Universal Basic Income Weakened by Its Own American Experiments

Among 122 pilot programs conducted in 33 states between 2017 and 2025, the four most rigorous studies show an average decline of 3.2 percentage points in employment. This data reveals a troubling methodological paradox: small experiments flatter the universal basic income thesis while large ones contradict it.

The exhaustive analysis conducted by the American Enterprise Institute catalogues 122 pilots that distributed $481.4 million to over 40,000 beneficiaries over eight years. Far from validating the instrument presented as a safety net against AI automation, these results reveal an empirical fragility that could undermine one of the decade’s most ambitious social projects.

The Essentials

Among the four studies with at least 500 participants, representing 55% of all beneficiaries, the average effect on employment is -3.2 percentage points
Only 52 of the 122 experiments published their results, 35 use randomized methods, and 30 measure employment effects
The average size of treatment groups is only 359 people, with an average attrition rate of 37%
The estimated average income elasticity of -0.18 confirms standard economic theory: more unearned income means less work at the margin

The Size Effect Reveals Methodological Limits

The 30 randomized pilots with published results display an average increase of 0.8 percentage points in employment. This figure, wielded by universal basic income advocates, masks a more nuanced reality. The average size of treatment groups reaches only 359 people, with a median of 151. This methodological fragmentation offers only a fragile foundation for rethinking the American welfare state.

Among the 122 unique pilots, 30 have fewer than 100 participants in the treatment group and 18 have fewer than 50. This miniaturization contrasts with the universalist pretensions of the project. More concerning still, the average attrition rate of 37% among the 26 pilots for which this data is measurable signals potential major biases in interpreting results.

The methodology reveals its own contradictions. Nearly all pilots that track results rely exclusively on survey data, exposing them to reporting and non-response biases. This dependence on self-evaluations weakens the scientific robustness of the conclusions.

Large Studies Contradict the Optimistic Average

The American Enterprise Institute’s analysis reveals an inverse correlation between sample size and positive results. The OpenResearch study financed by Sam Altman, with 1,000 beneficiaries receiving $1,000 monthly for three years, shows a decline of $1,800 in individual annual income (excluding transfers) and a decrease of 4.1 percentage points in labor market participation.

This experiment, one of the most methodologically rigorous, confirms predictions from standard economic theory. Participants reduced their working hours by 1 to 2 hours per week and their partners decreased their hours in comparable proportion. The largest increase generated by the transfer concerned time spent on leisure.

The experiment in 19 counties in Texas and Illinois, with payments of $1,000 monthly over three years, reports that “participants worked less and less over the course of the study.” Less work meant less income. Beneficiaries experienced an average reduction of $2,500 in their annual household income, transfers excluded.

The Disconnect Between Theoretical Hopes and Empirical Realities

Universal basic income advocates bet on the “trampoline effect”: researchers expected participants to eventually earn higher wages by accepting better-paying jobs, but this scenario never materialized. “But we find no effect on job quality,” concludes Eva Vivalt, assistant professor of economics at the University of Toronto and principal researcher of the study.

This disconnect reveals a fundamental analytical error. Beneficiaries primarily devoted more time to leisure activities, not to pursuing education, better jobs, or family care. “It is interesting to note that we do not observe those with children spending more or less time on childcare following the transfers.”

The absence of effect on entrepreneurship confirms this trend. Although researchers find that people report having more entrepreneurial intentions, this does not translate into actual entrepreneurial activity. This divergence between stated intentions and actual behavior underscores the limitations of survey-based methodologies.

Publication Bias in Favor of Positive Results Distorts the Debate

Pilots showing benefits receive media coverage, academic publications, and political attention. Pilots showing no effect or harm are less likely to be published or noticed. This creates a systematic bias in our understanding—we might be overestimating the benefits of universal basic income.

This phenomenon, documented by academic literature, partially explains why only 52 of the 122 pilots published their results. Experiments with neutral or negative results disappear in what researchers call the “file drawer effect,” distorting the overall assessment of universal basic income’s effectiveness.

The Stockton example illustrates this asymmetry. The one-year study showed that unconditional transfers increased beneficiaries’ full-time employment by 12 percentage points and decreased their measurable feelings of anxiety and depression. These results, widely publicized, contrast with the discretion surrounding more nuanced findings from larger-scale studies.

AI as New Pretext for an Old Idea

Current universal basic income policies are not really about reducing the state. They are mainly about its expansion because elites fear AI. This fear of automation justifies today the resurrection of an old economic concept, despite accumulating contradictory empirical evidence.

Artificial intelligence has become the latest excuse to revive one of the worst ideas in economic policy: a basic universal income. Recent articles in Newsweek, the LSE Business Review, and Fortune have all contributed to pushing the idea that AI might soon eliminate so many jobs that Washington should send a check to everyone.

This instrumentalization of technological anxiety diverts attention from the real stakes of work transformation by AI, which show that automation often stimulates human creativity rather than replacing it.

Budget and Political Constraints Ignored

Universal basic income also faces budget constraints. As Max Gulker noted in The Daily Economy, universal basic income is often sold through small pilots and vague moral language, but the national arithmetic is ugly. National debt is rapidly approaching $40 trillion.

The illusion of administrative simplicity masks a complex political reality. In theory, universal basic income proponents sometimes imagine replacing the welfare state with a simple money transfer. In reality, government programs rarely disappear. Bureaucracies defend themselves. Interest groups protect their privileges.

This institutional dynamic suggests that a universal basic income would probably be added to existing programs rather than replace them, aggravating the initial budget challenge. The 122 American experiments, despite their collective scale, provide no answer to this fundamental political question.

The Empirical Fragility of a Social Project

The AEI’s conclusions are appropriately cautious: these results may not generalize to a permanent, universal, and national basic income under current or future conditions. This methodological caveat should temper political enthusiasm.

After 122 local experiments, the case for universal basic income remains weak. The best evidence does not show an employment renaissance. The largest studies show employment declines.

The accumulation of contradictory data reveals a paradox: the more experiments multiply, the more they weaken their object of study. The $481.4 million distributed over eight years produced an unintended lesson about the limits of public policies founded on theoretical optimism rather than empirical observation. In the face of the economic challenges posed by AI, this lesson could prove more valuable than the monetary transfers themselves.