With the Advancement of Predictive Analytics, Consider Using FRCP 68 During Litigation

Settling on the courthouse steps happens every day. But if I’m doing that as a litigation manager, I’ve failed. I will have spent thousands, possibly hundreds of thousands, of dollars on motion practice, discovery, and trial preparation. The court will have used its scarce resources dealing with all that pre-trial practice, and now I’m settling — probably for a number that, had my adversary and I gotten serious about it sooner, would have been achievable months or years before. Most cases settle before trial. Settling earlier rather than later in the litigation process would conserve scarce judicial and client resources by avoiding the time and expense associated with motion practice and trial.

There is a procedural tool designed to encourage parties to settle, FRCP 68 (Offer of Judgment). It is the only procedural rule dealing specifically with settlement at the federal level (there are similar rules at the state level). The rule’s “plain purpose is to encourage settlement and avoid litigation” and it offers defendants that can accurately predict case value a potential cost and fee-shifting process that should encourage them to make early and attractive offers to plaintiffs. The operation of the rule may also cause plaintiffs to “think hard” about the value of their cases, if they have not done so prior to filing, when they respond to a Rule 68 offer from the defendant.

While the frequency of the defendant’s usage of the rule may be undercounted merely because of the realities of settlement negotiations and the high percentage of cases that settle, the rule seems to be underused by defendants and their counsel for a variety of reasons. Those reasons include the complexity involved with enforcing it, insufficient incentives for either party to take it seriously, the fact that only defendants are allowed to make offers under the rule, and the difficulty that attorneys and clients have negotiating early resolution of the fee-shifting cases most conducive to the rule’s operation. Another often overlooked reason why Rule 68 may be underutilized is the difficulty lawyers and clients have with valuing a case, particularly early in the litigation process when the rule would be most effective, and which would provide the best incentives for both parties to resolve the case.

This article will summarize the history of the rule and the interpretations courts have given it, and discuss some developments in predictive technology that may cause defendants to be more comfortable making early offers under the rule. Recent advances in predictive modeling technology could enable defendants to more quickly, efficiently, and accurately value their cases, and may therefore result in offers that are more attractive to plaintiffs. Use of the technology would not only provide litigants with earlier, and therefore less costly, resolution of claims, but preserve judicial resources as well.

Rule 68

The rule was included in the original Federal Rules of Civil Procedure enacted by Congress in 1938 and has been amended substantively only once in 1946. The rule seems fairly straightforward and reads as follows:

Rule 68 Offer of Judgment

1. Making an Offer; Judgment on an Accepted Offer. At least 14 days before the date set for trial, a party defending against a claim may serve on an opposing party an offer to allow judgment on specified terms, with the costs then accrued. If, within 14 days after being served, the opposing party serves written notice accepting the offer, either party may then file the offer and notice of acceptance, plus proof of service. The clerk must then enter judgment.
2. Unaccepted Offer. An unaccepted offer is considered withdrawn, but it does not preclude a later offer. Evidence of an unaccepted offer is not admissible except in a proceeding to determine costs.
3. Offer After Liability is Determined. When one party’s liability to another has been determined but the extent of liability remains to be determined by further proceedings, the party held liable may make an offer of judgment. It must be served within a reasonable time — but at least 14 days — before the date set for a hearing to determine the extent of liability.
4. Paying Costs After an Unaccepted Offer. If the judgment that the offeree finally obtains is not more favorable than the unaccepted offer, the offeree must pay the costs incurred after the offer was made.

Three potential Rule 68 outcomes

The United States Supreme Court established the modern interpretation of Rule 68 in the early 1980s in two cases: Delta Air Lines v. August and Marek v. Chesny. Justice Stevens in Delta Air explained that a Rule 68 offer could alter the traditional American Rule of cost allocation procedure under three scenarios: “(1) a judgment in favor of the defendant; (2) a judgment in favor of the plaintiff but for an amount less than the defendant’s settlement offer; or (3) a judgment for the plaintiff for an amount greater than the settlement offer.”

The Delta Air court concluded that “Rule 68 confines its effect to the second type of case — one in which the plaintiff has obtained a judgment for an amount less favorable than the defendant’s settlement offer.” Delta Air dealt with a “scenario 1” case and the Court held that FRCP 54(d), and not Rule 68, controlled how costs would be allocated in those types of cases.

Four years after Delta Air, the Marek decision dealt with a “scenario 2” case where the defendant’s Rule 68 offer was more generous than the plaintiff’s award at trial, and the Court concluded that under scenario 2, Rule 68 operated to block the plaintiff’s recovery of its post-offer costs and attorney’s fees. The Court reached this conclusion because the plaintiff’s underlying claim was based on a statute that defined the plaintiff’s attorney’s fees as “costs” which a victorious plaintiff could recover from the defendant (commonly known as a “fee-shifting” statute, because it “shifts” the traditional American Rule, which requires each party to pay their own attorney’s fees, and requires the defendant to pay the victorious plaintiff’s attorney’s fees). However, the Marek court held that because the plaintiff rejected the defendant’s higher-than-judgment Rule 68 offer, the plaintiff was blocked from recovering its post-offer costs and attorney’s fees under the underlying fee-shifting statute.

As an illustrative example, suppose a plaintiff makes a claim under a statute that defines “costs” to include attorney’s fees, and the defendant makes an offer under Rule 68 to settle the case for US$1,000. Then suppose the plaintiff rejects that offer, proceeds to trial and wins, but is awarded only US$500. Because the plaintiff was the prevailing party, absent the Rule 68 offer, the underlying fee-shifting statute would allow the prevailing plaintiff to recover its attorney’s fees from the defendant. But because the ultimate award was less than the defendant’s Rule 68 offer, Marek holds that Rule 68 operates to block the plaintiff from recovering its post-offer costs, including attorney’s fees, from the defendant. Under the reasoning in Marek, when a plaintiff brings a claim under a fee-shifting statute, Rule 68 gives the defendant an opportunity to block the plaintiff’s recovery of its fees, but only if the plaintiff receives a judgment that is less than the defendant’s Rule 68 offer. Some circuit courts have interpreted Marek to not only allow a scenario 2 defendant to block the plaintiff’s recovery of the plaintiff’s fees under a fee-shifting statute, but have held that the defendant may also recover its own fees from the plaintiff (a “reverse” fee-shift). Circuits adopting this interpretation of Marek look to the underlying statute only to determine whether it defines “costs” to include attorney’s fees, but do not refer to the underlying statute to determine the “prevailing” party. When interpreting the underlying statute and Rule 68 in this manner, these courts allow reverse fee-shifting.

Under “scenario 3,” where the plaintiff obtains a judgment that exceeds the defendant’s Rule 68 offer, the American Rule, Rule 54(d), and the underlying statute, but not Rule 68, will govern how costs and attorney’s fees are allocated between the parties. Under any of the three scenarios, the obvious challenges for both defendants and plaintiffs attempting to use Rule 68 is first predicting who is most likely to be the victorious party, and second, if the plaintiff is the victorious party, how much the ultimate award will be. Modern predictive technology may help both parties successfully address those challenges.

Formulating the offer

Regardless of whether the court allows costs, fee-blocking, or reverse fee-shifting, it is important for defendants to formulate an accurate and attractive offer, and do so early enough in the litigation process to maximize the amount of potentially shifted costs and fees. A savvy defendant should also factor its own anticipated defense costs into the offer, expecting to avoid incurring them if the offer is accepted — potentially including a “premium” in the offered amount that exceeds the plaintiff’s predicted recovery, thereby reducing the odds that the defendant will become a scenario 3 defendant after trial.

For example, assume the numbers in Marek: A US$100,000 offer, a US$60,000 award, US$140,000 post-offer fees, and US$32,000 in pre-offer fees (assume the defendant incurred the same amount of fees as the plaintiff). If the defendant, at the outset of the case, correctly predicts a US$60,000 award, and also anticipates incurring US$172,000 of its own legal fees, an acceptable offer to the defendant early in the case (before incurring substantial fees) may be any number below US$232,000. In other words, if the plaintiff accepted a US$232,000 offer early in the litigation, the defendant would be no worse off (and possibly better, by avoiding intangible litigation costs in the form of distraction and lost productivity) than it would have been spending its US$172,000 on going to trial and losing for US$60,000.

A defendant making a lower offer would have to consider the possibility that it undervalued the case, which would, under a fee-shifting statute, cause the defendant to pay not only its own fees, but also the plaintiff’s fees. This results in a dynamic where the lower the defendant’s offer, the higher the potential downside is if the defendant is wrong on its predicted outcome. Therefore, the more accurate the defendant’s prediction of the case outcome, the less risky his Rule 68 offers will be. However, as any litigation manager knows, it is very difficult to predict case outcomes.

Difficulty with predictions

Humans generally are not good at making predictions. Humans primarily make decisions using “gut feeling” or rules of thumb, more formally known as “heuristics,” based only on personal experience (a very limited data set) or data that is not highly correlated with an eventual outcome. In addition, unrecognized biases and emotions cause humans to ignore or misinterpret data that would otherwise suggest our predictions may be wrong.

Lawyers must make predictions and forecast outcomes in the face of uncertainty about costs, outcomes, and which facts and arguments may resonate with a jury, among a myriad of other factors that may affect the ultimate outcome of a case. One study found that when plaintiffs rejected a settlement in favor of going to trial, they fared worse than the settlement offer 61 percent of the time. When plaintiffs were wrong, it cost an average of US$43,000. In other words, the settlement offer the plaintiff rejected was, on average, US$43,000 more than the amount awarded at trial. Defendants, while faring worse at trial than the rejected offer only 24 percent of the time, paid more dearly for being wrong — US$1.1 million on average.

One need look no further than cases where Rule 68 is at issue to conclude that lawyers often get it wrong when predicting that the outcome at trial will be better than a settlement offer already on the table. For example, in Jordan v. Time Inc., the defendant made Rule 68 offers of US$15,000 and US$20,000, both of which were rejected by the plaintiff who was subsequently awarded US$5,500 at trial.

Our inability to quickly and easily access, synthesize, and analyze large amounts of data, and our often unrecognized human biases cause us to miss the mark on making accurate predictions. Predictive technology, enabled by an abundance of data and advances in computing power, is currently being used to make case outcome predictions, and will continue to evolve and improve to enable lawyers to make better and faster predictions, while reducing or eliminating human biases that negatively affect decision-making. A properly designed and routinely validated machine learning algorithm cannot only identify predictive patterns in large data sets of independent variables (which in a litigation context, are basically the facts of the case, and the circumstances influencing its resolution (e.g., judge, opposing counsel, jury pool, etc.)), but can also limit the effects of human biases in the analysis of that data. A discussion follows about how predictive technology uses large data sets previously inaccessible to humans, then discuss an example of a human bias (the “optimism bias”), which may be reduced or eliminated by use of good predictive models.

So much data

US Supreme Court Justice Holmes noted in 1897 that, “For the rational study of the law, the black-letter man may be the man, but the man of the future is the man of statistics.” Humans have survived by developing the ability to synthesize observations and make decisions about the future quickly and intuitively. We make the vast majority of our predictions based on informal heuristics developed using our own subjective experience and limited data sets. Those decision-making techniques, while serving us well from an evolutionary perspective, are often not effective as reliable predictors of the future.

In a civil litigation context, an in-house lawyer’s ability to accurately predict the future may represent their highest value. They need to understand whether or not they have a case, and if so, how likely they are to win or lose, and how much it will cost or benefit them. When smaller issues arise during the litigation, clients and lawyers frequently must assess the importance of the issue; whether certain facts will be meaningful or not, whether the demographic makeup of jurors would make a difference in the outcome, how likely a judge is to impose a penalty, among many others. “Could a subset of these predictions be improved by various forms of outcome data drawn from a large number of ‘similar’ instances? Simply put, the answer is yes. Quantitative legal prediction already plays a significant role in certain practice areas and this role is likely to increase as greater access to appropriate legal data becomes available.”

Utilizing data about outcomes and factors that judges and juries use to make their decisions, machine-learning algorithms can be created that are capable of analyzing large sets of data and identifying which independent variables, and in which combinations, are most predictive of the outcomes. Banks have used machine learning for years using attributes in consumers’ credit reports (number of loans, repayment histories, amount of debt, income levels, number of prior credit inquiries, among a variety of other attributes) to predict which loan applicants are more likely to repay, and to determine how to properly price loans based on that predictive model. These same methods are being deployed more frequently to make predictions about legal outcomes.

One example of how machine learning has been applied to predict a legal outcome is with the application of certain tax laws. In one study, researchers “trained” an algorithm to predict whether a Canadian worker would be classified as an employee or independent contractor. Canadian tax law, like that in the United States, uses various factors to make that determination. The researchers explained that “courts look to a number of factors including the level of control imposed by the hirer on the worker, who owns the tools and equipment that the worker needs to complete the work, the chance of profit and the risk of loss for the worker, and the level of integration of the worker into the hirer’s business.” They examined the process a worker’s or hirer’s lawyer would use to determine the answer to a hypothetical case that included factors supporting both sides of the employee/independent contractor determination. Noting that there have been over 600 decisions published which examine this issue over the past 20 years, they concluded that “[t]he volume of cases is a tremendous hurdle for lawyers in ensuring that they have covered all their bases and ensuring that their clients are getting the best advice. The amount of research required for a lawyer to fully investigate this question would be enormous.” After undertaking that enormous task, the lawyer would then need to predict how a judge, using the same facts and case research, may decide the issue.

Having loaded data from the 600 decisions into a database, the researchers “use machine learning technology to figure out the best way to assign weights to each of our variables and to figure out how the different variables interact with each other.” After completing the “learning” process, the researchers applied the algorithm to an out-of-sample set of cases and found that the algorithm “consistently gets more than 90 percent of predictions correct.” And rather than performing hours of research, and applying years of training and experience as a lawyer would under these circumstances, the algorithm, in a matter of a few seconds, predicts the likely outcome, and applies a level of confidence to its prediction. In the hypothetical case described by the researchers, which included factors supporting both possible worker classifications, the algorithm, using the factors and patterns “learned” by analyzing the prior 600 cases and outcomes, concluded the worker would be classified as an employee with 87.1 percent confidence. In another example, a set of researchers developed a tournament that pitted a computerized “classification tree” method of predicting US Supreme Court decisions against the predictions of “elite lawyers and law professors” and found that “the machine did significantly better at predicting outcomes than did the experts. While the experts correctly forecast outcomes in 59.1 percent of cases, the machine got a full 75 percent right.” Researchers have also successfully applied machine learning technology to predict patent dispute outcomes and securities fraud case outcomes. The same predictive technology and techniques could be used to attach an early win/loss prediction, and use past judgment amounts, to inform counsel’s valuation of a case in order to formulate and make a Rule 68 offer.

Human bias

In addition to employing often unreliable short-hand heuristics and hunches as their primary prediction tools, humans also contend with often unrecognized biases that negatively affect their decision-making and accuracy of their predictions. One bias that psychologists have identified which may be particularly evident in the litigation context is known as the “optimism bias.” This bias causes people to over-estimate the likelihood of the outcomes they desire, and therefore under-estimate the likelihood of the adverse outcomes. Lawyers may be particularly susceptible to optimism bias because of their ethical duty to zealously advocate for their clients, causing them to more readily adopt narratives that support their client’s best arguments and theories, potentially blinding them to other, possibly more important, believable, or persuasive narratives.

It seems obvious how an optimism bias could lead both plaintiff and defendant to faulty (and diverging) predictions about a case, each being biased toward predicting a win (the outcome they desire) and against predicting a loss. One lawyer has explained that “[t]he most obvious place this bias has an impact is in negotiating settlements. Settlement, as every litigator knows, is a lot like poker. Having an accurate idea of what’s in the other player’s hand is vital, as is knowing your own hand. If you don’t, or can’t, adequately understand the risks of litigation, you may end up pushing your client into litigating a case she would have been much better off settling, or using mediation, arbitration, or some other kind of dispute resolution.”

There are ways to control an optimism bias, including checking your assumptions and predictions with a diverse group of colleagues and other disinterested evaluators. And predictive technology, enabled by unbiased data, can help in-house counsel identify potential optimism bias in order to provide outside litigation counsel with firmer instructions regarding your decisions about settling vs. continuing to litigate, as well as potentially educate you both about the strength of the opponent’s case.

Better predictions and more data may increase use of Rule 68

Chief Justice Burger in Marek explained that Rule 68 “prompts both parties to a suit to evaluate the risks and costs of litigation and to balance them against the likelihood of success upon trial on the merits.” This evaluation and prediction exercise can be difficult to make early in the case and may be one of the reasons why more defendant’s counsel elect not to make a Rule 68 offer. But the development of predictive technology, which is currently being used to predict case outcomes, will advance to a point where quick and accurate predictions will be available to most lawyers.

Many defendants face recurring and similar claims because of the nature or size of their business, and many of those claims are often based on fee-shifting statutes where a Rule 68 offer could be particularly effective. In addition to compiling publicly available court decision data like the Canadian tax law researchers did, these defendants and their counsel would also have internal data about those claims. For example, a debt collector and its counsel would have data about the amount of settlements in past claims, the facts and circumstances of those past claims, the cost of defending cases through trial, the time and number of collection calls, the financial profiles of the borrower plaintiffs, and a myriad of other variables. These variables, in addition to the publicly available court data about outcomes in similar cases, might be used to build a machine-learning algorithm that could, given the factual inputs of a pending claim, predict more quickly and accurately than a human, the likelihood of a win or a loss, and potentially the amount of liability in the event of a loss. One writer predicts that “[t]he data held within law firms’ systems, along with publicly accessible data, will no doubt form the basis of future predictions ...” about legal outcomes.

The quickness and unbiased accuracy offered by this predictive technology will be very helpful to defendants interested in making an offer under Rule 68. In order to get the maximum potential benefit from a Rule 68 offer, the defendant offeror should make an offer as early in the litigation as possible, before the plaintiff incurs much of its costs and fees. There is little downside to the defendant to making an offer, even if a reverse fee-shift is not possible under the underlying statute. But the defendant in making the offer, and the plaintiff in responding to it (or electing to ignore it), needs to make a prediction about not only who is more likely to win or lose, but how badly the defendant may lose. If both the defendant offeror and the plaintiff offeree have access to good predictive technology, it may increase the successful use of Rule 68.

Conclusion

The Supreme Court recognized that the purpose of Rule 68 is to promote settlement. Nevertheless, the rule has been called the “settlement promotion tool that has not promoted settlements.” Courts, legislators, and legal scholars have wrestled for decades with how to apply it, and how to potentially change it to make it more effective. Nevertheless, defendants using the rule early, in response to a claim brought under a fee-shifting statute, in a jurisdiction that allows a successful Rule 68 defendant (a scenario 2 defendant) to reverse fee-shift, will need to make fast and accurate predictions about the case outcome in order to maximize the incentives offered by the rule. Making predictions about case outcomes is very difficult, and not conducive to quick and unbiased human analysis, which may be a reason why Rule 68 offers are not made, and accepted, more frequently. Modern predictive technology, enabled by easier access to a higher quantity of data may enable defendants and plaintiffs to more quickly and accurately predict the outcomes and value of their cases, which may prompt them to successfully take advantage of Rule 68 more frequently.