Michael Tanner’s Liberty Forum essay provides a characteristically insightful and level-headed overview of the 1996 welfare reform and its consequences. He admirably clears away the rhetorical fog that still envelopes our debates regarding welfare, and does not let either side get away with much. “Welfare reform was neither the disaster that its critics feared nor the success its supporters claim,” he concludes. “It was a step in the right direction—a small one, and one from which we can learn.” And he offers some key lessons we might build upon.

His proposals make very good sense. But I think they evince too much confidence in our knowledge of what sorts of reforms can avail us in confronting the kind of entrenched poverty that has most confounded our anti-poverty efforts in recent decades, and so they might lean too lightly on local experimentation. The next time our political system is again prepared to undertake a reform of federal welfare policy—which, let us be frank, may be some time from now, given contemporary political circumstances—the organizing principle of that reform should be decentralization and ongoing experimentation.

The case for greater welfare decentralization is not self-evident, to be sure. Decentralization and the empowerment of state and local governments have not always been favored by welfare reformers, even on the Right. Tanner himself evinces skepticism, accepting the federalism implicit in House Speaker Paul Ryan’s and Senator Marco Rubio’s recent reform proposals only grudgingly, as unavoidable concessions to political reality. And one of his criticisms of the now two-decades-old Personal Responsibility and Work Opportunity Reconciliation Act is that its funding stream was funneled through the states rather than directly to individuals.

This skepticism is reasonable. If you’ve ever been to Trenton, New Jersey, you can’t be too impressed with state governments. And in some important respects, the states are likely to be even worse than Washington at managing anti-poverty programs: State policymakers and administrators are frequently more corrupt, more subject to interest-group pressure, less professional, and less transparent than their federal counterparts. Conservative champions of federalism should not skirt these realities. So why should welfare reform involve greater decentralization?

The core of the answer, simply put, is that we don’t know how best to address the kind of entrenched poverty that poses the most serious challenges to our welfare system—and the welfare reform of the 1990s did not go very far to remedy our lack of knowledge.

There are many ways to conceive of the knowledge gap confronting welfare reformers, but the four-part typology of poverty offered by Robert Woodson, founder and president of the Center for Neighborhood Enterprise, provides one useful window on the problem. Acknowledging that any typology will of course be too crude to fully describe such a complex problem, Woodson posits a general outline of four distinct cohorts of Americans confronting poverty that he has encountered in his decades of anti-poverty work:

There is one cohort whose poverty is the result of an unexpected setback, such as the death of a breadwinner or the loss of a job. For these people, the welfare system can function as originally intended, providing temporary support until recipients can find their footing again.

A second cohort comprises those who have remained dependent on the system because the disincentives to marry and work embedded in its regulations make it a rational choice to avoid those stepping stones to self-sufficiency. They have “done the math” and calculated that it is not worth the loss of benefits to take the first steps toward upward mobility.

The third group is made up of the disabled, many of whom will always be in need of some support.

The fourth cohort consists of those in poverty because of the choices they make and the chances they take—for example, those suffering from alcoholism and other addictions, who choose to live with the consequences rather than pursue recovery.

No doubt there are many complicated combinations of these sources and causes of poverty, and there are other forms, too. I would, at the very least, add to this typology the many Americans who find themselves in the fourth cohort because they were born into impoverished and broken families and communities, and have had few models to follow and few opportunities or advantages from the very beginning. It would not be fair to say they remain poor because of choices they’ve made, but getting out of poverty would in their cases require both material resources they lack and a mix of discipline and wise decisionmaking that would be very difficult for anyone to muster, and all the more so for men and women who have rarely seen it done.

However we might extend and complicate Woodson’s typology, though, its basic categories certainly begin to suggest the challenge facing antipoverty efforts. Neither cash benefits nor many of our in-kind benefit programs can address such different forms and causes of poverty equally well, and efforts to better address one form can easily undermine efforts directed at other forms.

Moreover, Woodson’s fourth cohort—especially if it were extended as just suggested—poses the most complex challenge of all, and tends to resist even the best-conceived and most well-meaning efforts to help. In many parts of our country, broken communities, structural disadvantages, and poor choices interact to yield deeply entrenched and often inter-generational poverty that bespeaks far more than a shortage of money. And such complex dysfunction rarely looks the same across different places and circumstances. To paraphrase Tolstoy, healthy communities are often healthy in the same ways, but every broken, impoverished community is broken and impoverished in its own particular way.

Some targeted reforms of the welfare system can certainly improve that system’s ability to help the first and second of Woodson’s four cohorts. Changes to our disability programs could also improve our ability to offer meaningful assistance to the third. But it is the fourth cohort, whose members confront the most complex and deeply rooted problems of poverty, that is least well served by our existing system. And it is not easy to see how an extension of the logic of the welfare reform of the 1990s (which was particularly well-targeted to reach the second cohort Woodson describes) could by itself make a meaningful difference in the lives of those locked in that most pernicious cycle of poverty and brokenness.

A welfare reform that could reach that fourth group would need to look rather different. But we are unlikely to find any formula that would suffice, because by their nature the problems confronting these especially disadvantaged Americans are intertwined with specific local challenges and circumstances: from racial prejudice and ethnic animosities, to displacement of local economies driven by global economic trends, to the ravages of drugs and alcohol and more. The problem is not simply that federal reformers do not now possess the knowledge required to design a system that could address these difficulties, it is that such a system is essentially impossible in principle at the federal level because such knowledge is impossible to achieve and sustain in a centralized way in a society as vast as ours.

And knowledge, not just money, is required. After all, the challenge facing welfare reformers seeking to help this fourth cohort of Americans living in poverty is daunting: They have to find ways to help people who lack not only money but often also stable families, functional communities, and decent schools. They have to encourage work and responsibility while offering aid, and they often have to help people break bad habits or confront addiction or abuse while also respecting their dignity and independence. This can’t be done by a government check. Welfare in these circumstances often works best when it is accompanied by advice, by obligations, and by evident compassion at a personal level.

This requires neither a simple transfer of funds nor the right set of in-kind benefits. It requires help tailored to its beneficiaries and delivered to them with a human touch. So rather than generalized applications of centralized knowledge, it requires discrete applications of local knowledge to local circumstances—which would have to begin as experiments with different approaches to welfare assistance directed to specific instances of entrenched poverty.

Using public resources to let different institutions—from state social service agencies to local civic groups to churches and nonprofits—try different ways of meeting this daunting challenge in different circumstances is the logical response to the sheer complexity of the problem. Experimentation is what you do when you haven’t found the right solution and when it isn’t clear that any one solution will suffice.

The model of the 1996 welfare reform could well be built upon to better align the incentives created by some of our other welfare programs, and so to better help those Americans relegated to persistent poverty by the perverse incentives of our welfare system itself. Some degree of consolidation of benefits, and the transformation of some in-kind benefits into cash payments in some cases, could also help some of those who now rely on the welfare system for temporary assistance in confronting discrete and unanticipated life emergencies or economic downturns.

But when it comes to the most complex, persistent, entrenched, and unjust forms of poverty, the model of direct cash benefits will not suffice—and would, indeed, tend to encourage a perverse social model. Direct federal cash payments to individuals tend to isolate individuals from their communities and from responsibilities to others. They encourage an understanding of society as consisting of isolated individuals and a distant national government. But the most complex and entrenched forms of poverty often require the intervention of precisely the kinds of institutions that stand between the individual and the government, which is to say the mediating layers of our society.

Our welfare system should not seek to displace or render moot these mediating institutions, because they serve more and greater ends than the mere distribution of resources. By giving them some role in providing welfare assistance, and by allowing them to define that role through experimentation with different approaches to aid, our welfare system can help revive and reinforce these mediating layers. In impoverished communities, they are often badly in need of reinforcement.

By the same token, a direct relationship between such institutions—from churches and charities to private businesses, schools, and community groups—and the federal government would be neither wise nor practical. We would do better to allow state and local governments to further mediate that relationship, and therefore to limit the federal role to the provision of a distinct, predetermined level of assistance, which would flow through the states (for all their faults) and be deployed both by both state and local government agencies and by those institutions of civil society that seek to play such a role and prove themselves worthy of it.

The elements of our welfare system that now provide funds to meet fairly straightforward shortages of money should certainly be transformed into direct cash transfers wherever possible. The Medicaid program is one obvious target for such a transformation: It should become a premium-support benefit for the poor, allowing them to access the same private markets for health insurance as other Americans, ideally in the course of a larger reform that replaced Obamacare with a refundable tax credit for coverage purchased in a competitive insurance market regulated by each state.

The rest of our federal welfare system—with perhaps the exception of some benefits intended specifically for Americans with disabilities, but including the very great bulk of today’s in-kind federally funded benefits—should be replaced by a predetermined and predictable federal payment to the states, along the lines of the Flex Fund proposed by Senator Rubio (and noted by Tanner). And states should be encouraged to experiment with the provision of assistance—without being limited to any particular approach or form of benefit. The question of in-kind benefits versus cash payments or income supports, for instance, would not need to be resolved in some ultimate way at the federal level.

Some states will undoubtedly use these resources poorly, and some experiments will surely fail to yield meaningful improvements in the prospects of the poor. Trial and error involves error. But it is nonetheless the best way to seek practical knowledge we do not already possess, and to make possible ongoing, incremental improvements.

The next generation of welfare reformers will need to understand this as among their chief purposes: to discover and develop knowledge we do not yet have about how to help the poor and combat entrenched disadvantage. Setting that kind of goal does not come easily to policymakers, who are more inclined to seek ways to apply and impose what they deem to be their already superior expertise.

Moreover, an experimentally-minded approach to welfare reform will need to not only begin with the premise that we don’t have all the answers, but also to accept the proposition that no approach to combating entrenched poverty will work everywhere or all the time. It will need to grant that local solutions are not merely “pilots” for ideas that will someday be made universal if they work but are sometimes, and perhaps most of the time, simply what works here and now.

A 21st century welfare system will be diverse by design, and will be more successful the more diverse it is. It will aim not for uniformity but for results. Like the rest of our 21st century economy, it will be comfortable with fragmentation and with treating a vast array of options as itself a working answer, not merely as the path to a one-size-fits-all set of mandates.

Such a system would accommodate itself to the immense diversity of our vast nation, and so would aim to function from the bottom up whenever possible. It would seek to clarify and separate the roles of different levels of government and of the private sector—rather than confounding roles and responsibilities in a counterproductive “cooperative federalism” that ends up treating state governments, local governments, and private actors as mere appendages of a federal apparatus.

Diversity and subsidiarity are what modernization consists of in today’s society, and it is what the next welfare reform should consist of. This would make it far more ambitious than the reform undertaken in 1996—and could offer a model for public policy well beyond welfare.