In her 2017 Ely lecture, Esther Duflo adroitly noted that the model economist must be willing to be an “economist-plumber…in the pursuit of good implementation of public policy, she is willing to tinker.” This mentality has been absent in the conversation around the current US administration’s suggestion that the Supplemental Nutrition Assistance Program (SNAP) switch from vouchers (food stamps) to in-kind transfers (in this case, a literal box of food). Harvest boxes aren’t viable for several reasons—and their suggestion highlights the lack of dispassionate, evidence-based conversations about improving US benefits programs.

There may well be a real benefit to reforming SNAP, just one part of a patchwork of social safety net programs in the United States. The guiding question is in this case is, how can the flow of government support to low-income households most increase welfare, and nutrition, for each dollar spent? Tinkering can produce this knowledge, but tinkering also carries a risk. For example, one recent study found that a seemingly small reduction in monthly SNAP benefits—less than $25—can result in significant nutritional shortfalls for families: no food for a few days. Tinkering must be done using an experimental approach to generate quantitative evidence on a change’s potential benefits, or harm, before any sweeping nationwide changes are implemented. There are reasons to be careful before making  changes in people’s lives, as one change can spill over onto another—researchers have found children score lower on exams the farther they are in the month from receiving their food stamps.

The current administration’s proposal to convert nutritional aid into in-kind transfers is a form of tinkering with benefit restrictiveness. Transfers can be described on a scale of highly restrictive to totally unrestricted—receiving packaged food to receiving cash. In its current incarnation, SNAP vouchers would be considered a type of in-kind transfer, as they can only purchase food. Delivery mechanisms, frequency, conditionality, and benefit duration can vary broadly as well.

Unconditional Cash Transfers (UCT) are the most flexible type of benefit. They provide cash without any restrictions so that families decide how best to allocate resources. The benefit of this is intuitive: money can be anything, while shelf-stable milk can only be shelf-stable milk. UCTs also have the lowest implementation costs and remove the issue of fraud[1] from misuse of restricted transfers. Two recent studies provide evidence that unrestricted transfers have the best average outcome for individual’s economic success. Perhaps equally important in this political environment, they also do not seem to create the negative externalities that many U.S. critics fear and use as a reason to strike down UCTs in public policy.

In Kenya, Innovations for Poverty Action conducted research with the nonprofit GiveDirectly to test the impact of unconditional cash transfers among low-income households in one of the poorest parts of the country. The researchers varied recipient gender, payment timing, and transfer size. In all groups, there were positive impacts—the transfers increased in food security, asset ownership, and consumption, as well as psychological benefits.

In the U.S., recent research on the Alaska Permanent Fund—a yearly cash payment to Alaska residents—didn’t find any reduction in employment, a major concern from skeptics of UCTs. There were increases in part-time work, but it’s unclear whether those came from reduction in hours, or additions to the labor force, or both. In the case of SNAP, this can be understood as families continuing to spend money on food if they received cash (a result found by the USDA study).

Another common critique of unrestrictive cash transfers is that they incentivize harmful behavior such as smoking or drinking. While the purchase of tobacco and alcohol is prohibited by the current SNAP program, this argument is also not supported by evidence from the international context. Cash transfers reduce consumption of temptation goods such as tobacco and alcohol. Of course, the occasional individual could buy beer with a transfer, but the average individual benefits from greater freedom to pursue their goals. This could be paying off expensive debt to free up resources for more nutritious food next month, starting a college savings account, or even coming up with a security deposit for an apartment.

On the other end of the spectrum, there is some evidence that targeted, in-kind, transfers can provide benefits to those that need the services most. However, their effects seem to be highly dependent on the context of the market the goods or services are delivered in. For example, a study of school uniform donations in Kenya found that providing school uniforms reduced school absenteeism by 38 percent in a context where school uniforms were required to attend school and only about 60 percent of students had uniforms. But, in a context where the in-kind donation isn’t relevant, these transfers may not accomplish their goals and can even work against them. Take TOMS donations in rural El Salvador, a middle-income country where the vast majority of recipient children already owned shoes. Even though recipients liked the shoes, there were no significant improvements in foot health.

Until that hypothesis is tested, there is no way to know what that modification could do.

One could imagine that harvest boxes delivered to households in food deserts could help families eat more nutritious meals, in the absence of convenient places to purchase it. Until that hypothesis is tested, there is no way to know what that modification could do. If that was a modification of interest, preliminary research on recipient nutrition should be conducted to find out if that modification would be relevant, especially given the known capacity for harm in benefit reduction.

Ultimately, as Duflo noted, there is no silver bullet to these problems, but there is a solution to the knowledge problem: tinkering and testing modifications at all levels of program implementation. The US government needs to roll up its metaphorical sleeves, define the goals of these programs, and then start testing fixes, big and small. Their fixes must be rigorously evaluated to understand each moving piece—from which individuals qualify for the program to the wording of the communications.

That’s how to make nutritional assistance flow truer. That’s what a plumber does.