Charles Murray is a senior research fellow at the Manhattan Institute for Policy Research and the author of several books including Losing Ground: American Social Policy 1950-1980 (© 1984 by Charles Murray) from which this essay is adapted and reprinted with permission of Basic Books, Inc,
There are laws that explain why social programs not only do not but cannot produce the intended effects. Let me pose a problem in the form that Einstein used to call a “thought experiment.” Whereas Einstein used the device to imagine such things as the view from the head of a column of light, we will use it for the more pedestrian purpose of imagining the view from the office of a middle-echelon bureaucrat. Our task: tO think through how to structure a specific government social-action program so that it might reasonably be expected to accomplish net good. The experiment calls for us to put ourselves in the role of a government planner who must implement a new piece of legislation, The Comprehensive Anti-Smoking Act. The Act has several provisions common to the genre. It establishes a federal agency to coordinate the federal government’s activities related to the goal of less smoking. A large anti-smoking advertising campaign is planned. Federal matching funds are provided for school systems that teach courses on the perils of smoking. In addition to these initiatives, the legislation provides for direct, concrete incentives for people to quit smoking. A billion dollars will be appropriated annually for the indefinite future, to be used for cash rewards to persons who quit. We are in charge of designing this effort, with complete freedom to specify whatever rules we wish, provided they are consistent with constitutional rights. After five years an evaluation will be conducted to determine whether the number of cigarettes consumed and the number of smokers have been reduced by the program. The challenge in this experiment is to use the $1 billion in a way that (in our own best estimate) will meet this test. My proposition is that we cannot do so: that any program we design will either (1) have no effect on smoking or (2) actually increase smoking. I maintain that we are helpless to use the billion dollars to achieve our goal. The heart of the problem is designing a reward that will induce smokers to quit—and will not induce others to begin smoking, continue smoking, or increase their smoking to become eligible to receive the reward. Let us work through one scenario to illustrate the nature of the conundrum. Designing the Program Three sets of choices will decisively affect the success or failure of the program: choices about
There are laws that explain why social programs not only do not but cannot produce the intended effects.
Let me pose a problem in the form that Einstein used to call a “thought experiment.” Whereas Einstein used the device to imagine such things as the view from the head of a column of light, we will use it for the more pedestrian purpose of imagining the view from the office of a middle-echelon bureaucrat. Our task: tO think through how to structure a specific government social-action program so that it might reasonably be expected to accomplish net good.
The experiment calls for us to put ourselves in the role of a government planner who must implement a new piece of legislation, The Comprehensive Anti-Smoking Act. The Act has several provisions common to the genre. It establishes a federal agency to coordinate the federal government’s activities related to the goal of less smoking. A large anti-smoking advertising campaign is planned. Federal matching funds are provided for school systems that teach courses on the perils of smoking.
In addition to these initiatives, the legislation provides for direct, concrete incentives for people to quit smoking. A billion dollars will be appropriated annually for the indefinite future, to be used for cash rewards to persons who quit. We are in charge of designing this effort, with complete freedom to specify whatever rules we wish, provided they are consistent with constitutional rights. After five years an evaluation will be conducted to determine whether the number of cigarettes consumed and the number of smokers have been reduced by the program.
The challenge in this experiment is to use the $1 billion in a way that (in our own best estimate) will meet this test. My proposition is that we cannot do so: that any program we design will either (1) have no effect on smoking or (2) actually increase smoking. I maintain that we are helpless to use the billion dollars to achieve our goal.
The heart of the problem is designing a reward that will induce smokers to quit—and will not induce others to begin smoking, continue smoking, or increase their smoking to become eligible to receive the reward. Let us work through one scenario to illustrate the nature of the conundrum.
Designing the Program
Three sets of choices will decisively affect the success or failure of the program: choices about
• the size of the reward,
• conditions for receiving the reward, and
• eligibility to participate in the program.
What is a first approximation of a program that has a good chance of working?
Choosing the size of the reward. We know from the outset that the reward cannot be small. No one will quit smoking for pocket change, other than those who were going to quit anyway. On the other hand, the theoretical power of a cash reward is plausible—almost anyone would become and remain a nonsmoker in return for a million dollars. We settle on the sum of $10,000 as a reward that is an extremely powerful inducement to large numbers of persons.
Conditions for receiving the reward. We seek a middle ground between conditions that maximize the likelihood that a person has permanently quit smoking and conditions that make the reward so difficult to win that few will bother. Thus, for example, we reject plans that would spread the reward over several years. Eventually we decide to require that a person must remain smoke-free for one year. We make the award a one- time prize, so that people have no incentive to recommence smoking to qualify for another $10,000. A repayment scheme is added: People who begin smoking again will have to give up their award.
Eligibility to participate. The intent of the program is to appeal to the heavy smoker whose health is most at risk. On the other hand, it would defeat our purpose to limit eligibility too severely—to persons, for example, who have smoked three packs a day for twenty years—because in so doing we would disqualify many people in the vulnerable group of moderate smokers who are likely to become heavy lifelong smokers unless something is done. The compromise solution we reach is to require that a person have smoked at least one pack a day for five years.
Now let us consider the results.
After one year: We think ahead a year, and are pleased. The $10,000 reward has substantial effects on the people who are eligible for the program on day one—that is, persons who have smoked at least a pack a day for five years at the time the experiment begins. The effect is not unfailing; not everyone quits smoking to get the reward; and we must assume that not everyone who stops for a year is able to avoid a relapse. Some cheating occurs despite our precautions. But some people quit smoking permanently as a direct result of the program.
We recognize, of course, that we achieve the effect inefficiently. Thousands of persons in the target population quit smoking every year even in the absence of a monetary reward. Under the program, they collect money for doing what they would have done anyway. But the problem posed in our thought experiment says nothing about being efficient; the problem is only to create a program that reduces net smoking.
After two years: We think ahead two years, and are disturbed. For now comes time to examine the effects of the program on people who have been smoking a Pack a day but for a period of less than five years when the program begins.
We find that for all persons who have been smoking less than the required period of time, the program provides a payment to continue. For the person who has been smoking for exactly four years, the payment is $10,000 in return for smoking for one more year. Given that the smoking habit has its own attractions, the payment is exceedingly effective. In fact, we notice an unfortunate imbalance: For the person who has already smoked for five years (our target population), the inducement of $10,000 to quit must fight against the attractions of smoking and is not always adequate to achieve the desired result. For the smoker who has not reached this limit, the inducement to continue smoking is reinforced by those very attractions. Thus the effective power of $10,000 to induce continued smoking for one year in the one population is much greater than its power to induce cessation of smoking for one year in the other.
To this point, we have been concerned only with those who were already smoking at the pack-a-day level. Now we consider the effects of the program on smokers who had been smoking less than that amount. We find that a significant number of smokers increase their consumption to a pack a day, for the same reason. (Everyone who smokes nineteen cigarettes a day increases to twenty, almost everyone who smokes eighteen cigarettes a day increases to twenty, and so on.) This effect is strongest among those persons who think they “should” quit but who doubt their ability to quit without help. For them—through a process of plausible but destructive logic—it seems that the best way to do what they think they want to do (to quit smoking) is to smoke more.
Among those who are nonsmokers, the effects are entirely negative. A considerable number of teenagers who were wavering between starting or not starting to smoke decide in favor of smoking—they can enjoy smoking now, and then give it up when they qualify for the reward.
After five years: When we think ahead five years, we note a final logical by-product of the program. Quitting the habit after five years of smoking a pack a day is generally more difficult than quitting sooner and after lesser levels of smoking. Many people who try to stop when the fifth year is ended find that the $10,000 is no longer a sufficient inducement, though it may have seemed to them a few years earlier that it would be. The rules of the program have made heavy smokers out of people who would have remained light smokers and thereby have induced a certain number of people not only to smoke more and longer until they became eligible for the $10,000 but to become impervious to the effects of the reward once they do become eligible.
What is the net outcome? If 90 percent of the population had been smoking for five years when the program began, we might still argue that the program would show a net reduction in smoking. But only about 15 percent of the adult population smokes a pack a day or more. Let us estimate that a third of this number have been smoking at that rate for more than five years. If so, our plan has the potential for reducing smoking among five percent of the adult population and the potential for increasing smoking among 95 percent of the adult population. It is exceedingly difficult to attach numbers to the considerations we have just reviewed without coming to the conclusion that the program as specified would have the net effect of increasing both the number of cigarettes consumed and the number of smokers.
Back to Square One
When we reconsider the three parameters and try to select a combination that meets the challenge, the nature of their interdependence becomes clear. Suppose, for example, that we require a smoking history of at least ten years, and thereby, as intended, reduce the number of persons who are drawn into smoking just because of the reward. But such a step makes no difference in the calculations of those who have already been smoking more than five years (they are, in effect, operating under the logic of a five-year eligibility rule). Among those who have smoked less than five years, the change in the eligibility requirement has two counterproductive effects. First, persons who have smoked less than five years constitute a large proportion of smokers that the program should be reaching—younger, with more to gain from quitting. By extending the requirement to ten years, the program has been made irrelevant to many of them. For those who do think that far ahead, the effects will tend to be harmful, inducing a sense that there will be time to quit—and profit to be made—at a later point in their lives. Thus lengthening the eligibility period to ten years does not help; it makes matters worse.
As we ponder ways out of this bind, it becomes clear that the most dramatic reductions in smoking occur among persons who quit the soonest—a person who quits smoking at age sixty-five saves only a few years’ worth of smoking, whereas a person who quits at twenty saves decades. Why not focus our efforts among the very young? Even granting the tendency of the award to encourage smoking so as to qualify, perhaps this will be more than counterbalanced by the very long periods of “savings” that will result from each success. So we target the program at youth (perhaps by installing an age-eligibility criterion—the specific method makes no difference). But the results are even more disastrous. The qualification criteria must be loose, because only a tiny fraction of the teenaged smokers we want to reach have had time to smoke very long. The result, when combined with a significant reward for quitting, is that the inducement effect is overpowering. Even teenagers who have no desire to smoke at all find it worth inculcating the habit for a year (or whatever our time limit is reduced to). Once started, only a proportion of those who smoked only because the program existed and who fully intended to quit are actually able to quit. The age effect backfires: While it is true that inducing a youngster to quit (who otherwise would not have quit) saves decades of smoking, it is equally true that inducing a youngster to start costs decades of smoking, and we produce far more of the latter than the former.
Two Ways Out
We give up on a continuing program, instead, we propose that the program be made a one-time, never-to-be-repeated offer: Announce the program, give everyone who is already eligible a chance to enroll, but give no one a reason to start smoking or to increase their smoking in order to become eligible. State loudly and unequivocally that the program will never be repeated. We will at least achieve the success of the first year.
Theoretically, this scheme might (but only might) reduce net smoking. In practice, it is guaranteed that the program will be continued. A successful one-time effort will be refunded immediately and on a larger scale. Congress rarely cancels even a failed social program, let alone a successful one.
Ultimately, the logic of the situation drives us to the one configuration of awards that surely will reduce net smoking: we offer a dollar amount to everyone who does not smoke, but make them pay it back if they ever start. Since this will cost far more than a billion dollars a year, we seek permission to increase the budget, pointing out that, while it may be expensive, our way out will in fact reduce smoking, whereas the alternatives will not. But some unfriendly critic points out that all we need do is levy a fine on everyone who begins smoking (or who continues to smoke)that is equal to the reward we propose to offer for not starting. The effects on smoking will be essentially the same (a $10,000 penalty ought to have about as much effect as a $10,000 reward for persons at most income levels), and the government will get a lot of revenue to boot. This proposal is of course also rejected, on grounds that it is unfair to the poor.
As one experiments with different combinations of rules, it becomes apparent that the traps we encounter in the first approximations are generalizable. Any change in the parameters intended to reduce one problem raises a new one. Why should this be? Is it intrinsic to the process? Or is it a peculiarity of an example I carefully chose?
Laws of Social Programs: We cannot design programs that escape their influence.
At first glance, the smoking example seems most apt for a certain type of social program, the one that seeks to change behavior from X to Y—what might be called “remedial” social programs. But in fact it applies to transfer programs of all types. In all cases, the transfer is legitimized by the recipient’s being in a certain condition (whether smoking or poverty) that the government would prefer the recipient not be in. The burden of the smoking example is not that we failed to reduce smoking—to achieve the desired behavioral change—but that we increased the number of people who end up in the undesired condition. This charge applies to transfers in general.
The reasons why are not idiosyncratic. Let me suggest some characteristics we observed in the thought experiment that occur so widely and for such embedded reasons that they suggest laws. That is, no matter how ingenious the design of a social transfer program may be, we cannot—in a free society—design programs that escape their influence. Together, they account for much of the impasse we observe in the anti-smoking example and point to some important principles for designing social programs that work.
• #1. The Law of Imperfect Selection. Any objective rule that defines eligibility for a social transfer program will irrationally exclude some persons.
It can always be demonstrated that some persons who are excluded from the Food Stamps program are in “greater need” than some persons who receive Food Stamps. It can always be demonstrated that someone who is technically ineligible for Medicaid really “ought” to be receiving it, given the intent of the legislation.
These inequities, which are observed everywhere, are not the fault of inept writers of eligibility rules, but an inescapable outcome of the task of rule-writing. Eligibility rules must convert the concept of “true need” into objectified elements. The rules constructed from these bits and pieces are necessarily subject to what Herbert Costner has called “epistemic error”—the inevitable gap between quantified measures and the concept they are intended to capture. We have no way of defining “truly needy” precisely—not those who truly need to stop smoking, nor those truly in need of college scholarships or subsidized loans or disability insurance. Any criterion we specify will inevitably include a range of people, some of whom are unequivocally the people we intended to help, others of whom are less so, and still others of whom meet the letter of the eligibility requirement but are much less needy than some persons who do not.
Social welfare policy in earlier times tended to deal with this problem by erring in the direction of exclusion—better to deny help to some truly needy persons than to let a few slackers slip through. Such attitudes depended, however, on the assumption that the greater good was being served. Moral precepts had to be upheld. Whenever a person was inappropriately given help, it was bad for the recipient (undermining his character) and a bad example to the community at large.
When that assumption is weakened or dispensed with altogether, it follows naturally that the Law of Imperfect Selection leads to programs with constantly broadening target populations. If persons are not to blame for their plight, no real harm is done by giving them help they do not fully “need.” No moral cost is incurred by permitting some undeserving into the program. A moral cost is incurred by excluding a deserving person. No one has a scalpel sharp enough to excise only the undeserving. Therefore it is not just a matter of political expedience to add a new layer to the eligible population rather than to subtract one (though that is often a factor in the actual decision-making process). It is also the morally correct thing to do, given the premises of the argument.
• #2. The Law of Unintended Rewards. Any social transfer increases the net value of being in the condition that prompted the transfer.
A deficiency is observed—too little money, too little food, too little academic achievement—and a social transfer program tries to fill the gap—with a welfare payment, Food Stamps, a compensatory education program. An unwanted behavior is observed—drug addiction, crime, unemployability—and the program tries to change that behavior to some other, better behavior—through a drug rehabilitation program, psychotherapy, vocational training. In each case, the program, however unintentionally, must be constructed in such a way that it increases the net value of being in the condition that it seeks to change—either by increasing the rewards or by reducing the penalties.
For some people in some circumstances, it is absurd to think in terms of “net value,” because they so clearly have no choice at all about the fix they are in or because the net value is still less desirable than virtually any alternative. Paraplegics receiving Medicaid cannot easily be seen as “rewarded” for becoming paraplegics by the existence of free medical care. Poor children in Head Start cannot be seen as rewarded for being poor. Persons who are in the unwanted condition completely involuntarily are not affected by the existence of the reward.
But the number of such pure examples is very small. The paraplegic anchors one end of the continuum labeled “Degree of Voluntarism in the Conditions that Social Policy Seeks to Change or Make Less Painful.” The apparent unattractiveness of most of the conditions that social policy seeks to change must not obscure the continuum involved. No one chooses to be a paraplegic, and perhaps no one chooses to be a heroin addict. But the distinction remains: very few heroin addicts developed their addiction by being tied down and forcibly injected with heroin. They may not have chosen to become addicts, but they did choose initially to take heroin.
Let us consider the implications in terms of the archetypical social program for helping the chronic unemployed escape their condition, the job-training program.
Imagine that a program is begun that has the most basic and benign inducement of all, the chance to learn a marketable skill. It is open to everybody. By opening it to all, we have circumvented (for the time being) the Law of Unintended Rewards. All may obtain the training, no matter what their job history, so no unintended reward is being given for the condition of chronic unemployment.
On assessing the results, we observe that the ones who enter the program, stick with it, and learn a skill include very few of the hard-core unemployed whom we most wanted to help. The typical “success”stories from our training program are persons with a history of steady employment who wanted to upgrade their earning power. This is admirable. But what about the hardcore unemployed? A considerable number entered the program, but almost all of them dropped out or failed to get jobs once they left. Only a small proportion used the training opportunity as we had hoped. The problem of the hard-core unemployed remains essentially unchanged.
We may continue to circumvent the Law of Unintended Rewards. All we need do is continue the job-training program unchanged. It will still be there, still available to all who want to enroll, but we will do nothing to entice participation.
The alternative is to do something to get more of the hardcore unemployed into the program, and to improve the content so that more of them profit from the training. And once this alternative is taken, the program planner is caught in the trap of unintended rewards. Because we cannot “draft” people into the program or otherwise coerce their participation, our only alternative is to make it more attractive by changing the rules a bit.
Suppose, for example, we find that the reason many did not profit from the earlier program was that they got fired from (or quit) their new jobs within a few days of getting them, and that the reason they did so had to do with the job-readiness problem. The ex-trainee was late getting to work, the boss complained, the ex-trainee reacted angrily and was fired. We observe this to be a common pattern. We know the problem is not that the ex- trainee is lazy or unmotivated, but that he has never been socialized into the discipline of the workplace. He needs more time, more help, more patience than other workers until he develops the needed work habits. Suppose that we try to compensate-for example, by placing our trainees with employers who are being subsidized to hire such persons. The employer accepts lower productivity and other problems in return for a payment to do so (such plans have been tried frequently, with mixed results). Given identical work at identical pay, the ex-trainee is being rewarded for his “credential” of hardcore unemployment. He can get away with behavior that an ordinary worker cannot get away with.
May we still assume that the program is making progress in preparing its trainees for the real-world marketplace? Will the hardcore unemployed modify their unreliable behavior? What will be the effect on morale and self-esteem among those trainees who were succeeding in the program before the change of rules? It is tempting to conclude that the program has already ceased to function effectively for anyone anymore, that the change in rules has done more harm than good. But my proposition is for the moment a more restricted one: The reward for unproductive behavior (both past and present) now exists.
What of the case of a drug addict who is chronically unemployed because (let us assume) of the addiction? It might seem that the unintended reward in such a case is innocuous; it consists of measures to relieve the addict of his addiction, measures for which the nonaddict will have no need or use. If we were dealing with an involuntary disability—our paraplegic again—the argument would be valid. But in the case of drug addiction (or any other behavior that has its rewards), a painless cure generally increases the attractiveness of the behavior. Imagine, for example, a pill that instantly and painlessly relieved dependence on heroin, and the subsequent effects on heroin use.
Thus we are faced with the problem we observed in the thought experiment. The program that seeks to change behavior must offer an inducement that unavoidably either adds to the attraction of, or reduces the penalties of engaging in, the behavior in question.
We are now ready to tackle the question of when a social program can reasonably be expected to accomplish net good and when it can reasonably be expected to produce net harm. Again let us think in terms of a continuum. All social programs, I have argued, provide an unintended reward for being in the condition that the program is trying to change or make more tolerable. But some of these unintended rewards are so small that they are of little practical importance. Why then can we not simply bring a bit of care to the design of such programs, making sure that the unintended reward is always small? The reason we are not free to do so lies in the third law of social programs:
• #3, The Law of Net Harm. The less likely it is that the unwanted behavior will change voluntarily, the more likely it is that a program to induce change will cause net harm.
A social program that seeks to change behavior must do two things. It must induce participation by the persons who are to benefit, as described under the Law of Unintended Rewards. Then it must actually produce the desired change in behavior. It must succeed, and success depends crucially on one factor above all others: the price that the participant is willing to pay.
The more that the individual is willing to accept whatever needs to be done in order to achieve the desired state of affairs, the broader the discretion of the program designers. Thus, expensive health resorts can withhold food from their guests, hospitals can demand that their interns work inhuman schedules, and elite volunteer units in the armed forces can ask their trainees to take risks in training exercises that seem (to the rest of us) suicidal. Such programs need offer no inducement at all except the “thing in itself” that is the raison d’être of the program—a shapelier body, a career as a physician, membership in the elite military unit. Similarly, the drug addict who is prepared to sign over to a program a great deal of control over his own behavior may very well be successful—witness the sometimes impressive success rates of private treatment clinics.
The smaller the price that the participant is willing to pay, the greater the constraints on program design. It makes no difference to an official running a training program for the hardcore unemployed that (for example) the Marine Corps can instill exemplary work habits in recruits who come to the Corps no more “job-ready” than the recruits to the job-training program. If the training program tried for one day to use the techniques that the Marine Corps uses, it would lose its participants. Boot camp was not part of the bargain the job trainees struck with the government when they signed on. Instead, the training program must not only induce persons to join the program (which may be fairly easy). It must also induce them to stay in the program, induce them to cooperate with its curriculum, and induce them, finally, to adopt major changes in outlook, habits, and assumptions. The program content must be almost entirely carrot.
There is nothing morally reprehensible in approaches that are constrained to use only positive inducements. The objections are practical.
First, it is guaranteed that success rates will be very low. The technology of changing human behavior depends heavily on the use of negative reinforcement in conjunction with positive reinforcement. The more deeply engrained the behavior to be changed and the more attractions it holds for the person whose behavior is involved, the more important it is that the program have both a full tool kit available to it and the participant’s willingness to go along with whatever is required. The Marine Corps has both these assets. Social programs to deal with the hardcore unemployed, teenaged mothers, delinquents, and addicts seldom do.
Second, as inducements become large—as they must, if the program is dealing with the most intractable problems—the more attractive they become to people who were not in need of help in the first place. We do not yet know how large they must finally become. At this point, it appears that any program that would succeed in helping large numbers of the hardcore unemployed will make hardcore unemployment a highly desirable state to be in.
The Theoretical and Practical Result
The conditions that combine to produce net harm are somewhat different in the theoretical and the practical cases, but they come to the same thing. Theoretically, any program that mounts an intervention with sufficient rewards to sustain participation and an effective result will generate so much of the unwanted behavior (in order to become eligible for the program’s rewards) that the net effect will be to increase the incidence of the unwanted behavior. In practice, the programs that deal with the most intractable behavior problems have included a package of rewards large enough to induce participation, but not large enough to produce the desired result.
My conclusion is that social programs in a democratic society tend to produce net harm in dealing with the most difficult problems. They will inherently tend to have enough of an inducement to produce bad behavior and not enough of a solution to stimulate good behavior; and the more difficult the problem, the more likely it is that this relationship will prevail.