What Stockton showed.

In February 2019, the city of Stockton, California began sending $500 a month to 125 randomly selected residents.

No conditions. No work requirements. No documentation of how the money was spent. The money arrived by debit card. Recipients could spend it on groceries, rent, cigarettes, lottery tickets, or anything else. The program — called SEED, for Stockton Economic Empowerment Demonstration — had no opinions about what financial support should be used for.

SEED was designed from the beginning as a randomized controlled trial. A research team led by Stacia West at the University of Tennessee and Amy Castro at the University of Pennsylvania pre-registered outcomes before the first payment was sent. A control group of 200 additional Stockton households was enrolled, matched on demographics, and surveyed on the same schedule as the treatment group. The primary outcomes — employment, income stability, and health — were decided in advance.

The pilot ran for 24 months, through February 2021. The results were published in 2021 and 2023.

What the data showed

The most politically sensitive finding appeared first.

Critics of guaranteed income programs had long argued that unconditional cash transfers would reduce work — that people who received money without working for it would stop working. The SEED data showed the opposite.

At baseline, 28% of the treatment group held full-time employment. After 12 months, that share had risen to 40%. In the control group, the change over the same period was smaller: from 25% to 37%. The 5-percentage-point difference — full-time employment rising faster among people receiving $500 a month than among people who weren't — was statistically significant and replicated at the 24-month follow-up.

The mechanism appeared to be financial stability enabling risk-taking. Recipients reported being more willing to look for better jobs, take on freelance work, or pursue additional training because they had a cushion. The money reduced the cost of economic experimentation at the individual level.

Income volatility — the month-to-month fluctuation in household income that makes planning difficult and produces chronic low-grade financial stress — fell substantially in the treatment group. Recipients were more likely to report that their income was stable and predictable. The control group showed no comparable improvement.

Mental health outcomes were positive and large. Treatment group members showed significantly lower rates of anxiety and depression at 12 and 24 months than control group members. The effect size on mental health was larger than the effect size on most economic outcomes — which is consistent with findings from other cash transfer programs and suggests that the primary mechanism may be stress reduction rather than pure income supplementation.

Food security improved. Reported ability to pay for medical care improved. The share of recipients reporting that they could cover a $400 emergency expense without borrowing money increased.

What the data didn't show

The SEED study was not designed to evaluate all possible criticisms of guaranteed income programs, and it would be wrong to read it as if it were.

It could not tell us what would happen if guaranteed income were universal — if every household in Stockton, or every household in the country, received $500 a month rather than a randomly selected 4% of them. The control group's behavior reflects a world in which most people do not receive guaranteed income. A universal program would change the labor market, prices, and economic behavior in ways the SEED design cannot model.

It could not tell us what would happen at a higher benefit level, or a lower one, or a different schedule. The $500 figure was chosen for feasibility, not theoretical optimality. Whether $1,000 would produce different effects — on employment, on spending patterns, on the political economy of the program — is a separate empirical question.

It could not tell us whether the benefits are cost-effective relative to other interventions that target similar outcomes. Mental health programs, job training, and expanded childcare all show positive effects on some of the same outcomes SEED measured. The relevant policy comparison is not "guaranteed income vs. nothing" — it is "guaranteed income vs. other uses of the same public resources."

The political architecture of the debate

The SEED findings were published in a political environment that had already formed its conclusions.

On one side, advocates for guaranteed income used the positive employment finding to argue that the entire premise of work requirements — that unconditional transfers reduce labor — was empirically refuted. This was true in the specific context of SEED. It is not yet established as a general principle across contexts, benefit levels, and population sizes.

On the other side, critics argued that 125 households, funded by private philanthropy in a single California city, proved nothing about what a national program would do to fiscal balance sheets, labor markets, or political incentives. This objection is also legitimate. A pilot program run by a sympathetic mayor with private funding, selecting from a specific income band in a specific city, is not a policy at scale. Its findings are informative, not decisive.

Both reactions share the same structural flaw: treating a specific study as if it answers the general question, when it actually answers a narrower one.

What SEED showed, specifically: for low-income households in Stockton, California, receiving $500 a month for 24 months produced measurable improvements in full-time employment rates, income stability, mental health, and food security, relative to a matched control group that did not receive the payments.

That is a finding. It belongs in the evidence base. It informs the policy question. It does not resolve it.

The design achievement

Whatever one believes about guaranteed income as policy, the SEED study is a methodological achievement worth noting separately.

Most policy pilots do not run randomized controlled trials. They implement a program, measure outcomes before and after, and call the difference an effect. This design cannot rule out alternative explanations — that outcomes would have improved without the program, that the pre-period was unusually bad and regression to the mean explains the gains, that some external event changed conditions for everyone.

SEED's randomized design with a control group rules out most of those alternatives. The employment and mental health improvements were not shared equally by the control group — so they cannot be explained by broad economic changes. The improvements appeared in the treatment group specifically. That specificity is what makes the finding informative.

The pre-registration also matters. West and Castro committed to their outcome definitions and analysis plan before the data were collected. They were not searching for positive results across a large set of outcomes and reporting the ones that appeared significant. The outcomes that were pre-specified are the ones that were reported as primary.

For city officials and program designers working at the local level, the SEED model — randomized design, pre-registered outcomes, control group, 24-month follow-up, published results including null findings — is the standard to aim for, regardless of the program being tested.

What comes next

SEED was followed by a wave of similar pilots. By 2023, more than 150 guaranteed income programs had launched across the United States, most of them funded by pandemic-era local government surpluses and private philanthropy. The Mayors for a Guaranteed Income network grew to more than 100 cities.

Most of these programs were not randomized. Most will not produce evidence that can be distinguished from a general trend. The policy will be implemented at a scale and in a political environment far outside the conditions of the original pilots.

Whether any of what SEED found holds at that scale is genuinely uncertain. The evidence says: in this context, with this design, for this population, the effects were positive. It does not say: the effects will be positive everywhere, at any benefit level, regardless of fiscal and labor market context.

That uncertainty is not a reason to abandon the evidence. It is a reason to keep running experiments — with controls, with pre-registered outcomes, with an honest account of what each one can and cannot tell us.

Stacia West and Amy Castro's full data and analysis from the SEED project are publicly available through the University of Tennessee's Guaranteed Income Research Lab.