Litigating Toward Settlement Purchased

A generation of interdisciplinary legal scholars has worked to understand settlement’s propriety, timing, incidence, and causes. This attention is not surprising given that most filed civil cases settle (Galanter 2004; Eisenberg and Lanvers 2009). Indeed, settlement within formal litigation is but a part of an attenuated dispute resolution process that may entail several rounds of pre-complaint negotiation and continue well after the “case” has formally terminated (Clermont and Eisenberg 2002). Galanter minced no words when describing the landscape of disputes: “[T]he negotiated settlement of civil cases is not a marginal phenomenon; it is not an innovation; it is not some unusual alternative to litigation. It is only a slight exaggeration to say that it is litigation” (1985: 1).

Lawsuits begin with many known unknowns. A single filed complaint may advance multiple, often competing, theories, and causes of action. The plaintiff needs to have only a reasonable belief that her version of the facts is accurate, and along the way she will likely rely on discovery to know whether that faith is defensible. Though individual defendants might be expected to know what they have done, entities will have to devote time to investigation before knowing what—if anything—actually happened. The court’s perspective on the relevant doctrine may at times be predictable, but its view of the facts will likely be shaped by factors both remote from the merits and exceedingly difficult to observe. Even if the parties were not already adversely positioned, such informational gaps and barriers present formidable hurdles to the agreement of the parties and the settlement of their cases (Priest and Klein 1984).

What then can explain how cases progress toward settlement? The empirical evidence has frustrated those hoping to find a conclusive answer to this question. The settlement “rate”—the overall chance that a case in a given data set will settle—varies by jurisdiction, case type, time, lawyering, judicial demographics, and party characteristics (Eisenberg and Lanvers 2009). But this heterogeneity may obscure the role that information exchange through motion practice plays in motivating settlement in individual disputes. Indeed, one of just a handful of factors that distinguish formal litigation from dispute resolution outside of courts is that it provides the parties with a powerful tool—litigant-controlled motions—that enables them to force the exchange of information. In short, motion practice makes litigation dynamic and reflexive, permitting parties¹ to learn about their cases (Kritzer 1986).

In the case of discovery, that learning process is direct. It is therefore not surprising that much of the scholarship examining litigant learning and settlement, theoretical and empirical alike, has focused on the possible effect of discovery on case settlement (e.g., Shavell 1989; Farber and White 1991; Huang 2007). While discovery practice is the most salient way that this intra-litigation information exchange and learning takes place, the filing and resolution of other motions produces two additional sources of information: how the filing party understands the facts and law—their strategy for the case—and how committed they are to the case—the kind of resources they are willing to expend. When the court is induced to rule, even when that ruling does not end the case, the parties gain a fourth source of information—what the court thinks about the legal merits and facts. Because most accounts of settlement begin by assuming that it is uncertainty and/or asymmetric information about cases that prevents compromise (Bebchuk 1984; Priest and Klein 1984), the process of litigation can be a way in which education toward compromise happens.

In this study, we seek to further examine this progression of cases toward settlement by testing the effect of the filing and disposition of motions on the timing of settlement. We begin in Section 1 by developing a theory and five hypotheses for how information exchange through motions practice is likely to motivate case settlement in certain trial court cases. In particular, we expect that motion filing, content, and outcome, and movant identity may affect settlement timing in a case. In Section 2, we detail our data and methodology for carrying out our empirical tests. As we describe there, this study utilizes an original data set of 585 cases filed in federal district courts in which plaintiffs sought to pierce the corporate veil, thereby holding a defendant business entity’s individual owners liable beyond the company itself. To empirically analyze time to settlement in these cases (while accounting for both other, nonsettlement termination methods and ongoing cases), we estimate a series of competing risks duration models (one for each motion-specific hypothesis) in Section 3. After doing so, we find, among other things, that the mere filing of a motion in a case hastens settlement. In short, the time to settlement can be influenced by the dynamic characteristics of the litigation preceding it. In Sections 4 and 5, we discuss the significance of these results, including how they provide a controlled and uniquely detailed look at the mechanism of compromise via information exchange and litigation in the federal courts while also yielding evidence that this effect goes well beyond the traditionally studied discovery process.

1. Theory and Hypotheses

Early modeling of settlement, consisting largely of economics-based theoretical accounts, focused on how the interaction of the parties’ subjective probabilities of victory, divergent expectations, error-prone calculations of their likelihood of success, and asymmetrical information, along with the cases’ stakes and costs of litigation affect settlement rates (Posner 1973; Priest and Klein 1984; Bebchuk 1984). This research generally predicted that the farther apart the opposing litigants were in terms of these metrics, the less likely they should be to settle.

Over time, research has ventured to test these theoretical models quantitatively. This work has found that factors such as, for example, the parties’ experience and status (Waldfogel 1998), case importance or severity (Landes 1971; Elder 1989), and repeat attorney interaction (Johnston and Waldfogel 2002) can all play a systematic role in predicting the likelihood of case settlement. One notable line of this empirical research focuses on the specific effect of discovery in reducing informational uncertainties and asymmetries and thereby promoting case settlement. As Farber and White put it, “[t]he two-way exchange of information during the discovery process encourages resolution of cases without trial by making the information available to both sides more alike and by increasing the likelihood that both parties have the same expectation about the trial outcome” (1991: 201). Quantitative work, including studies of medical malpractice cases in the US litigation (Farber and White 1991) and general Taiwanese civil lawsuits (Huang 2007), provides strong support for this discovery-settlement effect.

Scholars have also considered, though to a much lesser degree, the duration of litigation and timing of case settlement. They find, for example, that institutionalized legal rules, such as those involving fee-shifting and fee structures (Fournier and Zuehlke 1996; Helland and Tabarrok 2003), comparative negligence rules (Kessler 1996), and liability standards (Chang and Sigman 2000) can influence the time to case termination and settlement.

Due largely to data limitations, only a small handful of settlement-timing empirical pieces have begun to make a connection between information exchange and the timing of case settlement. Spurr (1997) finds, for example, that cases settle more quickly when they have been referred to a settlement specialist during the litigation. Johnston and Waldfogel (2002) discover an effect on settlement timing due to the repeat interaction of attorneys and their corresponding ability to reach swifter litigation compromises. Similarly, Fenn and Rickman (1999) find strong effects on settlement timing based on parties’ costs and liability calculations, as well as on the presence of legal aid. Kotkin (2007) finds that settlement amount increases after the filing of a motion. These articles certainly advance our understanding of this process, but they are limited in scope by their measurement of litigation-specific variables at a single point in time (whether that point is case filing or settlement).

Following in the footsteps of this work on settlement timing, we seek now to gain a better theoretical understanding of the micro-level dynamics of litigation and how they affect when in the life of a dispute settlement is likely to happen. Settlement, as we have discussed, can occur at any moment after injury and takes different forms. It need not entail a transfer of wealth from the offending party to the disputant, and, famously, is almost always privately concluded without judicial intervention or sanction.² Indeed, most disputes are resolved via formal or informal compromise prior to the filing of a complaint, as the injured party either bears its losses or the injurer makes a pre-filing payment (Aubert 1967; Miller and Sarat 1980–81; Trubek et al. 1983).

When the parties fail to settle during the early stages of dispute resolution, injured parties can initiate formal proceedings.³ There are, we suggest, two general kinds of cases that meet this threshold. The first are ones in which the parties, through shared experiences or pre-filing communication, share an understanding of the facts and the putative plaintiff’s damages. However, they have disagreed as to whether the plaintiff has the resources, or motivation, to escalate the dispute and pursue litigation. To borrow a poker parlance, the defendant believes that the plaintiff is bluffing in his threat to involve the courts. The complaint is a response, turning up the Aces in the plaintiff’s hand (Clermont et al. 2010). Such cases are likely to end quite early in the life of a lawsuit, ordinarily without resorting to the formal litigation process (Rosenberg and Shavell 1985; Nielsen et al. 2010).

In other cases, there has been relatively little interaction between the parties, or at least they have not had the number and quality of discussions necessary to close the informational gaps between them. Because of this, the parties remain divided at case filing by their views of the facts or the legal merits. As has been classically described, such cases are marked by asymmetrically held information (Farber and White 1991). For compromise or settlement to become a reality, information exchange is crucial. While remaining mindful of the existence of the former group of filed cases (the “early settler, no litigation activity” lawsuits), we focus here primarily on this latter group of cases, an emphasis that allows us to more closely examine the relationship between information exchange and the timing of settlement once formal litigation commences.

To do this, we turn first to behavioral psychology, which provides an explanation for why cases in this latter group do not and should not settle at an even rate over time. Korobkin and Guthrie (1994), Rachlinski (1996), Guthrie (2000), and Korobkin (2002) have argued that individuals’ settlement decisions are influenced by their reference points. The simple version of the theory is that plaintiffs view the choice between settlement and nonsettlement as between two gains, one certain and one contingent. Defendants, who can only lose money in litigation, see two losses, with settlement being certain and trial merely possible. Given that choice, says prospect theory, plaintiffs should be risk averse and prefer settlement (the sure over the uncertain gain) while defendants should be risk seeking and prefer trial, even if the parties agree on what the relevant probabilities are (Rachlinski 1996).

As later work has explored, this reference setting can be complicated: if the plaintiff sees the litigation through the lens of how it looked before any injury, “gains” in settlement may actually feel like losses, and thus these uncompensated plaintiffs will prefer rolling the dice (Korobkin and Guthrie 1994). What is clear is that these reference points may change during the life of a lawsuit. Plaintiffs may adapt to the losses they have suffered (Bronsteen et al. 2008). Or the defendant, who we have assumed views the lawsuit as all loss, may see a preliminary ruling from a judge validating its position in part as a psychological victory: settlement preserves the gain (or at least avoids the loss). Thus, as the litigation develops, the parties’ basic orientation—“I’m currently ahead, the future is a choice between types of losses” or “I’m currently behind, the future can only be better”—can shift, and the shifts may be dramatic. These reference point changes will influence the settlement rate.⁴

Also relevant here is a distinct theory that relies on rational litigants but also incorporates insights from the valuation of options. Real options theory posits that the parties update their expectations about the variance of outcomes that might result from litigation, based on information coming from the court and from the other parties (Grundfest and Huang 2006). As variance increases, the price of the “option” to continue litigating increases; conversely, as the range of possible outcomes shrinks, so does the value of the option and thus the settlement price.

Notably, both psychological and real options theory depend on the parties updating their views of the case as it develops. This learning process relies crucially on the actions of the parties in filing motions. Consider the parties’ settlement posture before the filing of a motion to dismiss. They have asymmetric information about the facts, but are symmetrically unclear on how the court will respond to the uncertain legal theories, particularly if they have not litigated the issue before their assigned judge or magistrate in previous litigation. If the defendant does not make a motion, the playing field for settlement will be unchanged (or very close to unchanged) after the filing of the complaint.⁵ Factors present at filing will dominate settlement’s incidence and timing.

If the parties do engage in motion practice, the informational content of the case changes—even before the other party has a chance to respond and the court to rule. As we have suggested, motion practice provides at least four distinct kinds of information to the parties.⁶

The first—and most intuitive and commonly studied—is factual. A motion to compel, if granted, forces the disgorgement of documents or testimony from one party, updating the other party and reducing information asymmetries. Similarly, the documents appended to motions, even if previously disclosed, can highlight for a party the facts that search costs might have otherwise obscured (Huang 2007). Of course, almost all factual discovery occurs without motion practice, as the parties exchange documents, take depositions, and answer interrogatories without any court intervention. Indeed, motions to compel probably indicate that the ordinary mechanisms of information exchange have broken down. Factual discovery, with or without court intervention, is the form of information exchange that most empirical scholars to date have focused on, and it is a very important driver of settlement.

But discovery is not the only way that formal litigation can inspire compromise. A second kind of information is strategic. The filing of a motion to dismiss, for instance, tells the plaintiff about the defendant’s perspective on the case, including, for example, the cases that he will rely on throughout the litigation and his theory of nonliability. The responsive brief similarly informs the defendant about the plaintiff’s view of the law. Because various theories have distinct probabilities of relief—and require the parties to expend different amounts of resources—this opportunity for strategic learning is precious.

The third source of information is the kind of resources the parties wish to spend on the case. Learning about resource allocation is indirect and turns in part on the degree of polish in the case documents. Did the party spend money to be sure that the brief was carefully cite- and fact checked, the tabs pointing to the appropriate exhibits, the writing especially clear, and the arguments freshly generated? Lawyers probably know, looking at a brief, whether the other side is going all out to win the particular motion, or has made a filing in the usual course. This resourcing decision, in turn, can update the parties’ views of how much the case is worth, the likelihood of protracted litigation, and the other side’s reservation value in bargaining. Indeed, the mere filing of a motion may force a party to expend resources in response that it would rather have conserved. Motion practice thus may signal a party’s commitment to the litigation.

Fourth, motion practice may produce information about the judge. When the court rules, assuming that it passes on dismissing the action entirely, the parties learn a tremendous amount about how the court sees the case. Does the court see a nuisance suit, scorned, and humbled, but sent through to discovery? Will the judge be closely involved with the merits, pruning claims away? Which issues are set up for resolution at summary judgment, and what standards does the court provide about the scope of discovery? (Kritzer 1986)

These four kinds of informational exchange imply that different kinds of motions, made by various parties, resolved or left pending by the judge, at distinct times, will have different kinds of influence on the amount of total information available. This leads us to our hypotheses.

Hypothesis 1: The filing of a motion will increase settlement speed.

As we have explained, the mere filing of a motion should bring the parties to the bargaining table, by motivating learning and the updating of the parties’ subjective beliefs.⁷ Filing a motion implicates each of the first three sources of information exchange—strategy, resources, and facts. Thus, regardless of the disposition of the motion (granted or denied, in whole or part, or pending), we expect to see an uptick in the rate of settlement following its filing.

Hypothesis 2: Granted motions will increase the settlement speed rate more than denied motions.

Substantive motions that do not end the case—e.g., a plaintiff’s motion to certify a class or a defendant’s motion for partial dismissal or partial summary judgment—can be granted or denied. When granted, we would expect that the court will provide more information to the parties than if the court denied the motion, if only because the decision to say “no” may mean “not yet.” There is also evidence that denied motions to dismiss are almost never explained in written opinions (Hoffman et al. 2007). Thus, the amount of information conveyed by a judge in a granted motion will often exceed the amount of information conveyed by a judge in a denied motion.

Hypothesis 3: Plaintiff interstitial victories will prompt faster settlement than defendant victories.

At judgment, defendants generally win more cases than plaintiffs do—a function of multiple factors, including selection, disparities in resources, the burden of proof, and anti-plaintiff bias (Siegelman and Donohue 1995). For example, Parker (2006) found that racially based employment discrimination cases settled after plaintiff motion-level victories more often than after defendant motion-level victories, while Kaplan et al. (2008) found that plaintiffs recovered more in settlement than they did at trial and that workers whose claims were exaggerated tended to settle less often. This suggests that when plaintiffs do win at the interstitial stage (i.e., a victory that does not end the case) such wins will be more informative—in the sense of revealing the real strength of the plaintiff’s case—than when the defendants win such intermediate motions. In our data set, consisting of veil-piercing cases, we have an additional reason to expect this result: plaintiff interstitial victories will make it more likely that individual defendants’ personal assets will be exposed to a liability judgment. Given that defendants’ prior expectations in such cases is that their personal assets will be protected by the entity’s veil, any plaintiff success will be likely to provide an important and highly salient new piece of information about the case’s settlement value.

Hypothesis 4: Motions that turn on the application of law to fact will be more likely to influence settlement than those that are resolved on the pleadings.

The informational content of dispositions is likely to increase over time, as the factual picture develops and the judge pays more attention to the case.⁸ In part, this is a function of incentives: judges have a limited amount of time to spend on a given case and rationally will choose to do so in the factual position best developed for review and which is most likely to face an appellate challenge. Therefore, as compared with motions resolved on the pleadings (such as motions to dismiss, arbitrate, remand, or strike), motions that turn on application of law to fact (e.g., summary judgment, class certification, evidentiary rulings, and patent construction) will likely contain more information because the court has devoted more resources to their production. Being more informative, such resolutions will decisively influence the settlement process (Rave 2006).

Hypothesis 5: Motions seeking clarification of particularly uncertain law will be more likely to influence settlement timing than motions seeking clarification of certain law.

Since our theory depends on the positive influence of learning, we expect that motions targeting legal issues that are generally thought to be difficult or obscure will have a greater influence on settlement than motions targeting doctrines the parties know better. Our data set examines veil-piercing cases, an area of law famously decried as “rare, severe, and unprincipled” (Easterbrook and Fischel 1985) and an “unprincipled hodgepodge of seemingly ad hoc and unpredictable results” (Millon 2007). In our data, we can distinguish between motions that ask judges to resolve an issue related to piercing the veil—such as veil piercing-specific discovery, or a motion to dismiss targeted at the veil-piercing allegation—from motions that are not so specifically focused. We expect that veil-specific motions, because they potentially resolve a particularly uncertain doctrine, will have a more significant influence on the settlement rate than ordinary litigation practice.

We contend that these five motion-related hypotheses are causal in nature, since they meet the standard articulated by Holland: “[t]he key notion … is the potential (regardless of whether it can be achieved in practice or not) for exposing each unit to the action of a cause” (1986: 946), meaning that in an experimental setting, the presence and timing of different motions in our units of interest (cases) could be manipulated. Although concerns regarding correlation and reverse causation are inescapable in observational data studies like ours (Winship and Morgan 1999), we are inclined to discount that objection for our central hypotheses.⁹ With Hypotheses 1, for example, we expect that the filing of a substantive motion in a case, and the resulting information that it puts into circulation among the case actors, can cause the moment of settlement in the case to move up. The reverse of this, that imminent settlement causes motion practice and information exchange, is untenable since, as our cases without any motions or litigation activity reveal (discussed in further detail below), engaging in post-filing motion practice is not the most efficient way to settle cases when the parties have already approached a fully informed compromise position.¹⁰

We are less certain about the causal story for one particular kind of motion—those concerning discovery. Discovery motions can either seek to prevent disclosure (a motion for a protective order) or, more commonly, prompt it (a motion to compel). Thus, some granted discovery motions will enable crucial information exchange—for example, an order that denies a privilege claim and requires the disclosure of a particularly damaging document. But, more typically, discovery motions signal that the parties are failing to exchange information as they would in the ordinary case. This raises a problem not of causation but of selection.¹¹ Thus, it is very hard to know what inferences to draw from the filing or the grant of a discovery motion.¹² It is for this reason that we separately control for these motions.

2. Data and Methods

To test our hypotheses, we examined federal trial court litigation, from case filing through settlement. We collected cases in which plaintiffs sought to pierce the corporate veil: the database contains 697 separate district court cases filed from 2000 to 2005, a 78% random sample of all eligible and available federal veil-piercing cases within this period. To meet our criteria of “eligible and available” and be included in our sampling population, a case’s complaint had to be present in Westlaw’s Trial Pleadings Database, ¹³ filed in a federal district court,¹⁴ and include at least one veil-piercing claim.¹⁵ We exclude from this sample 112 cases that terminate via default judgment, bankruptcy, remand, and transfer or that contain lengthy, nonterminal stays.¹⁶ Of the remaining 585 cases, we focus primarily on the 452 that terminate via settlement.¹⁷

For coding events, parties, and activity in each case, we utilized the pleadings, the docket, and any other relevant case documents electronically attached to the docket.¹⁸ After gathering the primary pleading from Westlaw, we turned to PACER (“Public Access to Court Electronic Resources”) to retrieve the case docket and other case documents.

This data set of veil-piercing lawsuits consists largely of commercial disputes between at least three represented parties. In most such suits, the plaintiff advances a contract or commercial tort claim against a defendant corporation and its owners, seeking to impose liability both against the company (which may or may not be judgment proof) and the individuals. The veil-piercing claim thus operates as a derivative and remedial measure, much like a punitive damages allegation, rather than as a substantively significant cause of action.

While our data were identified based on the inclusion of a veil-piercing claim, two different metrics indicate that our underlying cases represent a broad array of lawsuits. First, and not surprisingly, the complaints serving as the origin for our data contain a wide variety of causes of action, including tort, contract, RICO, ERISA, and common law fraud claims. In addition, the cases in our data span a number of Nature of Suit (NOS) codes. At the time of filing, the filing party identifies a single NOS code for the case, which serves as a snapshot of what the filer believes best summarizes the case. Within empirical legal research, scholars frequently use these codes to identify case issue areas (e.g., Siegelman and Waldfogel 1999; Johnston and Waldfogel 2002; Galanter 2004). Figure 1 depicts the NOS code distribution for all cases filed in federal district courts (black bars; measured in 2007) and the distribution of codes within our data set (gray bars). Although our data are certainly not representative of all filed cases (as measured by NOS code distribution), we can see that nearly all major NOS categories are present, with the one real notable exception being prisoner petition lawsuits. Not surprisingly, the overall NOS distribution of our data is highly suggestive of a set of commercial litigation cases. Despite this variability in NOS codes and causes of action, however, we remain cognizant of the fact that our data are limited in their scope and generalizability, a topic that we return to further in the Section 4.1 below.

To operationalize our hypotheses, we required information on the timing of cases as well as on the motions, the parties, the judges, and other case characteristics. Our dependent variable of interest throughout this study, time to settlement, is measured as the number of months that a case is pending (“survives”) from its date of filing through the date of settlement. Summary statistics on case survival time as well as on our other variables are reported in Table 1. As the table indicates, cases in our data survive between 1 and 58 months, with a mean survival time of nearly 17 months.

To empirically analyze time to settlement, we use a series of duration models. Our model of choice, the competing risks regression model (Fine and Gray 1999), allows us to estimate the effect that individual case variables have on the time to case settlement. The model accounts for alternative (competing) termination modalities besides settlement, like trials and dispositive motions. Because the effect of case characteristics on settlement may depend on the existence of these other possible methods of case termination, failure to account for these nonsettlement terminations may bias our results (Box-Steffensmeier and Jones 2004). Competing risk models resolve this concern and, for this reason, have been used in other litigation-related duration work as well (Eisenberg and Farber 1997).¹⁹

A competing risks model also allows us to account for censored data. Due to inherent limitations from studying relatively recent filings, some cases in our data set had not terminated by the time we finished collecting data. The competing risks model is well designed to treat these type of censored data and avoid the bias inherent in dropping these cases altogether from the study. Within our data, the 29 cases not terminated by August 18, 2008 are included but, because they have not yet “failed” (but still hold the potential to do so going forward), are censored.

Our key independent variables revolve around the motion activity in our cases. Our measurement of each motion is discrete and time varying in nature. We capture these motions through two mutually exclusive and collectively exhaustive dichotomous variables: (1) one for the first month after the motion is filed (or resolved, depending on the motion) and (2) one for the set of all months thereafter (i.e., the second through the n-th month) until the case ultimately terminates. This measurement strategy allows us to more precisely capture the potential propelling effect (often referred to as the hazard ratio) that motions may (or may not) have on cases as they move toward settlement.

Because of this research design and the time-varying nature of these variables, our data set is constructed in an expanded format, where each month of each case has its own observation. As a result, while we have 585 cases, our data set contains 9931 case-month observations.²⁰

Motion made measures the presence of substantive (nondiscovery) motions in a case, regardless of how and whether those motions are resolved. Motion granted and motion denied measure the type of outcome for resolved, substantive motions in a case. Plaintiff motion granted and defendant motion granted quantify the presence of granted substantive motions where either a plaintiff or a defendant was the moving party. Application of law to fact motion denied and pleading-based motion denied measure the existence of denied motions in these two categories, where motions of the former type include summary judgment, class certification, discovery, evidentiary rulings, and patent construction and motions in the latter group include motions to dismiss, arbitrate, remand, or strike.²¹ Finally, veil-piercing motion made and nonveil-piercing motion made capture the presence of substantive motions that either do or do not center primarily on the veil-piercing claim(s) in the case. As we note above, we also separately control for discovery motion made. Each of these variables is coded as 1 when the motion and/or outcome of interest is present and 0 otherwise. Any time more than one such type of motion and/or outcome happens in a case, we record the first occurring instance.

In addition to our motion-level variables, we also control for a number of other variables concerning case actors, type, and strength that may affect the timing of case settlement or other termination.

Galanter (1974) sparked a large body of theoretical and empirical research examining the effects that litigant status and differences in resources due to that status have on the litigation process. The common argument (and empirical support) goes that imbalances in the resources and power of opposing litigants will leave the weak in a position of disadvantage and will lead to disproportionate outcomes for the stronger party (e.g., Kritzer and Silbey 2003; Collins 2004; Schwab and Heise 2011; Black and Boyd 2012). To measure this, we look for cases where the plaintiff is an individual and the defendant is an entity, contrasting such uneven matchups to cases where both plaintiff and defendant are entities.²² Therefore, entity v. entity measures cases where both the plaintiff(s) and the defendant(s) in the lawsuit are entities. Any time an individual party is involved on either side of this dispute (as evidenced through the case docket and pleadings), this variable is coded as 0.

To control for the possibility that Eisenberg and Lanvers (2009) observation that “strong cases tend to settle; weak ones do not” is also connected to the momentum of settlement, we include a variable for case strength in our models. Case strength is coded from a case’s complaint and takes on a value of 1 when plaintiffs argue that undercapitalization is a primary ground for piercing the corporate veil in the case. Previous work found this factor to be a strong indicator of overall case strength (Boyd and Hoffman 2010).

We also account for the effect that repeat play among attorneys (or the lack thereof) might have on the pace of a case. Attorneys familiar with one another will be more likely to seek to maximize shared gains and less likely to posture and exaggerate their positions, meaning that such repeat players will make settlement more likely (Johnston and Waldfogel 2002). The same effect may be present for the timing of settlement. We operationalize the repeat play effect by asking if the case includes local counsel. Local counsel appear in cases where the controlling lawyers are not admitted to the bar in the filing jurisdiction, and their role, including performing tasks like certifying papers, is rather ministerial in nature. Local counsel in the case is thus a proxy for the main attorneys being strangers to one another and, where present, may well delay case settlement. Local counsel participation is measured based on the presence of a local counsel listed among the case’s attorneys or through the presence of a pro hac vice motion, as recorded on the case docket and/or pleadings.

Many scholars have argued that females communicate and behave differently from males (Gilligan 1982), and that this translates to more empathetic and problem solving-based judicial behavior (Brudney et al. 1999; Martin et al. 2002). If female judges do indeed behave consistently with the gendered theories, when it comes to litigation and case management they may actively encourage cases to settle, both more frequently and more quickly than their male peers. To control for this possibility, we include female judge in our models, a variable that is coded as 1 when the district judge of record is a female.

On the theory that chief district court judges are aware of administrative constraints and are more likely than their district colleagues to worry about case load reduction pressures (Hettinger et al. 2006), cases assigned to district court judges with experience as a chief judge (chief judge) may settle more quickly than other cases. Any time a case has more than one assigned district judge over its life, we code this based on the judge at the time of case filing. We retrieved the information on these judges from the Federal Judicial Center’s Biographical Directory of Federal Judges.

We also control for a variety of case characteristics. These include whether the case was removed from state court (removal), what type of nature of suit category the case was classified into (see Figure 1 for categories), the type of jurisdiction asserted in the complaint (diversity jurisdiction or federal question jurisdiction), whether there was a jury demand by any party, and whether the case had any signs of litigation activity (a dichotomous, nontime varying variable indicating the presence of any substantive or discovery motion in a case).²³

3. Results

Figure 2 provides descriptive details on the distribution of settlement times in our data. The top panel’s dotted line depicts settlement times in our data from case filing to case settlement for cases without any recorded litigation activity—such as substantive and discovery motions. For these cases, the data are positively skewed, with settlement commonly occurring in the very early stages of the case. This is analogous to the recent findings in Nielsen et al. (2010) regarding early settlement in some cases. It is also consistent with our expectations regarding early settling, motion-free cases where the filing of a lawsuit serves as the final bargaining tactic for a plaintiff in already well-developed pre-filing negotiations between the two parties.

The solid black line in the top panel of Figure 2 displays just the opposite, showing the distribution of settlement timing for cases with litigation activity. In these cases, settlement timing varies greatly, with a relatively even distribution over time. It is in these cases that we expect motion practice to help foster information exchange and thereby motivate faster settlement. The bottom panel of Figure 2 depicts, again descriptively, the time to settlement from the filing of the first substantive motion to the settlement. This figure’s pattern closely resembles nonlitigation settlement timing. In other words, the bottom panel of Figure 2 provides preliminary, descriptive evidence that when a case has litigation activity, the filing of a substantive motion propels it toward settlement in a way that would not otherwise happen prior to motion filing.

For ease of substantive interpretation, we report the results in these models in Tables 2 and 3 as subhazard ratios, a statistic that is simply the exponential of the estimated coefficient. These subhazard ratios allow us to quantify the underlying increase (or decrease) in the likelihood of a case terminating by settlement in a fixed time after filing due to the added presence of these different types of motions. In terms of interpretation, a subhazard ratio over 1 indicates that a covariate has a positive effect on the risk of case settlement while a ratio below 1 indicates a negative effect.

The statistical significance and signs for the key results reported in the tables for these five models provide us initial evidence for each of our five hypotheses. As we can see, many of our motion variables have substantial statistical and substantive effects when it comes to the timing of case settlement. For example, as indicated in Table 2, Model (1), being in the ≥2nd month following the filing of a substantive motion increases the odds of settlement by 375% compared with the chance in that same case prior to the motion being filed. Interestingly, while the odds of a similar increase in settlement speed are also positive in the first month immediately following the motion filing, they do not reach statistical significance at the p < 0.10 level.

To better describe the substantive effect of motions being filed (≥2nd month), we plot the Cumulative Incidence Function (CIF) curve of our Model (1) competing risks regression in Figure 3. The CIF provides the probability of the failure of interest while accounting for the presence of other competing failures (Coviello and Boggess 2004; Kim 2007). Because of the time-varying nature of our key independent variable in Model (1) (motion filing), we are able to estimate the CIF before and after a motion is filed (while holding constant all other model variables). We plot the resulting CIF curves from these two estimates in Figure 3. As we can see there, prior to a motion being filed (black line), the probability of settlement (our “failure” of interest) remains below 0.10 even as a case enters its second year. However, once a motion is filed (gray line), we see an immediate uptick in the slope of the curve and the likelihood of case settlement. This, consistent with Hypothesis 1, is evidence of the propelling effect that we expect the filing of motions to have on case settlement timing.

Now, turning to tests of Hypotheses 2 through 5, we examine this motion effect in a more fine-grained way. Recall that in Hypothesis 2 we expect that motions granted will have a stronger effect on settlement timing than those that are denied. As Model (2) in Table 2 reveals, granted motions have an immediate effect (in the first 30 days after motion disposition), leading to an increase in the speed of settlement of 270%. A similar immediate effect is not present for denied motions. However, both granted and denied motions do have substantively important effects on settlement timing after that first month. Denied motions have a slightly larger effect compared with granted motions in this context (a 68% increase in settlement odds compared with 54%), but a Wald test reveals that this difference is not significant.²⁵

For Hypothesis 3, we can see in Model (3), Table 2 that, as expected, the plaintiff’s granted motions have a stronger effect on settlement timing than those from the defendant. Indeed, in the first month after a plaintiff’s substantive motion is granted, settlement speed is increased by over 450% (subhazard ratio of 5.538); after the first month of that successful plaintiff motion, the risk of case settlement increases by a more modest (but still significant) 54%. While positive, neither of the variables for defendant granted motions reach statistically significant levels.

In the context of Hypothesis 4’s results (found in Table 3’s Model [4]), neither motions applying law to fact nor pleading-based motions have a statistically significant effect on settlement timing in the first month after they are denied. After that first month, each type of motion has a positive effect on settlement timing (subhazard ratios of 1.5 and 1.25, respectively), effects that are not, statistically speaking, different from one another.²⁶

Finally, for Hypothesis 5, neither veil-piercing motions nor nonveil-piercing motions have, statistically speaking, an immediate effect on case settlement (Table 3, Model [5]). After the first 30 days, however, both types of motions increase the subhazard rate of settlement, with nonveil-piercing motions (a 126% increase in the rate) surprisingly doing so with more heft than those of a veil-piercing nature (a 42% increase).²⁷

Overall, then, we have strong support for two of our five motions-related hypotheses (1 and 3), modest support for Hypothesis 2, and no support for Hypotheses 4 or 5.

Turning briefly to our control variables, we find that some, but not all, have a statistically significant effect on the time to case settlement. As we observe in Models (1)–(5), discovery motions rarely have a statistically significant effect on the rate of case settlement and, when they do, that effect is quite modest compared with the effect of substantive motions (e.g., a 29% increase in settlement rate in Model [1]). This result, when paired with our theory and findings on substantive motions, provides an important caveat for previous research linking settlement timing to discovery practice (e.g., Farber and White 1991; Huang 2007).

We also find evidence that the sex of a case’s judge has a positive and statistically significant effect on settlement timing. In our five models, a female judge’s presence leads to anywhere between a 40% and 49% increase in settlement rate. While this sizable, consistent, and positive result is not likely to be surprising to those across fields of study that argue that women leaders and managers are more effective at fostering an environment of compromise and negotiation than men (Eagly and Johnson 1990; Rosenthal 1998), it does stand in sharp and impressive contrast to the notable published empirical work in the settlement arena that finds no difference in settlement rates among cases assigned to male and female judges (Ashenfelter et al. 1995). We hope that future work will explore this effect, and the theory behind it, in much greater detail.

We also find, as expected and confirmed in our descriptive results above, that cases without any litigation activity settle much more quickly than those with it (in each of our five models).²⁸ We graphically depict this effect in Figure 4 by once again plotting the estimated CIF curve for cases without any substantive or discovery motions. For sake of simplicity, we estimate this solely for litigation activity in Model (1). The resulting sharp, rapid uptick in the probability of settlement depicted in Figure 4 is quite impressive. And, more importantly, when the litigation activity figure is viewed alongside Figure 3 above, it goes a long way toward confirming our expectation that cases with and without litigation activity are qualitatively different from one another, particularly when it comes to information exchange and settlement timing.²⁹ While not filing any motions may be the fastest route to case settlement, most cases, in our sample and in trial courts more generally, lack the necessary litigation information to be positioned for this.

We also find that a jury demand is associated with as much as an 18% decrease in the rate of settlement in a given month in three of our five models, something that may reflect litigation uncertainty upon filing. Finally, negative, but statistically insignificant, results characterize our local counsel, case strength, litigant status (entity v. entity), and chief judge control variables. These (non) results may well indicate that these control variables are not settlement and/or termination-timing specific. They are not our main focus here, of course, but their performance certainly should prove intriguing for future work tackling the intricacies of the litigation process and its relation to settlement.

4. General Discussion and Limitations

4.1 Discussion

This study has allowed us to unpack the relationship between post-filing settlements, the dominant termination method for cases in trial courts today, and the litigation process. Our results confirm that a key mechanism of formalized litigation—motion practice—can and does influence settlement and the timing of when it occurs. Our federal district court data enable us to get a unique picture of the details of litigation and to systematically study the dynamic nature of the whole process, something that had previously been focused almost exclusively on discovery. Similarly, our methodology, including competing risks regression models of duration and time-varying motion-level covariates, allows us to capture the propelling effect that motions made and ruled on can have on the timing of case settlement.

In summary, our empirical modeling reveals that, as predicted, the filing of a substantive, nondiscovery motion speeds case settlement. In addition, we also find support for our expectations that motions that are granted are more immediately important to the settlement rate than motions denied and that plaintiff victories have a more substantial effect than defendant victories. These findings are substantial in both the direction of their effects and in their size.

Our work further illustrates the importance of continued study of the settlement process. Theoretical and empirical work on settlement is quite extensive, covering both domestic and foreign tribunals. Canvassing the literature, we find that researchers’ interest in the subject is largely driven by two goals. First, scholars seek to learn what factors motivate settlement, which is assumed (generally) to be a social good that ought to be subsidized. Investigations in this mode have generally assumed that the parties are passive and that changes in fixed case attributes are the levers which might be pulled in a more efficient manner. Our work demonstrates that motion practice can influence the timing of settlement, through unlocking information that the parties and the court otherwise would not share with one another. Although our study provides an important insight on this subject, we expect that future projects will employ larger, more globally representative, data sets that more closely mirror the NOS distributions depicted in black bars in Figure 1 and that deal with a more diverse set of issues. These future studies could further illustrate the interaction of motion type, timing, and settlement. As time passes and electronic case resources become more comprehensive and cheaper to study, this kind of work will be increasingly easy to perform.

Second, scholars seek to learn about settlement because they really wish to know about the effect of various changes in the legal regime on case outcomes—who is winning and who is losing. Here, our work on the timing of settlement fits in a recent tradition that notes that early settlements may result in important distributional consequences for the parties (e.g., Nielsen et al. 2010). Data on the timing of settlement has the advantage of being relatively easy to collect, unlike information on how the parties have divided the pie. Our finding that parties can motivate settlement through their own efforts might help to focus research on new ways of looking at “victory”. For instance, a settlement that results from party A making and winning a motion might be more likely to be felt (by A and B) to be A’s “victory” than a settlement that happens absent motion practice.

The analysis of settlement’s timing could also be used to examine the influence of lawyering: sophisticated lawyers might be extracting more value from litigation (in terms of efficient compromise) than less sophisticated lawyers, through tactically timed motion practice. Scholars of settlement to date have largely discounted these active roles for lawyers and lawyering, preferring to see both clients and lawyers as passive obstacles to settlement. Employing this framing of the “problem” of settlement, policymakers have focused on changing the law, or helping judges to learn techniques, to encourage compromise. But our research suggests that the parties have an important role to play, and that the structure of the procedural rules in particular might influence the timing of settlement in unexpected ways.

4.2 Limitations

Although we believe this work makes substantial strides in providing systematic evidence into how the dynamics of litigation like motion practice can affect settlement, we would be remiss to not mention a few of its limitations. Most importantly, apart from the causation concerns discussed above, the data set is limited to veil-piercing cases, and as such is not representative of all federal cases, let alone all litigation. We see in our data set proportionately fewer torts, civil rights, and constitutional claims than the typical federal filing, and proportionately more contract, labor and intellectual property suits. It is possible that litigation of these commercial disputes is not generalizable to other kinds of cases, which, in state court in particular, ordinarily resolve a personal injury claim. Settlement in personal injury cases may be even more strongly driven by attorney agency costs, as large “settlement mills” control many filings (Engstrom 2009). This would tend to result in stronger effects for attorney characteristics, including some that are unobservable on the face of the docket.

Similarly, our data set is limited in time and scope, with cases filed from 2000 through 2005 and which had their initial complaints available on Westlaw’s PLEADINGS database. Although we have no reason to expect that the latter limitation creates bias, it might be that the introduction of electronic docketing—when coupled with other products of the digital revolution—has worked a fundamental change on settlement practice. Certainly, e-discovery has increased the costs of civil litigation, making settlement an ever-more-attractive alternative to continued litigation. Thus, as compared with cases filed before digital docketing, our sample might overemphasize the importance of motion practice in settlement, as settlement is more common today than it used to be (Clermont and Schwab 2004; Galanter 2004).³⁰

At the same time, we observe neither the amount of settlements, nor the demands of the parties. Previous work found these two factors very important in determining the likelihood and nature of any eventual compromise (Kaplan et al. 2008; Schwab and Heise 2011).

We do not discount these limitations. However, even after considering them, we believe that our approach advances this area of research by empirically examining the effect of litigation motion practice, including substantive, nondiscovery motions, on the intra-case timing of settlement. Only through the recent advent of new technologies and an increase in systematic electronic record-keeping across the federal trial courts have the data utilized for this kind of dynamic study of litigation become available. As additional technologies, measurements, and methods continue to emerge, we suspect that future work will be well positioned to continue to refine this approach and address some of the concerns about our work that we state above.

5. Conclusion

Cases settle because the parties choose to compromise rather than contest. Until recently, the content of these settlements was a black box that impeded our understanding of the civil justice system. It has become clear that “future scholarship on the American civil justice system will inevitably have to include rigorous work on settlement and settlement behaviors” (Schwab and Heise 2011: 934). Through the use of a unique hand-collected data set of federal trial court cases, this project helps to illuminate the mechanism of settlement by focusing on the parties’ use of the litigation process, and particularly motions, to obtain information from one another and the court. Our most substantial and important finding is that motion practice can spur settlement even in the absence of a judicial ruling. This in turn suggests that the parties may use the processes of litigation to exchange information with one another, meaning that procedural rules that discourage motion practice may also slow settlement. The findings thus reveal a more nuanced, dynamic, and reflexive view of litigation and compromise than that previously reported.

Funding

Temple University Beasley School of Law and the Baldy Center for Law & Social Policy at the University at Buffalo

We thank Jane Baron, Ryan Black, Kevin Clermont, Tracey George, Craig Green, Michael Heise, Greg Mandel, Andrew Martin, Harvey Palmer, Rafael Pardo, James Spriggs, and the editors and anonymous reviewers of the Journal of Law, Economics, & Organization for valuable comments and advice. We are grateful to Temple Law School’s Clifford Scott Green Research Fund and the Baldy Center for Law & Social Policy at the University at Buffalo for supporting our research. A previous version of this article was presented at the 2010 Annual Conference on Empirical Legal Studies, Branstetter Litigation & Dispute Resolution Program’s 2012 New Voices in Civil Justice Workshop at Vanderbilt Law School and at Houston Law School. Law students Ajay Nair, Andrew Seaberg, Joshua Bugay, Casey Coyle, Courtney Morgan, Dana Becker, Devin Hager, Terrence Grugan, Isaac Pineo, Jason Ballenger, Jedidiah Vandermark, Lucy Moore, Laurice Smith, Pong Chulirashaneekorn, Michael Sirkin, Stephen Rudman, Matthew Canan, Robert Blumberg, and Meriliee Arevalo and graduate student Jacqueline Sievert provided research and coding assistance. As with all our projects, the authors contributed equally to this work and name order is alphabetical.

References