 20.2 9.7 20.8 Party appealing:

Defendant 12.7 6.6 41.5 19.7 8.1 28.4 Plaintiff 11.3 7.1 21.5 22.2 12.2 14.8 (N)

8,038

965

549

176

21,415

4,476

2,143

444

NOTE: SOURCES: (State) U.S. Dept. of Justice, Bureau of Justice Statistics, Supplemental Survey of Civil Appeals, 2001 (ICPSR

4539); (federal) Clermont & Eisenberg (2001, 131, tbl.1). A. The Vanishing Appeal?

If state civil complaints that result in a trial on the merits are becoming increasingly rare over time,28 rarer still, as Table 1 illustrates, are the 12% of trial verdicts that initiated the appeals process. Fewer cases still (6.8%) resulted in an appellate court decision. Assessing whether the 12% appeal and the 6.8% appeal-to-completion rates are large or small and how they might trend over time requires context. Regrettably, a helpful empirical context eludes. Our cross-sectional data provide only a snap-shot of the state civil appeals world rather than a picture of trends over sustained periods of time. With direct context lacking, we turn instead to various sources of indirect context.

The federal civil appeals rate provides one plausible reference point to assess our observed state civil appeals rate (12%). At first glance, Clermont and Eisenberg’s federal appeals rate of just over 20 percent29 suggests that our state appeals rate (12%) may be small both in comparative and absolute terms. Structural differences between the streams of federal and state appeals, however, help explain the difference in federal and state appeals rates. Important differences in the types of claims pursued in state and federal 27 Id. at 54 (arguing that, aside from possible precedential concerns, selection effect applies “indistinguishably to trail and appellate disputes”). 28 Galanter, Vanishing, supra note 2, at 509 tbl.5 (analyzing state courts of general jurisdiction for 10 states from 1992-2002). 29 Clermont & Eisenberg, Advantage, supra note 6, at 131 tbl.1.

8

appellate courts might account for appeals rate differences. For example, to access the federal court system litigants must either raise a federal question30 or plead for damages that exceed $75,000 and establish diversity among parties.31 Either jurisdictional requirement might route systematically higher stakes cases into federal courts and away from state courts. Just as higher stakes case tended to be litigated more than their lower-stakes counterparts, higher stakes cases might similarly exert upward pressure on decisions to appeal.32 Therefore, if federal trials typically involve higher stake claims than those pursued in state trials, we should expect that the federal appeals rate to exceed the state appeals rate.

A steady diminution of trial activity over time might also contribute to differences in federal and state appeals rates. Marc Galanter’s important study of trial activity and rates documents a profound and sustained reduction in civil trial activity in federal and state courts.33 In light of a decline in trial activity as a percentage of case filings, one might expect a similar decline in litigants’ appetite for post-trial appeals. Insofar as the Clermont and Eisenberg’s study of federal appeals involves a slightly earlier sample of (1988-97) than our study of state appeals (2001-05), our finding of a lower state appeals rate might owe to a timing gap between the two studies and capture a downturn in appeals activity. This alternative explanation, however, assumes that federal and state trial and appeals rates trend similarly in both magnitude and direction. At least one study of federal trials and appeals between 1987 and 1996 found that the federal appeals rate remained essentially constant during that decade, thereby potentially calling into question the assumed time trend.34 The influence of motor vehicle cases on state courts also contributes to a discrepancy between state and federal appeals rates. Motor vehicle cases account for 35.1% of our pool of state trials, by far the more popular case type. When it comes to appeals, however, motor vehicle’s rate (4.0%) is the lowest across all individual case types. Consequently, the sheer number of motor vehicle cases and its low appeals rate combine to depress the overall state appeals rate (12%). We can increase (albeit somewhat artificially) the overall state appeals rate from 12% to 16.3%, and bring the overall rate closer to the federal appeals rate (20.9%), by eliminating motor vehicle cases from our sample.

30 28 U.S.C.A. ' 1331. 31 28 U.S.C.A. ' 1332 (a). 32 See generally RICHARD A. POSNER, ECONOMIC ANALYSIS OF LAW 557 (1992) (noting that cases with larger stakes at issue are more likely to result in litigation); Samuel R. Gross & Kent D. Syverud, Getting To No: A Study of Settlement Negotiations and the Selection of Cases for Trial, 90 Mich. L. Rev. 319, 352 (1991) (noting the influence of the size of a case’s “stakes” on a party’s willingness to go to trial). 33 Galanter, supra note 5, at 462-63 tbl.1 (noting a decline in federal civil trials as a percentage of federal civil dispositions), 507 tbl.4 (same for state courts). 34 Theodore Eisenberg, Appeals Rates and Outcomes in Tried and Nontried Cases: Further Exploration of Anti-Plaintiff Appellate Outcomes, 1 J. Empirical Legal Studies 659, 668 fig.1 (2004). Eisenberg’s finding of an approximately 40 percent tried case appeals rate owes to some extent to differences in case matching methodology. For a discussion see id. at 665-67.

9

Regardless of whether the overall state civil appeals rate (12%) is more plausibly characterized as large or small, the overall rate masks important variation across party and trial types as well as case types. These observed variations largely track variations observed in the federal system. As Table 1 illustrates, defendants who lost at trial were more likely than plaintiffs to initiate an appeal (12.7% v. 11.3%). Our findings comport with party appeals rates in the federal context, though Professors Clermont and Eisenberg report a slightly larger gap between defendant- and plaintiff-initiated appeals (22.2% v. 19.7%). State litigants were more likely to appeal bench than jury trial decisions (14.0% v. 11.4%). Although Clermont and Eisenberg reported a small appeals rate gap between trial types and concluded that “nothing striking distinguishes jury trials from judge trials, from the overall vantage,”35 the appeals rate gap between trial types in our study achieved statistical significance.36 The highest appeals rate combination belonged to defendants who lost at a bench trial. The combination associated with the lowest appeals rate involved plaintiffs who lost in front of a jury.

In addition to party and trial type variation, as Table 2 shows, appeal rates also varied, sometimes substantially, across case types. Although 12% of all state tried cases were appealed, professional (non-medical) malpractice, employment contracts, and product liability cases were appealed at rates far above the mean (30.9%, 29.5%, and 25.6%, respectively). As previously discussed, motor vehicle cases were the least likely case type to generate an appeal (4.0%). If differences in case stakes helps explain differences in the federal and state appeals rates, similar differences might also explain variations in appeals rates across state case types. To put the point slightly differently, the appeals rate for professional malpractice claims may exceed that of motor vehicle claims because of differences in the financial stakes contested in such cases.

35 Clermont & Eisenberg, Advantage, supra note 6, at 130. 36 Chi-Square test statistic= 9.088; p= .003.

10

TABLE 2: STATE APPEAL AND REVERSAL RATES, BY CASE CATEGORY AND TYPE

Case Category

Appeals Filed

Appeals Filed (%)

Appeals Completed

Appeals Completed (%)

Overall Reversal Rate (%)

Defendants’ Reversal Rate (%)

Plaintiffs’ Reversal Rate (%)

D-P Reversal Difference

All cases 965 12.0 549 56.9 32.1 41.5 21.5 .000 Motor vehicle 114 4.0 54 47.4 25.9 28.0 24.1 .766 Assault, slander, libel 48 15.4 27 56.3 44.4 73.3 8.3 .001 Product liability 30 25.6 21 70.0 33.3 30.8 37.5 1.00 Dangerous premises 92 10.0 54 58.7 27.8 29.0 26.1 1.00 Medical malpractice 129 15.2 73 56.6 27.4 33.3 23.9 .424 Prof. malpractice 21 30.9 14 66.7 42.9 75.0 0.0 .010 Other tort 65 18.0 45 69.2 24.4 36.0 10.0 .079 Employment contract 83 29.5 52 62.7 50.0 61.5 38.5 .165 Fraud 78 18.2 41 52.6 31.7 39.3 15.4 .164 Lease 26 13.8 13 50.0 23.1 50.0 0.0 .070 Seller plaintiff 118 15.3 55 46.6 32.7 44.4 10.5 .015 Buyer plaintiff 89 16.8 50 56.2 30.0 38.5 20.8 .224 Other contract 36 17.1 20 55.6 40.0 50.0 25.0 .373 Property 36 20.5 30 83.3 26.7 27.3 26.3 1.00 Tort

490

9.2

288

57.7

29.5

38.2

20.8

.002

Contract 430 17.8 231 53.7 35.9 46.3 21.6 .000 Property 36 20.5 30 83.3 26.7 27.3 26.3 1.00 Jury trial

703

11.4

407

57.9

33.7

42.3

22.8

.000

Judge trial 262 14.0 142 54.2 27.5 38.7 18.8 .013 (N) 965 965 549 549 176 120 56

NOTE: The “Defendants’ Reversal Rate (%)” and the Plaintiffs’ Reversal Rate (%) columns illustrate where an appealing defendant and plaintiff, respectively, persuaded an appeals court to reverse the trial court judgment. The final column, “D-P Reversal Difference” tests the statistical difference between the defendants’ and plaintiffs’ reversal rates. SOURCE: U.S. Dept. of Justice, Bureau of Justice Statistics, Supplemental Survey of Civil Appeals, 2001 (ICPSR 4539). B. Appeal “Drop-Outs” If comparatively few litigants in state trials pursued an appeal, fewer still pursued an appeal to a decision on the merits. Of the 965 cases that initiated the appeals process, just over one-half (56.9%) pursued and appeal to completion.37 Appeals that dropped-out prior to decision were resolved through case settlement, withdrawal, or pretrial disposition. Although we discussed previously how state and federal litigants’ appetites for initiating an appeal may differ,38 once the appeals process was launched the appeals-to-completion rate in our study is similar to the rate found in a study of federal appeals.39

As Table 1 illustrates, although state appeals-to-completion rates varied across party and trial types as well as case categories, they varied less noticeably than appeals rates. Although defendants who lost at trial were more likely than plaintiffs to initiate an

37 By “completion,” we mean an appeals court decision. 38 See supra Sub-Part III.A. 39 Eisenberg, Appeal Rates, supra note 6, at 660 (finding that “about one-half” of the federal appeals pursued to a conclusion on the merits).

11

appeal, defendants were slightly less likely to pursue their appeals to a completion on the merits (6.6% v. 7.1%). Regarding trial types, bench trials were more likely to be appealed and to be appealed to completion (7.6% v. 6.6%) than jury trials.

Table 2 reveals how appeals-to-completion rates varied across case types. With the overall 56.9% appeals-to-completion rate serving as a reference point, product liability, other tort, and property appeals were more likely pursued to completion by litigants. In contrast, motor vehicle and seller plaintiff appeals were more likely to be resolved through settlement or pre-(appellate) trial disposition.

Any focus on those appeals pursued to completion risks ignoring the sizable number of appeals initiated but resolved prior to appellate court decision through settlement, withdrawal, or pre-trial disposition. What might explain the overall 43.1% appeals melt-rate? Table 3 charts state civil appeals’ attrition from trial court decision through a state’s initial appeals court and to a state’s highest appellate court. Interestingly, only 24 (or 0.3%) cases from our sample persisted through an accepted appeal by a state’s highest court.

Clearly, legal strategy plays an important role in explaining why litigants initiate but do not complete appeals.40 By launching an appeal a party can impose additional litigation costs on an opponent, delay final disposition of a case, and encumber a trial court award.41 While trial courts decisions disposed of the overwhelming majority (86%) of state civil claims, for some percentage of litigants a trial court judgment is best understood as an opening bid that influenced (but not necessarily resolved) the settlement process that persists even after a notice of appeal has been filed.

Setting aside purely strategic appeals, some degree of settlement activity owes to the normal forces at work in litigation. Moreover, on some level, appeals settlement activity reflects an increase in formal institutional activity designed to promote settlements during the appeals process. For example, state appeals courts in Michigan operate a settlement program42 which became fully operational in February 1998.43 Whether appeals settlement programs, such as Michigan’s, stimulate cases to settle that otherwise would not is not clear. Nevertheless, such programs illustrate institutional commitments to appeals case settlement.

40 See, e.g., Lynn M. LoPucki & Walter O. Weyrauch, A Theory of Legal Strategy, 49 Duke L.J. 1405, 1416 (2000) (noting various strategic aspects incident to the Texaco-Pennzoil appeal). 41 See, e.g., Harlon L. Dalton, Taking the Right to Appeal (More or Less) Seriously, 95 Yale L.J. 62, 85 (1985)(noting various reasons motivating a party’s decision to file an appeal). 42 See MCR 7.213(A). 43 For a discussion see James N. McNally, Lessons Learned in the Court of Appeals Settlement Program, 79 Mich. B.J. 488, 489 (2000).

12

TABLE 3: STATE COURT APPEALS & APPELLATE OUTCOMES

Total civil appeals (number)

IAC appeal withdrawn, t’ferred, or dismissed

IAC decision

IAC decision (affirm)

IAC decision (reverse)

IAC decisions appealed to COLR

Cases accepted by COLR

All trials 965 416 549 373 176 183 24 Judge trials 262 120 142 103 39 37 3 Jury trials 703 296 407 270 137 146 21 Party appealing: Defendant

551

262

289

169

120

98

14

Plaintiff 414 154 260 204 56 85 10 Tort cases

499

211

288

203

85

93

17

Contract cases 430 199 231 148 83 83 7 Property cases

36 6 30 22 8 7 0

NOTE: A state’s initial appeals court (IAC); a state’s appellate court of last resort (COLR). SOURCE: U.S. Dept. of Justice, Bureau of Justice Statistics. Supplemental Survey of Civil Appeals, 2001 (ICPSR 4539). C. Appeals Outcomes

Theory and folk-lore influence important perceptions about the outcome of appeals. Selection theory implies that, similar to trial outcomes, appeals outcomes (reverse or affirm) should be roughly similar and distribute equally between plaintiffs and defendants.44 Attorney folk-lore also emphasizes jury verdicts’ insulation from reversal by appellate courts.45 Neither theory nor folk-lore, however, satisfactorily explains our major findings. Appeals courts were far more likely to affirm than reverse a trial court decision (67.9% versus 32.1%). Moreover, appeals courts were more inclined to reverse jury than judge decisions. Finally, defendants were more successful than plaintiffs in reversing an adverse trial court decision. Selection theory predicts that state trials (and appeals) should approximate fifty percent in terms of the two basic outcomes.46 At the state trial court level selection theory fares reasonably well as a predictor of outcome distributions—at least at the aggregate level—as they do not obviously favor plaintiffs or defendants.47 However, aggregate win rates in trial courts mask critical variation across case types. In 2001, for example, plaintiffs win rates ranged from a high of 72.7% (mortgage foreclosure cases) to a low of 26.8% (medical malpractice cases).48 44 Professors Priest and Klein, among others, previously articulated and developed this point. See generally Priest & Klein, supra note 26, at 54 (arguing that, aside from possible precedential concerns, selection effect applies “indistinguishably to trail and appellate disputes”). 45 See, e.g., George A. Somerville, Standards of Appellate Review, in Appellate Practice Manual 25 (Priscilla A. Schwab, ed.) (1992). 46 See Priest & Klein, supra note 26, at 17. 47 See, e.g., BJS, Trials, 2001, supra note 15, at 4 tbl.5 (reporting an overall plaintiff state trial win rate of 55.4%). 48 Id.

13

Regardless of how well one might conclude that selection theory predicts state trial outcomes, selection theory falters when it comes to explaining state appeals outcomes. As Table 1 makes clear, we find an overall 32.1% reversal rate. Notably, our observed reversal rate is closer to the federal appeals reversal rate (20.7%) than the (approximately) 50% rate selection theory predicts. Thus, the filtering mechanisms that sorted cases pre-trial along probabilities of success stumbled when it came to filtering decisions to appeal. Clearly, if litigants properly assessed their respective prospects for success on appeal the reversal rate should be closer to fifty percent.

Perhaps even more dramatic is the starkly asymmetric distribution of appellate outcomes between plaintiffs and defendants and jury and judge trials. As Table 1 shows, defendants were far more likely than plaintiffs (41.5% versus 21.5%) to successfully reverse an adverse trial court decision. Indeed, from the perspective of a plaintiff victorious at trial, the appeals process offered a chance to retain victory not far from what a coin-flip would predict. For defendants, by contrast, victories at trial were far more secure. Moreover, as Table 2 illustrates, the reversal rate difference favored defendants in both trial types and all but one (product liability) case type. As the final column in Table 2 illustrates, the overall defendant and plaintiff reversal rates difference (41.5% v. 21.5%) achieved statistical significance, along with a few individual case types.49

Table 2 also conveys the level of variation across case types in terms of the defendant advantage at the appellate level. In five of the 14 case types considered, the defendants’ reversal rate met or exceeded 50%. In none of the case types considered did the plaintiff reversal rate exceed 40%. Thus, defendants emerged from the state appellate process in a far better position than they left the trial court.

Finally, although judge trials were more likely to involve an appeal than jury trials (14% versus 11.4%), Table 1 shows that, contrary to conventional wisdom,50 appellate courts were kinder to judge than jury decisions (33.7% v. 27.5%). Moreover, the distribution of reversal rates enjoyed by defendants and plaintiffs differed as well across trial types. As Table 2 illustrates, these asymmetrical differences achieved statistical significance.

On balance our findings comport with prior research and evidence an appeals court tilt favoring defendants, especially defendants that lost in a jury trial.51 Because the evidence is descriptive, however, it can easily mask an array of complex and confounding factors, including the trial court’s behavior, the decision to appeal, and posttrial settlement conduct. To better ensure we do not prematurely dismiss selection theory, we need to formally model both the decision to appeal and the appellate case outcome.52

49 The case types that achieved statistical significance include: assault, slander, and libel cases, professional (non-medical) malpractice, and seller plaintiff cases. 50 See infra notes 3-4 and accompanying text. 51 Clermont & Eisenberg, Advantage, supra note 6, at 138. 52 See, Subsection IV, infra.

14

IV. Evaluating Competing Explanations for a Defendant Advantage by Modeling Appeals The asymmetrical distribution of appeals outcomes requires explanation. Two general theories complete to explain our central finding—an appellate court tilt favoring defendants, especially defendants that lost at trial in front of a jury. As both approaches have been previously explained in significant detail elsewhere,53 what follows is a summary.

The attitudinal hypothesis emphasizes the possibility that either trial courts exhibit a pro-plaintiff bias or appeals court perceive (or mis-perceive) trial courts as pro-plaintiff. Decisional bias is implied by differences in trial and appeals court outcomes as well as the starkly asymmetrical distribution of appeals outcomes between plaintiffs and defendants. This hypothesis assumes that bias can influence appellate (and trial) outcomes, and this assumption enjoys empirical support.54 What might plausibly account for an appellate court bias against plaintiffs and what are we to make of it? First, trial court bias favoring plaintiffs may exist. One source of potential bias flows from natural empathy with a victim, an impulse to make victims whole, and a willingness to redistribute wealth from a comparatively deeper economic pocket to an aggrieved plaintiff. In addition, unlike appellate courts, trial courts do not deal with a “cold” record and are less concerned with opinion-writing and the future effect of their decisions.55 Second, even if trial court bias does not exist in fact, appellate courts may assume that such a bias exists. Regardless of whether appellate courts accurately or inaccurately assume that trial courts favor plaintiffs, appellate courts may tilt in defendants’ favor to off-set a perceived trial court bias favoring plaintiffs. Although the starkly differing reversal rates suggest that either trial courts tilt in favor of plaintiffs to a degree that requires correction by appeals courts or that appeals courts tilt toward defendants because they assume (perhaps erroneously) a trial court bias, or some combination of both, our results (and research design) do not lead us to one explanation over the other. Findings from research that focuses on just this question (that is, whether the trial courts exhibit bias or whether appeals courts assume trial court bias) provide little support for the actual trial court bias explanation.56 As a consequence, appeals courts’ mis-perceptions of trial court adjudicators provide a more persuasive explanation for our starkly differing reversal rates.

A second plausible hypothesis—the selection hypothesis—posits that trial and appellate courts systematically face different streams of cases and that these different 53 Clermont & Eisenberg, Advantage, supra note 6, at 141-49. 54 See, e.g., LoPucki & Weyrauch, supra note 39, at 1484-85 (2000) (noting the documented influence of an array of non-legal factors in judicial decisionmaking). 55 See, e.g., Michael E. Tigar, FEDERAL APPEALS, 2D 8 (1993). 56 See, e.g., Richard Lempert, Why Do Juries Get a Bum Rap? Reflections on the Work of Valerie Hans, 48 DePaul L. Rev. 453, 454-45 (1998); Michael J. Saks, Public Opinion About the Civil Jury: Can Reality Be Found in Illusions?, 48 DePaul L. Rev. 221, 229-30 (1998); Neil Vidmar, The Performance of the American Civil Jury: An Empirical Perspective, 40 Ariz. L. Rev. 849, 868-71 (1998).

15

case streams explain our different appellate reversal rates. Systematically different cases might flow from plaintiff and defendant’s differential stakes at issue in a lawsuit. Selection theory implies that the side with comparatively more at stake should be willing to settle their weakest cases and, in so doing, increase their win rate for those cases fully adjudicated.57 The plaintiff win rate at trial was approximately 55.4% and it remains possible that the slightly higher plaintiff trial win rate might evidence plaintiffs’ higher stakes.58

Problems for the selection hypothesis quickly arise, however. Just as plaintiffs’ higher stakes in cases should boost their trial win rates, the selection hypothseis also predicts that the side with lower stakes (defendants) would appeal more.59 Indeed, although defendants did, in fact, appeal slightly more than plaintiffs (12.7% v. 11.3%), this difference did not achieve statistical significance.60 Moreover, although defendants appealed more than plaintiffs, defendants appealed to conclusion slightly less than plaintiffs (6.6% v. 7.1%), and, again, at a level that was not significantly different.61

More important, however, is that plaintiffs, whose higher case stakes should make them more selective about what cases to push at trail, should be equally careful about only pushing strong cases on appeal. If so, appeal reversal rates for plaintiffs would approximate the defendants’ reversal rate, if not exceed it. As our findings (in Table 2) make abundantly clear, however, this was plainly not the case as defendants enjoyed a far higher appeals reversal rate than plaintiffs (41.5% v. 21.5%), and the difference in defendants’ and plaintiffs’ reversal rates was strongly significant.62 Consequently, selection theory appears to exert little explanatory force in the appeals context. Moreover, to the extent that any selection effect exists, the selection does not appear to pivot on parties’ perceptions about an appeal’s legal strength. A more rigorous assessment of these competing theoretical explanations about the appeals process requires empirical testing. To do so we constructed a model of the appeals process. Efforts to model appellate outcomes must first account for the selection bias generated by the necessary decision to appeal a trial court verdict and, second, for the outcome of the appeal itself. To account for a selection effect generated by the predicate decision to pursue (and complete) an appeal and, in turn, how it might influence the outcome of an appealed case, we used a Heckman probit model. A. Testing hypotheses by modeling the decision to appeal and the appeals outcome What variables should appellate models include? With respect to the decision to appeal, parties’ perceptions (correct or not) about how appellate courts react to jury trials compared to bench trials, and to plaintiff trial court wins compared to defendant wins, 57 Clermont & Eisenberg, Advantage, supra note 6, at 146. 58 See, e.g., BJS, Trials, 2001, supra note 15, at 4 tbl.5 (reporting an overall plaintiff state trial win rate of 55.4%). 59 See Clermont & Eisenberg, Advantage, supra note 6, at 146. 60 Chi-Square test statistic= 3.718; p= .054. 61 Chi-Square test statistic = .575; p= .448. 62 Chi-Square test statistic = 25.095; p= .000.

16

likely inform litigants. To capture this possibility, we include dummy variables signaling whether a case was tried to a jury or judge, and whether a plaintiff or defendant prevailed.

We also expect that a state and particular case type’s “reversal culture” influences a decision to pursue an appeal. For each state63 and case category we computed an overall reversal rate. For example, the reversal rates in Georgia and New Jersey were 12.5% and 56.3%, respectively. Similarly, reversal rates for various case types ranged from 50% (employment contract cases) to 23.1% (lease cases). Holding all else constant, we expect that parties seeking to reverse a trial court decision would be more reluctant to pursue an appeal in states with lower reversal rates or involving case types with lower reversal rates. In addition to varied reversal rates, case types themselves also exercise important influence over other factors. For example, case types influence the routing of cases to either judges or juries and display sometime sharply different trial outcomes.64 As well, case types also influence various trial facets, such as case disposition time.65 Given case types’ ranging influences on the civil justice system, we also anticipate that case types influence the appeals process. Given our expectation, we include a dummy variable for each of our 14 case types. Because individuals, corporations, and governments vary in their appetite for litigation, it is prudent to assume that such variation persists into the appeals process. To account for this variation we use dummy variables to identify cases involving corporate-plaintiff, corporate-defendant, and governmental-plaintiff. Finally, because litigants’ decisions about whether to appeal may vary over time, we include the year the lawsuit was initially filed to help control for any linear time trend in our data. Finally, to account for those states that include more than one of the sampled large counties we include a dummy variable as a check on possible differences between those states and states with one sampled county. Efforts to model the outcome of an appeal call for a similar, though slightly different, set of independent variables. Analyzing the determinants of appellate court outcomes (reversals) is our principal interest and, among the array of plausible independent variables, our primary focus involves trial type (judge v. jury) and trial outcome (defendant or plaintiff victory). Our descriptive results, presented in Tables 1 and 2, suggest that appellate reversals were more likely for defendants that lost in jury trials. Other independent variables, such as case types, party types, and year appeal filed, are obvious candidates for inclusion given these variables’ import for the decision to file an appeal as well as trial court outcome. To account for possible state-level effects on the appellate outcome, we include dummy variables for each state, as well as a dummy variable signaling those states that have more than one sampled large county. Finally, because outcomes of individual appeals within a state might not be independent of one

63 The “other states” dummy variables includes six states (GA, HI, MA, NC, VA, and WI) where no more than one appeal was reversed during the time frame of our study. 64 See, e.g., Clermont & Eisenberg, Transcending, supra note 5, at 1137-38, 1167-70. 65 See, e.g., Heise, Justice Delayed, supra note 23, at 839-42.

17

another, we treat appeals as clustered at the state level, resulting in adjusted standard errors. C. Discussion Table 4 reports results of the key variables in the appeals outcome portion of our selection model (1). (The full models are reported in the Appendix tables, along with summary statistics for the variables.) Our selection model includes all cases and our findings show that, similar to our descriptive findings, plaintiffs who won jury trials were more likely to be reversed on appeal.66 Focusing only on jury trials, a test of the difference between the dummy variables “defendant won jury trial” and “plaintiff won jury trial” achieved statistical significance67 further suggesting important differences between how appeals courts treat defendant and plaintiff trial court victories. To assess our central findings’ robustness we repeated our analyses using two alternative statistical models (logit and probit). Our findings are robust across all models and largely confirm our descriptive accounts.

TABLE 4: MODELS OF STATE APPELLATE OUTCOME AND THE DECISION TO APPEAL

(1) Heckman

(2) Logit

(3) Probit

Appeal Outcome Equation: Trial outcome variable

Defendant won bench trial (ref) (ref) (ref) Plaintiff won bench trial .174 .076 .048 Defendant won jury trial .099 .065 .068 Plaintiff won jury trial .532** 1.009** .452** (Other variables and the decision to appeal equation are reported in the Appendix)

Rho .429 --- --- N 7,998 7,998 7,998 N (outcomes) 961 Log likelihood -3105.719 -743.226 -743.859 Pseudo R2

--- .117 .117

NOTES: Dependant variable in outcome equation is reversal of trial court decision; dependant variable in selection equation is whether an appeal was filed. ** p <0.01. We estimated the models using the “heckprob”, “logit”, and “probit” commands, respectively, in Stata v.9.2.

SOURCE: U.S. Dept. of Justice, Bureau of Justice Statistics. Supplemental Survey of Civil Appeals, 2001 (ICPSR 4539). 1. Implications for Competing Hypothesis Our results support two principle findings, one positive and the other negative. First, results from the appeals outcome model support the descriptive findings and 66 Coefficient = .532; p= .001. 67 p< .001.

18

evidence an appellate court tilt favoring defendants, especially defendants that lost in a jury trial. Notably, the influence of a plaintiff jury trial victory on appellate outcome persists across alternative models. Second, in addition to our main positive finding that supports the attitudinal hypothesis, a notable negative finding emerged as well. Specifically, the absence of a selection effect from our selection model further weakens (albeit indirectly) the plausibility of the selection hypothesis. Taken together, our results provide more support for the attitudinal explanation for our observed appellate outcomes. Moreover, both principle findings generally comport with existing empirical research, including prior work exploring federal civil appeals.

a. Attitudinal Hypothesis In all three models a plaintiff jury trial victory correlates with an appeals court reversal at a statistically significant level. This finding comports with what the descriptive analyses implied as well as with prior research on federal appeals, though with less breadth. Clermont and Eisenberg found an appeals outcome tilt favoring defendants who lost judge and jury trials.68 Results from our state appeals study suggest a similar appellate court pro-defendant tilt, but our finding is limited to defendants that lost in jury trials. Consequently, our finding of a pro-defendant bias in the appeals process—while consistent with federal appeals’ findings—is more narrow and targeted.

An additional wrinkle provides further texture to our argument about appeals courts’ mis-perceptions of the trial process. Appeal reversals distribute unevenly between both trial types. Specifically, as Table 1 illustrates, appellate courts were more likely to reverse jury than judge trial decisions. Combined with findings presented in Table 4, we found that while a defendant jury trial victory was quite stable, a plaintiff jury trial victory was far more likely to be reversed on appeal. Thus, appellate courts’ perceptions about erroneous jury decisions only extended to jury decisions favoring plaintiffs and not to defendants. Do our findings imply that jury’s exert a level of pro-plaintiff bias that accounts for the overall pro-plaintiff bias at the trial court level? This scenario is unlikely, especially insofar as trial court judges retain the legal authority to over-ride jury decisions in civil cases. As well, a considerable and growing empirical literature suggests that, in general, judges and juries act in comparable ways when confronted with comparable cases.69 If they do, we are left to explain why the appellate process treats judge and jury decisions, in general, and plaintiff and defendant jury trial outcomes, in particular, differently. Our findings suggest that appellate court mis-perceptions about jurors’ bias towards plaintiffs—rather than outright jury bias at trial—are the more likely explanation.

b. Selection Hypothesis

68 Clermont & Eisenberg, Advantage, supra note 6, at 152 tbl.4. 69 See, e.g., Theodore Eisenberg et al., Judge-Jury Agreement in Criminal Cases: A Partial Replication of Kalven and Zeisel’s American Jury, 2 J. Empirical Legal Studies 171 (2005) (noting level of judge and jury agreement in criminal case outcomes).

19

Those favoring a selection-effect hypothesis as the explanation for differing appellate outcomes suggest that there are systematic differences between the types of cases that plaintiffs and defendants appeal and, more importantly, that these case-level differences lead to different streams of appeals pushed by plaintiffs and defendants. Our findings provide direct evidence supporting the attitudinal hypothesis as well as indirect evidence dampening the selection hypothesis’ plausibility. The combination of our positive and negative findings points us away from the selection hypothesis as an explanation for appellate outcomes. Results from our selection (Heckman) model (1) dis-favor the selection hypothesis as rho (ρ), a measure of the correlation in the error term in the separate selection and outcome equations, is statistically insignificant.70 Consequently, one cannot reject the hypothesis that the error terms from the selection and outcome equations are uncorrelated. To explain this point differently, our results on appellate outcomes were not a consequence of selection. Had rho (ρ) achieved statistical significance, the selection (Heckman) model would be necessary for our data. Because rho (ρ) is not statistically significant, however, simple logit or probit models may suffice as well. In the event that our data do not call for a selection (Heckman) model and to generate external checks on our selection model results we re-ran our selection and outcome equations as separate logit (2) and probit (3) models. As Table 4 illustrates, our central finding persists across these alternative models. Obviously, neither the logit nor the probit models speak to any potential selection effect. Rather, findings from these two alternative models—specifically, the statistically significant influence of plaintiff wins in jury trials on the outcome of appeals—increase our confidence in the robustness of this critical finding.

The diminution of selection theory’s efficacy increases the persuasiveness—albeit indirectly—of the attitudinal theory. Although the finding here is a negative one—moored in statistical insignificance—its consistency with prior findings makes our negative claim slightly more important.71 Because our findings offer only an indirect challenge to the selection hypothesis, plausible counterarguments–especially those that our data and research design do not squarely address—warrant careful consideration. For example, selection theory proponents may suggest that plaintiffs start with systematically weaker cases and then

70 Coefficient = .429; Chi-Square test statistic = 1.75; p= .186. 71 A further technical word of caution is warranted with respect to interpreting a statistically insignificant finding. The power of a statistical test is the likelihood of detecting an effect of a specific size at a specified significance level. If the test used is not very powerful, the likelihood of detecting a statistical effect is diminished. Thus, perfectly designed and executed studies may fail to detect socially important differences “simply because the sample sizes are too small to give the procedure enough power to detect the effect.” STANTON A. GLANTZ, PRIMER OF BIOSTATISTICS 178 (4th ed. 1997). Therefore, it is important to consider a test’s power when one claims that no significant effect has been detected. Here, however, our sample size is large enough to reduce the potential that it accounts for the insignificant findings regarding our selection and outcome models. See also John Blume & Theodore Eisenberg, Judicial Politics, Death Penalty Appeals, and Case Selection: An Empirical Study, 72 So. Cal. L. Rev. 465, 491 n.83 (1999).

20

litigate them less effectively than defendants. We are unaware of any empirical support for such claims, however. Moreover, we do not find such claims persuasive from a theoretical perspective. After all, at a general level plaintiffs and defendants face similar (albeit not identical) economic incentives. These incentives should discipline plaintiffs and their attorneys and discourage weak cases. To the extent that the litigation process itself discourages case filings,72 the pool of claims that begin the litigation process might be systematically stronger than the universe of potential legal claims. As well, our study involves appealed cases. By definition these cases have persisted through the entire trial process, including the pre-trial motion stage which, by design, also helps screen frivolous cases. Despite our pool of appealed cases having survived multiple layers of filtering we still found evidence of a pro-defendant appellate court tilt. This finding conflicts with the “plaintiff weak case” counterargument.

i. Evidence of an Unexpected Selection Effect?

Although we remain confident that our findings more plausibly support an attitudinal rather than a selection explanation for the asymmetrical distribution of state appellate court reversals, we are not prepared to discard the selection theory altogether. Indeed, while selection theory may not explain appellate outcomes within the state context or, according to Clermont and Eisenberg, the federal context, selection theory may help explain the slight differences between the state and federal appellate settings.

Although differences between our study of state appeals and the Clermont and Eisenberg study of federal appeals are small, two of the differences warrant note as they implicate selection theory.73 First, the federal and state appeals settings generated slightly different appeal and reversal rates. Second, results from our more formal models were slightly narrower than the corresponding results from federal appeals study. Specifically, where Clermont and Eisenberg find that appellate reversals correlate with plaintiff judge and jury trial victories,74 we find a correlation only with plaintiff jury trial victories (Table 4).

As we explained previously,75 structural differences between federal and state cases likely contribute to the slightly different findings. We believe that selection effects generate distinctive streams of federal and state cases and these distinctive streams of cases react differently in the appeals context. Within the federal and state appeals settings, evidence of a selection effect does not emerge. Slight differences between the federal and state appeals studies, however, hint at a possible selection effect. Thus, it is not the case that the selection hypothesis finds no support in our study. Rather, we believe selection effects might present in unexpected ways.76 With the benefit of 72 See, e.g., David M. Trubeck et al., The Costs of Ordinary Litigation, 31 U.C.L.A. L. Rev. 72 (1983). 73 Despite small differences in findings from the two studies, however, it remains important to keep in mind that any differences are overwhelmed by the similarities. 74 Clermont & Eisenberg, Advantage, supra note 6, at 152, tbl.4. 75 See supra Subpart III.A. 76 Indeed, Clermont and Eisenberg found evidence in their study of federal appeals consistent with the selection theory. Clermont & Eisenberg, Advantage, supra note 6, at 153 (noting findings for a sub-pool of personal injury cases).

21

empirical studies of the federal and state appeals settings we are now in a better position to speculate whether and, if so, why this might be so. At this point, however, further empirical work is necessary to explore this possibility with empirical rigor. V. Conclusion Similar to findings from studies of federal appeals, findings from our analysis of state appeals evidence state appellate courts’ greater propensity to disrupt jury rather than judge verdicts and plaintiff trial victories more than defendant trial victories. With respect to the asymmetrical distribution of state appellate court reversals, our findings favor an attitudinal rather than selection explanation. That is, similar to the federal setting, in the state appeals setting the degree of reversals that tilt favorably to defendants that lost in a jury trail likely evidences appellate courts’ misperceptions regarding jury trial outcomes. We are not prepared to conclude, however, that the selection hypothesis has no place. Indeed, selection effects likely influence the appellate setting but in unanticipated ways that can be detected only by comparing appellate outcomes between state and federal systems. Although the attitudinal hypothesis strikes us as more persuasive in explaining the distribution of reversals within the state and federal systems, the selection hypothesis provides a plausible explanation for the admittedly small differences in results between the state and federal systems. Practical implications flow from our findings, not least of which is that the common wisdom regarding the sacrosanctity of trial jury verdicts, especially those favoring plaintiffs, may be misplaced. Findings from our study of state civil appeals should inform not only litigants but appellate judges as well.

22

TABLE A1: MODELS OF APPELLATE OUTCOME AND THE DECISION TO APPEAL

(1) Heckman

(s.e.)

(2) Logit

(s.e.)

(3) Probit

(s.e.)

Appeal Outcome Equation: Trial outcomes

Def. won bench trial (ref) (ref) (ref) Plaintiff won bench trial .174 (.184) .076 (.252) .048 (.105) Def. won jury trial .099 (.164) .065 (.308) .068 (.134) Plaintiff won jury trial .532** (.164) 1.009** (.313) .452** (.137) States

AZ (ref) (ref) (ref) CA .065 (.472) .767 (1.011) .304 (.371) CT -.032 (.507) .174 (1.171) .039 (.439) FL .237 (.476) .571 (1.098) .269 (.417) IL .480 (.521) .801 (1.026) .340 (.376) IN .323* (.149) .650* (.327) .257 (.133) KY .214 (.143) .627* (.318) .350** (.129) MI -.109 (.173) .093 (.576) .070 (.245) MN .712** (.161) 1.088** (.308) .450** (.129) MO -.005 (.153) .493 (.319) .210 (.130) NJ .623 (.497) .414 (1.001) .176 (.365) NY .069 (.157) .024 (.338) .057 (.132) OH -.053 (.503) .389 (1.047) .161 (.388) PA .370 (.464) .825 (1.056) .351 (.387) TX .251 (.493) .708 (1.017) .269 (.371) WA -.117 (.158) .770* (.334) .324* (.138) (other states) -.682** (.182) -1.271** (.425) -.520** (.167) Multi-county state -.261 (.445) -.466 (.956) -.171 (.343) Case types

Motor vehicle (ref) (ref) (ref) Dangerous premises .375 (.206) 1.302** (.383) .474** (.146) Product liability .790* (.318) 2.297** (.465) .934** (.208) Assault, slander, libel .829** (.250) 2.172** (.373) .862** (.156) Medical malpractice .412 (.246) 1.629** (.450) .617** (.174) Prof. malpractice 1.052* (.438) 2.982** (.628) 1.250** (.288) Other tort .529 (.301) 1.871** (.463) .713** (.184) Fraud .536 (.373) 1.832** (.557) .720** (.227) Seller plaintiff .414 (.303) 1.624** (.454) .623** (.180) Buyer plaintiff .486 (.293) 1.623** (.466) .651** (.187) Employment contract 1.180** (.282) 3.053** (.457) 1.286** (.192) Lease .344 (.433) 1.348 (.727) .534 (.290) Other contract .712* (.329) 1.977** (.517) .793** (.205) Property 1.028** (.276) 2.569** (.455) 1.049** (.188) Year case filed

-.059

(.042)

-.181**

(.062)

-.083**

(.029)

Litigant characteristics Government plaintiff

.285

(.324)

.712

(.618)

.318

(.272)

Corporate plaintiff .150 (.092) .407* (.178) .179* (.077) Corporate defendant -.009 (.113) .110 (.184) .056 (.079) Constant

115.537

(84.123)

355.314**

(123.602)

162.456**

(58.139)

N 7,998 7,998 Log likelihood -743.226 -743.859 Pseudo R2

.117 .117

23

TABLE A1: MODELS OF APPELLATE OUTCOME AND THE DECISION TO APPEAL (CONT.)

(1) (2) (3) Heckman (s.e.) Logit (s.e.) Probit (s.e.) Decision to Appeal Equation: Trial outcomes

Def. won bench trial (ref) (ref) (ref) Plaintiff won bench trial -.242** (.079) -.433** (.143) -.241** (.079) Def. won jury trial -.104 (.089) -219 (.159) -.103 (.089) Plaintiff won jury trial .115 (.090) .221 (.162) .115 (.090) State appeals rev. rate -.008* (.003) -.017* (.007) -.008* (.003) Case type appeals rev. rate .048** (.006) .091** (.010) .049** (.006) Multi-county state .075 (.095) .160 (.182) .075 (.095) Case types Motor vehicle (ref) (ref) (ref) Dangerous premises .322** (.085) .711** (.172) .322** (.085) Product liability .579** (.149) 1.145** (.254) .579** (.149) Assault, slander, libel -.171 (.137) -.250 (.240) -.171 (.136) Medical malpractice .667** (.077) 1.350** (.153) .666** (.077) Prof. malpractice .369* (.156) .686** (.262) .368* (.156) Other tort .846** (.080) 1.682** (.158) .846** (.080) Fraud .513** (.083) 1.037** (.151) .514** (.083) Seller plaintiff .414** (.085) .849** (.158) .414** (.085) Buyer plaintiff .512** (.093) 1.053** (.178) .512** (.093) Employment contract --- --- --- --- --- --- Lease .834** (.195) 1.652** (.371) .834** (.195) Other contract .032 (.117) .117 (.212) .033 (.117) Property .891** (.187) 1.735** (.338) .891** (.187) Year case filed -.050** (.017) -.093** (.031) -.050** (.017) Litigant characteristics Government plaintiff

.246

(.217)

.437

(.381)

.254

(.217)

Corporate plaintiff .126 (.085) .236 (.154) .126 (.085) Corporate defendant .136** (.050) .238* (.094) .136** (.050) Constant 97.161** (33.599) 180.174** (61.515) 97.445** (33.629) Rho .429 (.283) N 7,998 7,998 7,998 N (outcomes) 961 Log likelihood -3105.719 -2683.038 -

2684.861

Pseudo R2

--- .087 .086

NOTES: Dependant variable in outcome equation is reversal of trial court decision; dependant variable in selection equation is whether an appeal was filed. Other states consists of the six

states (GA, HI, MA, NC, VA, and WI) that had no more than one appeal reversal. Robust standard errors are in parentheses. * p <0.05; ** p <0.01. We estimated the models using the “heckprob”, “logit”, and “probit” commands, respectively, in Stata v.9.2. SOURCE: U.S. Dept. of Justice, Bureau of Justice Statistics. Supplemental Survey of Civil Appeals, 2001 (ICPSR 4539).

24

TABLE A2: DESCRIPTIVE STATISTICS OF VARIABLES IN SELECTION MODELS

Variable n Mean S.D. Min. Max. Reversed trial court 965 0.182 0.386 0 1 Defendant won bench trial 8,038 0.083 0.275 0 1 Plaintiff won bench trial 8,038 0.150 0.357 0 1 Defendant won jury trial 8,038 0.375 0.484 0 1 Plaintiff won jury trial 8,038 0.392 0.488 0 1 AZ 8,038 0.050 0.218 0 1 CA 8,038 0.157 0.364 0 1 CT 8,038 0.020 0.141 0 1 FL 8,038 0.072 0.258 0 1 GA 8,038 0.016 0.124 0 1 HI 8,038 0.003 0.055 0 1 IL 8,038 0.053 0.225 0 1 IN 8,038 0.015 0.123 0 1 KY 8,038 0.019 0.137 0 1 MA 8,038 0.030 0.170 0 1 MI 8,038 0.048 0.214 0 1 MN 8,038 0.025 0.156 0 1 MO 8,038 0.018 0.134 0 1 NC 8,038 0.013 0.113 0 1 NJ 8,038 0.077 0.266 0 1 NY 8,038 0.038 0.192 0 1 OH 8,038 0.050 0.219 0 1 PA 8,038 0.097 0.295 0 1 TX 8,038 0.126 0.332 0 1 VA 8,038 0.031 0.172 0 1 WA 8,038 0.026 0.159 0 1 WI 8,038 0.016 0.124 0 1 Motor vehicle 8,038 0.351 0.477 0 1 Assault, slander, libel 8,038 0.039 0.193 0 1 Product liability 8,038 0.015 0.120 0 1 Dangerous premises 8,038 0.115 0.319 0 1 Medical malpractice 8,038 0.106 0.308 0 1 Prof. malpractice 8,038 0.008 0.092 0 1 Other tort 8,038 0.045 0.207 0 1 Employment contract 8,038 0.035 0.184 0 1 Fraud 8,038 0.053 0.225 0 1 Lease 8,038 0.023 0.151 0 1 Seller plaintiff 8,038 0.096 0.295 0 1 Buyer plaintiff 8,038 0.066 0.248 0 1 Other contract 8,038 0.026 0.160 0 1 Property 8,038 0.022 0.146 0 1 Year case filed 8,038 1998.91 1.341 1985 2001 Government plaintiff 8,028 0.008 0.091 0 1 Corporate plaintiff 8,028 0.156 0.363 0 1 Corporate defendant 8,001 0.425 0.494 0 1 State appeals rev. rate 8,038 37.084 15.248 0 100 Case type appeals rev. rate 8,038 29.589 6.069 23.077 50 Multi-county state 8,038 0.682 0.466 0 1

NOTE: SOURCE: U.S. Dept. of Justice, Bureau of Justice Statistics. Supplemental Survey of Civil Appeals, 2001 (ICPSR 4539).