During the recent presidential campaign, “crime in the streets” emerged as a crucial political issue. All three major candidates vowed to make the nation’s capital—where the crime rate is particularly high and painfully visible—a model of safety for our other urban centers. Now President Nixon faces the task of transforming slogans into programs.

In his first major policy statement on crime, President Nixon proposed a law “whereby dangerous hardcore recidivists could be held in temporary pre-trial detention when they have been charged with crime and when their continued pre-trial release presents a continued danger to the community.” By thus supporting “preventive detention”—albeit in cautiously vague terms—the President has provided substantial impetus to a movement which could have far reaching significance not only in the District of Columbia, but throughout the country.

The conditions giving rise to the call for preventive detention are not difficult to understand. A person suspected of committing a crime cannot stand trial on the day of his arrest; he must be given time to consult with his lawyer and prepare a defense. Although this should rarely take more than a few days, the delay between arrest and trial has been growing, until it is now almost as long as two years in some cities and a year in most other cities, including the District of Columbia. This is the consequence primarily of our unwillingness to pay for needed increases in judicial machinery.

At the same time there has been a growing sensitivity to the plight of the indigent accused, who are unable to raise even modest bail; this is reflected in a 1966 bail reform law which authorizes federal judges to release most defendants without requiring money bail. The net result of bail reform and increased delays in court has been that more criminal defendants spend more time out on the street awaiting their trials than ever before. This has led to an increase—or at least the appearance of an increase—in the number of crimes committed by some of these defendants between arrest and trial. And so, in an effort to stem this tide of increasing crime, many political leaders including Senators as diverse in their political views as Roman Hruska and Joseph Tydings, have focused their attention on the defendant awaiting trial. The slogan “crime in the streets” has found its first political victim.

The resulting proposals for preventive detention vary: some are limited to the District of Columbia, while others apply to all federal courts; some would seem to authorize the confinement of a very large number of defendants, while others are narrower in their scope; some include methods for shortening the time interval between arrest and trial, while others seem satisfied to leave things pretty much as they are now.

But they all have one point in common: they permit the imprisonment of a defendant who has not been convicted, and who is presumed innocent, of the crime with which he stands charged, on the basis of a prediction that he may commit a crime at some future time. These predictions would be made by judges on the basis of their appraisal of the suspect’s dangerousness, after study of his prior record and the crime for which he is being tried. The proponents of preventive detention hope thereby to identify and isolate those defendants awaiting trial who account for the apparently high incidence of serious crime. The opponents of preventive detention, a heterogeneous group which includes not only Senator Ervin of North Carolina but the American Civil Liberties Union, maintain that, under our system of criminal justice, which is characterized by the presumption of innocence,” conviction for a past crime is the only legitimate basis for confinement; they are fearful that acceptance of this “novel” approach to crime prevention might be an opening wedge leading to widespread confinement of persons suspected, on the basis of untested predictions, of dangerous propensities.

Before the claims for and against pre-trial preventive detention can be fairly evaluated, this misunderstood device must be placed in its historical and contemporary setting. Predicting who will commit crimes has long fascinated mankind. In the eighteenth century Cesare Lombroso, an Italian criminologist, thought that he could detect the criminal type by observing the configuration of bumps on the head; Sheldon and Eleanor Glueck, my colleagues at Harvard and pioneers in the prediction of juvenile delinquency, maintain that they can spot potential criminals at an early age by observing aspects of their family life; and some biologists now assert, on the basis of rather flimsy evidence, that they can identify potential criminals by examining the chromosomal structure of their cells.

One can sympathize with these efforts to predict and prevent crimes before they occur, rather than to wait until the victim lies dead. Indeed, Lewis Carroll put in the Queen’s mouth an argument for preventive detention that Alice found difficult to refute. The Queen says:


“There’s the King’s Messenger. He’s in prison now, being punished: and the trial doesn’t even begin till next Wednesday: and of course the crime comes last of all.”

“Suppose he never commits the crime?” asked Alice.

“That would be all the better, wouldn’t it?” the Queen responded….

Alice felt there was no denying that. “Of course it would be all the better,” she said: “But it wouldn’t be all the better his being punished.”

“You’re wrong….” said the Queen. “Were you ever punished?”

“Only for faults,” said Alice.

“And you were all the better for it, I know!” the Queen said triumphantly.

“Yes, but then I had done the things I was punished for,” said Alice: “That makes all the difference.”

“But if you hadn’t done them,” the Queen said, “that would have been better still; better, and better, and better!” Her voice went higher with each “better,” till it got quite to a squeak….

Alice thought, “There’s a mistake somewhere—“

And there is a mistake somewhere, but it is not where the opponents of preventive detention have sought to locate it.

The debate over pre-trial preventive detention has proceeded on the assumption that confining people on the basis of predictions of future crime is unprecedented in this country (and throughout the civilized world). Opponents of the proposal are fond of quoting Justice Jackson’s dictum in a case where the government sought to deny bail to a number of Smith Act defendants:

If I assume that defendants are disposed to commit every opportune disloyal act helpful to Communist countries, it is still difficult to reconcile with traditional American law the jailing of persons by the courts because of anticipated but as yet uncommitted crimes. Imprisonment to protect society from predicted but unconsummated offenses is so unprecedented in this country and so fraught with danger of excess and injustice that I am loath to resort to it….1

But Justice Jackson’s history is simply incorrect: “imprisonment to protect society from predicted but unconsummated offenses” is quite common in this and every other civilized country. A Justice of the Supreme Court of Burma came closer to the truth when he observed that “preventive justice which consists in restraining a man from committing a crime which he may commit but has not yet committed…is common to all systems of jurisprudence.”2

No system of jurisprudence has ever required that its law enforcers always sit back and wait until the spear has been thrown, or even until the gun has been loaded. Societies have differed in their techniques of crime prevention, but for centuries people throughout the world have been imprisoned “to protect society from predicted but unconsummated offenses.” That preventive justice was part of the English common-law tradition was a source of great pride to Blackstone: “And really it is an honor,” he said, “and almost a singular one, to our English laws, that they furnish a title of this sort, since preventive justice is upon every principle of reason, of humanity, and of sound policy, preferable in all respects to punishing justice, the execution of which…is always attended with many harsh and disagreeable circumstances.”3

The “preventive justice” to which Blackstone was specifically referring consisted of confining persons about whom there was “a probable suspicion that some crime is intended or likely to happen,” unless they could find “pledges or securities for keeping peace, or for their good behavior.” This “humanitarian” device led to the confinement of large numbers of “vagabonds” and “strangers.”

During both world wars, Great Britain promulgated regulations explicitly authorizing the preventive detention of certain persons suspected of “hostile origin or association.” During the Second World War the United States employed one of the grossest forms of preventive detention known to history: the mass transfer and internment of Americans of Japanese descent, allegedly on the basis of a prediction that otherwise some of them would sabotage our war effort on the West Coast and become victims of racial violence. The Supreme Court’s approval of that device4 laid the foundation for a statute, now on the books, which authorizes the detention, during a declared internal security emergency, of any person whom “there is reasonable ground to believe…probably will engage in or probably will conspire with others to engage in acts of espionage or of sabotage.”5 A lawsuit has just been filed challenging the constitutionality of this provision.

Nor is preventive detention limited to Blackstone’s England or to wartime. Indeed, every American jurisdiction permits pre-trial detention of at least one category of criminal suspects: those charged with capital offenses. And it should be recalled that at the time of the enactment of the Eighth Amendment—which has been construed to prohibit excessive bail in non-capital cases—many, if not most, felonies were capital; accordingly, many, if not most, accused felons were detained pending completion of their trials.


The most widespread form of preventive detention employed in the United States today is commitment of the mentally ill. More than half a million mentally ill persons who have not been convicted of crime are imprisoned in state hospitals without adequate treatment, and, often, with little hope of eventual release, on the basis of psychiatric predictions that unless confined they would do violence to themselves or to others.

Peace bonds—the Blackstonian technique of “preventive justice”—are still authorized in many parts of the United States. In a Pennsylvania case, a defendant was charged with assault and battery. He stood trial and was acquitted by a jury. Despite this, the judge required him to post a bond of $1,000 “to keep the peace for two years.” In its opinion vacating the bond as inconsistent with the right to trial by jury, the State Supreme Court cited data indicating that during ten previous years “478 men, after acquittal of criminal charges, were compelled to serve an aggregate of over 600 years in…prison in default of bonds aggregating $613,200.”6

Juvenile statutes authorize confinement of young persons who have not yet committed criminal acts, but who are thought likely to become criminals. Indeed, in his separate opinion in the Gault case, Mr. Justice Harlan cited figures indicating that between 26 and 48 percent of the 600,000 children brought before juvenile courts “are not in any sense guilty of criminal misconduct.”7 Some sex psychopath laws also authorize detention of persons who have never been convicted of crimes, but are thought likely to engage in sexual misconduct. Many states permit the incarceration of so-called “material witnesses”—that is, persons not themselves charged with crime, who may be important witnesses at another’s trial, and who are thought likely to flee the jurisdiction unless confined. An alleged witness to the assassination of Dr. King was recently imprisoned under such a statute.

Another widespread American practice is the so-called “preventive arrest,” recently described in a report to the Commissioners of the District of Columbia as follows:

…A person trying front doors of stores, or peering into parked cars, in the early hours of the morning; a person “known” to the police as a pickpocket loitering at a crowded bus stop; a “known” Murphy game operator talking to a soldier or a sailor—such persons may be arrested…largely in order to eliminate at least temporarily the occasion for any possible criminal activity. The principle upon which such arrests are made appears to be: if the individual is detained until 10 or 11 A.M. the following day, at least he will have committed no crime that night….”8

The Courts are beginning to place limits on police discretion in some such situations. For example, the arrest of a person found wandering around toilets on the ground that there is cause to suspect that he is about to perform an act of homosexuality might not be sustained as a justifiable exercise of police discretion. But many legislatures have made it unnecessary for the police to justify such arrests by reference to as yet uncommitted crimes. They make the “suspicious” act itself a crime justifying arrest. Thus a number of states now make it a crime “to loiter in or about public toilets” or “to wander about the streets at late or unusual hours without any visible or lawful business.”9

A recent judicial decision has recognized that the “basic design” of vagrancy statutes is one of “preventive conviction imposed upon those who, because of their background and behavior, are more likely than the general public to commit crimes, and that the statute contemplates such convictions even though no overt criminal act has been committed or can be proved.” The real issue in the enforcement of vagrancy statutes was viewed as “whether our system tolerates the concept of preventive conviction on suspicion.” The United States Court of Appeals for the District of Columbia recently struck down such statutes as unconstitutional, observing that “statistical likelihood” of a particular person’s or group’s engaging in criminality “is not permissible as an all out substitute for proof of individual guilt.”10

But “statistical likelihood”—gross and impersonal as that sounds—is all we ever have, whether we are predicting the future or reconstructing the past. When we establish rules for convicting the guilty, we do not require certainty; we only require that guilt be proved “beyond a reasonable doubt.” And that means that we are willing to tolerate the conviction of some innocent suspects in order to assure the confinement of a vastly larger number of guilty criminals. We insist that the statistical likelihood of guilt be very high: “better ten guilty men go free than one innocent man be wrongly condemned.” But we do not—nor could we—insist on certainty; to do so would result in immobility.

What difference is there between imprisoning a man for past crimes on the basis of “statistical likelihood” and detaining him to prevent future crimes on the same kind of less-than-certain information? The important difference may not be one of principle; it may be, as Justice Holmes said all legal issues are, one of degree. The available evidence suggests that our system of determining past guilt results in the erroneous conviction of relatively few innocent people.11 We really do seem to practice what we preach about preferring the acquittal of guilty men over the conviction of innocent men.

But the indications are that any system of predicting future crimes would result in a vastly larger number of erroneous confinements—that is, confinements of persons predicted to engage in violent crime who would not, in fact, do so. Indeed, all the experience with predicting violent conduct suggests that in order to spot a significant proportion of future violent criminals, we would have to reverse the traditional maxim of the criminal law and adopt a philosophy that it is “better to confine ten people who would not commit predicted crimes, than to release one who would.”

It should not be surprising to learn that predictions of the kind relied upon by the proponents of preventive detention are likely to be unreliable. Predictions of human conduct are difficult to make, for man is a complex entity and the world he inhabits is full of unexpected occurrences. Predictions of rare human events are even more difficult. And predictions of rare events occurring within a short span of time are the most difficult of all. Acts of violence by persons released while awaiting trial are relatively rare events (though more frequent among certain categories of suspects), and the relevant time span is short. Accordingly, the kind of predictions under consideration begin with heavy odds against their accuracy. A predictor is likely to be able to spot a large number of persons who would actually commit acts of violence only if he is also willing to imprison a very much larger number of defendants who would not, in fact, engage in violence if released.

This brings me to an obvious fact that is often overlooked in evaluating the accuracy of predictions. In order that the evaluation be fair, there must be information about both sides; we must not only know how many crimes committed by defendants out on bail were prevented; we must also know how many defendants were erroneously imprisoned. Either of these alone tells you very little. It is no trick at all to spot a very high percentage of defendants who would commit acts of violence while awaiting trial: you simply predict that all or almost all will do so. (Of course, the number of erroneous confinements would be extraordinarily high, but most or all of the crimes would have been prevented.) Conversely, it is easy to avoid erroneous confinements if that is your only aim: simply predict that few or none of the defendants will engage in violence pending trial. (In that case, you would prevent very few, if any, of the potential crimes, but the number of erroneous confinements would be minimal or non-existent.)

The difficult task is to select a category which includes the largest number of defendants who would commit violent crimes and the smallest possible number who would not. If it were possible to select a category which included all those, and only those, who would commit such crimes, there would be little problem. But since this is impossible, a choice must be made. It must be decided how many defendants we should be willing to confine erroneously in order to prevent how many acts of violence. This will in turn depend on the nature of the violence to be prevented and the duration of the contemplated confinement: we should be willing to tolerate fewer erroneous confinements to prevent predicted purse-snatching than predicted murder; and fewer again if the trial is a year off than if it can be completed within two weeks of the arrest.

Another reason why predictions of the future are less reliable than reconstructions of the past concerns the processes by which human beings make decisions. Participants in judicial decision-making—lawyers, judges, even jurors—have some sense of what it means to decide whether a specifically charged act probably was or probably was not committed. The participants bring to their decisions some basis for sorting out the relevant from the irrelevant, the believable from the incredible, the significant from the trivial. And this basis—though often rough and intuitive—is far more than the judge is likely to bring to the process of predicting the future.

It is true that all judgments about human events, whether past or future, rest upon a superstructure of assumptions about how people behave; all decision-making requires a theory. What I am suggesting is that participants in the judicial process are better equipped by their experience to construct and employ theories about what probably occurred in the past than theories about what is likely to occur in the future. Put another way, we are all historians, but few of us are scientists. Perhaps Lewis Carroll’s Queen had a “memory” that worked equally well both ways: she remembered “things that happened the week after next” even better than things that happened yesterday. But Alice spoke for most of us when she said that her memory “only works one way…I can’t remember things before they happen.”

The most serious danger inherent in any system of preventive detention is that it always seems to be working well, even when it is performing dismally; this is so because it is the nature of any system of preventive detention to display its meager successes in preventing crime while it hides its frequent errors. This has been demonstrated in other areas where detention rests on predictions of dangerousness. One such area—which I have studied in detail and about which I have written elsewhere12—is the confinement of the mentally ill on the basis of psychiatric predictions of injurious conduct. It has long been assumed that these psychiatric predictions are reasonably accurate; that patients who are diagnosed as dangerous would have engaged in seriously harmful conduct had they not been confined. The accuracy of these predictions has never been systematically tested, since patients predicted to be dangerous are confined and thus do not have the opportunity to demonstrate that they would not have committed the predicted act if they were at liberty.

Accordingly, the psychiatrist almost never learns about his erroneous predictions of violence. But he almost always learns about his erroneous predictions of non-violence—often from newspaper headlines announcing the crime. The fact that the errors of underestimating the possibilities of violence are more visible than errors of overestimating inclines the psychiatrist—whether consciously or unconsciously—to err on the side of confining rather than of releasing. His modus operandi becomes: When in doubt, don’t let him out.

Recently the accuracy of psychiatric predictions has been called into considerable question. A decision of the United States Supreme Court in 1966, Baxtrom v. Herald,13 resulted in freeing many mentally ill persons predicted to be dangerous. Grave fear was expressed for the safety of the community. But follow-up studies now indicate that the predictions of violence were grossly exaggerated, and that very few of the patients have done what the psychiatrists predicted they would do if released. Similar studies in Baltimore support this conclusion of extreme overprediction.14

The same phenomenon is likely to plague efforts to predict violence pending trial if a preventive detention statute is now enacted. Judges, like psychiatrists, will rarely learn about their erroneous predictions of violence; for these defendants, being confined, will not have an opportunity to demonstrate that they would not have committed the predicted crime. But every time a judge makes an erroneous prediction of non-violence—every time he decides to release someone who then does commit a violent act—he learns about his “mistake” swiftly and dramatically.

Thus, if a statute is enacted authorizing pre-trial preventive detention on the basis of judicial predictions of violence, we will never know how many defendants are being erroneously confined. And as more and more information is accumulated, most of it concerning defendants who were erroneously released, judges will keep expanding the category of defendants to be detained. There is evidence that this is already being done by some judges today. During its recent hearings, the Senate Subcommittee on Constitutional Rights considered a study of the decisions of two trial judges in the District of Columbia on pre-trial release and detention; one judge routinely, if unlawfully, detained all suspects whom he regarded as “bad risks”; the other judge routinely released most suspects. Over the period studied, the “tough” judge detained about half the defendants who came before him (144 out of 285); while the “lenient” judge detained only about one fifth (46 out of 226). Of those released by the tough judge, twelve were charged with offenses—either felonies or misdemeanors—while awaiting trial; while of those released by the lenient judge, sixteen were accused of such offenses.

In other words, in order to prevent about four more crimes (some of them misdemeanors), the tough judge had to confine almost a hundred more defendants. Moreover, of the 144 persons detained by the tough judge, thirty-six subsequently had their cases dismissed and another “large percentage of them” were acquitted. Most of the defendants in this latter group were therefore the victims of a compounded legal error: not only did they not commit the predicted crimes; they were not even guilty—or so the process determined—of the past crime with which they stood charged.

Now it may be that eventually criteria for confinement can be refined to the point where such errors are minimized. Perhaps the high rate of violent crime by certain categories of released defendants will permit a high degree of crime prevention without too many erroneous confinements. It is claimed, for example, that a very high percentage of defendants charged with armed robbery in the District of Columbia—some place the figure as high as 34 percent, others as low as 11 percent—commit new felonies while awaiting trial. But if a statute were to be enacted now authorizing the confinement of all persons awaiting trial who, on the basis of specified criteria, were predicted to commit violent crimes, then the development of such refined criteria would be seriously retarded.

It must never be forgotten that many years of experience administering an untested system will not always increase the accuracy of that system. Many years of experience are often only one year of experience repeated many times. The unknown mistake of the past becomes the foundation for a confident, but erroneous, prediction of the future. This was demonstrated many years ago, in a famous “experiment” conducted by the Harvard psychologist Thorndike, who had a student throw darts repeatedly at a board to test the thesis that aim improves with experience, but blind-folded the student and never told him when he hit or missed the target. Needless to say, his aim did not improve with “experience.” Nor would the accuracy of judicial predictions necessarily improve simply as a result of judges spending more and more years meting out preventive detention without any accurate way to test their predictions.

The time is not yet ripe for resolving definitively by legislation the dilemma of pre-trial preventive detention. We have just begun to understand what the problem is, but we do not yet have enough information to know what the optimal solutions are. We have not even tried other—less drastic—amelioratives, such as speedier trials, more supervision for released defendants, and perhaps even increased penalties for crimes committed while out on bail. If such solutions were tried, the problem of crimes committed by released defendants might become a very small one indeed. What must be avoided is a simple solution that freezes knowledge at its existing low state. And this is precisely what would occur if Congress now enacted a statute authorizing confinement of all defendants predicted to commit acts of violence while awaiting trial. What must be encouraged is an approach that is tentative; that continues to gather information; and that is in the nature of an experiment.

What I suggest is that any proposed criteria for confinement be tested to determine how accurate—or inaccurate—they are in predicting violence. This could be done in a number of ways. Judges might be asked to apply the criteria being tested—for example, those in the statute proposed by Senator Tydings15—and to predict on the basis of those criteria which defendants awaiting trial would engage in violent crimes. All of the defendants would then be released, even those who the judges think should be detained. Careful studies should then be conducted to determine how accurate the judges were in their predictions. There is, however, a serious problem with this kind of test. If the judges know that everyone will be released and their actions observed, they might be extremely cautious about predicting violence, more cautious than if they knew that the predicted criminals would be safely confined.

This suggests a variation which would increase the accuracy of the test, but at a substantial cost in human liberty. The judges would again decide who should be detained on the basis of the criteria being tested, but, this time, only some—say half—of those selected for detention would be randomly released and observed. The other half would be detained. This suggestion is not free of difficulties either, for the status of experimentation under the law is far from clear. But I am confident that a workable and constitutional approach can be devised.16 Then we can see how many of these defendants would, in fact, fulfill the predictions.

What we learn about our ability to predict may be discouraging to those who advocate preventive detention. But it is far better to know the discouraging truth than to build a house—especially one with bars—on untested assumptions.

This Issue

March 13, 1969