In Wal-Mart Stores, Inc. v. Dukes, Justice Scalia registered his disapproval of using statistics to litigate liability in a class action, writing

The Court of Appeals believed that it was possible to replace such proceedings with Trial by Formula. A sample set of the class members would be selected, as to whom liability for sex discrimination and the backpay owing as a result would be determined in depositions supervised by a master. The percentage of claims determined to be valid would then be applied to the entire remaining class, and the number of (presumptively) valid claims thus derived would be multiplied by the average backpay award in the sample set to arrive at the entire class recovery — without further individualized proceedings. We disapprove that novel project.

(Emphases added, internal citation omitted.) Several months later, Connecticut Law Professor Alexndra Lahav wrote a spirited defense of the practice of "Trial by Formula," in the Texas Law Review, titled, aptly enough, "The Case for Trial by Formula."

I would love to say that Professor Lahav’s argument is sound as far as it goes, but it goes a little too far. What do I mean by that?

For the most part, Professor Lahav argues that "Trial by Formula" (which she takes to mean statistical sampling in litigation) is an excellent way of ensuring equality of outcome in mass tort litigation. As she writes:

The problem with this understanding of injury valuation is that the tort system does not approximate the actual damages suffered by the plaintiff. The tort system is an institution that is supposed to monetize injuries, yet injuries are not readily monetizable. What the tort system does is assign a value to the damages suffered by the plaintiff. The amount of money damages the system assigns to injuries is contextual and cultural. This means that tort values are comparative; the value assigned to a given injury is dependent on values assigned to other injuries. The cultural contingency of tort damages is the reason that the amounts awarded in tort cases are sometimes controversial. This is also the reason that critics of the tort system are able to say that the system is unpredictable. The problem of valuing injury is not limited to the trial context. In settlement, even if one is able to accurately discount the amount of damages by the probability of the defendant being found liable, the damages assigned to a plaintiff (the amount that is to be discounted) will still be contested.

(Emphasis in original.) To ensure an accurate valuation of damages, Professor Lahav argues that courts should be more rigorous in their statistical methods, an argument that I (and most lawyers) would have no problem with.

To the extent that Professor Lahav argues that statistical sampling may help to smooth out the variations in damages awards, I think she has a strong case. And while I can certainly see where there are sound strategic arguments on the other side (who chooses the sample? for example), she has at least helped to frame an issue that both plaintiffs and defendants might agree on. (And, in many cases, they do. This is why matrix settlements have become popular in mass torts.)

The problem with her argument is that–at least implicitly–she does not confine herself to using statistical sampling to measure damages. Instead, she appears to also want to use it to determine liability. She tips her hand in two places. The first is in her discussion of several class actions that used statistical techniques, not just to determine the amount of damages, but also to determine the fact of injury for different plaintiffs:

In the late 1990’s, a few trial courts experimented with binding statistical adjudication procedures. In Hilao v. Marcos, a federal court used statistical methods to adjudicate a class action brought on behalf of persons who suffered human rights abuses under the regime of Ferdinand Marcos in the Philippines. A special master conducted on-site depositions in the Philippines, and based on these he recommended a recovery schedule to a jury, which then adopted his recommendations (for the most part). The Ninth Circuit upheld this procedure. Around the same time, a U.S. District Court judge in Texas tried 160 asbestos cases and was prepared to use these verdicts to extrapolate to the remainder of asbestos cases consolidated before him. The Fifth Circuit quashed his efforts, holding that the extrapolation of the results of the sample verdicts violated the defendant’s due process right and the Seventh Amendment. No trial court has followed in the footsteps of these innovators and the appellate courts continue to express hostility to mandatory statistical adjudication of this type.

(Emphasis added, footnotes omitted.) The second place is more explicit, when she discusses how one might use statistical sampling to root out fraudulent claims:

Trial by Formula has the potential to resolve many other problems that plague modern litigation. For example, commentators have repeatedly lamented patterns of baseless claiming in mass tort litigation. Sampling offers a way of addressing the phenomenon of fraudulent claims and creating incentives to curb them.

(Emphasis added, footnotes omitted.) In both of these cases, the problem that the courts (and the plaintiffs) have worried about is not the statistical determination of damages once liability has been established, it is the statistical determination of liability itself. This is the same issue that Justice Scalia had in the Dukes case. To repeat his specific issue (as opposed to the summation):

A sample set of the class members would be selected, as to whom liability for sex discrimination and the backpay owing as a result would be determined in depositions supervised by a master. The percentage of claims determined to be valid would then be applied to the entire remaining class …

It isn’t the use of statistical sampling to determine damages that defendants (or the Court) worries about. It’s the use of sampling to determine liability that causes problems, because statistical sampling cannot tell which plaintiffs are actually entitled to relief and which are not. Professor Lahav argues that "Trial by Formula" works because we want equality of outcome–treating like cases alike. But no defendant–or anyone else concerned with due process–wants unlike cases treated alike, particularly when the difference between the cases is that in one the defendant is actually liable and in the other it is not. That’s the bridge too far, and that is the one that Professor Lahav and others either don’t notice–or won’t admit–they are crossing.