Do Cash Transfers Make People Work Less? The Sign Is Set by Which Constraint Binds

Precise question

For low-income adults receiving unconditional cash, what is the average causal effect on labor supply, and what determines whether that effect is negative, zero or positive?

The question is deliberately narrower than "does a universal basic income work?" Almost every serious argument about income floors turns on the same empirical premise: recipients will work less, so the program costs what it pays in lost output plus the transfer. That premise is testable. It has been tested, repeatedly, with randomization, on both sides of the income distribution and in both directions. The answer is not yes, and the reason it is not yes is more useful to a designer than a single number would have been.

Two distinctions have to be held throughout, because most apparent contradictions dissolve when they are separated. The first is between margins: whether someone works at all (participation) and how much they work once employed (hours) can move in opposite directions. The second is between transfer designs: money that is added to existing income and money that is withdrawn as earnings rise produce different behavior, because only the second changes the net return to working.

Why it matters

Unconditional cash is not a fringe instrument. It sits underneath welfare reform in Europe and North America, negative income tax experiments, humanitarian and refugee support, disaster relief, agricultural safety nets, and a large family of development programs in low- and middle-income countries. The same design is now being proposed, in several countries, as a response to labor-market disruption from automation and AI.

Every one of those debates reaches the same fork. If recipients work less, cash is partly self-financing in reverse: the fiscal cost exceeds the program's budget. If recipients keep working, the cost is the budget and the question becomes distributional and moral rather than macroeconomic. That is why the labor-supply estimate is not a technicality. It is the hinge.

It matters for a second reason that is easy to overlook. If the sign of the effect turns on the withdrawal rate rather than the payment level, then the most important design decision in an income floor is not how much to pay but how fast to take it back. That is an actionable finding, and it is the kind that a program designed to look generous might get wrong by default.

Finally, labor supply is not welfare and not output, and treating it as either has produced real errors. The largest randomized U.S. experiment found that household income excluding the recipient fell by about $1,700 a year and that partners were less likely to be promoted or to move to a better job, while leisure rose. In the other direction, a quasi-experimental study of Brazil's Bolsa Família expansion found employment rose and mortality fell by 14 percent. The two outcomes moved in opposite directions across settings, and the labor effect alone cannot adjudicate between them.

Best current answer

The general claim fails. The strongest single piece of evidence against it is a Bayesian hierarchical meta-analysis of 115 studies covering 72 unconditional cash transfer programs in 34 low- and middle-income countries, with follow-up windows mostly between 12 and 48 months. At the median total transfer amount, it estimates that unconditional cash raises labor-force participation by 3.0 percentage points (95% credible interval 1.5 to 4.5), and by 5.9 points (2.2 to 9.7) when expressed per unit of monthly payment. Monthly income rises by $1.3 per $100 transferred (0.9 to 1.8). The authors state plainly that the income result "clearly reject[s] dependency theories that predict negative impacts on income," and that cash transfers generate a strong positive effect at the extensive margin.

The hours margin is different and should not be conflated with it. The same meta-analysis puts the effect on hours worked at about +0.3 hours per week (-0.2 to 0.8), a point estimate that is positive and an interval that contains zero. On hours, unconditional cash in poor settings is indistinguishable from no effect on average.

In high-income settings the picture is consistent but smaller. A preregistered German trial paid 1,200 euros a month for three years to 107 treated participants against 1,580 controls, with outcomes measured in linked social-security administrative records. It found no extensive-margin response: the point estimate on employment was −0.003 with a standard error of 0.025, and the confidence intervals exclude employment effects larger than about five percentage points in either direction. There was no effect on job transitions in either direction, a small statistically insignificant shift toward part-time work covering about 3% of recipients, self-reported hours down about 1.3 per week, and an implied fiscal excess burden of roughly 7.5% of the transfer.

A larger U.S. trial — 1,000 low-income individuals randomized to $1,000 a month for three years against 2,000 controls receiving $50 — found a 4.2 percentage point decline in labor market participation, about 1 to 2 fewer hours per week for participants and a comparable reduction for partners, and roughly $1,900 a year less individual income excluding the transfer. Crucially, the authors looked for offsetting productive activity and did not find it: the largest increase in time use was leisure, and they could rule out even small improvements in job quality.

So the two high-income trials differ — one null on participation, one negative at about four points — and both differ from the positive aggregate sign in poorer settings. The reconciliation is not statistical noise. It is a mechanism.

Mechanisms: which constraint binds

Standard labor-supply theory offers two forces. The income effect makes a recipient better off and therefore able to afford less work. The substitution effect makes each additional euro earned worth less, because it is clawed back, so working more yields less net gain. With a flat unconditional payment, only the income effect operates and it predicts less work. With means-tested payments, the substitution effect operates too and can dominate.

That framework assumes a recipient who is choosing freely between work and leisure. For most of the people in the relevant samples, that is the wrong starting point, and the empirical pattern says so.

Where subsistence and health constraints bind, cash expands work. The Bolsa Família expansion is the cleanest large-scale test. Exploiting a change in the generosity of the world's largest cash-transfer program for the extremely poor, and using administrative microdata covering low-income households in Brazil, Best, Lobel and Pinho Neto find the reform raised employment by about 5 percent while sharply improving health: hospitalizations fell 8 percent and mortality 14 percent, which the authors estimate saved roughly 1,000 lives. They formalize this as "productive inclusion" — when households face binding subsistence and health constraints, cash relaxes them, raises effective productivity and expands labor supply. The mechanism predicts a positive sign, and it is confirmed with a large positive sign on both employment and earnings.

Where the net return to earning collapses, cash shrinks work. The clearest high-income demonstration is the B-MINCOME trial in Barcelona, which randomized 1,200 households across ten disadvantaged neighborhoods between 2017 and 2019. Two years in, the main recipient in a treated household was 10.2 percentage points less likely to work than their control counterpart (p = 0.003), a 22% relative reduction against a 47% control mean, and it survived correction for multiple inference. The decisive design detail is the withdrawal schedule. Some households lost transfers euro for euro as income rose; others faced a withdrawal rate of 25 to 35 cents per euro earned. Negative employment effects roughly doubled under a 100% phase-out, and government cost per euro of benefit rose from 12 cents to 34 cents. The authors estimate an income elasticity of −0.07 and a substitution elasticity of 0.15, and find the response concentrated among recipients with care responsibilities, which suggests labor substituted for unpaid care rather than for leisure.

The same underlying logic explains the Australian evidence. Jones and Marinescu find that Alaska's universal, permanent dividend had no effect on aggregate employment and increased part-time work by 1.8 percentage points (17%), attributing the null to a general-equilibrium offset: the dividend raised local demand for labor and nontradable services, cancelling the individual incentive to supply less. Working with the same program, Feinberg and Kuehn estimate nonlabor income elasticities of −0.10 to −0.15 for men, −0.09 to −0.14 for single women and −0.11 to −0.18 for married women. Both can be right because they measure different things: individual labor supply falls, aggregate employment does not.

Why the meta-analytic sign is positive but the U.S. sign is negative. The comparison is not a contradiction if the populations differ in what cash buys. For a household below subsistence, the binding constraint is often the ability to pay for transport, a data plan, childcare, tools or a wage delay — the prerequisites for working at all. For a household in the bottom of the U.S. distribution that is already employed, the binding constraint is more often the price of leisure, which cash buys directly. The literature on adult labor market outcomes has long argued that missing markets, liquidity constraints, risk constraints and price effects make the textbook prediction unreliable, and that the labor-leisure trade-off shows up most clearly for the elderly, who are also the group whose reduced work is least often treated as a social problem.

A fourth mechanism sits inside the household. Follow-up analysis of the U.S. guaranteed-income experiment finds the effect is not confined to the recipient: partners reduce labor supply by a comparable amount, household income excluding the recipient falls about $1,700 a year, partners are less likely to be promoted or to move to a better job, and other household members are more likely to enroll in education or training. A transfer to one adult partially finances the whole household's reallocation of time, which means the recipient-only labor effect understates the household-level response.

Competing explanations that are not mechanisms

Four rival accounts must be ruled out before the binding-constraint account can be accepted.

Measurement artifacts. Participation, hours, earnings and formal employment are different outcomes measured by different instruments, and they disagree. In the U.S. trial, survey data showed a 2.2 percentage point employment decline while administrative unemployment-insurance records alone showed 6.6 points, with 3.9 in the pooled estimate. Contract, temporary and gig work may not appear in administrative records at all. If the same asymmetry runs through the literature — informal work undercounted in poor settings, gig work undercounted in rich ones — a genuine positive sign could be attenuated in the first case and a genuine null inflated in the second. The direction of the bias is not settled.

General equilibrium. Individual responses need not aggregate. Alaska is the clearest demonstration that demand-side effects can cancel supply-side ones at the state level. Most trials are too small to see this. A positive participation effect from a national transfer could coexist with a different national outcome, and a trial-level estimate cannot settle it.

Selective entry. Everyone in these experiments volunteers. The German trial drew from 2,048,370 online applicants and kept single adults aged 21 to 40 with monthly personal income between 1,100 and 2,600 euros who were not long-term unemployed — a band deliberately chosen to approximate a universal program, and a sample deliberately unlike the general population. The U.S. trial enrolled in two states through two nonprofit operators. Volunteer samples plausibly differ in exactly the ways that determine labor supply response.

Non-randomized evidence masquerading as evidence. The Bolsa Família result is a difference-in-differences design on administrative data, not a randomized trial, and is subject to the usual parallel-trends and composition concerns. It is also the study most often cited for the positive sign, so it carries weight it should not carry alone.

Strongest counterevidence

The negative and null results deserve more than a mention.

• B-MINCOME is a real randomized trial with a large relative effect, and its negative employment result persisted at least six months after transfers stopped. Something durable happened, and the withdrawal-rate finding suggests it was design-driven rather than behavioral weakness.
• The U.S. guaranteed-income trial found no offsetting productive activity at all. Leisure rose. This is the strongest available evidence against the charitable reading that reduced work is really investment in a job search, a business or training — at least at a three-year horizon, for this population.
• Swedish lottery winners reduce hours worked substantially when they receive large unearned windfalls, and American data on idiosyncratic changes in unearned income find larger responses still. Wealth shocks of that magnitude are rare in welfare policy, but they bound how much one can assume that income effects are always small.
• The Alaska individual-level elasticities of −0.10 to −0.18 sit comfortably inside the range implied by the U.S. trial, and a matched analysis of a guaranteed-income program for artists in New York found earnings from both arts and non-arts work falling even as artists shifted hours into their own work. The reallocation is real; whether it is welfare-improving is a separate question the labor data cannot answer.

Evidence quality and limitations

The evidence is stronger than the perennial public debate suggests, but its weaknesses are specific and worth naming.

The randomized evidence is genuinely good at the study level: preregistration in the U.S. and German trials, administrative linkage in Germany, low survey attrition in the U.S. trial, and consistency across independent measures. The Barcelona trial had no pre-analysis plan and no preregistration, and it reports the extensive margin only, with no hours or earnings data.

The weakness is at the synthesis level in high-income countries. A 2024 Campbell systematic review of guaranteed basic income interventions in high-income countries located 27 studies from 10 experiments conducted between 1968 and 2020, of which nine were randomized. It rated at least one domain at "some concerns" for every study and "high risk" for 25 of 27, driven by attrition and by analysis and reporting bias. That is a verdict on the literature's ability to support confident pooled claims, not on any individual experiment.

Four structural limits bound everything above.

1. Nobody has tested the policy parameter of interest. As Hoynes and Rothstein argue, most experiments pay in addition to existing benefits, because administrators will not forfeit them. A universal basic income financed by taxation would mostly replace existing transfers, changing both the net income of recipients and the fiscal cost. The pilots estimate the supplement parameter; the policy is about the replacement parameter. The Campbell review states plainly that no true universal basic income has ever been implemented in a high-income country.
2. Durations are short. The major experiments run three years, sometimes with follow-up windows of 12 to 48 months. Behavioral responses, human capital investment and employer responses all plausibly need longer.
3. Scale is missing. Trials have hundreds to a few thousand participants. Wage compression, price responses, occupational licensing and firm-level substitution operate at population scale and are invisible here.
4. Welfare is almost never estimated. Labor supply is an input to the cost-benefit calculation, not the benefit side. Only the Bolsa Família work attempts a formal welfare evaluation, and its own evidence shows that health and mortality gains can dominate labor-market effects in the same program.

Calibrated confidence

• ~90% that the average effect of unconditional cash on adult labor-force participation is not negative, and is positive by roughly three percentage points in low- and middle-income settings. This rests on the largest randomized synthesis available, with credible intervals clear of zero, corroborated by a large quasi-experimental program and by a shift in the composition of work.
• ~85% that labor supply alone cannot settle whether an income floor is worth funding. The same randomized program moved leisure, income and household outcomes together, and the only study estimating welfare moved health and mortality in the same direction as employment.
• ~80% that for means-tested programs in high-income settings, the withdrawal rate is the dominant design lever for participation effects. One well-run trial establishes the ordering; generalizing across programs is less certain.
• ~75% that binding subsistence, health and search constraints, rather than preferences about leisure, explain why the sign differs across income levels. The mechanism is coherent, consistent with the cross-setting pattern and confirmed in a program with a health channel, but no single experiment varies the constraint while holding the payment fixed.
• ~35% that any sign identified in these samples predicts the effect of a universal, permanent, tax-financed basic income in a high-income country. The replacement parameter, the financing, the permanence and the universality have each never been tested together, and the Campbell risk-of-bias profile means even the closest proxies are soft.

Reversal conditions

The account would be wrong, or badly incomplete, if any of the following occurred.

• A preregistered trial with multiple randomized arms that varies the withdrawal rate while holding the payment level fixed, measures employment in administrative data, and runs five years or more in a high-income setting. That would convert the phase-out finding from one program into a design law.
• Evidence that the positive participation estimate in low- and middle-income settings is an artifact of undercounted informal work. A re-analysis of existing randomized trials using administrative records capturing informal employment would do it.
• A national or near-national transfer evaluation showing that the trial-level sign does not survive scale, whether through general equilibrium or employer substitution.
• A credible estimate of the welfare cost and benefit of a specific universal design including health, care and human-capital channels. Labor supply is currently the wrong statistic for this, and the question is being answered without it.
• Replication of the U.S. trial's null result on offsetting productive activity in a population with weaker labor markets, or the opposite — a finding that three years is simply too short to observe investment paying off.

The cheapest experiment worth running is the first one. Everything else in this literature is a shadow of it.

Decision implications

For anyone designing an income floor, the evidence supports five conclusions that do not depend on resolving the open disputes.

1. Do not treat "people will work less" as a settled input. It is wrong as a general statement and wrong in the direction that matters most for programs in poorer settings.
2. Treat the withdrawal rate, not the payment level, as the primary design lever on participation. A flat supplement and a means-tested guarantee are different policies with different signs.
3. Measure participation and hours separately, and measure them with administrative data where possible. Administrative and survey estimates of the same effect can differ by a factor of three.
4. Do not use labor supply as the benefit side of the ledger. The strongest evidence pairs lower labor supply with higher leisure and lower household income in one direction, and higher employment with lower mortality in the other.
5. Treat every existing pilot as an estimate of the supplement parameter, not the replacement parameter. A universal program that replaces existing benefits is a different intervention and the field has no estimate of it.

References

1. Baird, S., McKenzie, D., & Özler, B. (2018). The effects of cash transfers on adult labor market outcomes. IZA Journal of Development and Migration, 8, 22. https://doi.org/10.1186/s40176-018-0131-9

2. Bernhard, S., Bohmann, S., Fiedler, S., Kasy, M., Schupp, J., & Schwerter, F. (2026). Basic income and labor supply: Evidence from an RCT in Germany. INET Oxford Working Paper Series, No. 2026-08. https://www.inet.ox.ac.uk/publications/basic-income-and-labor-supply-evidence-from-an-rct-in-germany

3. Best, M. C., Lobel, F., & Pinho Neto, V. (2026). Cash transfers and productive inclusion: Evidence from Bolsa Familia. NBER Working Paper 35006. https://doi.org/10.3386/w35006

4. Cesarini, A., Lindqvist, L., Notowidigdo, M. J., & Östling, R. (2017). The effect of wealth on individual and household labor supply: Evidence from Swedish lotteries. American Economic Review, 107(12), 3917–3946. https://doi.org/10.1257/aer.20151589

5. Crosta, T., Karlan, D., Ong, F., Rüschenpöhler, J., & Udry, C. R. (2024, revised August 2026). Unconditional cash transfers: A Bayesian meta-analysis of randomized evaluations in low and middle income countries. NBER Working Paper 32779. https://doi.org/10.3386/w32779

6. Feinberg, R. M., & Kuehn, D. (2018). Guaranteed nonlabor income and labor supply: The effect of the Alaska Permanent Fund Dividend. The B.E. Journal of Economic Analysis & Policy, 18(3). https://doi.org/10.1515/bejeap-2018-0042

7. Golosov, M., Graber, M., Mogstad, M., & Novgorodsky, D. (2023). How Americans respond to idiosyncratic and exogenous changes in household wealth and unearned income. The Quarterly Journal of Economics. https://doi.org/10.1093/qje/qjad053

8. Hoynes, H. W., & Rothstein, J. (2019). Universal basic income in the United States and advanced countries. Annual Review of Economics, 11, 929–958. https://doi.org/10.1146/annurev-economics-080218-030237

9. Jones, D., & Marinescu, I. E. (2022). The labor market impacts of universal and permanent cash transfers: Evidence from the Alaska Permanent Fund. American Economic Journal: Economic Policy, 14(2), 315–340. https://doi.org/10.1257/pol.20190299

10. Navarro-García, M. (2026). Causal effects of minimum income schemes on poverty reduction: Systematic review. International Journal of Economic Policy Studies, 20, 337–366. https://doi.org/10.1007/s42495-026-00172-y

11. Noonan, D. S., & Woronkowicz, J. (2026). Testing the empirical validity of the work preference labor supply model: Evidence from a guaranteed income program for artists in New York. Journal of Cultural Economics. https://doi.org/10.1007/s10824-026-09594-0

12. Rhodes, E., Broockman, D. E., Vivalt, E., Krause, P. K., Miller, S., & Bartik, A. W. (2026). Income shocks and intrahousehold dynamics: Evidence from a guaranteed income experiment. NBER Working Paper 35589. https://doi.org/10.3386/w35589

13. Rizvi, A., Kearns, M., Dignam, M., Coates, A., Sharp, M. K., Magwood, O., Labelle, M., & Elmestekawy, S. (2024). Effects of guaranteed basic income interventions on poverty-related outcomes in high-income countries: A systematic review and meta-analysis. Campbell Systematic Reviews, 20. https://doi.org/10.1002/cl2.1414

14. Vivalt, E., Rhodes, E., Bartik, A. W., Broockman, D. E., Krause, P., & Miller, S. (2024, revised August 2026). The employment effects of a guaranteed income: Experimental evidence from two U.S. states. NBER Working Paper 32719. https://doi.org/10.3386/w32719

15. Verlaat, T., Todeschini, F., & Ramos, X. (2025). The employment effects of a means-tested guaranteed income policy. Journal of Public Economics, 248, 105420. https://doi.org/10.1016/j.jpubeco.2025.105420