It is one thing to claim—as the politicians of both parties now do—that the replacement of welfare by work would be a good thing for recipients, for taxpayers, and for the general reputation of public assistance to the poor. It is quite another question whether that transformation can actually be accomplished, and what it would then take to accomplish it. In particular one is entitled to ask: What jobs will former welfare recipients find, and how will they find them?

This elementary distinction between desirability and feasibility is often neglected in political debate. During the rhetorical maneuvering that led to the welfare “reform” bill passed in the summer of 1997 everyone seemed to be devoted to ending “welfare as we know it” but no one was prepared to describe how the new system would actually function. (Very likely “none of the above” would have been the most popular answer if the question had been asked.) Some time will pass before the shape of the new system is visible. The legislation left the main decisions to the individual states, which may well pass the buck to the large cities where most of the problem is, which may in turn pass the buck to the bishop of Caesarea.

That particular question is not my subject here because I am not trying to understand the consequences of any particular legislative proposal. (That has already been done for the 1997 bill by the Urban Institute, with scary results that do not seem to faze the bill’s sponsors a bit, as well as by Peter Edelman in a recent Atlantic Monthly article.1 ) My intention is quite different from theirs. It is, first, to describe in theoretical but common-sense terms the consequences of withdrawing welfare benefits and forcing the former recipients into the labor market. What will become of them? Where will the jobs come from that they are supposed to find and occupy?

I will also examine the results of some experimental “workfare” initiatives on the part of several states, in order to get a quantitative grip on the employment and earnings prospects of former welfare beneficiaries and their successors. Finally, I will speculate briefly about what would be required for a successful transformation of welfare into work. My conclusion is going to be that we have been kidding ourselves. A reasonable end to welfare as we know it—something more than just benign or malign neglect—will be much more costly, in budgetary resources and also in the strain on institutions, than any of the sponsors of welfare reform have been willing to admit. And the reasons are based on normal economics.

On the question of job availability there are two extreme positions to consider. The first is very optimistic: there is no problem. The jobs are there; they are always there. It is only necessary that those who seek them be willing to accept realistic wages. Former welfare recipients, having nowhere else to go, will do just that. They will be paid what their productive capacity justifies, and that may be more than we think. The demand for labor is elastic; that means even a small reduction in going wage rates will generate a substantial expansion of job openings. And the implied clear presentation of a route to self-betterment will lead unqualified workers to acquire the education and training they need to move up the ladder. The small residue of genetic or accidental incompetents—the true paupers—can be left to private or public charity.

There is nothing illogical or incoherent about this story. It could apply in some worlds. I have to say that I do not think it describes our world, the sort of world that generated the 1982 recession in the United States and a decade of 10 percent unemployment rates, now even higher, in the main countries of Europe. It would be irresponsible, almost Alfred E. Newmanesque, to depend on this idealized story to smooth the transition to welfare as we will come to know it.

There is another extreme theory that sees only rigidity where the first sees flexibility. It comes to deeply pessimistic conclusions. In this story, the total amount of employment is determined almost entirely by macroeconomic factors. Certain broad characteristics of the private economy, together with national monetary and budgetary policies, determine, within narrow limits, the aggregate expenditures of the final purchasers of goods and services. Most of the time the aggregate volume of production is limited by the amount of spending available to support it. The step from aggregate production to aggregate employment depends only on current productivity, a remote and slow-moving part of the macroeconomic equation.

It follows that the labor market is like a game, or several games, of musical chairs. (At the birthday parties of my childhood it had the more picturesque name of Going to Jerusalem.) When the music stops, the players scramble for the available chairs. Since there are fewer chairs than players, the losers are left standing. They are, you might say, the unemployed. If the game were repeated, the losers might be different people, but the number of losers is determined entirely by the number of players and the number of chairs. Adding more players—which is what forcing welfare beneficiaries into the labor market would do—can only increase unemployment. Some former welfare recipients will find jobs—perhaps many will because, among other reasons, they are hungry—but only by displacing formerly employed members of the assiduously working poor.


I think that this story does not give enough credit to the adaptability of real market systems. Anyone who believed it would have a hard time explaining the fairly long periods during which the US economy accommodates a growing labor force while the unemployment rate fluctuates within a fairly narrow range. The only possible explanations would be very good luck or very good policy, and you would have to be pretty gullible to find either one to be a plausible account of history.

Then how would a large-scale substitution of work for welfare play itself out in the real-world system of imperfect labor and product markets? A more accurate understanding will lie somewhere between the extremes I have just sketched. It will have to allow market forces to operate with some effectiveness, but will also respect the power of macroeconomic conditions over aggregate expenditure and output. This territory is still being fought over by mainstream economists.

Any effective transformation of welfare into work, if it means anything, must mean that a substantial number of unqualified people will be looking for work who were previously not doing so. Some of them will find jobs just by being in the right place at the right time; they might have done so earlier if they had tried. These jobs will represent a net addition to aggregate employment. One sometimes gets the feeling that this is what some members of Congress visualize, and all that they visualize. If so, they cannot be right. There is absolutely no reason to believe that our economy holds a substantial number of unfilled vacancies for unqualified workers. The machinery of adjustment must be something more elaborate. Here and later, it is worth keeping in mind a point recently emphasized by Christopher Jencks.2 There are substantial cash costs associated with going to work; the largest costs are incurred when the mothers of small children take jobs. For that reason, many welfare recipients who do find work will find themselves worse off, perhaps substantially so.

The most immediate route by which the ex-welfare population can find jobs is by competing with and displacing other unqualified workers who are already employed, either by being in some way a more suitable employee or, more likely, by offering to work for less than the incumbent is getting. Unqualified workers are presumably excellent substitutes for one another, so only a very small wage cut would be needed. But pure displacement is just musical chairs: more players and the same number of chairs.

More important is the possibility that competition for jobs by ex-welfare recipients and their successors will drive down the wage for unqualified workers by enough to induce some employers to hire them to replace slightly more qualified incumbents who do the job better but have to be paid more. Since bottom-end workers are less than perfect substitutes for second-level workers, the fall in the unskilled wage will have to be perceptible to make the switch profitable for employers. There is displacement going on here too, but it is somewhat better than one-for-one because unqualified workers are, by definition, less productive than second-level workers. Also, a broader wage reduction for lowest-level and second-level workers has a better chance of expanding the number of employment opportunities available in that segment of the labor market. So there would be a small gain in total employment, but it comes at the expense of the earnings and job prospects of previously employed second-level workers. (This talk of discrete levels of skill is just an artificial simplification of a more complex process of job search by individuals and occasional matches with firms. It helps keep the discussion orderly.)

In principle, the process does not stop there. The erosion of the wages of second-level, slightly skilled, workers makes them more competitive with third-level, slightly more skilled workers. The fact that some second-level workers have been displaced into unemployment may lead to a further bidding-down of third-level wages as the competition for jobs intensifies. So the costs of adjusting to the influx of former welfare recipients spreads to the working poor, the working just-less-poor, and so on, in the form of lower wages and heightened job insecurity.


There is, of course, a long, branching hierarchy of skill levels in a modern economy. Each level is subject to competition from those just above and below, especially below. But one would naturally expect the degree of displacement to attenuate as one gets further and further away from the relatively unqualified former welfare recipients whose appearance on the job market is the source of the disturbance. By the time you get to the very top of the food chain, say the Princeton philosophy department, no one will be feeling any pain, and in fact the tenured members may be able to get their yardwork done more cheaply. The adjustment costs will be concentrated at the bottom of the job hierarchy and the bottom of the income distribution. Of course it could be said that those are the very people who have been protected from competition all along by the unreformed welfare system. It is not a remark I would choose to make myself, but there must be some truth to it.

All this reshuffling in the labor market must have macroeconomic implications. The relevant question is whether any of them hold the promise of an easier transition to a world in which work has replaced welfare. Suppose we imagine the displacement and wage-reduction process to have worked itself out completely. The result is a lower economy-wide real wage. Can we expect that interim fact to generate enough net new jobs to accommodate the addition to the labor force created by the end of welfare as we know it? Or will there just be more unemployment?

It may help here to think about two rather different varieties of unemployment. These correspond roughly to the two extreme theories I sketched to introduce this discussion. One sort of unemployment arises because there is not enough demand for the products of labor. Spending on goods and services is somehow inadequate. This is often called “Keynesian” unemployment. The other sort arises because, through one mechanism or another, wages are too high. Business firms could produce and sell more, but it would be unprofitable for them to do so. The way to expand production and employment is to have lower costs; for the economy as a whole that means mainly lower labor costs. This is often called “classical” unemployment.

Classical unemployment may diminish in response to wage reduction, though the process may be more complicated than this simple statement indicates. Keynesian unemployment may not respond; there is even a danger that a transfer from wage earnings to profits might result in lower total spending, which would have a perverse effect on jobs. (This, too, is more complicated than it sounds.)

To come back to our particular problem, the issue is whether lower wages, on average, will more or less automatically provide new jobs to be filled by former welfare recipients and their successors. That turns on the responsiveness of the aggregate demand for labor to the real wage. There has been quite a lot of research on that very question, because the answer is of great general importance. I think it is fair to say that the measured responsiveness has been disappointingly small. (I say “disappointingly” because life would be easier if small real-wage changes could induce substantial shifts in employment, and if the same were true of other prices and the associated quantities.) It is not easy to characterize the range of estimates numerically, but it would not be far off to say that as much as a 2 or 3 percent change in the real wage level would be needed to elicit a 1 percent change in the demand for labor (in the opposite direction of course). This result could be taken to reflect the relatively small weight of classical unemployment in the total, in the United States at least; or it might be telling us something about the working of labor-market institutions. In any case, the implication is that it would take a reduction of 3 to 5 percent in the average real wage to generate net new jobs equal to two thirds of the adult AFDC population.

Is this a lot or a little? The first thing to say is that the required change in the national average wage is not the figure that matters. I hope I have made it clear that the competition for jobs set off by welfare reform would be concentrated at the lower end of the job hierarchy. It is certain that a perceptibly larger reduction in the wages of unskilled and semiskilled workers would have to take place if the bottom end of the labor market had to absorb an additional million and three quarters relatively unqualified workers. I would not want to say more than that. But it seems likely that unskilled wages would have to fall by considerably more than 5 percent in order to make jobs available to those 1.75 million workers. If conventions of equity or propriety or the existence of a statutory minimum wage should prevent the required reduction in unskilled wages, the consequence would be higher unemployment. Either way, the working poor will pay.

The more important observation is not numerical at all. Apart from magnitudes, the argument leads to the conclusion that the burden of adjusting to any genuine replacement of welfare by work will fall primarily on low-wage workers, especially those virtuous ones who have been employed all along. The burden will take the form of lower earnings and higher unemployment, in proportions that are impossible to guess in advance. It would be too drastic to imagine that the process might lead to the growth of a distressed class of very-low-wage workers and, through the workings of altruism, to the reconstruction of welfare as we knew it. There are alternatives, to be discussed briefly later on. But I hope it is not drastic at all to doubt that many reasonable people who favor welfare reform have had in mind the imposition of nontrivial additional impoverishment on the industrious working poor.

Completeness requires me to mention one other way in which macroeconomic forces might ease this problem. Just because the addition of a million and a half or so new workers to the labor force represents some potential unemployment, perhaps the Federal Reserve might see it as some additional protection against inflation. Any consequent easing of monetary policy—or other macroeconomic policy, if there were any—could lead to lower interest rates, economic expansion, and better job prospects. I think this is a forlorn hope, however. Wage-induced inflation does not come from excessive tightness in the market for unskilled labor, but from better-skilled, higher-wage, sometimes unionized workers, if it comes from the labor market at all. The economy will not be more inflation-proof, and will have to work it all out on its own.

I say this despite the tendency for low-end wages to rise proportionally more than high-end wages in good times. My guess is that this may happen because higher-end wages are more likely to be governed by long-term agreements, explicit or implicit, whereas low-end wages are free to respond to immediate market forces. It seems wholly unlikely that unskilled wage-push plays much of an independent inflationary role. An influx of former welfare recipients into the job market will not give the Federal Reserve much of a cushion against the economy’s overheating.

Beyond these rather general considerations, there is some more direct evidence about the probable fate of welfare recipients forced into the labor market by the withdrawal of support. Most of it comes from the “workfare” experiments designed and operated by many states during the past decade. The most useful for my purpose are those that were conducted as genuine experiments, with participants assigned at random to the program itself or to a control group. The intake into the process consisted entirely of participants in or applicants to AFDC: those chosen for the experiment were subject to the particular workfare program being tried out, while others in AFDC were subject to the normal regulations. Differences in outcomes can thus be imputed to the effects of whatever mandatory requirements were imposed by the workfare program being tested. It cannot be assumed that these experiments anticipate the likely outcome of an all-out imposition of time limits, work requirements, or simply the closing-down of AFDC. They do give us some quantitative insight into the likely fate of welfare recipients tossed into the open labor market.

I shall use as my main example the California GAIN (Greater Avenues for Independence) experiment. It is the largest and best documented of the state initiatives; and Manpower Demonstration Research Corporation has collected and analyzed data extending out to three years after experimentals’ first exposure to the program.3 Longer-term observations are still to come.

The program itself is complex and I will give only a brief and crude description of it. Upon assignment to the program, a welfare recipient or applicant who lacks a high school diploma or a GED certificate (General Educational Development, probably worth very little), or scores low on a basic skills test, or is deficient in English, is assigned to one or another basic education scheme. Others, and those who finish their basic education, move on to an organized job-search activity. This includes training sessions in which groups are taught basic job-seeking and interviewing skills, and then a supervised job search, with access to telephone banks, job listings, and some counseling. This goes on for about three weeks. Those who do not find a job in this way proceed to formulate an individual employment plan, working with a counselor. The plan will entail further activities, like vocational training, unpaid work experience, and so on. These activities then alternate with job search.

The question is: What is the subsequent labor-market history of those subject to these requirements, particularly but not only as compared with the controls who simply carry on as before? We can answer this question for six miscellaneous California counties. One of them, Riverside, between Los Angeles and San Diego, is especially interesting because its program was much more successful than the others and was conducted by its staff in a very energetic and aggressive way.

Here are the key results, taking all six counties together, including a total of 17,677 people who were selected for the experiment and 5,114 controls. During each of the three years of follow-up, about 40 percent of the experimentals had some employment. These were not steady jobs; in the last quarter of the third year, only 28.5 percent had any employment, and of course the proportion employed in any month or week would be still smaller. All told, 56.7 percent of those who took part in the experiment held a job at one time or another during the three-year period. Almost 51 percent of the controls had some employment during that time, so the net impact of the GAIN program was to increase the fraction ever employed by 6 percentage points. This difference is statistically significant, but it is fairly small.

The conclusion to be drawn is this: in California, in the economic conditions of the early l990s, about a third of welfare recipients held a job at one time or another during any year; participation in the GAIN version of workfare increased that fraction by 4 to 6 percentage points. One cannot be sure that this small margin is an indicator for the future, but the burden of proof is on anyone who thinks that welfare recipients forced into the labor market will be very successful in the search for jobs.

I mentioned that Riverside County seemed consistently to get better results than any of the other five counties. It is worth seeing how much better, as an indicator of the best that might be hoped for. In one sense the comparison is a source of optimism. Riverside did do better than the other five counties; so it does matter how a welfare reform program is conducted. Activism pays off. That is the good news. The bad news is that even the Riverside results suggest that the job prospects for former welfare recipients are pretty grim.

Two thirds of the people selected for the Riverside experiment held a job at some time during the first three years of their exposure to GAIN, 10 percentage points more than the average for all six counties. And the difference seems to have nothing to do with the Riverside area itself, because the control group in Riverside had the same experience as the statewide average. So the conduct of the program made the difference. But the Riverside advantage diminished year by year and, besides, although it is big enough to be noticed it is not big enough to solve the problem.

I could report on similar studies of the work-welfare experiments conducted by a dozen other states. But the basic message would be unchanged. The various states have tried slightly different programs, in slightly different economic environments, and naturally they produce slightly different results. But none of them offers grounds for optimism about the ability of welfare recipients to find and hold jobs, or to earn a decent living. (Some are more pessimistic in their implications than the California GAIN experiment.)

Instead I shall describe briefly a much smaller and more casually studied episode in Michigan, because it reproduces more nearly the effects of a pure-and-simple end to welfare benefits.4 Until October 1991, the state of Michigan had funded a program called General Assistance that paid cash benefits of $160 a month to non-elderly poor adults without dependent children. The authors of a study of this program note that this population was probably more rather than less able to find and keep jobs than the standard AFDC population. General Assistance was ended in October 1991. (Most of the recipients had been receiving, and continued to receive, other benefits.)

A representative sample of 426 ex-recipients of General Assistance were interviewed two years after the program had ended, and were asked about their labor-market experience in the meanwhile. About 65 percent of them had worked at a regular job or at casual labor at some time during the period. This was the same for those with less than a high school degree and those with a high school diploma, a GED Certificate, or more. The better-educated group held significantly steadier and better-paid jobs, however. For instance, 46 percent of them were employed in the month of the survey, compared with 28 percent of the high school dropouts, at average hourly wages of $6.07 and $4.78, respectively. Their total earnings in the month before the survey averaged $596 for the better-educated and $377 for the less well-educated, which implies that the two groups averaged about 100 and 80 hours of work, respectively, in that month. (Full-time work would be about 160 hours.)

Those who worked in the survey month, even the high school dropouts, earned more than the old General Assistance benefit of $160. But it could not be said that they earned a living. It would be a gross overestimate even to multiply $377 per month by twelve to get $4,500 because a third of the sample never worked at all during the two years, and very few of those who worked were able to work steadily. The high school-educated did better, but for the same reasons, $7,200 a year would considerably overestimate the earning capacity even of those who succeeded in finding work at all.

The indications from Michigan and California are much the same. Without some added ingredients, the transformation of welfare into work is likely to be the transformation of welfare into unemployment and casual earnings so low as once to have been thought unacceptable for fellow citizens.

More microscopic, almost ethnographic, observations only add depth to this picture. William Julius Wilson has powerfully documented the disappearance of jobs from poor, black inner-city neighborhoods like the South Side of Chicago.5 Katherine Newman, herself an anthropologist, followed up all job openings filled by four fast-food franchises in Harlem in 1993, and interviewed those who got the jobs and those who applied but failed. (There were fourteen applicants for each job filled.) The winners in this sweepstakes were better educated and better connected than the losers, but even the losers were more experienced and more educated on average than the typical welfare recipient. Three quarters of the losers were unemployed when interviewed a year later, although most of them had continued to look for work.6 It is impossible to believe that the forced influx of ex-welfare beneficiaries into these labor markets could do anything but make a bad situation worse. (By the way, preliminary results from the Canadian Self-Sufficiency Project confirm this pessimism.)

The proper conclusion from this analysis is not that the substitution of work for welfare, however desirable it may be, is infeasible in practice. That might be so if the only alternative to “welfare as we know it” were simply to walk away from it. More to the point, I think, is the conclusion that a decent welfare-to-work transition will require a more complicated—and more expensive—set of changes. Two conclusions, in particular, seem to me to follow from the argument I have made.

The first is that an adequate number of jobs for displaced welfare recipients will have to be created, either through some version of public-service employment or through the extension of substantial special incentives to the private sector (profit and nonprofit). Appeals to businesses to hire welfare recipients voluntarily are a form of abdication of responsibility and even subsidies to employers are likely to run into real problems of management. There will have to be a determined and expensive effort to increase the demand for unskilled and unqualified labor.

William Julius Wilson has advocated that the government create something like the WPA of the New Deal years. I can see the point of that. Pretty clearly there are major needs to improve infrastructure in urban and rural communities that could be met with little or no trespassing on the private sector, and with intensive use of unskilled labor. But there are two ways in which this suggestion seems to fall short of the need. Wilson is thinking mainly about males trapped in inner-city ghettos without employment opportunities. But the AFDC population is primarily female, often women with children. Unskilled construction labor may be a mismatch if the goal is to insert that group into the world of work and to build up marketable skills.

The second deficiency is related. In an economy that has been durably tending toward the production of services instead of tangible goods, focusing on heavy construction is like trying to make water flow uphill. It would seem more useful to create an employment track that led to work habits and skills normally in demand in the service sector. This would also be a better match with the proportions of women and men in the welfare population. There are no big-time models for such an effort, but some institutional ingenuity might find a way.

The main point, however, is not the design of a particular scheme. It is, as Wilson sees, the need for purposeful creation of jobs, in numbers, places, and forms that are suitable for the people who will fill them, and that can provide the sort of experience that may eventually have cash value in the open labor market. Any scheme that will do the trick will be costly, in budgetary dollars and in the need to invent and to staff institutions of a kind for which we have little experience or even intuition. The task is even harder than it sounds, because it involves swimming against the current. There has been in recent years a huge shift in demand away from unskilled labor. The source appears to have been mostly technological but the source is less important than the fact, and the fact suggests that the labor market will not naturally welcome an influx of unskilled workers.

The second conclusion I want to draw goes back to the notion of “packaging”—i.e., the recognition and acceptance of the fact that many welfare recipients will simply have to combine earnings and public assistance if they are to lead tolerable lives.7 Suppose we succeed in managing a transition from welfare to work. The evidence implies inescapably that the jobs obtainable by former welfare recipients will pay very low wages and pay them irregularly. (The irregularity inheres partly in the job and partly in the situation of the jobholder, as we have seen.) I think it is legitimate for taxpayers to want welfare recipients to work, but not so legitimate to want them to live at the miserable standard their earning capacity can provide, least of all if children are involved. The implication is that packaging will have to continue, and should be planned for. This means, by the way, that time limits are incompatible with the substitution of work for welfare.

A good many welfare beneficiaries today either alternate between working and welfare or do both at the same time. That pattern will have to be recognized as normal, even as a good thing under the circumstances. It should be regularized and institutionalized to see that the incentives point in the right direction and that justice is done, and to guard against corruption. (Corruption is also a danger in any scheme of public employment.)

The institutional details can be important. Here we have the advantage of an already functioning mechanism, the Earned Income Tax Credit. It could be calibrated to provide a tolerable standard of living for ex-welfare recipients—and others—who work hard and play by the rules, to use another of those phrases. Employers should understand that they benefit from the EITC too, because, like any subsidy, it puts a little downward pressure on the market wage.

The object of this mixed system should be to achieve a reasonable equilibrium between the norms of self-reliance and altruism. The real trouble with welfare as we knew it is that it tended to erode both. My suggestion is that a mixed work-welfare system, with an adequate supply of jobs, stands a chance of reinforcing both self-reliance and altruism, but such a system will not come cheap. There has been no sign yet that the United States is willing to put the necessary money where its mouth is.

Suppose nothing special happens. Welfare “reform” follows the script, without any amelioration. What will we then think about it? The welfare rolls will diminish. Governors will point to this decline with pride. Congressmen and senators (and presidents?) will nod their satisfaction. No one will ask what has happened to the former welfare recipients or to the working poor. If anyone asks, there will be no answer. There will be no data. As the Princeton economist Alan Krueger pointed out to me, the relevant experiments will not have been performed; the administrative system tracks only recipients, not the would-have-beens. They may be living with relatives who cannot afford them, or on the street, or in shelters. The need for relevant data is not just the peculiar craving of academic social scientists. It is the lifeblood of rational social policy and its evaluation.

This Issue

November 5, 1998