I. THE CONVENTIONAL FRAMEWORK

Stein endorses a fairly conventional framework in which the central concerns of evidence law—or at least Anglo-American evidence law, to which he devotes his full attention—are the following:

Stein argues that genuine evidence rules serve one or more of these objectives and he explains that his book “focuses on the rules, principles, and doctrines that forge the instrumental relationship between evidence and fact-finding” (3; emphasis omitted). Of course, he acknowledges that some rules that are lumped together for expositional or pedagogical purposes with the foregoing rules serve primarily other purposes, such as the protection of privacy, while other rules that, at least in part, address these intrinsically evidential concerns are found in other doctrinal categories, such as criminal law and contract law (3–9, 25–31).Footnote ³

This is all insightful yet largely familiar stuff. Experienced evidence scholars will generally agree that many if not most of the rules of admissibility—the hearsay rule, character evidence rules, the original document rule, authentication rules, and so on—are focused primarily on the first of Stein's three objectives, the enhancement of accuracy in fact–finding, although they may have ancillary goals as well, while many other nominally evidence rules, such as rules of privilege, are designed to serve fundamentally different, often competing purposes. Similarly, it will generally be agreed that the principal rules regulating the burden of proof are focused on the third of Stein's three objectives, the apportionment of the risk of error. Finally, few will disagree that in a world of scarce resources, the second of Stein's objectives—essentially cost containment—must serve as a constraint on the achievement of the other two objectives, especially the first.

It should be noted as well that Stein remains conventional in rejecting radical factual skepticism (56–63) and in expressing strong doubts about claims of freestanding, noninstrumental evidential rights (as distinguished from freestanding rights that are extrinsic to evidence theory but have effects on evidentiary practices; 31–33). Stein does not devote extensive discussion to these tangential topics, and I find myself largely in agreement with his conclusions, so I will not discuss them further here.

Within this conventional framework, it is useful to isolate these three quintessentially evidence goals—error–risk minimization, error–risk allocation, and process–cost minimization—as marking off a distinct body of thought and practice that calls for creative theorizing. This is the task Stein sets for himself. He argues for a substantial realignment of our understandings of the rules of evidence—so conceived—and their relationship to these important evidence goals.

II. THE RELATIVE PRIORITY OF TRUTH–FINDING AND RISK ALLOCATION

One of the recurring themes in Stein's book concerns the relative priority of truth-finding and risk allocation as normative bases and organizing principles for evidence law. He takes it that the conventional or traditional view posits truth-finding as the dominant goal of evidence law (e.g., xi, 64). In contrast, his theory posits that “the key function of evidence law is to apportion the risk of error in conditions of uncertainty, rather than facilitate the discovery of truth” (x).

This claim is puzzling when one remembers Stein's list of the three intrinsic objectives of evidence law, the first of which is “enhancement of accuracy in fact-finding or, in other words, minimization of the risk of error.” That certainly sounds like the truth-finding function. If one reduces the risk of error, one increases the likelihood that the verdict will correspond with the truth. Conversely, if one takes steps to increase the chance of such a correspondence, one thereby minimizes the risk of error. Indeed, in rejecting radical skepticism, Stein defends the “rationalist” view that “identifies accuracy or, in Bentham's terms, rectitude of decision as [adjudicative] fact-finding's principal objective” (56). So why is it that Stein in other passages seems to relegate truth-finding to a seemingly subordinate position?

It turns out that this subordination can easily be misunderstood by the reader of Stein's book. In another, unfortunately brief passage, Stein provides a clue to the precise nature of this prioritization: “Adjudicative fact-finding instantiates practical reasoning. This goal oriented reasoning aims at producing the best decision available, not at finding the truth for its own sake” (34; emphasis added).

He goes on to explain how accurate fact-finding is crucially important for the proper application of substantive norms (34–35, 55). Thus what Stein subordinates is not the facilitation of the discovery of truth or the enhancement of accuracy but rather the idea that the truth should be sought “for its own sake.” Presumably, “for its own sake” means as an end in itself, without regard to its implications for the application of substantive legal norms. Understood in this way, the first of Stein's three intrinsic evidence law objectives is unaffected, for it refers to truth-finding in an instrumental sense, as the necessary predicate for the application of substantive law, not as a goal in and of itself. This instrumental value of truth-finding, it would seem, is not subordinated to the allocation of the risk of error.Footnote ⁴

In fact, the opposite is true, even according to Stein. He “divides the realm of evidence law into two distinct domains, epistemological and moral” (11). In the former we find his first objective, enhancement of accuracy of decision. The governing norm here, Stein writes, is the “best evidence principle.” Under this principle:

In the moral domain we find Stein's emphasis on the apportionment of risk of error. The governing norms here are a complex mixture of utilitarian and nonutilitarian principles. The important point here is that Stein believes these principles to be a matter of political morality, operating in a different domain from the epistemic principles of the first domain and requiring some kind of political settlements (13, 16).

One can (and I do) question the idea that epistemological considerations are nonmoral,Footnote ⁶ but it would still be plausible that the moral considerations involved in Stein's epistemological domain (the domain of error-risk minimization) are distinct from the moral considerations involved in the domain of error-risk allocation. In any event, the important point for present purposes is how these two domains interact. According to Stein, after all reasonable efforts to enhance accuracy have been taken, real-world adjudication always suffers from the presence of a residual uncertainty. At that point, risk allocation is unavoidable (34–35). As he explains:

This is basically correct, both as a description and as a prescription. As a description, the rules of admissibility are applied temporally and analytically prior to evidence evaluation; in a jury trial, for example, admissibility decisions are temporally and analytically prior to jury deliberations based on admitted evidence. Admissibility rules seek to optimize the information available to the trier of fact relative to the goal of accuracy. The tribunal then assesses evidence under the guidance of the burdens of proof, which allocate the risk of error. As a prescription, a decision-maker ordinarily should worry about allocating irreducible uncertainty under the burden of proof only after taking all cost-justified measures reasonably available to reduce the risk that needs to be allocated.

But, ironically, we have now reached an inversion of the priorities Stein first seems to assert. So long as truth-finding is understood as instrumentally valuable, then Stein's own motto gives priority to truth-finding, to cost-justified efforts to reduce uncertainty about the facts of substantive legal significance. In the ideal world, and as an aspirational target in real-world cases, the application of the best evidence principle (or other appropriate epistemic principles) would leave no uncertainty upon which principles of allocation might operate. Risk allocation is an analytically residual concern, albeit an undeniably important one in the practical world.

What, then, are we to make of Stein's original proposition, apparently subordinating the conventional emphasis on truth-finding to the inherently more political risk-of-error allocation? There is less here than meets the eye. The reason is that, as we have seen, the subordination refers only to truth-finding as an end it itself, but there is no “conventional” emphasis in evidence law or theory on truth-finding as an end in itself. There is, of course, much judicial rhetoric extolling the importance of truth-finding, but I am aware of none—and Stein cites none—that declares adjudicative truth-finding to be important primarily as an end in itself. There is, on the other hand, much conventional theory emphasizing the instrumental importance of truth-finding as it conduces to the proper implementation of the substantive policies of the law.Footnote ⁷

Still, there are two related senses in which Stein may be right to emphasize the priority of risk allocation even with respect to the instrumental value of truth-finding. I am not sure that either is what Stein intended, but they are compatible with his overall theory, and I suspect he would agree with one or both of them. First, one might argue that policy-makers and analysts need to pay more attention to issues of risk allocation, not because risk reduction is less important but because the law and theory of risk reduction are fairly well developed already, whereas the law and theory of risk allocation are poorly developed by comparison. Second, it is possible that many or most of the specific rules of evidence conventionally categorized as truth-oriented or accuracy-enhancing are better understood as risk-allocation rules.Footnote ⁸ As already noted, Stein acknowledges that some rules of evidence, those derived from the best evidence principle, are truth-oriented. His claim, therefore, might be that as compared to the best evidence principle, risk allocation concerns explain a relatively large portion of evidence rules, larger than “conventionally” believed. Truth-finding is not then subordinated as a goal but only as the interpretive lens through which to see these specific rules. If true, this could make such rules less susceptible to being abandoned—as unnecessary or ineffective to truth-finding—if their risk-allocation function remains well conceived. We examine these ideas more fully in the following sections.

III. RISK ALLOCATION AND THE PRINCIPLE OF MAXIMAL INDIVIDUALIZATION

The most distinctive contribution that Stein injects into the largely conventional framework reviewed in Section I primarily consists of a set of claims about the importance (within the genuinely evidential domain) of risk allocation, the norms of risk allocation, and the relationship of risk allocation to the other two objectives intrinsic to evidence law. He argues that fact-finding involves “three distinct types of risk-allocation decisions”:

1. (i) formulation of the general standards and burdens of proof that set the appropriate probability thresholds for factual findings;

2. (ii) determination of the quantitative sufficiency of the entire evidential base;

3. (iii) determination of the qualitative sufficiency (or adequacy) of each individual item of evidence that joins the evidential base. (122)

The first of these is generally accepted as involving risk allocation, although Stein adds some new features into the analysis. The core of Stein's claim to fresh insight lies in the other two.

In all these areas, Stein pursues a blend of utilitarian and rights-based principles (1–2, 17). As he puts it:

Moreover, because such principles and rules concern political morality—no less than the law of crimes, torts, or contracts—policy-makers should not avoid the question of determining their content by committing decisions to fact-finders, according to Stein (119). This point is generally understood with regard to the first category of risk-allocation norms, the standards of proof. But, Stein believes, the same cannot be said for norms of quantitative and qualitative sufficiency. These need to be authoritatively formulated by institutions with political authority (133).

After an extended discussion of the difficulties of drawing inferences from evidence under conditions of inevitable uncertainty, Stein states his principal contribution to the normative theory of evidence law in the form of a principle that should govern adjudication, the principle of maximal individualization (“PMI”). Curiously, Stein does not clearly state this principle itself, other than to give it a name. Instead, he states two “specific requirements” that either collectively constitute it or else follow from it:

This canonical statement raises a host of interpretive questions.Footnote ⁹ On its face, it appears that the first requirement speaks to the question of admissibility of evidence, while the second speaks to the question of the sufficiency of evidence. But this interpretation leads to odd results. Does the first requirement mean what it says? Must all “case-specific” evidence be “received” and therefore admitted when offered? Apparently not, for Stein argues in support of a variety of conventional admissibility rules that do sometimes bar case-specific evidence, and he does so (in part) on the ground that PMI requires such rules (see Section V below).

The better interpretation is revealed by the context of Stein's discussion of PMI. It seems that the two requirements of PMI are directed at what Stein calls “quantitative” and “qualitative” weight respectively, and the primary implications of each requirement concern the adequacy of proof, a matter closest to and with most direct implications for questions of the sufficiency of the evidence (121). Unlike the first requirement, the second requirement at least specifies an “or else” clause; violation of the requirement precludes a finding against the litigant against which the argument is deployed. If a similar consequence applies to the first requirement, then that requirement might be more clearly expressed as something like this: “If any (admissible) case-specific evidence is missing or not considered, then the case must be resolved against the party bearing the burden of proof.” That still seems odd on its face because it does not allow in any obvious way for the play of the adversary system, which in many contexts must be understood at least to allow the party bearing the burden of proof to omit unfavorable case-specific evidence that the opponent is likely to be able to introduce, even if the opponent chooses not to do so.Footnote ¹⁰ Moreover, this reconstruction of Stein's first requirement suggests a question: Is a dismissal of the case also the consequence if the fact-finder violates the first requirement's duty to “consider” evidence once it has been received? If so, how is violation of that duty to be determined? Are juries to draft opinions? Are jury deliberations to be observed? Or are juries simply to be reminded of their duty to consider all such evidence?

Stein does not address these questions. He is focused instead on another set of questions: How does PMI relate to the allocation of the risk of error? On its face, the two requirements appear to add considerable obstacles to the party bearing the burden of proof. The quantitative and qualitative weight requirements operate in the context of and thus qualify the allocation of the risk of error specified by the “general standards and burdens of proof that set the appropriate probability thresholds for factual findings” (122). That poses other questions: Why are quantitative and qualitative weight requirements needed? How do they work? Why do they not create excessive burdens when added to the general standards of proof? To these questions Stein does offer some answers, and we examine them in the next section.

Before we turn to that, however, one point should be noted. As Stein puts it:

The upshot, I believe, is that one is to take PMI, as well as its two “specific requirements,” as a kind of hortatory ideal that is relaxed for various reasons and to varying degrees in different contexts. This is in tension with Stein's assertion that PMI is a “rigid” standard. Moreover, the various reasons for relaxing this rigid standard then become the heart of the matter.

IV. RISK ALLOCATION, BURDENS OF PROOF, AND THE PROBLEM OF EVIDENTIAL WEIGHT

Stein's most extensive contribution relates to the jurisprudence of burdens of proof. Here Stein augments the conventional understanding that burdens of persuasion (standards of proof such as “by a preponderance of the evidence” or “beyond reasonable doubt”) allocate the risk of error. He does this (especially in chapters 3, 6, and 7) with a series of claims about the importance of a separate consideration, the matter of evidential “weight.” Building on the work of people such as John Maynard Keynes and L. Jonathan Cohen, Stein's contribution is another in a series of suggestions that the adequacy of proof cannot be expressed simply in terms of the idea that the probability that the claimant's allegation is true exceeds some numerical threshold.

The starting point here is a body of straightforward decision-theoretic analyses that have provided such a threshold criterion (15–16). For example, if P(C|E) denotes the probability of the claim, C, given the evidence, E, then the standard result is that minimizing the expected costs of error requires a decision for the claimant if and only if P(C|E) is greater than

where D(+) is the utility loss attributable to a false-positive decision, that is, an erroneous decision for the claimant, and D(−) is the utility loss attributable to a false-negative decision, that is, an erroneous decision against the claimant.Footnote ¹¹ Thus, when the disutility of a false positive is equal to the disutility of a false negative, as is generally thought to be the case in civil trials, this critical probability is 0.5, in accordance with the civil standard of proof by a preponderance of the evidence. But in criminal cases, when the disutility of a mistaken conviction, D(+), is much larger than the disutility of an erroneous acquittal, D(−), this critical probability approaches 1. For example, if D(+) = 10 × D(−), then the critical probability is 0.91. Up to a point, Stein endorses this kind of explanation and applies it to procedural devices such as affirmative defenses and presumptions (143–153). But Stein (sometimes) diverges from this utilitarian perspective because of the matter of evidential weight. The remainder of this section examines this important topic.

Against the decision-theoretic explanation of our proof standards or any other explanation that translates the burden of persuasion into a threshold criterion of probability (viz., P(C|E) > p*), various paradoxical cases have been presented in the literature, most of which have the feature that crucial evidence, evidence that appears to make the probability exceed the indicated threshold, is “naked statistical evidence.” For example, in the so-called “Blue Bus paradox,” a person is injured by a bus in a hit-and-run accident in a town in which 80 percent of the buses belong to the Blue Bus Company and 20 percent belong to the Red Bus Company. The injured person sues Blue Bus Company for damages, and no other evidence is presented. (In some versions, including Stein's, it is assumed that no other evidence is available [78]. This turns out to be quite important.) Arguably, P(C|E) for this case exceeds the 0.5 threshold for civil cases suggested above, and yet a verdict for the claimant seems troublesome. According to Stein—he cites only one case of limited applicability—a verdict for the claimant would be “widely regarded as both counterintuitive and legally impermissible” (78).Footnote ¹²

Stein proposes a solution to this important puzzle based on PMI, which—as already observed—has two components, once expressed by Stein as follows:

PMI is said to be a principle about evidential weight, which Stein explains in several disparate passages, including the following:

The two components of PMI refer to this latter dimension of probability. As noted earlier, they correspond to what Stein calls the “quantitative” and “qualitative” sufficiency of probability estimates.Footnote ¹³

The core insight here is easily understandable. Suppose one wants to test the hypothesis that a coin is “fair,” that is, that the probability of getting heads on a single flip is 0.5. One test consists of four flips. Upon conducting this test, the count is two “heads” and two “tails.” Based on this information, one plausibly asserts that the evidence supports the hypothesis. Alternatively, suppose a twenty-flip test, after which the count is ten “heads” and ten “tails.” Again, one plausibly asserts that the evidence supports the hypothesis. Despite the fact that the frequency of heads in each of the two tests is the same, there is something different about the two tests and about the strength of our conclusions. The second seems to be weightier. In Stein's terms, it is quantitatively weightier. The abstraction of the example has shielded us from the element of qualitative weight, but in fact one would like to know as much as possible about the conditions under which the flips were conducted and reported in order to assure oneself that one is really measuring the propensities of the situation and not, for example, the results of a reporting process that was rigged so as to falsely report an even number of heads and tails regardless of the actual results. That is a qualitative weight concern.

Stein claims that acceptance of PMI resolves the paradoxes. He asserts that in situations like the Blue Bus paradox, a verdict for the claimant, while not inconsistent with the probability ascription, would be unwarranted because in such cases the weight criterion, PMI, is not satisfied. Stein explains:

Of course, the long-run prediction here is not necessarily true. If Red Bus employs worse drivers, then the 80 percent accuracy prediction may be wrong. But I take Stein's point to be simply that it would not be irrational to anticipate an 80 percent accuracy rate based on the available evidence. However that may be, presumably the kind of evidence the absence of which, according to Stein, deprives claimant's case of the necessary weight is testimony of an eyewitness, subject to cross-examination, who could identify the offending bus as blue or (perhaps) a witness who could explain the import of tire tracks that could be traced differentially to the Blue Bus Company rather than the Red Bus Company, or some other such evidence.Footnote ¹⁴ In technical terms, without some such evidence, the plaintiff has failed to discharge the burden of production.Footnote ¹⁵

The above-quoted paragraph reveals a recurring problem in Stein's analysis. He often writes about such cases as being amenable to probability ascriptions that while rational, are nonetheless “completely unevidenced.” How this is possible is never quite successfully explained. In a lengthy argument (65–91), Stein claims, for example, that if there is only one winning ticket in one thousand tickets of a lottery, then without any additional information other than that a given ticket is randomly selected from the total set of one thousand tickets, a fact-finder cannot rationally accept the proposition that the ticket selected is a loser:

Examples like this one are the source of Stein's phrase “completely unevidenced,” but the argument, though repeated, remains mysterious to this reviewer.Footnote ¹⁶ In the Blue Bus case, for example, the assumption is made that it is evidenced that 80 percent of the buses in town are operated by the Blue Bus Company and that the claimant was hit by a bus. Part of that evidence is circumstantial and overtly statistical, to be sure, but it is certainly not a void. Notice, moreover, that Stein retreats from his “completely unevidenced” claim when he acknowledges that a probability can rationally be ascribed in the Blue Bus example but states that such an ascription “would not carry much weight” (emphasis supplied). Conspicuously, Stein does not here write “would carry no weight.” Stein's conclusion that a directed verdict against the claimant is appropriate in the Blue Bus case because claimant's allegation that he was hit by a blue, not a red, bus is “completely unevidenced” is in truth an assertion that the weight of the evidence in the Blue Bus case, as hypothesized, is insufficient to warrant verdict for the claimant. That, however, merely reasserts the supposedly dominant intuition about that case; it leaves open the question of how we are to determine at what level of weight a verdict for the claimant would be warranted. Categorizing inferences as “weighty” or “not weighty,” as “high-weight” or “low-weight,” does not answer this question; it merely states a binary (yes/no) conclusion about those inferences that must be justified, if they are to be justified, by reference to some consideration other than the degree of weight.

In a broader sense, however, Stein's claim is that if we direct a verdict against the claimant in a case like the Blue Bus case, it is because we believe that the weight of the evidence (as distinct from the probability of the claimant's allegations being true) is so low that the risk of error should fall on the claimant, and that allocating the risk of error in this manner is inherently a decision of political morality (48). I am inclined to agree (although I may well disagree with Stein about what it takes for weight to be inadequate). Furthermore, I think Stein has it right—and importantly so—when he maintains that “[f]actual findings need to satisfy the appropriate probability standard and the appropriate weight standard at once” (120). Stein thus rejects the more extreme claims by the British philosopher L. Jonathan Cohen to the effect that the decision criterion for trials must be articulated wholly in terms of weight.Footnote ¹⁷ Stein clearly believes that such an exclusively weight-based criterion, at least if universally applied, would ignore legitimate utilitarian goals (105–106). Stein also rejects the opposite monistic approach, which would claim that considerations of weight can be adequately subsumed within the assessments of probability made by the fact-finder.Footnote ¹⁸. As far as I can tell, Stein does not clearly articulate a reason for this. A powerful argument for rejecting this monism, consistent with Stein's emphasis on our inability to know the exact import of missing evidence (e.g., 44, 120), is that adjustments of probability on account of missing evidence are only an incomplete and sometimes inappropriate substitute for being able to consider that evidence itself.Footnote ¹⁹

Of course, what is at stake is much more than the resolution of highly unusual cases like that of the Blue Bus paradox. Stein rightly perceives that the drawing of inferences from “rudimentary” evidence presented at trial (e.g., witness W testifies that X is true) requires the application of generalizations about the way things happen that are inherently statistical, even when they are articulated in qualitative, nonnumerical terms, as when a trier of fact invokes the generalization that “disinterested witnesses generally tell the truth.” Stein elaborates on how these inferences from rudimentary evidence and generalizations pose essentially the same kind of problem that was perceived in regard to the hypothetical cases of overtly numerical, “naked statistical” evidence (65–80). He writes:

Thus the problem addressed is a ubiquitous challenge to the validity of adjudication. And again, PMI is offered as the solution. Among other things, case-specific evidence is critically important for selecting the most useful generalizations that inform inferences. As Stein puts it, “[t]he more case-specific evidence, the better” (72). How, then, do considerations of political morality resolve the question of the tipping point, the question of how much case-specific evidence, how much weight, is necessary to allow a verdict for a claimant?

In answering this question, Stein takes us down a confusing path. Recall the first component of PMI: “fact-finders must receive and consider all case-specific evidence pertaining to the case.” Juxtapose this with his insistence, quoted in the last paragraph, that “[s]ome of the case-specific evidence that fact-finders need for their decisions is always missing.” If the practical significance of the demand that fact-finders consider all case-specific evidence pertaining to the case is that in default of such consideration, a verdict for claimant is unwarranted, as in the Blue Bus case, then necessarily all claims must be dismissed. That is a very high price to pay for the want of perfection in the inferential warrants that can be developed at trial, one that Stein repeatedly denies we should be willing to pay (48, 121).

The central conceptual dilemma here is whether we take PMI, or any other articulation of the weight requirement, as a requirement that is preferential. That is, in considering whether the failure to introduce or consider some potential item of case-specific evidence is fatal to the claim, are we to take into account the availability of that evidence and excuse its absence if it is unavailable to the claimant through no fault of the claimant, thus allowing the claim to go forward to a verdict that is nonetheless considered warranted. Or is our weight requirement strict in the sense that the want of certain evidence precludes a verdict because it produces a fatal inferential gap the importance of which does not depend on why that gap exists. The structure of the best evidence principle, which Stein maintains governs the epistemological domain, is certainly preferential in this sense.Footnote ²⁰ Is this also a feature of the principle of maximal individualization? Unfortunately, it is difficult to determine Stein's position on this critical point.

On the one hand, Stein insists that PMI is a “practical standard” (100) and that “[a]pplications of PMI maximize case-specificity to the extent feasible” (103). Thus he seems to be clear that availability does matter. For example, he states:

So it seems that PMI demands only the maximal practicable individualization, which will depend on the availability of evidence in the particular case, not some strict version of maximal individualization, however that might be conceived. But let us return now to the Blue Bus case. Stein agrees that a verdict for the claimant would be unwarranted and he posits that this result is explained by PMI. Yet, as noted above, his formulation of that hypothetical case includes the condition that no additional evidence—case-specific or otherwise—is available. That would suggest that PMI is exhausted and the case is left to be resolved by the burden of persuasion, understood according to the P(C|E) > 0.5 threshold. Something is amiss. The ambivalence continues in a later passage, when Stein seems to retract his previous assertion by arguing that maximum “feasibly available” evidence does not satisfy the weight requirement so as to permit a verdict based on the resulting probability assessment under the controlling probabilistic standard of proof (122–124).

The resolution of this seeming paradox is that Stein takes PMI to have two branches, one that is preferential and one that is not—a distinction orthogonal to the distinction between the two requirements of PMI. This is suggested by the following language, which appears much earlier in his book than most of the passages quoted above:

Thus Stein appears to be saying that PMI in its epistemological branch is preferential, but PMI in its supraepistemological (i.e., moral, political) branch is not preferential. The nonpreferential demand is added to the preferential one.

What, then, is the nature of this nonpreferential branch of PMI? I read Stein as suggesting that somewhere between an evidence package that produces absolute certainty about the litigated events (which never obtains in practice) and a package that includes all evidence that is practically available (the preferential ideal, which can obtain in practice) is a package that is complete, one that includes all potentially available evidence in the case. Weight is then measured by reference to the gap between the evidence actually considered and this “complete” package of all potentially available evidence. This idea coheres with a comment Stein makes in explanation of the theory of weight put forward by Professor Cohen:

Although Stein does not explicitly endorse Cohen's idea, he does not reject it either. His implicit endorsement is suggested by the unqualified references in PMI to “all” case-specific evidence and to “maximal” individualized testing, standing alongside Stein's rejection of the “Benthamite precept ‘gather all relevant information that can practicably be obtained”’ (122–123). In any event, if this notion of a complete evidence package—still short of certainty about the material events but more extensive than the package of evidence that is practically available—could be made to work, then Stein's various claims in regard to the Blue Bus paradox could at least be reconciled, more or less. The idea appears to be that if the complete evidence package, one of ideal composition, could be taken into account, then the trier of fact could determine the “true” probability to apply to the appropriate standard of proof. Stein writes, for example, that “the relative strength or cogency of a probability estimate depends on the size of the factual ground that its supporting evidence covers,” which suggests that evidence could in principle cover the entire ground, rendering the probability assessment perfectly cogent (82).Footnote ²¹ Short of this, however, there is risk of error in the determination of the probability, which risk must be allocated among the parties.

Unfortunately, Stein gives us nothing by way of an explanation of how this counterfactual notion of complete evidence can be worked out as a theoretical tool of practical value. For example, if the hypothesis that a coin is fair could be tested by evidence of ideal composition—presumably, a test with an infinite number of flips—then one would know the “true” probability of the coin coming up heads. But such a test is obviously not practical. If, then, the maximum feasible test for the relevant decision context involves, say, one hundred flips, on what grounds could we specify an ideal package consisting of a number of flips greater than one hundred but less than infinity? Coin tossing is a randomized process (indeed, intentionally so), but similar problems attend the notion of a complete or ideal evidence package in relation to deterministic phenomena. Here the problem is that the obvious candidate for an ideal package of evidence entails certainty about the event rather than about the probability of the event, a point that Stein seems to recognize (82–83, 88–89, 123). Of course, such certainty is incompatible with idea of inductive or abductive inferences involved in trials and cannot be achieved in real cases. We are, therefore, driven to the conclusion that PMI cannot demand such an ideal evidence package, yet Stein does not tell us how to construct a notion of ideal evidence that is short of such certainty without limiting PMI to a preferential principle.

Moreover, even if the concept of the ideal or complete evidence package could be coherently constructed so as to be the object of a practical yet strict PMI, it would still present the problem of explaining why such a PMI would be morally or politically superior to a preferential principle. Suppose the question is whether the want of testimony by some witness not presented at trial is fatal to a nonpreferential weight requirement. How are we to distinguish between a witness to the litigated events who is dead by the time of trial (quite accidentally, let us assume—or else a preferential analysis might be implicated) and a witness to the litigated events who never existed? Presumably someone such as Cohen or Stein has in mind a concept of complete evidence that would demand the consideration of the testimony of the first witness but not the second.Footnote ²² (If the first is not demanded either, then we are back to a preferential-weight requirement.) What epistemological or moral difference exists, however, between a witness who once existed but no longer does and a witness who never existed at all? Both involve a counterfactual situation that could, if somehow realized, yield important information. What makes one counterfactual problematic but not the other? And if both possibilities violate PMI, then no case will ever satisfy it, because one can always hypothesize additional relevant evidence that would, if taken into account, reduce our uncertainty about the case.

I do not think adequate answers to these questions can be given, and Stein does not purport to provide them. Assume, then, that availability does qualify the demands of PMI, as some of Stein's statements suggest, and that he is simply inconsistent in claiming that the Blue Bus case should not be allowed to go forward because of a deficiency in evidential weight.Footnote ²³ That would call for further explanation of the meaning and contours of “availability.” Indeed, one would like to have seen Stein explore how considerations of political morality determine answers to the crucial question of why and in detail how availability is a test that limits the demands of PMI. One would like to have seen distinctions among morally pertinent categories related to the reasons that can be given by a party for not presenting certain evidence, such as: specific evidence that once existed but was destroyed (by whom? with what motive?); specific evidence that exists but is legally privileged (with the privilege in the hands of whom?); specific evidence that exists but is available only at exorbitant cost; and specific evidence that is more readily available to one party than the other.

These, and perhaps other categories, are the ingredients by which a moral theory for allocating the risk of error can be constructed. Had Stein focused on these considerations, he might have discovered that the problem here is not that low-weight cases should not go forward but rather that cases that are less weighty than practically necessary should not go forward. And that entails the idea that even a relatively weighty evidence package may be inappropriate for adjudication if it can be made yet weightier by expedients that are practical given the nature of the controversy. Rules imposing and enforcing duties to increase case weight may involve allocating the risk of error, but when this is so, such risk allocation is secondary, a tool used to reduce the aggregate risk of error, rather than the rules' primary goal.Footnote ²⁴ Stein does not pursue this kind of analysis, perhaps because he believes (incorrectly, I think) that “[e]quitable apportionment of the risk of error . .nbsp;. is not one of the grounds that can justify suppression of evidence under [the accuracy-maximizing] criterion” (15).Footnote ²⁵

Stein's general discussion of PMI thus leaves things quite cloudy. Just how stringently PMI must be applied and why that is so are questions left unspecified by any discernible criterion more specific than, in effect, “to the extent that it allocates the risk of error properly.” One hopes that this gap will be filled in the last three chapters, two of which (chapters 6 and 7) are specifically named as suggesting answers for criminal and civil litigation, respectively. On the other hand, one is forewarned by caveats that appear throughout the earlier portions of the book, including these:

What I think these caveats mean is that while Stein is not prepared to offer a strict demonstration to justify PMI as a risk-allocation norm or its applications in particular contexts, he is prepared to argue that rules and principles for the allocation of the risk of error should incorporate PMI in some way that qualifies the standard utilitarian result and that doing so can account for many of our considered moral intuitions and established rules about the sufficiency of evidence in civil and criminal cases.

Chapter 6 begins by arguing that the decision-theoretic analysis is inconsistent with British case law articulating the standard of proof in criminal cases, in that courts resist the specification of an exact critical probability with which to compare P(C|E) and draw an implicit distinction between two kinds of risk of erroneous conviction, a distinction that the decision-theoretic model cannot accommodate:

It is open to question whether courts endorse this distinction in the way Stein means itFootnote ²⁶ and whether the decision-theoretic model can accommodate such a distinction.Footnote ²⁷ But Stein's claims are ultimately normative, and he proceeds to argue that the utilitarian framework should be rejected (it makes no subsequent appearance in the chapter) in favor of two sets of requirements operating together, PMI and what he calls the “equal-best” standard: “The legal system may justifiably convict a person only if it did its best in protecting that person from the risk of erroneous conviction and if it does not provide better protection to other individuals” (175).

As Stein notes, this equal-best standard, while plausible, does not preclude a conviction on naked statistical evidence, if that is all the prosecution can provide, yet Stein asserts that such a constraint was established by his discussion of PMI earlier in the book (176). As already suggested, however, all that the earlier discussion shows is that if one is to dismiss a case on the basis of inadequate weight, then that must be because principles of political morality—principles that Stein does not fully provide and that his caveats admit he cannot fully provide—require that the risk of error be placed on the prosecution, regardless of how high the fact-finder estimates the probability of guilt to be.Footnote ²⁸ Undeterred, Stein draws the strong conclusion that:

Somewhat surprisingly (given Stein's commitment to the idea that “[f]actual findings need to satisfy the appropriate probability standard and the appropriate weight standard at once” [p. 120]), that the evidence is required to generate a probability of guilt “close to certainty” is said to arise not from the decision-theoretic balancing of false positives and false negatives but rather from PMI's (first) requirement that the evidential support “cover the factual grounds of the accusation” (183) and eliminate “all evidenced and case-specific—as opposed to merely abstract and theoretical—scenarios in which the defendant is innocent” (178). In other words, weight is the only thing that counts in this context.Footnote ²⁹

One of Stein's examples of how this is supposed to work concerns the alibi. W offers testimony that defendant was with him during the relevant time period. According to Stein, this testimony must be specifically and completely refuted, or an acquittal is in order (179, 184). Thus evidence (such as a tape-recorded conversation between W and defendant) that W and defendant concocted the alibi will “completely refute” the alibi, perhaps in combination with other evidence. (Stein does not address what should happen if the authenticity of the tape recording is challenged and the jury plausibly takes the odds of its being authentic to be, say, two to one. Is that enough to be part of a “complete” refutation?) But without such rebuttal evidence specifically targeting the alibi, the prosecution must fail no matter how strong the other evidence, including eyewitnesses placing the defendant at the scene of the crime, fingerprint evidence connecting defendant with the crime, and so forth—presumably including multiple unimpeached eyewitnesses to the criminal act identifying the defendant as the perpetrator. This other evidence, Stein admits, will reduce the probability that the alibi is true, sometimes to a figure that is minuscule, but cannot in any event “completely refute” it without evidence “specifically” rebutting the alibi. That cannot be right. The appropriateness of conviction cannot turn on the fortuity of whether the prosecution happens to have evidence “specifically” rebutting the alibi as opposed to evidence that renders it extremely unlikely, albeit more indirectly.Footnote ³⁰

Another of Stein's important implications (curiously not presented in the section on burdens of proof) is this:

That one will also come as a shock to prosecutors: if an important witness should accidentally die, at least without having made an utterance on the matter that would be admissible under the hearsay rule, the prosecutor's case evaporates no matter how strong the remaining evidence. Prosecutors' reactions aside, that does not seem to be a plausible result. To return to the questions posed earlier: What conceivable principle of political morality would distinguish between two very strong cases, otherwise identical, on the basis of the fact that a potentially important witness died accidentally in one case whereas he or she never existed in the other? If no such principle can be articulated, PMI—at least as Stein applies it—must collapse from a serious case of reductio ad absurdum.

With regard to civil cases, Stein articulates a principle of equality of risk allocation based on the duty of the state to show equal respect and concern for all citizens (214–219). In its purest form, argues Stein, the stakes of the litigation would be divided in accordance with the probability that the claim is true upon the evidence presented, P(C|E), without regard to how high P(C|E) is; that is, the claimant would receive P(C|E) × D, where D is the damages or other amount in controversy, even if P(C|E) were less than 0.5. Stein illustrates that this award would place equal risk of error on each party. Nonetheless, Stein rejects this decision rule for two reasons. First, it would not minimize errors or error costs of the decision itself or a long run of similar decisions (144–145). Second, it would have a long-run corrosive effect on the concept of entitlements (219–221). Based on these consequentialist concerns, Stein arrives at the traditional all-or-nothing rule that the claimant recovers if and only if P(C|E) > 0.5, subject to certain exceptions. The most important exception is that the defendant wins if the probability, on the evidence, of an affirmative defense being true is greater than 0.5, although Stein interestingly suggests that this should apply only to affirmative defenses that are “excusatory” in nature (based on the similar argument he makes in regard to defenses in criminal law) (151).

All this is interesting, although it breaks little new ground.Footnote ³¹ What I find most curious here is that there is no talk of an analogue to the “equal-best” standard, yet there is no explanation of why such a standard should not also apply in civil cases. Why should not the legal system do its best to prevent erroneous imputation of civil liability on a defendant or an erroneous refusal to grant corrective measures to a plaintiff? A preferential PMI could be useful in this regard; it could be employed to create incentives to produce evidence in accordance with differentials in the availability of such evidence to the parties. But again, Stein's PMI is not limited to the preferential sense I have defined. On the other hand, it is hard (though perhaps not impossible) to construct a strict PMI that would treat plaintiff and defendant equally. (It might look like this: plaintiffs lose all claims, and defendants lose all defenses; that treats plaintiffs and defendants equally in some sense, but it is hardly comforting to deserving plaintiffs.) Presumably, Stein avoids an equal-best standard in civil cases in order to avoid such untoward consequences of a strict PMI.

A number of matters concerning the sufficiency of evidence, such as the sufficiency of naked statistical evidence in light of PMI, are discussed in chapters 6 and 7 under the confusing rubric of “pre-emption” (199–207, 238–241). Stein initially defines “pre-emptive” strategies in terms of preferential admissibility rules associated with the best evidence principle (134–135). Moreover, these preemptive doctrines are repeatedly said to state a “preference” (198, 238), which again suggests that PMI might be a preferential principle. But apparently what Stein means by a preference against naked statistical evidence, for example, is that no inference can be based solely upon such evidence, which is a “strict” rule in my terms, and a sufficiency rule as well. To be sure, in the civil context he writes about the law's “general preference” for “case-specific over naked statistical evidence” (238). This suggests that he might recognize exceptions to the ban on naked statistical evidence, but the exact nature of such exceptions is poorly developed. Stein simply maintains that “[f]act-finders . .nbsp;. can rely upon naked statistical evidence when it yields a resilient probability datum (strictly speaking, such evidence would not then be ‘naked statistical’)” (239).

If I understand this correctly, Stein's ban on naked statistical evidence remains unqualified.Footnote ³² There is no obvious way, moreover, to define when a probability datum is “resilient.” For Stein, “resilient” is just another way of articulating the concept of “weighty” (47–48). But, as noted above, weight (and thus resiliency) is not an all-or-nothing characteristic; it comes in degrees, or at least gradations. So Stein's exception for evidence that yields a resilient probability datum is just an exception for evidence that is weighty enough relative to the goal of proper risk allocation. Yet, as shown above, Stein has provided us no coherent theory of how weighty is weighty enough for this purpose. Nor has he explained why demanding more weight than is required by the epistemological (preferential) PMI would be a better way to adjust the allocation of the risk of error than, say, increasing the threshold probability, p*, required under the standard of proof.

To be sure, as interpretive claims, Stein finds some support for his theory in the existence of a handful of corroboration requirements that remain, mostly in criminal cases (208, 242–243). Most of the rules Stein discusses, such as the rule that precludes a conviction based solely on the accused's confession, can be understood simply as formalized judgments, based on juridical experience, that no reasonable person could find that P(C|E) > p* without corroborating evidence—that is, not a doctrine of weight at all.Footnote ³³ But at other times such rules must be acknowledged as requiring proof to meet a more demanding standard. Requiring that there be two witnesses to a particular event (an overt act of treason, for example) is surely some kind of weight requirement and it affects the allocation of the risk of error, so this would seem to be the kind of rule that best supports Stein's thesis. Stein struggles to explain corroboration rules as consequences of PMI: “A credible piece of case-specific evidence has to corroborate the generally suspicious evidence in a way that removes the suspicion-throwing generalization [e.g., ‘Accomplice confessions under interrogation are inherently untrustworthy.’] from the case at hand” (208). But how can a single credible piece of case-specific evidence satisfy a strict PMI phrased in terms of “all case-specific” evidence and “maximal individualized testing”? And how does it remove the suspicion-throwing generalization from the case. The only way to do that would be to exclude the confession, whether or not a corroborating witness is offered, yet that is not the rule. Still, one can agree that there is an important, if limited, role to be played by such rules and that they involve explicit regulation of the risk of error without accepting Stein's broader thesis that regulation of the weight of evidence using a strict PMI is or should be a prominent part of the law of evidence.

Obviously I am unconvinced by Stein's PMI, at least in the strict (nonpreferential) form that on balance he seems to endorse. Nonetheless, there remain in these passages important reminders of the significance of evidential weight, as distinct from assessed probabilities, and the need for a jurisprudence that will address the risk-allocation consequences of decisions about the adequacy of weight in particular cases. Insofar as these arguments encourage us to take more seriously the problems of evidential sufficiency, they are a good thing.

V. RISK ALLOCATION AND ADMISSIBILITY RULES

In Section I, I noted that conventional thought about evidence law places those genuinely evidential rules that regulate the admissibility of evidence—like the hearsay rule, character evidence rules, the original document rule, and authentication rules—in what Stein calls the epistemological domain, because they are thought to be accuracy-enhancing. Stein emphasizes that such rules have an important risk-allocation component, one that determines the content of such rules (131–132).

The risk-allocation element of admissibility decisions depends on the type of admissibility rule. Two types of admissibility rule come within the genuinely evidential domain that Stein has marked out. The first kind is any rule that excludes evidence α because the entire package of evidence is considered better (in some sense) without α than with α, all other evidence being equal. We can write this as V(E and α) < V(E), where V is the valuing function and E represents the evidence other than α. (Stein does not use this formalism himself.) A paradigmatic case is evidence of bad character used to show conduct in conformity with the character trait (183–189, 227–228). The second kind is any rule that excludes evidence α as an inducement for the proponent to present evidence β instead, because V(E and α) < V(E and β). Paradigmatic is the original document rule, which excludes copies of documents when offered to prove the contents of the original (135). I have elsewhere called the former “taint” rules (because evidence α is viewed as tainting the evidence package, without regard to what might be put in its place) and the latter “preferential” rules (because they express a preference for evidence β over evidence α, even though each might well be a helpful addition to E).Footnote ³⁴ Stein calls these, rather confusingly, “exclusionary” rules and “pre-emptive” rules, respectively (134–135), even though they both involve the exclusion of proffered evidence.

Conventional theory takes the valuing function, V, to be based on error-risk-reduction. That is, V(E₁) > V(E₂) whenever evidence E₁ tends to produce more accurate verdicts than E₂. Stein, however, proposes that in determining the admissibility of evidence, the valuing function should be understood as one based on the degree of achievement of an optimal error-risk allocation. That is, V(E₁) > V(E₂) whenever evidence E₁ tends to produce a more appropriate distribution of the risk of error than E₂. Stein argues that whenever an item of evidence, α, is excluded (in his technical sense of excluded under an “exclusionary” rather than a “pre-emptive” rule), this allocates the risk of error associated with the loss of that evidence upon the proponent; whenever such evidence is admitted, this leaves the fact-finder free to allocate the risk of error associated with that evidence as the fact-finder thinks best (134). When evidence is excluded under a preemptive rule, it is because “the pre-emptive strategy identifies the initiator of the second-best inquiry as the bearer of the risk of error” (135). In other words, it's a form of penalty for, or at least incentive against, the selection of second-best evidence.Footnote ³⁵

Where, then, does this shift of focus to error-risk allocation take Stein in terms of the proper content of admissibility rules? In order to follow his conclusions, it is best to start with an idea that he presents only very late in the book. He argues that a rule of free proof admitting all relevant evidence would seem roughly to equalize the risk of error for each side (225). It is not clear to me why that would be, and in fact Stein rejects it elsewhere (229–230). But it does seem to be his conceptual starting point. It is after all one of his objects of criticism as a part of the “free proof” regime. The question then becomes: Under what circumstances and for what reasons ought we to disturb this (very) rough form of equality? Stein gives different answers depending on whether the case is civil or criminal.

In criminal cases, the risk of error is to be skewed against the government and in favor of the accused in two ways. First, PMI—specifically its second component—should be applied rigorously to evidence offered by the prosecution (183). Again, that qualitative weight component is articulated as follows: “[F]act-finders must not make any finding against a litigant, unless the argument generating this finding and the evidence upon which this argument rests were exposed to and survived maximal individualized examination” (100).

In order to assure that the fact-finder makes no such improper finding, each item of the prosecution's evidence that is not susceptible to maximal individualized examination must be excluded (104).Footnote ³⁶ Second, the best evidence principle—not a strict, nonpreferential PMI—is to be applied to evidence offered by the accused. Thus defense evidence may be excluded only if it is not the best evidence reasonably available to the defense on the particular point (193).Footnote ³⁷

I have space to consider only one example. According to Stein, exposing a witness to the possibility of cross-examination is the classic example of a demand of PMI. This in turn means that the admissibility of hearsay for the government must be strictly limited (189). Stein distinguishes between “proceeding-statements” (when a declarant speaks in a legal proceeding of some sort) and “event-statements” (made during an event outside of legal proceedings).Footnote ³⁸ Under PMI and the “equal best” principle:

In contrast, “[a] defendant is . .nbsp;. entitled to adduce in his or her defence any evidence—hearsay or non-hearsay—if that evidence is the best evidence available” (195). Presumably this means that hearsay is admissible for the defense as long as the declarant is also called to testify, or the declarant is not available to be called, or, perhaps, the circumstances indicate that the hearsay is better evidence than declarant's testimony would be. Stein does not elaborate on the point.

This is puzzling on both ends. While the demands on the prosecution are fairly tight, Stein's scheme does not impose maximal individualized testing whenever it allows a substitute for cross-examination before the present trier of fact, which it does as to both proceeding-statements and event-statements. On what ground is this done? It cannot be inferred from PMI? As Stein puts it, such substitutes are “almost as effective as direct cross-examination of the declarant” (192). But how can “almost as effective” be good enough in the context of a nonpreferential PMI that is “rigidly” enforced? Similarly, if the accused is to be allowed every chance to create reasonable doubt, why place upon him or her the constraints of even the best evidence principle? Stein does not provide answers to these questions, even though his language generally suggests that his answers are derived from previously stated principles.

His language notwithstanding, the scheme Stein endorses simply cannot be derived from PMI or the “equal best” principle, at least not without the help of auxiliary principles that Stein does not articulate. Based on what he presents, all one can say is that this is how stringently Stein thinks PMI should be applied in order to get the right skewing of the risk of error in criminal cases. He may be right about that, but his arguments do not demonstrate that he is. To be sure, it is probably fair to say that we do have a system in which the use of hearsay by the prosecution is in general more tightly constrained than the use of hearsay by the accused, especially when one takes constitutional limitations into account.Footnote ³⁹. And this surely has something to do with our concern to skew the risk of error in favor of the accused. That far, Stein is probably correct. But it will take a much more detailed and nuanced treatment of the topic than Stein has given us to establish that the contours of this difference (either as it is or as Stein thinks it should be) are best explained by PMI, especially a strict, nonpreferential PMI.

In civil cases, once again his governing principle is equality. According to Stein, the rough equality of free proof must be improved by or supplemented with “corrective equality mechanisms”: “These mechanisms render inadmissible any evidence that creates inequality between the litigating parties by imposing a non-reciprocal—and, therefore, inequitable—risk of error upon its opponent” (225; emphasis omitted).

Here we must be careful. There are two senses in which evidence might be immune from “utilization prospects” on behalf of the opponent. On the one hand, the evidence might be so powerful that the opponent is in fact unable to utilize it to her advantage despite a full panoply of opportunities. Clearly that is not what Stein thinks should render evidence inadmissible, for that would produce the perverse result of excluding the strongest pieces of evidence. On the other hand, the evidence might be such that there is no opportunity for the opponent to develop utilization prospects. This is what Stein has in mind. The actual development of such opportunities will depend on the situation, and the fact that no such development is possible on the particular state of the evidence cannot be reason to exclude the proponent's evidence.Footnote ⁴⁰

Again my example is hearsay. Stein argues that in civil cases, properly inadmissible hearsay is hearsay that is “one-sided in its content and not susceptible to adversarial scrutiny by in-court cross-examination and other appropriate means” (228). As Stein points out, evidence offered by a party almost inevitably is one-sided in that it favors its proponent's case (228), so the nonreciprocal quality that generates exclusion lies in its “untestability.” His paradigmatic example of inadmissible hearsay runs as follows:

The argument seems plausible. It rests on the idea that the opponent cannot cross-examine Heather and get her to qualify, nullify, or even reverse her direct testimony. Nonetheless, the argument is ultimately unconvincing. One can easily imagine the following falsifying evidence and arguments, to the possibility of which the proponent is exposed by introducing the testimony of its witness (whom I will call W₁):

1. (i) Testimony by W₁ on cross-examination that in fact Heather did not say what he reported on direct examination and that the claimant had bribed W₁ to give this testimony, but that he, W₁, has had a change of heart when reminded by the cross-examiner of the costs of a perjury conviction.

2. (ii) Evidence that W₁ has been convicted of perjury on two prior occasions, a fact known to the claimant.

3. (iii) Testimony by W₂ that he observed the declaration by Heather (the time and place of which would have been identified by W₁, or else the fact-finder would have strong reasons to make a negative inference against the claimant anyway) and that Heather did not say what W₁ claims she said, but rather she said the opposite, a fact that W₂ also communicated to the claimant.

4. (iv) Testimony by Heather that she never said what W₁ claims she said, that she said the opposite, and that she had communicated this fact to the claimant.

5. (v) Testimony by W₃ that Heather said to the claimant that the defendant intended to honor the agreement.

6. (vi) Testimony by W₄ that Heather hates the defendant and promised to “get even” with him, even if that meant lying under oath.

7. (vii) Testimony by W₅ that Heather has a financial interest in the claimant's business.

8. (viii) Closing argument by the defense pointing out that the claimant did not call Heather to testify and gave no explanation for why that is so.

I could go on, of course, but the point is made. By introducing the hearsay (if that were allowed) as opposed to Heather's testimony (assuming the latter is an available option), the claimant is exposed to impeachment of W₁ that would not otherwise be possible. (Either way, Heather is subject to impeachment.)Footnote ⁴¹. By introducing the hearsay as opposed to introducing neither the hearsay nor Heather's testimony, the claimant is exposed to impeachment of both W₁ and Heather that would not otherwise be possible. (If Heather is unavailable, then offering her hearsay may also involve its proponent's inability to offer rehabilitation testimony that would be available only from Heather herself.) In either context, many of these potential exposures constitute what Stein calls “adverse utilization prospects,” the presence of which he considers to be sufficient for admissibility (230). That is, they go beyond reasons simply to ignore W₁'s testimony and become reasons to infer against the claimant in the case, inferences that would not be appropriate had claimant not offered the testimony of W₁ in the first place. In other words, a proponent's reliance on this hearsay—indeed, any hearsay—is testable in a variety of ways, some of which involve potential falsification that “projects untrustworthiness upon the proponent's case” (230). Under Stein's criterion, therefore, all hearsay should be admissible, at least in civil cases. But Stein insists that the current elaborate hearsay regime for civil cases roughly matches the demands of his corrective equality idea (231–235).

Of course, none of this shows that hearsay is as good as the declarant's in-court testimony under oath and subject to cross-examination, especially if “good” is measured in terms of accuracy enhancement. In terms of risk allocation, particular hearsay might involve somewhat more nonreciprocal risks than live testimony, and that might be grounds to exclude the hearsay under a preferential exclusionary rule, where the proponent's choice to go with specific hearsay reflects an intentional skewing of the risk of error. Such advantage-taking may need to be thwarted. But this is an unavoidably comparative and context-sensitive assessment that must look to whether the proponent can present evidence that provides comparable enlightenment to the fact-finder with less unfair skewing of the risks.Footnote ⁴² Stein writes in noncomparative (nonpreferential) terms, with some hearsay imposing reciprocal risks and some imposing nonreciprocal risks, independent of the availability of the declarant or other alternative modes of proof. The cases of hearsay (or any other kinds of evidence) with perfectly reciprocal error risks must be few in number and very difficult to identify. Moreover, the risks of error generated by preferring live testimony over hearsay could and very likely will shift in unpredictable ways.

The final point about admissibility is the most important, for both civil and criminal cases. In deciding whether to admit or exclude particular evidence (or how to shape a rule for the exclusion of classes of evidence), there are potential conflicts between risk reduction and risk allocation. The choice that maximally reduces risk may not be the same as the choice that optimally allocates the risk associated with the evidence in question, however optimality is measured. As explained in Section II, even Stein seems to give priority in cases of conflict to risk reduction, at least up to a point. Given an appropriate allocation of the risk of error by the standard of proof coupled with an intelligible sufficiency criterion of weight, why not allow the marginal changes in risk allocations associated with the admission of particular items of evidence to be governed by rules designed to reduce the aggregate risk of error?

VI. CONCLUSION: RISK ALLOCATION AND OPPOSITION TO “FREE PROOF”

Stein proceeds from a premise repeated throughout the book: epistemology cannot answer questions of risk allocation; such questions fall in the domain of political morality (e.g., p. 48). As a consequence, Stein's principal claims, as examined in the foregoing sections, all serve to make the case against the deregulatory trend of modern evidence law. The greater the importance and scope of risk-allocation decisions, the stronger are the grounds for opposing the trend toward free proof and insisting on authoritative legal resolution of allocation standards, at least as long as one believes that risk-allocation decisions should be made by law-making officials or institutions rather than fact-finders.

I am inclined to believe that Stein proceeds from an unnecessarily narrow conception of the scope of epistemology. That subject or discipline is broad enough to encompass questions of risk allocation. But pressing this point is unnecessary. Whether or not one takes epistemology as capable of speaking to questions of error-risk allocation, such questions are still matters of political morality, at least when they concern the adjudication of legal cases. So the real question is what follows from that point. We have reached a mixed assessment of Stein's claims.

In at least one important respect, Stein is correct about the underappreciated quality of risk allocation in the context of burdens of proof. There is a distinct component of burdens of proof associated with weight. Courts only fitfully and largely unofficially endorse that component as a distinct requirement, and when courts do not recognize and implement this component as a matter of law, it is left to fact-finders, including juries, to implement their own risk-allocation preferences in regard to questions of weight, perhaps by ignoring the issue entirely. That is not an entirely satisfactory state of affairs. Stein's main concern seems to be that such value judgments, reflecting political morality, should be made authoritatively as a matter of law, not left to the ad hoc judgments of fact-finders as if this were purely a matter of commonsense and logic. One can, however, agree with his conclusion—that the question of weight is underregulated—without endorsing his reasons.

One can accept the proposition that jurors (or trial judges deciding without a jury) have a legitimate if subsidiary role in making judgments of political morality; their legitimate role is not limited to “fact-finding.” Such a role is commonly recognized, for example, in allowing juries to make determinations of so-called “mixed questions of law and fact,” such as whether a defendant's conduct was negligent under the abstract “reasonable person” standard. There are many other examples.Footnote ⁴³. This potentially undermines a major portion of Stein's thesis, at least in the form he presses it. Despite this, one might believe that some questions of the adequacy of evidential weight should be resolved by higher-level law-making authorities because of the importance of precedential value, because of the superior authority of such law-making authorities to assess questions of weight, or because appropriate weight regulation as a matter of law (which litigants can anticipate and learn from better than when reacting to ad hoc reactions of fact-finders) may assist fact-finders to perform their task better than would leaving the matter in the hands of fact-finders themselves. In particular, if PMI were understood as a preferential doctrine that attends to the reasons that may be given for the omission of particular evidence, then law-makers who understand (better than juries) the economics and institutions of pretrial discovery devices would be in a better position to make determinations of weight and perhaps to develop a common law on the matter.Footnote ⁴⁴

If PMI or some other weight criterion is understood as preferential in the sense described here, then Stein's subordination of truth-finding is brought into question even in the context of regulating the sufficiency of evidence. A preference based on the availability of evidence makes sense primarily as a tool to enhance accuracy. (If there is another purpose, it is not clear what it would be.) A preference enforced by placing the burden of production on an appropriate party will have a risk-allocation effect, to be sure, but the purpose that is served by such an allocation is to improve the evidential base upon which the fact-finder must make its decision. Truth-finding is primary; risk allocation is secondary.

A similar point can be made in the admissibility context. To be sure, the implications of a strict PMI in the admissibility context are not easy to discern. One might readily agree with Stein's gross conclusions—strictest constraints on prosecution evidence, weakest constraints on evidence offered by the accused, and intermediate constraints on the evidence offered by civil parties—but getting to these results from his premises, and especially getting to the specific doctrines he endorses, is another matter entirely. As illustrated above in Section V, one can reach dramatically different prescriptions from his premises. Put another way, his theory is simply not worked out in enough detail and with enough rigor at this point to assess it fully.

Stein is correct that there are reasons to be skeptical of a totally free-proof regime, one with no admissibility rules other than those authorizing discretionary exclusions for irrelevance or waste of time and with only the barest of rules regarding the burden of proof. But in truth there are few who suggest otherwise. It is worth remembering, for example, that perhaps the strongest proponent of free proof, Jeremy Bentham—the theorist against whom Stein argues most strongly (108–116)—did not actually endorse such a free-proof regime, even within the domain of evidence law proper that Stein stakes out. Bentham accepted that admissibility rules were needed to control lawyers' strategic excesses in an adversary system. For example, he endorsed rules barring what he called “makeshift” evidence, a category that subsumes the hearsay rule and the original document rule, so long as these are understood as preferential.Footnote ⁴⁵ This, of course, is just the kind of “pre-emptive” admissibility rule that Stein also endorses, one that allocates the risk of error associated with the loss of the (excluded) second-best evidence on the party who proffers it in place of the best evidence.Footnote ⁴⁶ In this respect at least, the importance of allocating the risk of error has long been appreciated, even if it has not been fully articulated. But notice that such preemptive rules allocate risk in a manner disconnected from the allocation of risk in the case as a whole; they operate at a retail level, based on the circumstances surrounding the offer of particular evidence. Once again, the allocation of risk is made in service of the goal of truth-finding.Footnote ⁴⁷ The same conclusion may pertain to those admissibility rules that Stein calls “exclusionary.” While the exclusion of particular evidence does entail an allocation of risk, that does not necessarily mean that risk allocation is the goal of the exclusion. The goal might well be error minimization, as under the conventional view.

Then, too, Bentham's most important target, the idea that dominated early nineteenth-century evidence law, was the vast complex of incompetency rules, which barred all manner of “interested” persons from testifying and generated a labyrinth of distinctions between the competent and the incompetent.Footnote ⁴⁸ Against that backdrop, Bentham's quest for free proof makes a great deal of sense. Interestingly, Stein does not defend the old competency rules; he barely even mentions them.

In the final analysis, Professor Stein has produced for us a challenge. He has put a shot across the bow of the continuing, if glacially slow, trend toward the elimination of most rules of evidence or their replacement with doctrines of judicial discretion. In recent decades, that trend is sustained largely by two factors that seem to be in ironic tension—the decline in the use of the jury and the accumulating empirical evidence that juries do surprisingly well with the evidence we give them. If distrust of juries as a theory of evidence rules is either muted or empirically unfounded, then alternative functions for the rules of evidence must be articulated and assessed—or else the rules should go. Stein has begun to articulate a rationale based on the allocation of the risk of error that is not done simply in aid of truth-finding or as a part of our probabilistic standards of persuasion. Perhaps in the future he will make his case more persuasively.