Why Welfare Research Fails

Case studies:

New Jersey's "Family Cap" Welfare Experiment

The Urban Institute's Portentous Prophecy

A paper prepared for the 1998 meetings of the Eastern Evaluation Research Society. Cape May, NJ. August 1998.

by Ted Goertzel

Sociology Department

Rutgers University

Camden NJ 08102


This and related papers can be found at:


People expect from the social sciences the knowledge to understand their lives and control their future. They want the power to predict, not the preordained unfolding of events, which does not exist, but what will happen if society selects one course of action over another."

Edmund O. Wilson (1988: 181)

The welfare reforms of the 1990s provided an excellent opportunity for the social sciences to help our society make an important policy choice. Poverty and welfare have been extensively studied by many of the country's best social scientists (Jencks and Peterson, 1991; Bane and Ellwood, 1994; Wilson, 1996). The Clinton administration recruited several outstanding social scientists, including Peter Edelman, Mary Jo Bane, Daniel Ellwood and Wendell Primus, to help formulate its policies. There was a strong political consensus that reform was necessary, several states passed reforms in the early 1990s under federal waivers, and in 1996 a comprehensive federal reform was passed (Bryner, 1998).

But the reforms the welfare experts wanted were not politically feasible, and they somberly warned of dire consequences if the Clinton administration’s compromise was adopted. Edelman, Bane, Ellwood and Primus resigned their government jobs in protest against the Clinton administration's compromise. In an article called "The Worst Thing Bill Clinton Has Done"(The Atlantic Monthly, March 1997), Peter Edelman predicted dire consequences:

there will be suffering. Some of the damage will be obvious -- more homelessness, for example, with more demand on already strapped shelters and soup kitchens. The ensuing problems will also appear as increases in the incidence of other problems, directly but perhaps not provably owing to the impact of the welfare bill. There will be more malnutrition and more crime, increased infant mortality, and increased drug and alcohol abuse. There will be increased family violence and abuse against children and women, and a consequent significant spillover of the problem into the already overloaded child-welfare system and battered-women's shelters. One of the most respected think tanks, the Urban Institute, issued a highly publicized report warning that 1.1 million children were in danger of being thrown into poverty by the 1996 federal legislation (Urban Institute, 1996). Secretary of Human Services Donna Shalala personally hand delivered a copy of the report to President Clinton, in the hopes of persuading him not to sign the reform. Welfare advocates organized massive demonstrations and letter writing campaigns, protesting what they saw as a cruel assault on America's most vulnerable citizens.

But the lobbying was to no avail. President Clinton and the other advocates of reform were not persuaded that the social sciences could foretell the future. Congress and the general public seemed more receptive to the arguments of conservative critics such as Martin Anderson (1978), Charles Murray (1994), Robert Rector (1995) and Michael Tanner (1996) who argued that the welfare system exacerbated poverty rather than ameliorating it. These authors had not done nearly as much original research as the professional welfare researchers, and the tone of their writing is more polemical. Yet many legislators thought their policy judgment was better, perhaps because they made more realistic assumptions about human nature.

It is too early for a definitive assessment of the impact of the 1996 welfare reform. Some of the liberals continue to predict disaster (Center on Hunger and Poverty, 1998). It will be several years before definitive statistical data are available, and even then analysts will disagree about its significance. But policy makers cannot wait until the final statistics are in. They must respond to information they receive from journalists and human service workers in the field. So far, that information has generally been positive, and President Clinton has been pleased with the news.

The newspapers are filled with reports such as a New York Times interview with Marcia Harris, a caseworker in Brooklyn's Southside neighborhood. The Times reporter finds that "clearly, an immediate disaster has not occurred. Ms. Harris has not recommended that a single Southside child go into foster care because women unable to cope with new welfare requirements or cutbacks had put their children in danger. And she has seen striking instances in which mothers have risen to the challenges." (January 4, 1988, page 17). The 1998 Wall Street Journal Almanac, summarizing the national reports, finds that "the first year of the nation's experiment with welfare reform brought encouraging results, as the number of people on welfare continued to decline… Preliminary data in states such as Iowa and Wisconsin, which are ahead of most in forcing recipients from the rolls, showed little evidence of homelessness or destitution among former beneficiaries. Instead, many former recipients are reporting incomes that approximate levels they were receiving while on public aid." (page 142).

As always, the social scientific data is slower than the journalistic reports, but the latest statistics I can find confirm the journalistic impression. The most potent policy change, welfare time limits, was not retroactive and had been applied statewide only in Connecticut and Texas as of March, 1997. Wisconsin and several other states had applied time limits in parts of the state under waivers. Nevertheless, caseloads, have declined substantially, with major declines in many states which have yet to impose time limits (29% in New Jersey, 32% in Pennsylvania and 17% in New York). Nationwide, the number of recipients was down 20%, and the number of families 19%, between August, 1996, when the reform was passed, and September, 1997, the most recent figures available. (Statistics are from the homepage of the Administration for Children and Families, http://www.acf.dhhs.gov/ ).

The poverty statistics are slower here than they are for caseloads, and as yet there are no data for 1997. The most recent data show that, "the poverty rate for people residing inside central cities declined significantly between 1995 and1996, from 20.6 (+/-0.8) percent to 19.6 (+/-0.8) percent." (Poverty in the United States: 1996 by Leatha Lamison-White, U.S. Census Bureau).

The decline in crime has been remarkable nationwide, for reasons which are much debated. Certainly there is no sign of the increase that Peter Edelman anticipated. In Camden, NJ, a city with a very high welfare concentration, crime has been sharply down in the last few years, according to recent figures we have obtained from the Camden police department. Most striking is the decline in domestic crimes, which some had thought welfare reform would exacerbate..

It seems more and more likely that the welfare researchers will be proven wrong in their warnings of disaster. But why? Certainly, no one questions their integrity or motivations. One can only second E. J. Dionne's (Philadelphia Inquirer, September 18, 1996) praise of their intentions: "they remind us that there are people who come to politics not to make money, not to get famous, but to try to move government policy in directions they believe to be in the public interest." But good intentions are not enough. From social scientists, we need the best possible information about what will happen if society chooses one course of action rather than another. And by this standard, welfare research has room for improvement.

This paper seeks to contribute to that process by offering a critical analysis of some of the problems with welfare research, and by offering some suggestions for improvement. Two case studies are discussed in some depth as illustrative of broader problems: the evaluation of the New Jersey "family cap" welfare reform and the Urban Institute's 1996 forecast.

Limitations of Standard Research Designs. Most welfare research is executed carefully and professionally, often with good samples and thorough statistical analysis. The problems are usually not in the execution of the studies, but in limitations imposed by the research designs. Welfare research follows a number of typical research designs, many of which have serious weaknesses:

? Evaluations of specific welfare programs. This is probably the most common kind of welfare research, because funding sources usually require that interventions be evaluated. Usually, however, the researchers find that the variations in programs have little impact because of the overwhelming influence of the social and economic environment in which the participants live. For social scientists, this should not be surprising. If social science theories are correct, it is unlikely that peoples' lives can be transformed by changing a few rules in a social program or by offering them some additional services, without changing the social environment in which they live. While these studies are valid for the contexts and programs in which they were done, one cannot generalize from them to predict the impact of much larger and more powerful policy changes.

? Comparisons of states. A good deal of the research has examined the consequences of policy differences between states. The problem with this approach was that, at least until 1996, the policies in the various states usually did not differ enough to provide information about what would happen if major changes were imposed, such as time limits. This approach may be more productive after 1996, since the policies in the various states are much more varied than they were in the past. And the Urban Institute has secured $30 million in funding for a comprehensive database on state policies and trends. A fundamental difficulty here, however, is that states are not homogeneous in socio-economic terms. The primary goal of welfare reform is to help impoverished ghetto communities, and state-by-state comparisons are not the best way to study these communities.

? Trend analyses of statistical data. These analyses usually rely on national data (or sometimes state data) instead of focusing on the communities of most concern, where the data are more limited and their deficiencies more obvious. One problem with these studies is that policies were consistent for a long time, which made it impossible to assess what would have happened if policies had changed more. This research method may be very useful in assessing the impact of the more radical changes in the 1990s. Another fundamental problem, which is seldom seriously addressed, is the fact that poverty statistics depend on income data gathered by survey interviewers. This information is likely to be highly inaccurate for welfare dependent communities where residents have reasons to conceal their incomes. Martin Anderson drew our attention to the inaccuracy of income data on the welfare population in 1978, and pointed out that everyone continues to use the official figures because they are available and are legitimated by being "official." This continues to be the case. A National Research Council (1995, p. 403) review of the four major government income surveys reports that "all four surveys clearly experience net underreporting of income." Kathryn Edin and Laura Lein's (1997) outstanding ethnographic study of the finances of single mothers in ghetto communities shows that "almost all poor single mothers supplement their regular income with some combination of off-the-books employment and money from relatives, lovers and the fathers of their children... Secretiveness is especially common among welfare recipients, almost all of whom have non-welfare income that they conceal from the welfare department"(from the forward by Christopher Jencks, p. xi). Yet, all of our "official" poverty statistics are based on the assumption that these women give honest answers to survey interviewers.

? Ethnographic Studies. Qualitative, anecdotal accounts based on intensive interviews and observations in ghetto communities are very helpful for giving insight into the lifestyles and mental sets of ghetto residents. But the samples are small and not statistically representative, and researchers tend to write about those recipients who fit their ideological agenda. Welfare advocates studiously ignore the interesting and insightful ethnographies of welfare critics such as Leon Dash (see his Rosa Lee and When Children Want Children), while conservatives ignore the liberal ethnographies. The welfare population is so diverse that one can usually find ethnographic evidence to illustrate any point one wishes to make.

? Community Studies. Comprehensive studies of ghetto communities, combining quantitative and qualitative data, are very rare, although this is the best research design for understanding the complexity of poverty in the setting where it occurs. William Julius Wilson's When Work Disappears, is a good example, because it focuses on poor neighborhoods in Chicago and uses both interview and statistical data. Wilson argues that the loss of jobs in old industrial cities is the "root cause" of the persistence of poverty in affluent societies. The concept of "root cause," however, is misleading. Social phenomena have multiple causes, and once they are established they may continue even when the "root cause" disappears. In our work on Atlantic City (Society Magazine, May/June 1997), John Cosby and I found that a massive influx of good jobs, to the point where employers were desperate for workers, had remarkably little impact on the poverty of a city with an entrenched ghetto culture. A combination of jobs with welfare reform and social services, however, has had more success.

Research on the community level, combining qualitative and quantitative methods, and using time series as well as cross-sectional data, give us the best chance of understanding the complex dynamics of poverty and social change where it really matters.

Why Experiments Fail: The New Jersey Case. Experiments have inherent limitations that make it highly unlikely that they will answer the important questions about welfare policy. This is an unfortunate conclusion, since, in principle, experimental designs provide the most rigorous test of causal hypotheses. But excellent researchers have conducted welfare experiments, and the results have been disappointing for several reasons.

? It is impossible to isolate participants in an experiment from changes in the larger community and society, which are likely to be more important than the variables that can be experimentally manipulated.

? Experiments take a long time, and by the time results are ready the policy context is likely to have changed.

? There are ethical and practical difficulties in experimenting with peoples' lives, which can often be overcome only by limiting the effectiveness of the experimental intervention.

? Being a participant in an experiment may change peoples' behavior simply because they know they are being observed - the "Hawthorne effect".

The New Jersey welfare experiment, which is still ongoing, provides an excellent example of these problems. An ambitious research design was mandated by the federal government as part of the waiver process that authorized New Jersey to deviate from the AFDC rules in effect when the reform was passed in 1992. The centerpiece of this design was a rigorous experiment, which exempted a random sample of recipients from the provisions of the reform. These women (called the "control group") were to be compared with a larger matched sample of women who experienced the full provisions of the New Jersey reform.

The New Jersey reform attracted considerable national attention because of the "family cap" provision, which provided that women who conceived an additional child while on welfare would not receive an increase in their cash grant. The reform was, however, a complex piece of social legislation (Goertzel and Hart, 1995), which changed a number of important policies at the same time, making it very difficult, if not impossible, to statistically distinguish any impacts of the "family cap" from the effects of other provisions.

The experimental design, however, proved incapable of detecting the impacts of the reform as a whole, let alone its specific provisions. Despite the best of intentions, the researchers were unable to maintain the integrity of the research design. The Division of Family Development "discovered a few instances where cash benefits were capped for cases in the control group after the birth of another child." (Camasso, Harvey and Jagannathan, 1996, p. 10). This was apparently a small number of cases, and it might have been possible to adjust for the damage statistically, although it undoubtedly blurred the distinction between "experimental" and "control" groups in some participants' minds.

More critically, a survey of participants showed that "62% of the control group subjects who participated in the survey did not recognize their control group status." Although this was explained to them at the beginning of the experiment, and they signed a form saying that they understood, they apparently viewed it as just another form they had to sign so they could get benefits. The researchers quite properly concluded that "we have no way of ensuring that this small group of clients (fewer than 4,500 cases) would base their actions and decisions on the premise that they were not affected by the Family Development Program, especially in view of the wide publicity given to the "family cap" provision." (Camasso, Harvey and Jagannathan, 1996, p. 11).

Properly implementing an experiment of this magnitude requires an exceptional effort on the part of a state's welfare caseworkers, who are already overburdened with administrative tasks and who are often more concerned with getting benefits to their clients than with carrying out research projects. Even if the caseworkers made the exceptional effort required to assure that every participant continued to be aware of the provisions governing her participation, however, the experiment would be of doubtful validity. The women who participated in the experiment would be in a quite different situation from that of the large majority of welfare recipients, who would receive no such frequent reminders from their caseworkers. The New Jersey experiment thus lacked both internal validity (the integrity of the experimental procedures) and external validity (the ability to generalize to women not participating in the experiment).

The negative results of the experiment - no significant difference between experimental and control groups - prove nothing other than the failure of the research design. The time series data, on the other hand, showed a remarkable decline in births to welfare mothers as well as a sharp decline in welfare caseloads and new welfare cases in the state. Recently released data from the New Jersey Department of Health shows a sharp decline in the birth rate in New Jersey's poorest cities. There has also been a statewide decline in births to young teenagers, and especially minority teenagers. There has continued to be a steady decline in infant mortality. Employment is up, as it is throughout the nation. Thus, all the social indicators are positive and the reform is generally considered to be successful by policy makers and state officials responsible for administering it.

Perhaps surprisingly, the New Jersey welfare reform was also considered to be effective by the women participating in it. In the survey which was part of the evaluation research, most of the respondents said they knew about the "family cap" provision, and 66.5% agreed with the statement "do you think that this family cap, as you understand it, is a fair rule?" Fully 86.1% agreed that "the family cap encourages people to take responsibility for their actions," while only 36.5% agreed with the reform's opponents that "the family cap interferes with the woman's right to have a baby." Most respondents said that they were not personally affected by the "family cap" since they did not plan on having any more children, but they felt that it was a good policy which would stop a lot of teen pregnancies.

Unfortunately, these findings were not available when the U.S. Congress was debating imposing the "family cap" as a national requirement, and there were urgent pressures for release of findings from the experiment. A congressional committee asked for data, and was given some inconclusive preliminary findings, and made a decision to leave the "family cap" issue up to the states to decide.

Ironically, the vigorous opposition to the "family cap" by welfare advocates was an essential factor in its success. It drew tremendous attention to the measure, causing many women to exaggerate its importance, and causing a general belief that welfare was changing and could no longer be relied upon. Gary Young and I (Goertzel and Young, 1996) found that there was a marked downturn in welfare caseloads and births to welfare mothers as soon as the measure was passed, almost a year before it legally took effect.

The "family cap" was a hot issue when the experiment was designed, but the policy environment changed so quickly that it seemed minor by the time the first results were available. With the imposition of welfare time limits, women knew that they would have to get off welfare and become self-supporting long before any new children were grown. This reality completely overwhelmed concerns about the "family cap" in their minds. By the time data from the experiment began to become available, the issue was of little importance. The experiment will apparently continue for several more years, at considerable expense to the taxpayers, despite the fact that it no longer seems very relevant.

Why Predictions Fail: The Urban Institute Case.

On August 1, 1996, Senator Pete Domenici predicted that the new welfare reform would "encourage and make people work," reestablish the idea that "parents should take care of their children," and "change the culture of welfare." Senator Daniel Patrick Moynihan, who had opposed the reform, vehemently disagreed, predicting that it "would substantially increase poverty and destitution," and cause "2.6 million persons to fall below the poverty line." (Bryner, 1998, pp. 172-173, quoting from the Congressional Record). So far, Senator Domenici's more hopeful prediction fortunately seems to be closer to the mark. But Senator Moynihan is one of the few social scientists in the U.S. Senate, and its leading expert on welfare issues. His forecast was based on the most authoritative social science estimate available at the time. In a highly publicized report, distributed through press releases and the World Wide Web, the Urban Institute had argued that: the proposed welfare reform changes would increase poverty and reduce incomes of families in the lowest income group... We estimate that 2.6 million more persons would fall below the poverty line as a result, including 1.1 million children. (Urban Institute, 1996, p. 1)

Senators who supported the reform dismissed the Urban Institute's analysis as advocacy in disguise. Many of them undoubtedly remembered the Club of Rome's 1972 warning that the world was running out of oil, which was based on a similar methodology. Both warnings sprung from labyrinthine computer programs tended by prestigious scientists. The 1972 report frightened a susceptible public because almost no one understood it well enough to refute it, and because mishandling of the Arab oil embargo had created a short-term shortage of gasoline. By 1996, computer modeling had lost its mystique, and "garbage in, garbage out" had become part of the conventional wisdom. Welfare advocates rallied around the prediction, even distorting it on placards in demonstrations that claimed that a million children would starve if the reform were passed. But policy makers were not persuaded.

The computers that make these projections are programmed with a maze of interlocking equations that do their best to simulate the complexity of the real world. This is an admirable undertaking, going beyond the limitations of studies that isolate a few variables in a complex system. The Urban Institute's "microsimulation model" is a marvel of mathematical modeling. It begins with data from the approximately 60,000 households interviewed each march in the Census Bureau's Current Population Survey. This, unfortunately, may be a "garbage in" error since the data is known to be inaccurate for welfare recipients who have every incentive to understate their income. The computer takes this imperfect data from each family and computes economic outcome for the society as a whole. If the data and the equations were perfect, the results of the calculations would match up exactly with the real world. But they never do, so the analysts "align" the model until the results come out right.

In a technical paper published in 1989, two of the authors of the Urban Institute's model candidly stated that alignment is necessary because otherwise, "the discrepancy between the simulated and actual figures could cause some users of model results to suspect the model's overall validity." They further warn that "clearly, alignment is not an exact science. Alignment choices will depend on a model's input data base, the scope of the simulation, the way a model is used, and a modeler's individual preferences." (Giannarelli and Young, 1989).

Microsimulation modeling works, to the extent that it does, because it is carefully aligned to make things come out right. The Urban Institute does this about as well as it can be done, so their results can be interpreted as saying "if this reform were to be implemented right now, with only certain specified changes such as an increase in employment by single mothers, these would be the consequences." The point of welfare reform, however, is to cause much more significant social changes. When it succeeds, the outcomes are quite different.

If welfare time limits encourage more young couples to marry before having babies, the poverty rate among their children will go down. If the reforms cause more young women to delay pregnancy until they have finished school, or to defer having an additional child until they and their partner can support it, poverty among their children will go down. These changes are quite possible. Indeed, there is good evidence that they are already happening. In 1995, the rate of births among unmarried women declined for the first time in nearly two decades, according to the National Center for Health Statistics. The teen-age birth rate also dropped, for the fourth year in a row. (Holmes, 1996).

This data was available at the time, but the Urban Institute's scientists decided not to include estimates of "behavioral changes" in their computer model. Study director Sheila Zedlewski explained that, "we did not incorporate marriage and fertility responses because there is no consistent, accepted evidence in the empirical literature that AFDC affects these behaviors"(personal email communication). Since changes of this sort have not been tried before, of course there is no conclusive evidence of their impact. In effect, the Urban Institute's researchers said, "since we assume that the reform will fail in its major social objectives, and things will stay much as they are, we predict that it will be harmful." By insisting on incorporating only findings that had been rigorously demonstrated in the past, they guaranteed a negative outcome.

From a forecasting point of view, the Urban Institute's study was defective because it included no testable numerical predictions. At no point in time will it be possible to say definitively whether their prediction was correct or not, since it was actually a statement about the future behavior of their model rather than about trends that can be observed in real life. When they said that poverty among children would increase, what they actually meant is that it would be higher than their model tells them it would have been without the reform. Without making it explicit, they were actually making a ceteris paribus argument, a statement of what would happen if nothing else were to change.

This is an understandable, even a necessary, approach, since the poverty rate depends on variables that are outside the scope of their analysis, such as the unemployment rate. But forecasting cannot improve unless welfare forecasters, like weather and economic forecasters, submit their predictions to empirical tests. The Urban Institute researchers could have issued a set of conditional forecasts, with different predictions depending on the behavior of the economy or other factors. If they had done so, their prediction would be testable by other researchers. Since they did not do so, their prediction is technically unfalsifiable. Only they can tell us how much alignment it will take to make the model fit the trends that have actually occurred.

Predicting Policy Impacts. If social science is to be useful in guiding policy decisions, we must be able to make useful predictions of consequences. If these predictions are to have scientific value, they must be testable. Writers often fudge this point. When they make their predictions, they seem like straightforward predictions. When they do not work out, however, they point out conditions and qualifications that were not explicit at the time. For example, Peter Edelman's predictions, quoted at the beginning of this paper, seem straightforward enough. He seems to be saying that infant mortality and crime, for example, will go up in response to the 1996 welfare reform bill. A careful reading of his article, however, does not reveal whether he meant that these problems would get worse compared to the past, or if he meant that they would be worse than they would have been without the reform. The first would be easily testable, but it is highly unlikely to come true. Infant mortality has been declining quite steadily for many years, a trend which is probably too strong to be deflected by welfare changes. Rates of violent crime are decreasing markedly around the nation, again for reasons that are much broader than welfare policy.

Edmund O. Wilson, however, did not challenge the social sciences to predict the future, which we clearly cannot do, but only to advise as to the impact of policy choices. Ceteris paribus predictions are appropriate for this purpose, but only if they are testable. And they cannot be tested in the absence of a full causal model that enables us to account for all of the other variables involved. Without such a model, it will always be too easy to make excuses, such as "my prediction of disaster would have been realized, except for the fact that unemployment is low." This kind of argument would be fine if the analyst had stated, in advance, "welfare reform will be a disaster in an epoch of high unemployment, but will be successful in the context of a good labor market."

Making these kinds of predictions requires looking at policy in the context of broad historical trends, such as those in the following chart. Understanding the turning points in these trends is difficult, and requires a good deal of historical knowledge. It also involves some very controversial judgments about the direction of causal influences.

For example, Paul Peterson (Jencks and Peterson, 1991, p. 11) has observed that, "poverty in the United States had been declining steadily between 1940 and 1960, two decades in which the welfare state expanded hardly at all. Yet when the welfare state expanded in the 1970s, progress toward eliminating poverty came to a halt. What is more, poverty increased among young families and inner-city residents."

The massive increase in welfare caseloads in the late 1960s came about because of legal and social activism, which took AFDC out of the hands of social workers and made it an entitlement (Bane and Ellwood, 1994). As soon as that happened, the decline in the percentage of children in poverty ended. This is the kind of observation which welfare opponents, such as Charles Murray, use to argue that welfare causes poverty. Of course, this depends on income statistics that are questionable - it may simply be that more people found it paid to conceal their income. Infant mortality and other health indicators continued to improve during this period.

Of course, other variables may account for the increase in poverty. There was a leveling off of the median real family income and of the real income of the families at the 20th percentile point. Poverty also varies with the unemployment rate, as one would expect, although this does not account for the dramatic turning point in poverty in 1968. These dramatic historical turning points may also reflect general changes in the zeitgeist. Strauss and Howe's (1991) analysis of generational shifts in American culture suggests that 1968 was the coming of age of the idealistic, crusading baby boom generation, which provided the political force behind the social programs which are being questioned by today's "reactive," cynical Generation X . Of all the writers on poverty, the only one I know of who associated it with the 1960s generational culture was Myron Magnet, in his The Dream and the Nightmare: The Sixties Legacy to the Underclass. He argues that most middle class youth had the resources and support to recover quickly from their experimentation with countercultural values, but many underclass ghetto residents remained mired in self-defeating antiestablishment, antiwork values.

If this generational analysis is correct, we may be reaching a turning point in the mid 1990s when a number of social trends will change at once. Employment and incomes will rise, and poverty will resume the downward trend, which was cut off with the increase in welfare dependency at the 1968 to 1971 turning point. The much remarked upon decline in crime, for example, can be seen as part of this generational shift. This success in lowering crime rates has come as quite a surprise to the nation's criminal justice researchers. In the latest edition of a standard text Samuel Walker (1998, p. xxi) states the paradox: "The basic argument of this book, from the first edition to the present one, is that popular crime control policies do not work... Yet, if they do not work, then how do we explain the reduction in crime?"

In welfare research as in criminal justice research, experimental evaluations of specific programs do not seem to provide a valid basis for predicting the effects of major policy changes. All the research shows that nothing works; yet things suddenly get better when there is a generational shift in the zeitgeist. Of course, the politicians who happen to be in power when the shift occurs are quick to take credit for the solving problems that seemed intractable to their predecessors. It may seem unfair that the Mayor Giuliani gets credit for the decline in crime in New York City or President Clinton gets credit for the national decline in welfare dependency. But timing is important in politics and in public policy, and fame and fortune come to those who anticipate trends. This is an arena where social scientists are often at a disadvantage, because their laudable insistence on validating their findings with solid data and research means their focus is on the events of a few years ago rather than the emerging trends.

Some Suggestions for Future Research. To be more useful for policy makers, research must focus on the community and societal impact of reform, not just on the effect on a specific cohort of recipients. The most important impact of reform may be in discouraging women from premature pregnancies and beginning a period of welfare dependency. The following suggestions may be helpful:

Works Cited

Anderson, Martin. Welfare: The Political Economy of Welfare Reform in the United States. Hoover Institution, 1978.

Bane, Mary Jo and David Ellwood. Welfare Realities. Harvard University Press, 1994.

Bryner, Gary. Politics and Morality: The Great American Welfare Debate. Norton, 1988.

Camasso, Michael, Carol Harvey and Radha Jagannathan. An Interim Report on the Impact of New Jersey’s Family Development Program. Submitted to the State of New Jersey, Department of Human Services, July 29, 1996.

Center on Hunger and Poverty. Are States Improving the Lives of Poor Families? Tufts University, 1997.

Edin, Kathryn and Laura Lein. Making Ends Meet. Russell Sage, 1997.

Giannarelli, Linda and Nathan Young, "Alignment of Microsimulation Models to Target Data," Proceedings of the Social Statistics Section of the 1989 Meetings of the American Statistical Association, pp. 300, 303.

Goertzel, Ted and Gary Young. "New Jersey's Experiment in Welfare Reform," The Public Interest, fall 1996, pp. 72-80.

Goertzel, Ted and John Hart. "The Politics of Welfare Reform in New Jersey," in The Politics of Welfare Reform (Donald Norris, ed.), SAGE Publications, 1995.

Steven Holmes, "Study Finds Decline in Teen-Age Births," New York Times, October 5, 1996.

Jencks, Christopher and Paul Peterson, eds. The Urban Underclass. Brookings Institution, 1991.

Magnet, Myron. The Dream and the Nightmare: The Sixties Legacy to the Underclass, Morrow, 1992.

Murray, Charles. Losing Ground. Basic Books, 1984.

Rector, Robert and William Lauber. America's Failed $5.4 Trillion War on Poverty. Heritage Foundation, 1995.

Strauss, William and Neil Howe. Generations: the history of America's future, 1584-2069, Morrow, 1990.

Tanner, Michael. The End of Welfare. Cato Institute 1996.

Urban Institute. Potential Effects of Congressional Welfare Reform Legislation on Family Incomes. 1996. Posted on http://www.urban.org.

Walker, Samuel. Sense and Nonsense about Crime and Drugs, 4th edition. West Wadsworth, 1998.

Wilson, William Julius. When Work Disappears: The World of the New Urban Poor. Knopf, 1996.

Wilson, Edmund O. Consilience: The Unity of Knowledge. Knopf, 1988.