THE BARE FACT that a cause-effect relationship links certain social policies to some of the trends we examined in part II has been established. It was most clearly established, oddly, in an ambitious attempt to discredit the notion that such links exist. Most of our inquiry in the following chapters will wrestle with the dynamics of the causes and effects at work, but it is worth our while to document first the persuasive evidence that causes, not just coincidences of timing, are at work. For that, we turn to the story of the Negative Income Tax experiment.
In the account in chapter 2 of the disillusionment about the first War on Poverty programs, I mentioned the role of the program evaluators who reluctantly brought the bad news. The evaluators constituted only one small part of the story, however. When OEO was setting up shop in 1965, social scientists of all sorts were reaching out from the campus to become part of the excitement in Washington, bringing with them a tool kit of new methods with which they hoped to help build the Great Society.
The theory behind their techniques was not new. But in the absence of computers, applications had been limited. Quantitative social science had been for the most part restricted to research questions involving only a few variables and small samples; the computational burden otherwise was too great.
By the mid-1960s, the technological lid had been lifted. The social scientist could answer real-life questions that intuition and simpler methods could not. What is the effect of Policy X on, for example, monthly income when one holds sex, race, age, prior education, and ten other variables constant? With the computer to do the work, the social scientists knew how to reach an answer.
It turned out to be harder than it looked, of course—the theoretical power of the leading techniques (mainly forms of regression analysis and factor analysis, at the time) was greater than their practical utility when put to work. A few colossal blunders were made and, in some instances, discovered only after the government had acted on the basis of the erroneous analyses.1 But theory and practice continued to improve as the quantitative researchers learned from their mistakes. By the mid-1960s, any social scientist who was not at least conversant with sophisticated statistics was unable to follow much of the professional literature that, only a few years earlier, had seldom included anything more complicated than a cross-tabulation. Graduate students were routinely performing analyses that had been beyond the reach of any scholar only a few years earlier. And it was all happening at just the moment when the hardnosed idealists of the OEO were gearing up for their assault on poverty. A partnership between the policymakers and the newly equipped social scientists was sought by both sides. When die-hard southern congressmen ranted that these social programs were leading the country to perdition, the social scientists would have the evidence—cool, impartial, conclusive evidence—that no such thing was happening. Whence the origins of the great experiment that followed.
The foundation of the scientific method is the controlled experiment. The investigator takes two identical sets of subjects, exposes one but not the other to a specified stimulus or condition, and observes the subsequent differences between the two. No other procedure is as elegant in its assumptions, and no other evidence of causation carries as much weight.
By 1966–67, the planners at OEO thought they needed this demanding level of proof. Power at OEO had passed from the apostles of opportunity to the structuralists. It was increasingly taken for granted that some form of guaranteed annual income was the only way that the War on Poverty was going to be won. But the president was refusing to introduce such legislation. Even if he could be won over, passing the legislation was going to be exceedingly difficult. There was opposition, of course, from the remaining congressional conservatives, and moderates were on the fence. They were sympathetic to the goals of a guaranteed annual income, but they were worried about its negative effects on the “work ethic” and about how they could justify their support of such a program to working-class constituencies who would see it as a welfare giveaway. Somehow, proof must be established that a guaranteed income would not cause people to reduce their work effort, get married less often, divorce more quickly, or do any of the other things that the popular wisdom said it would cause them to do.
The OEO’s vehicle for providing the proof took the form of the most ambitious social-science experiment in history. No other even comes close to its combination of size, expense, length, and detail of analysis. It went under the unprepossessing name of the Negative Income Tax (NIT) experiment.2 It began in 1968, ultimately used 8,700 people as subjects, and lasted for ten years. (A planned twenty-year subsample was cancelled in 1980.) It resulted in a body of literature that, as of 1980, included more than one hundred published titles and countless unpublished reports.3 Its cost ran far into the millions; accountings vary.
Very briefly, a negative income tax provides payments to persons whose income falls below a certain floor. As implemented in the NIT experiment, it strayed far from its intellectual origins. Conservative economists, not liberal social reformers, had first advocated a negative income tax as a replacement for the existing welfare system.4 They took for granted that any form of welfare payments produced work disincentives. The point of an NIT was not to get people off welfare, but to fill the welfare system’s functions with somewhat fewer unwanted side-effects. The negative income tax was not felt to be especially “good"; it was just considered better than the alternatives.5
By the time that the NIT experiment was ready for implementation, such thoughts had been lost as the NIT concept was embraced by new proponents for whom it had nothing to do with efficiency, but instead had become “a pervasive way of thinking about poverty as an economic phenomenon [and] a device for trying to solve the problem of welfare dependency . . .,” as Heclo and Rein put it.6 More colloquially, the proponents of the NIT in the Johnson administration were out to slay the folk beliefs that welfare makes people shiftless. The NIT, properly redesigned, would provide work incentives and get people off the welfare rolls.
A disquieting element about the experiment was that it would be, as one close observer put it, “sponsored, designed, and even administered by ‘believers.’ ”7 Critics watched the work suspiciously for signs that the data were being cooked. But OEO had no such fears for its own integrity. “[I]t was the faith in the method of investigation, not the disinterestedness of the initial impulse, which would rescue their partisanship and lend credibility to the final results,”8 In the end, this faith in the method was justified.
The procedure followed the classic experimental paradigm. In each site, a sample of low-income persons was selected and randomly split into two groups: the “experimental” group and the “control” group. The members of the experimental group were told that for a specified number of years (usually three) they would have a floor put under their incomes.9 The benefits varied among participants, to test the sensitivity of the results to the generosity of the guaranteed income. The most common benefit level put the floor at approximately the official poverty line. The members of the control group received no benefits.
For the next ten years, results dribbled in. The New Jersey and Pennsylvania sites opened the experiment between 1968 and 1972. Then came tests with rural populations in Iowa and North Carolina, from 1970 to 1972. A predominantly AFDC population was tested in Gary, Indiana, between 1971 and 1974. The largest, longest, and best-evaluated experiments were in Seattle and Denver from 1971 to 1978, the “SIME/DIME” of so many research reports (Seattle Income Maintenance Experiment/Denver Income Maintenance Experiment). As the results appeared, they were subjected to methodological critiques. Experimental and analytic procedures were tightened for the next round, results were compared across sites, data were reanalyzed, and finally, by the end of the 1970s, a body of results was established that was broadly accepted as valid.10 With rigor and in enormous detail, the scientists validated not the sponsors’ hopes but their fears. The results were more or less what the popular wisdom said they would be.
The key question was whether a negative income tax reduced work effort. The answer was yes. The reduction was not the trivial one that NIT sponsors had been prepared to accept, but substantial. In the SIME/DIME sites (which produced neither the largest nor the smallest changes, but probably the most accurately measured ones), the NIT was found to reduce “desired hours of work” by 9 percent for husbands and by 20 percent for wives.11 “Desired hours of work” was measured by actual employment after factoring involuntary work reductions out of the calculation.
The 9-percent reduction for husbands was not in itself a fearsome number. If the reduction had consisted of husbands who were working a few hours less—giving up overtime opportunities, for example—to increase the time they could devote to other worthy pursuits, these results would have been defensible, even agreeable to the sponsors of the experiment. But the reduction in SIME/DIME and the other experiments appeared to have consisted primarily of men who had opted out of the labor market altogether.12 In the most detailed examinations, it was found that “reductions in the probability of employment are due primarily to reduced rates of entry into employment.”13
The results among husbands were disappointing but not the most troubling. As the analysts dug deeper into who was being affected and how, they found that the groups who showed the largest negative effects were precisely the ones who were in a position to cause the most long-term damage to the goal of reducing poverty.
The first of these groups was wives. At least from the Second World War (and much earlier for some groups) through the early 1960s, wives represented for poor families a source of marginal income that could push a family out of the poverty trap and into a more secure long-term future— either by continuing to work indefinitely, or by providing income that permitted the husband to upgrade his skills, move to another labor market, or make some other investment in long-term gains that requires a short-term expense.14 Thus the 20-percent reduction in work hours compared with the control group not only was quite large but implied that substantial numbers of families were, whether they knew it or not, climbing off one of their most promising ladders to prosperity.
The second group of special interest was young males who were not yet heads of families (“nonheads,” in the jargon). They were at a critical age in their lives: about to enter into the responsibilities of marriage and just establishing themselves in the labor force. If they were to escape from poverty, this was the moment to start. The NIT had a disastrous impact on their hours of work per week: down 43 percent for those who remained nonheads throughout the experiment, down 33 percent for nonheads who married.15
Perhaps they were going to school to better themselves? No, the possibility was investigated and rejected. Perhaps it was only a temporary effect? No, on the contrary, the response seemed to be stronger in the five-year experiment than in the three-year experiment. The investigators summed it up with perhaps excessive restraint:
The reduction in work effort by male nonheads who become husbands is clearly important. These males are reducing their work effort at the time when they are undertaking family responsibilities. Not only is their response important in the current period, but the reduction in work effort may also have long-term effects on their labor supply behavior.16
The effects on work effort were associated with collateral effects among all the subpopulations. Perhaps the most striking was the increase that NIT produced in periods of unemployment when a member of the experiment lost his or her job. Such periods lengthened by nine weeks (27 percent) for husbands, fifty weeks (42 percent) for wives, and fifty-six weeks (60 percent) for single female heads of families, in comparison with the control group.17
Does welfare undermine the family? As far as we know from the NIT experiment, it does, and the effect is large. In the SIME/DIME sites, the dissolution of marriages was 36 percent higher for whites receiving the NIT payments than for those who did not; for blacks the figure was 42 percent. In the New Jersey site, there was no difference among the white families in the experiment, but black family breakup was 66 percent higher in the experimental group than in the control group, and in the Spanish-speaking sample it was 84 percent higher.18 In one experiment, Gary, no effect was observed. When researchers looked into the possible reasons why, they found that, in Gary, couples were under the impression that if they split up, they would lose their NIT payments.19
The results were exhaustively analyzed, as researchers checked out the alternative explanations. None worked. The only salient difference that seemed to explain the substantially higher rates of marital instability in the two groups was the “treatment” itself, the NIT.20
The results I have just reviewed are noteworthy as they stand. But the true negative effects of the NIT were considerably larger than the data indicate. A variety of biases tended to suppress the negative effects.21 We cannot be sure how large the understatement was; it was surely substantial. The notes refer the reader to some of the analyses of this issue, Martin Anderson’s being the most exhaustive.22 Two of the most important of the reasons for the bias are these:
First, the observed effects were not obtained through a comparison with a “pure” control group (one experiencing no work disincentives), but in comparison with a population that was receiving all the normal welfare benefits of the 1970s, which were extensive and growing during the same period that the NIT experiment was conducted. The reductions in hours worked, the lengthened periods of unemployment, and every other effect are reductions over and above the effect of the work disincentives in the existing system.
Second, the great majority of the participants in the NIT experiment knew from the outset that they could count on the payments for only three years. Presumably people are less likely to burn bridges behind them if they know that the guaranteed income ends in three years than if it is legislated for life. Insofar as the subsamples given five-year and (in a very few cases) 20-year guarantees permit estimates, common sense is borne out —the longer the guarantee, the greater the negative effects of the NIT.23
The NIT experiment made a shambles of the expectations of its sponsors. But at the same time it was being conducted, the disincentives it would later demonstrate were being woven into the fabric of the welfare system.24 For our purposes, the NIT experiment directly answers the question we posed about causation, at least for the outcomes relating to welfare, work, and marriage: The only time we have been able to put the question to a controlled test, the causal effect was unambiguous and strong.