New Jersey Negative Income Tax Experiment

Summary

The New Jersey Graduated Work Incentive Experiment was the first large-scale randomized evaluation of a guaranteed-income program anywhere in the world. Initiated under the Johnson administration's War on Poverty and continued through the Nixon administration, it was an extraordinary commitment of resources to answer a single empirical question: how much would poor families reduce their work effort if given a guaranteed income? The political debate at the time presumed the work-reduction effects would be catastrophic. The experimental results were more nuanced: labor supply did decline, but by less than half what critics had predicted, and the largest reductions were among secondary earners (typically wives) who used the transfer to spend more time on child care and household work rather than on leisure. The study established methodological norms (multi-arm randomization across program parameters, long-duration treatment, panel data collection) that subsequent income-maintenance experiments — Seattle/Denver, Gary, Manitoba Mincome — built on directly. Although Nixon's Family Assistance Plan failed in the Senate, the experiment's findings did not. Subsequent US welfare design pivoted toward the Earned Income Tax Credit — a phased-in subsidy that preserves work incentives — rather than the flat NIT model, an explicit response to what the NJ trial had shown.

Research question

"If poor families receive a guaranteed minimum income that phases out as earnings rise (a negative income tax), do they reduce their labor supply, and by how much? Are the work-disincentive effects large enough to make the policy infeasible?"

Methodology

Result

Mixed

Husbands reduced hours worked by approximately 5–6%; wives reduced hours by 20–30% (off a low base); overall family labor supply fell by 5–8%. Effects were considerably smaller than the policy debate had assumed but were measurable and statistically significant.

Evidence strength

Strong

Randomized trial, replicated across multiple sites or studies.

Replication status

Replicated