WHEN AND TO WHOM DOES THE PRESUMPTION OF INNOCENCE APPLY?

I begin with the easiest of the puzzles facing us: At what points along the path marking the sequence of events in the criminal justice system and to which actors in that sequence does the PI apply? One might have supposed, naively as it turns out, that a person is presumed innocent from the moment police attention first falls upon him as a suspect until the end of his trial, when a verdict has been pronounced (supposing things progress that far). One might have supposed, equally erroneously, that all those charged to make decisions about the defendant (judges, grand juries, prosecutors, and so on) are under some duty to presume his innocence.

While such a broad construal of the meaning of PI is common (at least in theory) in several countries, the Supreme Court has made it very clear that in American criminal procedure the PI applies only to the trial itself and, within the trial, only to the jury or other trier of fact. In 1979, the Supreme Court considered a class-action suit from various persons in New York State who had been denied bail and jailed pending trial. Their suit claimed that the conditions in the correctional facilities housing them violated their constitutional rights, among them the right to be presumed innocent until proven guilty. The Supreme Court, in a majority opinion written by then–Justice Rehnquist, found that the presumption of innocence simply did not apply to extratrial events:

The Court's conclusion about pretrial confinement was derived from a general analysis of role of the PI. In the view of the Court, the PI is not an omnibus claim of the innocence of everyone charged with a crime. The Court deliberately chose to characterize the PI narrowly as a mechanism for localizing the burden of proof during a criminal trial itself. “The presumption of innocence,” the majority wrote, “is a doctrine that allocates the burden of proof in criminal trials; it also may serve as an admonishment to the jury to judge an accused's guilt or innocence solely on the evidence adduced at trial and not on the basis of suspicions that may arise from the fact of his arrest, indictment, or custody, or from other matters not introduced as proof at trial.”¹⁷

If the PI is but a mechanism for allocating the burden of proof and warning jurors against drawing adverse inferences from the indictment against the defendant, then there is not the slightest warrant for supposing that the PI governs events beyond the trial itself. The courts took a similar line when Congress passed a law making it illegal for someone under indictment for a serious crime to buy a firearm. When the law was challenged on the grounds that it violated the presumption of innocence to which all persons are entitled, the courts insisted that the law was constitutionally sound and that the presumption of innocence did not apply outside of the trial itself, adding, for good measure, that “the indictment of an individual for a crime punishable by imprisonment for a term exceeding one year is so often indicative of a propensity for violence.”¹⁹

While there are many critics (including some on the High Court itself) of this restricted construal of the provenance of the PI—critics who fancy that the presumption of innocence follows the defendant right through the sequence of decisions made by the justice system—this bit of wishful thinking simply cannot be right. Running through the events in a criminal proceeding are a host of decisions that cannot be squared with a broad construal of the meaning of the PI. Consider just a few: The police who arrest Jones clearly do not regard him as innocent, for probable cause precludes the arrest of persons thought likely to be innocent of a crime. The grand jury that indicts Jones likewise no longer presumes his innocence, since the indictment indicates that it is satisfied that he probably committed the crime. When the now-arrested and charged Jones requests a bail hearing, the presiding judge cannot be presuming Jones's innocence in any robust sense, since if he did, he would not be demanding that Jones plunk down a tidy sum to assure his appearance at trial nor, even more vividly, would the judge have the authority to lock away the presumedly “innocent” Jones until the completion of his trial.

We do not generally deprive persons whom we genuinely believe to be innocent of their liberty whom we genuinely believe to be innocent. Indeed, in many jurisdictions, bail can be denied to a defendant—even if he is facing a murder charge—only if the prosecution can persuade the judge that the evidence creates a “strong presumption” of Jones's guilt.²²

Once Jones's trial gets underway, the PI still has a quite limited scope in terms of those who are bound by it. The prosecutor obviously is not constrained by the presumption of innocence; to the contrary, it is his responsibility to depict Jones convincingly as the perpetrator of the crime. The trial judge is likewise forced to reveal his hand as the trial proceeds in ways that cannot easily be squared with the notion that the judge presumes Jones's factual innocence throughout the ordeal. Specifically, once the prosecution has presented its case-in-chief, the judge is frequently called upon by a defense motion to dismiss the case against Jones for lack of evidence (thereby producing a directed verdict of not guilty). If the judge permits the trial to proceed, as he usually will, this can only be because he believes that the prosecutor has presented a case that could persuade a reasonable person of Jones's guilt beyond a reasonable doubt. That is, the trial judge who does not dismiss a case at this stage has already determined—well before the jury adjourns to its chambers for deliberation—that it would be entirely reasonable to believe firmly that Jones committed the crime and therefore unreasonable to believe that he did not. The judge's determination that the case for Jones's guilt may rise to proof beyond reasonable doubt cannot easily be reconciled with the assumption that the judge believes that Jones did not commit the crime (although the judge's ruling at this point is still compatible with a jury finding of “not guilty”).

In short, the only participants in the legal process who would appear to be bound to believe in Jones's innocence until jury deliberation begins are the jurors themselves; and the only judicial proceeding in which the presumption of innocence unequivocally operates is the trial itself. (And if the analysis in the next section is correct, not even jurors are obliged to assume that Jones is innocent in the sense that he did not commit the crime with which he is charged.)

One might be inclined to argue—and some courts have—that there is an attenuated sense of the PI that kicks in well before a criminal trial begins.²⁴

KINDS OF INNOCENCE AND GUILT

Having narrowed our focus to the trial and, within it, to the jurors, we can now begin the tricky task of exploring exactly what it is that jurors are supposed to assume or presume. It is possible and will prove profitable to distinguish two distinct notions of innocence and guilt:

Obviously, a similar distinction can be drawn between material and probatory guilt:

There is a salient asymmetry between the two pairs of distinctions. It consists in the fact that (1) while a finding of guilt_p sustains (fallibly) the assertion of guilt_m (that is to say, the legal system assumes that someone proved to be guilty is genuinely guilty), (2) a finding of innocence_p (a “not guilty” verdict) warrants no inference about innocence_m. That, indeed, is why juries do not make findings of innocence but of “not guilty,” where the latter implies nothing more than lack of proof sufficient to warrant conviction.

There can be little doubt that the layperson's (and hence the juror's) sense of “innocence” is innocence_m. When a judge instructs jurors that they should presume the defendant to be innocent until proven otherwise, they will almost certainly construe that as committing them to believing that the defendant did not perpetrate the crime, especially since the judge will rarely gloss the meaning of the PI, supposing (as more than one court has put it) that: “The term [sic] presumed innocent has a self-evident meaning comprehensible to the lay juror.”²⁵

Interestingly, however, the layperson's notion of innocence, innocence_m, is an almost empty category in Anglo-Saxon law. From the moment of arrest forward, virtually all the relevant judicial decisions about the accused fail to include a finding of innocence_m among their possible outcomes. Whether we have in mind a prosecutor's decision not to press charges, a grand jury's finding of no bill, a judge's decision to dismiss a case at arraignment, a jury's acquittal, a judge's directed verdict, or an appellate reversal of a conviction, nowhere can we find an outcome that would enable us to infer that the accused was found or believed by the relevant decision-maker to be innocent_m, that is, that he (probably) did not commit the crime. Instead, every one of these decision points involves asking whether the case against the accused rises to the standard of proof considered proper to that sort of decision. At no point along this spectrum does the failure of the state to satisfy the relevant standard of proof justify an assertion of innocence_m.²⁶

A judge's refusal to issue a warrant for someone's arrest reflects a decision on his part not that the person in question is innocent_m but that the evidence offered to justify the arrest fails to establish probable cause. A judge's granting of bail over the objections of the state means the state has not produced enough evidence of guilt to justify pretrial detention. The grand jury's no bill says merely that there is insufficient evidence to indict. The judge's directed verdict says that a reasonable person could not believe that the evidence proved the defendant's guilt beyond a reasonable doubt. The jury's acquittal says that the state's case against the accused failed to persuade them beyond a reasonable doubt. None of these findings renders the hypothesis that the defendant in question is innocent_m more likely than not.²⁷

Curiously, almost the only way a criminal defendant can secure a decision that actually implies his innocence_m is first to win an acquittal in a criminal case and then to bring and win a suit for malicious prosecution. Virtually no other judgment in the criminal justice system represents a finding of innocence_m.²⁸

Indeed—and the irony should not go unnoted, for it is significant—the only place in a typical criminal proceeding where the innocence_m of the defendant is ever alluded to, by law and by tradition, is in the judge's opening instruction on the presumption of innocence. That instruction typically is couched in such terms as “The defendant is presumed to be innocent” (Virginia Model Instruction) or “The defendant stands before you presumed to be innocent of any crime” (Massachusetts).²⁹

Numerous scholars agree with Tribe that factual innocence should be presumed by jurors.³¹

On those rare occasions when a judge slips up and suggests to jurors that their task might be to settle whether the accused is guilty or innocent (as opposed to guilty or not guilty), higher courts invariably cry foul, insisting (as the First Federal Circuit has on several occasions) that “jurors are called upon only to decide whether the prosecution has proven the defendant guilty beyond a reasonable doubt, not whether the defendant is innocent.”³²

By way of explaining to the jury why he was instructing them on the presumption of not guilty rather than innocent, the Flores judge said specifically: “There is a difference between being innocent and not guilty. A person can be found not guilty and still not be innocent of the crime charged.”³⁵

It is worth noting for future reference the tensions implicit in the cases just mentioned. In the first series of trials, we learned that judges must not suggest that jurors have to decide whether the accused is innocent or guilty but rather whether he is guilty or not guilty. But when, as in Flores, a judge tells jurors that they should presume the defendant is not guilty (rather than innocent), his chain is firmly yanked, and higher courts insist that it is innocence_m, not lack of guilt, that must be presumed. In short, what juries have to decide (whether the defendant is not guilty) is something wholly different from what they have to presume (the defendant is innocent_m). In a nutshell: the defendant herself has pleaded not guilty (that is innocent_p not innocent_m; if the jury acquits her, they will be affirming her innocence_p, not her innocence_m. Why, then, should the jury presume her innocence_m? This tension will prove to be a salient and resilient one coming back to haunt us more than once.

Much of this is perfectly familiar terrain. Still, I have my doubts about whether either the courts or academic lawyers have carefully thought through what is entailed by the fact that nowhere does the system routinely make a finding about the innocence_m of the defendant. The point where the implications of this odd state of affairs show up most vividly is in our understanding of what the PI means, what work we expect it to do for us, and what protections it can offer to the defendant.

I begin with a critique of the view that the trier of fact must believe at the outset of a criminal trial that the defendant is innocent_m, that is, that he did not commit the crime of which he is charged. On this view of things, it falls entirely to the state—if it is to secure a conviction—to persuade the trier of fact to change her belief from one of innocence_m to one of high confidence in the guilt_m of the accused.

PROBLEMS WITH THE SUPPOSITION OF MATERIAL INNOCENCE

There are severe conceptual problems associated with this way of seeing things. For instance, how strong must the juror's initial confidence in the innocence_m of the defendant be in order to satisfy the demands of the PI? For technical (but no less compelling for that) reasons, the juror cannot begin the trial assigning a probability of 1.0 to the defendant's innocence, for it would then be impossible for any evidence rationally to shift the juror's opinion.³⁸

Leaving technical probabilities to one side, there are other curious features to a presumption of innocence_m. In essence, the juror is told that he must adopt an assumption—presumably that means a belief—that the defendant did not commit the crime. He is given no evidence whatever for this proposition and is simply instructed to add it to his doxastic repertoire. How realistic is it to suppose that jurors—or anyone else, for that matter—can appropriate their beliefs in this fashion? Can we will ourselves to believe X, absence any evidence or arguments for the truth of X, simply because a judge has said that such is our duty? I doubt it. It would be one thing, and a familiar one, if the judge were to say, “For the sake of argument, let us suppose that Jones is innocent_m of the charges against him.” All of us know how this conceptual ritual works and we have had plenty of practice in suspending disbelief for the sake of seeing where a certain line of inference might lead. But suspension of disbelief is less than what the PI enjoins. It demands not merely that we approach the case as if the defendant were innocent_m, but that we accept the view that he is innocent_m until such time, if ever, as the prosecutor persuades us otherwise beyond a reasonable doubt. I dispute that any rational person can will himself to believe an empirical proposition of that sort simply because a judge tells him that such is the law.

The salience of the conundrum of believing in the innocence_m of the accused shows up more vividly if we consider a common-enough sort of situation, one in which jurors have acquitted a defendant but claim that they believe him to have committed the crime. Obviously, this is a perfectly coherent position to take: the prosecutor persuaded them that the defendant actually committed the crime but the proof did not rise to the level of proof beyond a reasonable doubt. Can this be squared with the presumption of innocence_m? Jury instructions invariably insist that jurors must presume the defendant to be innocent (and remember that innocence_m is what is at stake here) until and unless they are persuaded beyond a reasonable doubt of the defendant's guilt_m. Here is how the model federal jury instruction puts the point:

This instruction clearly precludes the possibility that a juror might say to herself: “I believe Jones committed the crime (and is thus guilty_m but I am not persuaded of that beyond a reasonable doubt.” Clearly, there is something fishy going on when the rules of the system in effect preclude this type of situation. What the courts want is that jurors should not vote to convict Jones unless his guilt has been shown to the relevant standard. Fair enough. But when they attempt to produce this outcome via an instruction to the effect that jurors should believe/suppose Jones innocent_m until and unless they are convinced that his guilt has been shown beyond a reasonable doubt, the courts are confusing the logic of guilt_m with that of guilt_p. The point is subtle but important. In the case of any verdict to convict, it is plausible to suppose that the jurors reach a point in their deliberations where they have decided that Jones was not innocent_m but where they have not yet decided that their confidence in his guilt reached the point of proof beyond a reasonable doubt. That is, they believed that Jones committed the crime but they are not yet sufficiently sure to convict him. If that is so, why on earth should jurors be instructed to believe that Jones is innocent_m “until such time, if ever, as you as a jury are satisfied that the government has proven him guilty beyond a reasonable doubt”? This instruction simply does not reckon with the fact that the path from believing Jones is innocent_mto believing Jones is guilty_pinevitably passes through the stage of believing Jones is guilty_mbut not guilty_p.

On this material innocence model, it is as if at time t, a juror must believe that Jones did not commit the crime and at time t+1—the next moment—he not only believes that Jones committed the crime but is firmly convinced of it beyond a reasonable doubt. I doubt very much that rational agents reason thus. If, as seems reasonable to suppose, a typical juror gradually moves from believing that Jones did not commit the crime (innocent_m to believing that he probably did commit the crime (guilt_m, to, finally, being morally certain that he committed the crime (guilt_p, then there must be a time when jurors have abandoned their belief in Jones's innocence_m but have not yet embraced the belief that the likelihood of Jones's guilt_m is strong enough to convict him. This jury instruction excludes such a possibility, insisting that Jones must be presumed innocent_m until such time, if ever, as jurors are persuaded of his guilt_m beyond a reasonable doubt.

Moreover, and picking another irritating nit, if jurors are really supposed to believe Jones innocent (either materially or probatorily) until the moment when the jury has collectively reached a verdict, no juror could ever vote to convict in a straw poll, for that would violate the instruction to “presume Jones innocent until the jury is satisfied that he is guilty beyond a reasonable doubt.”⁴⁰

Leaving the instructions aside, a third problem with the presumption of innocence_m is the unclarity about the content of the belief meant to be conjured up in the juror's mind by the PI. Imagine that Jones is on trial for aggravated assault with a deadly weapon against Smith. If a juror accepts the judge's admonition to presume Jones's innocence_m, what specifically should she believe? Is it enough to believe that Jones assaulted Smith but without a deadly weapon? Is it sufficient to believe that Jones probably shot at Smith but did so unwittingly? Would it do for the juror to believe that Jones shot at Smith and intended to harm him but did so in self-defense? Any one of these beliefs, if true, would be sufficient to win a not-guilty verdict for Jones (and, indeed, enough to imply his innocence_m).

Or does the PI require the juror to believe that {Jones never assaulted Smith at all & that he never aimed and fired a gun at him & that he did not intend to harm him & that he was acting in self defense & with diminished mental capacity & without negligence}? That is, must the juror begin the trial disbelieving every element of the case that the prosecution must prove or is it sufficient that she disbelieve at least one? The conventional answer, of course, is the former. But if we take that seriously, then the PI actually requires the juror to believe much more than a literal construal of “presuming Jones to be innocent_m” would demand. After all, Jones is innocent_m so long as any key part of the charge against him is false. Worse, it is impossible to believe all of these things at once (shooting at Smith in self-defense obviously rules out not shooting at him at all).

Interestingly, model jury instructions address none of these concerns. They do not tell the juror how strongly he should believe in the defendant's innocence_m, or how the juror is to adopt a belief for which he has no evidence whatever, or how many of the elements of the indictment he must disbelieve.

As intriguing as these puzzles may be, they do not yet bring us to the root problem with PI, which goes deeper than worries about accepting beliefs on no evidence, or arbitrarily assigning probabilistic degrees of belief to assertions of innocence_m, or parsing innocence across the elements of the crime. I think the core mistake in the promiscuous uses of “innocence” that we see alike in jury instructions and in academic treatments of the PI is the supposition that the PI per se has anything to do with hypotheses about the innocence_m of the defendant. Properly understood, I argue below, the presumption of innocence requires of jurors not a belief about innocence_m at all but rather a belief about the innocence_p of criminal defendants at the outset of a trial. Like the judge in the Oklahoma case mentioned above, I maintain that the PI is a presumption of innocence_p (that is, guilt not yet established), not one of innocence_m.

This is, perhaps, a controversial thesis, so we must approach it cautiously. To begin with, let us remind ourselves what an acquittal means. It is not an assertion of innocence_m; to the contrary, it is fully compatible with guilt_m. As appellate courts have repeatedly emphasized, an acquittal is neither less nor more than an assertion by the trier of fact that the case offered by the prosecution failed to satisfy the relevant standard of proof.⁴¹

If we want a fair trial for a defendant, we expect the trier of fact to begin the trial by supposing that, short of hearing powerful proofs from the prosecutor, he will end up acquitting the defendant. In short, we want him to assume at the outset that he has no conclusive proof of guilt. But it is not enough to assume that there is no conclusive proof at the outset; what must be assumed is that at that moment there is no proof at all. The appearance of the accused in the dock, the juror must suppose, provides no evidence in and of itself pertinent to questions of guilt or innocence. Nor do whatever snippets of information about the case that the juror may have come by through other sources provide evidence of guilt or innocence. If the prosecutor can shift this belief that the defendant's guilt is unproven by presenting compelling inculpatory evidence (not discredited by the defense), then the defendant should be convicted. If he cannot, the juror ends the trial where she began—believing that the defendant's guilt is unproven—although at the end, unlike at the beginning, she need not suppose that there is no evidence of guilt at all. That is, a fair and conscientious juror should begin the trial not necessarily with a presumption of innocence_m but with one of full innocence_p (no inculpatory evidence). To argue that in addition to or instead of supposing innocence_p, the unbiased juror in a fair trial must likewise suppose the innocence_m of the defendant is neither necessary nor appropriate.

It is not necessary because innocence_m is not presupposed by even the most defendant-favorable outcome of a trial: an acquittal. It is pointless and gratuitous to insist that the juror at the beginning of a trial must adopt a belief about innocence_m much stronger than what he will later assert if he votes to acquit. There is no conceivable reason for saying that the defendant should on the opening day of her trial be considered innocent_m in a more robust sense of that term than she is when his trial ends in acquittal. Yet that is precisely what the presumption of innocence_m requires. By contrast, the presumption of innocence_p makes no stronger an assumption about the defendant's innocence at the opening of a trial than an acquittal implies at its end.

The assumption of innocence_m is no more appropriate than it is necessary. After all, a juror at the outset has no basis for either affirming or denying the innocence_m of the defendant. He does not know the defendant, he has no idea what sorts of exculpatory and inculpatory evidence will arise in the course of the trial, and so on. Worse, the assumption of innocence_m requires the juror to deny what he already knows: to wit, that persons do not get to the trial stage in a criminal proceeding if the probability of their guilt_m is vanishingly small. Insisting that the juror must go into a trial utterly convinced that the defendant is innocent_m belies the elaborate, winnowing structure of the legal system; it acts as if the selection of whom to put on trial were conducted by some sort of randomizing mechanism; and it insults the intelligence of jurors. By contrast, instructing the juror to assume that she has seen as yet no evidence of the defendant's guilt correctly mirrors the juror's own epistemic situation and thus is an injunction that she can act on in good conscience.

We can note straightaway that this way of construing the PI avoids the several awkward problems, already mentioned, associated with a presumption of innocence_m. In the latter case, for instance, someone needs to decide what probability or degree of confidence the juror should initially assign to the hypothesis that the defendant is innocent_m. There is no nonarbitrary solution to that problem. Do we face the same difficulty with a presumption of innocence_p? Some scholars, most notably Judge Richard Posner, think that we do. Specifically, and in the context of a discussion of possible juror bias, he has argued that: “Ideally we want the trier of fact to work from prior odds of 1 to 1 that the plaintiff or prosecutor has a meritorious case.”⁴²

I would suggest that the ideal juror is one who has no beliefs at all about which side is likely to prevail (which is not the same as believing, with Posner, that the odds are even). Indeed, the assignment of equal odds appears (at least to me) to stack things in the prosecutor's favor. Here is why: We all suppose that the criminal standard of proof is very high. Just for the sake of argument, let us suppose that it weighs in at about 90 percent. To win a case, the prosecutor must persuade the jury that the apparent guilt of the defendant is greater than this value; anything less demands an acquittal. Now if, following Posner, a juror supposes that the prosecutor is as likely to win the case as he is to lose it, that must mean that the juror believes that the apparent guilt of the accused is as likely to be above 90 percent as it is likely to be less than 90 percent. This, in turn, implies—unless we have a fiendishly bizarre distribution of degrees of apparent guilt among a large class of defendants—that most defendants are guilty_m (even if only half of them are assumed to be guilty_p). That may be true as a statistical generalization about the cases that actually go to trial, but surely we do not want jurors going into a trial believing that most defendants are guilty_m, especially not when (as now) the standard of proof is defined in terms of subjective conviction about guilt. In sum, Posner's stipulation of the ideal juror as assigning equal initial odds to guilt_p and innocence_p is not only unmotivated but is a clear violation of the spirit and probably the letter of the PI.

It would be far better to say that the fair-minded juror is one who is open-minded about the eventual outcome of the trial and feels under no obligation nor even in a position to assign odds—even or otherwise—to the trial's outcome. (We must disabuse ourselves of the Bayesian-inspired notion that one does or one should assign probabilities here and now to a vast collection of future states of affairs. Just as I have no opinion whatever about whether it is likely to rain next September 23 at noon in Kuala Lumpur, so a reasonable juror can—and should—be unwilling to assign odds at the opening of a trial to its possible outcomes.) What is important is that the juror concedes that she has no proof now about guilt and that therefore she lacks any clue about which side will eventually prevail. This is a patently true description of her situation, and any non-comatose juror should accede to its truth; she can simply take it—as indeed she should—as an obviously correct characterization of the epistemic situation in which she finds himself. Nor, if we take this route, need we ask the juror to believe something for which she has no evidence (such as, for instance, that the defendant is innocent_m. What the PI, rightly construed, requires is that the juror believes something that she should have no trouble believing, since it correctly reflects her situation: that she has yet to see any proof of the guilt of the defendant—nor, come to that, has she seen any exculpatory evidence.

To be clear, I am not claiming that we can be utterly indifferent to the views that (actual or prospective) jurors may harbor concerning the innocence_m of the defendant. High confidence prior to trial that the defendant is guilty_m is clearly a sign of bias and should be grounds for juror exclusion for cause. But, at the risk of propounding heresy, I submit that high confidence in the innocence_m of the defendant–of precisely the sort enjoined by the standard reading of the PI—is just as troublesome a source of bias. Anyone who goes into a trial believing that she already has enough information about a case to have a settled and firm opinion about guilt or innocence is suspect as a trier of fact. This is not to gainsay that rational and suitable jurors may have, on the strength of pretrial publicity and the like, tentative hunches about the innocence_m or guilt_m of the defendant. So long as such persons will grant that they have no evidence of guilt and that the verdict must depend only on the proof offered at trial, they are qualified as jurors. The crucial point is that the juror must be willing to accept without reservation the thesis of innocence_p, regardless what tentative hypotheses she may or may not have entertained about guilt_m and innocence_m.

THE CORE FAILING: CONFUSING JUROR CONFIDENCE WITH A STANDARD OF PROOF

I have been railing in this essay against the view that jurors must begin a trial with a strong belief in the innocence_m of the accused. I suggest rather that what they must believe is that the defendant's guilt has not yet been proved and that his appearance as a defendant constitutes no evidence of his guilt. Since jurors are never called upon to affirm the defendant's innocence_m, I argue that presuming such innocence is at best gratuitous. If in retrospect this seems fairly obvious, we need to try to figure out why it has seemed so natural for so long to think that a presumption of innocence_m is essential to a fair trial.

I conjecture that the principal factor operating here has been the ubiquity, especially in Anglo-Saxon criminal law (but not without its counterpart in Roman-law countries, as we will see below), of a certain model of juror/judge deliberation and belief change that I call the theory of the subjective standard of proof.

The received picture, simultaneously normative and descriptive, of jury and juror decision-making goes something like this: The trier of fact begins (and should begin) a trial with a certain assumption about the high probability of the defendant's innocence_m (and accordingly assigns a very low probability to his guilt_m). As the trial proceeds, each juror is thought to be constantly revising her estimate of apparent innocence_m in light of evidence presented by both defense and prosecution. As final deliberation begins, the only question to be faced by the juror is whether her by-now multiply revised estimate of the probability of guilt_m reaches or exceeds the standard of proof. If she is firmly convinced about guilt_m, if she harbors no reasonable doubts about guilt_m, if her belief in guilt_m is the sort of belief that she would base important life decisions upon, then—she will have been instructed by the judge—a conviction is called for. Otherwise the defendant should be acquitted.

The instant familiarity of this picture makes it difficult to take a couple of steps back and see just how strange this story is. And nothing about it is more peculiar than the fact that it offers no free-standing standard of proof. Instead the system tells the jury: At the end of the trial, think carefully about the evidence; note what level of confidence you have reached about the defendant's guilt_m; if you are really, really confident that he committed the crime, then you should convict; if you are less than deeply persuaded, acquit.

Not to mince words, this is a travesty of a system of proof. A standard of proof—in every area in which proof is called for outside the law (including natural science, clinical trials in medicine, mathematics, epidemiological studies, and so on)—is meant to tell the investigator or inquirer when she is entitled to regard something as proved, that is, when the relation of the evidence or the premises to the sought-after conclusion warrants the acceptance of the conclusion as proven for the purposes at hand. In the criminal law, by contrast, that issue is either wholly ignored or shamelessly finessed. Instead of specifying that the juror's level of confidence in guilt_m should depend on whether a robust proof has been offered, the criminal law makes the standard of proof parasitic on the inquirer's (that is, the juror's) level of confidence in defendant's guilt_m. We have a proof, says the law, so long as the jurors are strongly persuaded of the guilt_m of the accused. Never mind how they arrived at their high confidence, we have a proof. This gets things precisely backwards.

In case it is not already clear why this perverts the logical order of things, imagine that we were henceforth to say to mathematicians that they have a legitimate proof of a theorem so long as they are firmly convinced the theorem is true. Or suppose we told epidemiologists that they have a proof of the existence of a causal link between A and B provided that they are highly confident that such a link exists. Such a proof policy would be a laughing stock. One ought not say to any trier of fact: “You have a proof of X provided you are firmly convinced of X.” To the contrary, we say: “You have no entitlement to be strongly convinced of X unless and until you have a proof of X,” adding for good measure that “your firm convictions about X count for nothing absent an acceptable proof of X.” And we then proceed to tell them what a proof of X would look like; that is what it is to have a standard of proof. The point of having a standard of proof is precisely to tell one whether one's subjective confidence is justified.

In marked contrast, proof practices in the criminal law make the very existence of a proof of guilt parasitic on the prior existence of a firm conviction about guilt_m. The language used to capture this subjective conviction varies from one era to another and from one jurisdiction to another. Sometimes it is an “abiding belief” in guilt_m that is called for. At other times it is a “moral certainty” about guilt_m, or a “settled belief” about guilt_m, or a “satisfied conscience” or a “firm conviction.” In Roman-law courts, the corresponding word games involve things like “an intimate conviction” of the guilty of the accused or “a sense of certainty, freely arrived at.”

What these and other familiar variants share in common is the use of the existence of high personal confidence in the prosecution's story as the touchstone for deciding whether guilt has been proven.⁴³

“Well,” one might reply, “doesn't the rule of proof beyond a reasonable doubt specify a genuine standard of proof?” The answer, as a little reflection will show, is a resounding no. As I have argued in detail elsewhere,⁴⁵

I return to explore this point below, but a brief digression is in order, because I begin this section by saying that it is targeted at explaining how the presumption of innocence_m took root. If we ponder the character of what I call the received view, it is easy to see why someone who conceives of a trial and the decision to convict or acquit along these lines might think it important that jurors should begin a trial believing in the innocence_m of the accused. After all, if the end point of a trial, when it leads to a conviction, is a deeply held belief that the defendant committed the crime (and that is the current “standard,” insofar as we can be said to have one), then the most powerful protection that can be afforded to the defendant is an initial belief in innocence_m. To secure a conviction, the prosecutor must convert this initial belief in innocence_m into a firm belief in guilt_m. The current system requires a presumption of innocence_m because it will ultimately hang the decision to convict not on whether a powerful proof of guilt has been offered but on whether the jurors have sufficiently shifted their subjective estimates of guilt_m. The higher the initial estimates of the probability of innocence_m, the harder it will presumably be to shift them to a firm conviction of guilt_m.

But suppose, just for the sake of argument, that the criminal law had a genuine standard of proof, one that did not depend on the jurors' subjective evaluation of guilt but on the establishment by the prosecutor of a powerful inferential link between the evidence presented and the guilt of the accused. In these circumstances, guilt would not depend on jurors' introspection of their confidence in guilt_m but on a determination by them of whether the standard had been satisfied. Here, obviously, a presumption of innocence_m would be irrelevant. What would be essential, in order to give the defendant every conceivable advantage, is a presumption of innocence_p.

It is fair to ask: What would it mean to have a legal standard of proof that was not parasitic on subjective juror assignments of high confidence or high probability to the guilt_m of the accused? For a start, let us remind ourselves of the standard of proof that dominated Romano-German law from the late Middle Ages through the Enlightenment. Put simply and ignoring the nuances, the formula for securing a conviction in criminal trials in those days was two-pronged: the production of two credible eyewitnesses to the crime or a plausible confession from the defendant. (I hasten to add that I am not endorsing this standard; its use here is simply illustrative.)

With this sort of standard in place, the subjective belief of the judge (obviously, juries were not used as the trier of fact) about the innocence_m or guilt_m of the accused was neither here nor there. The standard of proof specified what was needed to condemn the defendant. To determine whether the accused should be convicted, the judge could (and was supposed to) ignore his own private hunches about guilt_m or innocence_m and simply ascertain whether the benchmark had been met. Supposing that the judge was intellectually honest, supposing that he could apply the standard without prejudice, it did not matter what his subjective beliefs about guilt_m and innocence_m were. In a system like this, a presumption of innocence_m had no role and needed none. The judge did have to suppose innocence_p (that is, no proof of guilt) until such time as the standard of proof was satisfied, but his private beliefs about guilt_m and innocence_m were irrelevant to the decision he had to make.

While this was a sorry standard of proof (in the sense that two witnesses or a confession are not very compelling markers of guilt_m, nor is their absence a reliable sign of innocence_m), it must be said that this was, at least, a standard of proof in the strict sense of the term. The question we have to ask ourselves is whether we can fashion out of contemporary practices something like it.

The fact is that there are already several interesting possibilities on offer that might do the job. They simultaneously provide us a way of (1) breaking out of the bind generated by the dependence of the current system on purely subjective estimates of the strength of belief about guilt_m; and (2) making it crystal clear that what the trier of fact must assume is innocence_p rather than innocence_m. Consider, for instance, a standard of proof along the following lines:

Or, taking a cue from some of Ron Allen's interesting proposals:⁴⁶

For present purposes, I am endorsing neither of these proposed standards. I cite them because they both exhibit one key virtue lacking in existing practice in criminal trials: to wit, they tell the juror what features he should be looking for in the evidence to hand that would justify him in convicting the accused. They do not ask the juror to engage in a meta-analysis of his own degree of confidence in the guilt_m or innocence_m of the accused. (It may be, of course, that many cases that would satisfy standard A or B would also leave most jurors with a firm conviction about the guilt_m of the accused. One would even hope that this was the case. But it is not necessary.)

Nor would either standard represent a radical break with historical practices in common law. For a very long time, specifically until the 1954, when the Supreme Court eliminated it from federal trials, there was a standard jury instruction associated with cases involving circumstantial evidence. It followed the lines of the circumstantial evidence instruction still used in California state courts. It reads in part:

The specific language of this rule is both redundant (since the second sentence is entailed by the first) and infelicitous (since condition (2) should read something like “is inconsistent with the theory that the defendant is innocent of the crime”). But such untidiness can be cleaned up easily enough. What this rule is clearly driving at is the notion that (in cases involving circumstantial evidence) a conviction is in order only if the evidence rules out every reasonable hypothesis except that of the guilt_m of the defendant. This instruction has the virtue of going directly to the question of the nature of the evidence and avoids dalliance with characterizing how the jurors may feel about the matter. It tells jurors that they should not convict unless the evidence before them has refuted every version of the events in the case they can think of that would be exculpatory for the defendant. It says to the jurors:

We should note that standard C does not ask jurors to take a reading on their confidence about guilt_m. Rather, it asks them whether the credible evidence presented at trial excludes every hypothesis of innocence_m that they can imagine. Allen's proposed standard, B, and A, involving the element of surprising evidence, are clearly in the same family. Instead of continuing the use of the circumstantial evidence instruction, the Supreme Court incredibly decided that telling jurors that they could convict only if they were “firmly convinced” of the guilt_m of the accused was the right instruction and that the circumstantial-evidence standard was the wrong one!⁴⁸

Obviously none of the three standards under discussion depends upon the juror's subjective estimate of the likelihood of guilt_m of the accused. Indeed, all would allow for a situation in which a juror is obliged to convict even if she (perhaps perversely) still believes in the innocence_m of the accused, despite the fact that the standard has been satisfied. One might imagine that jurors would bridle at the suggestion that they could be obliged to convict someone whom they regard as innocent_m. But this dilemma is not different in kind from the current situation in which jurors often find themselves obliged to declare “not guilty” someone whom they believe to be guilty_m. If jurors can grasp (as they obviously do) that the demands of proof must trump their private beliefs where guilt_m is concerned, they can surely learn that their private beliefs about innocence_m may likewise have to give way before a plausible standard of proof.

It is also worthy of our attention that the adoption of any one of these candidate standards would bring a consideration of the defendant's innocence_m back into play explicitly in a trial. Each instructs the jury to consider the hypothesis that the defendant actually did not commit the crime and to see whether that hypothesis is genuinely excluded (or otherwise severely compromised) by credible evidence. If one or the other of these standards were in play, no longer could it be said that the notion of innocence_m is invoked in the PI and disappears from view thereafter.

Confronted with proposed standards of proof such as A, B, or C, one's first reaction is to say that things cannot possibly be as simple and formulaic as they suggest. We are dealing, after all, with highly complex human events and (in the case of jury deliberation) obscure thought processes. One's next, more constructive instinct is to conceive of cases where A through C would fail to convict the guilty_m or, worse, where they might condemn the innocent_m. Such cases are not hard to conceive. But that is beside the point. Outside of mathematics and formal logic, we are not dealing with certainty or deductive relations between premises and conclusions. Instead, we face uncertainty and fallible inferences. It is in the nature of ampliative rules of inference that they will sometimes lead to erroneous conclusions. No standard of proof can fully eliminate the uncertainty or render the kind of abductive inferences used in legal reasoning truth-preserving.

The germane question concerns the relative frequency of mistakes that A to C would lead to, compared to the present mechanism of making the outcome of a trial depend on unregimented and undisciplined juror instincts about guilt_m and innocence_m. If we suppose (as many legal scholars do) that the current de facto standard amounts to a level of confidence of somewhere between 90 percent and 95 percent confidence needed for a conviction, then we already know that there is likely to be a substantial number of erroneous verdicts. We know, too, that when judges are asked to rate the level of accuracy of jury verdicts, they generally report that juries acquit seemingly guilty_p defendants about 20 percent of the time.⁴⁹

Fortunately, we do not have to settle that fascinating question here. The object of describing these possible standards of proof is not to advocate their use in particular but to show two basic points about the logic of a trial, to wit, (1) that a standard of proof need not be cashed out in terms of subjective confidence of belief in guilt_m on the part of jurors; and, accordingly, (2) that a presumption of innocence_m need play no role in a trial involving such a standard. With a proper standard of proof, it is innocence_p, not innocence_m, that must be presumed if we aim to give the defendant the benefit of the presumption of innocence to which he is entitled.

In sum, the instincts of the much-maligned judge in Flores were sound: In a system with a genuine standard of proof, jurors should be required to presume “not guilty” (innocence_p rather than innocence_m. What he did not reckon with was that he was working in a system that had foregone the task of formulating a robust standard of proof, falling back instead on the “gut” assessment of guilt_m by jurors using their own internal compasses.

Before leaving this topic, there is at least one significant issue that needs to be cleared up. It concerns the question of precisely what a presumption of innocence_p would look like. In the Flores case, the judge instructed jurors that they should presume that the defendant was “not guilty” until proven otherwise. What drew the ire of appellate judges about this instruction, I surmise, was that it could be construed as enabling jurors to suppose that although the defendant's guilt was not yet proven, there was already plenty of inculpatory evidence available (say, in the form of an indictment).⁵⁰

THE PI AND OTHER KEY CONCEPTS AND PRECEPTS

Having settled that innocence_p is the proper focus of the PI, we can now pursue the query raised at the beginning about the relation of this presumption, so understood, to other key concepts of the criminal law, especially the notions of the prosecutorial burden of proof and the standard of proof.

The first point of interest to note is that there is a high level of redundancy between these three concepts. The standard of proof can be readily enough formulated as follows: The defendant must be acquitted unless the state establishes that the evidence of the defendant's guilt exceeds proof beyond a reasonable doubt (or whatever other standard may be in play). That is to say, the “standard” of proof specifies a level of proof, it gives a decision rule for conviction or acquittal, and (in the version I have just stated) it spells out that the burden of proof falls on the prosecutor. On this analysis, a separate burden of proof principle is unnecessary. Alternatively, one could formulate the standard of proof more narrowly along these lines: The defendant must be acquitted unless the evidence of the defendant's guilt exceeds a certain level. This formulation leaves it less clear where the burden of proof falls and would need supplementation by a separate rule to the effect that the state bears such a burden.

In my view, not much hangs on whether we go with the first formulation or the second. But supposing that we accept one or the other, what tasks remain to be discharged by the PI? The localizing of the burden of proof (one traditional function associated with the PI) is already accommodated. The important work remaining to be done by a presumption of innocence in these circumstances is that of making it entirely explicit that the defendant begins the trial without prejudice. That is to say, it is the “clean slate” element of the PI that is not already present, at least not explicitly, in the standard of proof and the burden of proof.

So conceived, the PI is an important prophylactic device, cautioning jurors to attach no probatory significance to the fact that the defendant sits (metaphorically) in the dock. The “innocence” that jurors need to assume is not a material innocence but an epistemological one: The trial begins with jurors knowing (and, ideally, presuming) nothing about whether the defendant committed the crime and, more important, presuming nothing about whether the evidence to be presented by the state will prove conclusive. A conscientious juror need not believe at the outset that the defendant is innocent_m nor need he believe it highly unlikely that the state will be able to prove its case. But neither should he believe that the defendant is guilty_m nor that the state will probably be able to make out its case with ease. It is enough if he believes that that case has yet to be made and that the defendant's status qua defendant contributes nothing to the case for guilt. Above all, though, he needs to understand that the standard for proof of guilt is very demanding and that nothing short of achieving that standard will suffice for conviction. (This moral is tied more closely with the instruction on the standard of proof than with the PI.)