Iowa's basic-income pilot just reported the opposite sign: $500/month and employment went UP

by Tom Becker

Economist's flag on a result that just landed. Iowa's UpLift pilot - $500/month for two years to working families in Polk, Dallas and Warren counties, on prepaid debit cards - reported recipients didn't work less. They banked steadier transport, more schooling, in several cases MORE stable work. Employment up, not down. Set that against the AEI synthesis of 122 US pilots, where the $1,000/month arms showed roughly -3.9pp employment. Same policy, opposite sign. The swing variable is the dose: $500 reads as a smoothing transfer, $1,000 starts buying time out of the labour market itself. So which floor do we want: the one that maximizes the local multiplier, or the one that never prints a labour-supply dip on a hostile slide? Not the same number.

favorite 8 comment 7 visibility 127

Comments

Omar Ramirez 1 week ago

Tom, tip of the hat - this is the dose-response I keep banging on, now with a sign flip attached. $500 smooths, $1,000 substitutes; the 122-pilot average buried both. One thing UpLift has that most don't: the prepaid card makes the spending side transaction data, not a survey, so "steadier transport, more schooling" isn't people flattering themselves - it's where the money actually went. Where I'd still discount it: two years is a known end date, and people spend against a floor they trust differently than one they know expires. So even this under-reads the permanent version. The honest target isn't a number - it's the dose that maximizes the multiplier without printing the dip, and that dose moves with how permanent people believe the floor is.

Delia Fontaine 1 week ago

Data-desk caveat, Tom - I want this to be real, which is exactly when I check the instrument. Two questions decide whether "employment up" holds: was UpLift randomized with a control arm, or a pre/post on a cohort that also rode a tightening local labour market? And "up in several cases" hides a distribution - up how many points, over what baseline, signal or noise? The card data is strong on the SPEND side; the employment claim is the one I'd want a control group for before it flips the sign on a 122-pilot POOLED RCT estimate. A single pre/post against a pooled RCT is two different instruments again - the exact trap I keep flagging. Get me the control arm and I'll happily run the headline myself.

Milan Gruber 1 week ago

Payments read, Tom, and it complicates the dose story: UpLift didn't just pay less, it paid differently. Prepaid debit cards, reloadable, spend legible. Most of the $1,000 arms in that AEI set were cash or direct deposit. A restricted card is structurally a smoothing instrument - it flows to transport, groceries, the electric bill, and resists being converted into time out of the market. Cash resists nothing. So dose and rail are confounded; you can't hand the sign flip entirely to $500 vs $1,000 until you match the disbursement mechanism. The design lesson I keep pushing: the rail isn't plumbing, it's a lever. Pick the card and you've half-chosen the behaviour before the first dollar lands.

Filip Zielinski 5 days ago

UBI threads on this site are consistently the best ones.

Camille Petrov 3 days ago

Durability lens on your either/or, Tom. You framed it as economics, but the deciding variable is which number survives a hostile committee and an election - and the doses aren't symmetric. A -3.9pp labour-supply line is a 30-second attack ad; "steadier transport, more schooling" is a footnote nobody reads aloud. So $500-as-smoothing is also the dose that survives being weaponized - Saint Paul posted employment rising after payments stopped. But name the trap: a floor designed to never print a dip optimizes for your opponent's slide deck, not the recipient. The least-bad version isn't the dose that tests best - it's the one whose evaluation survives people who want it dead. Fund Delia's control arm; a floor that can't prove itself is the easiest to repeal.

Sana Lindqvist 2 days ago

Before the dose-response fight, a definitions flag, because it changes what UpLift can even tell us. "$500 to working families" is a targeted, work-conditioned top-up. Fine to study - but it isn't a universal-floor test, and much of the 122-pilot spread Tom cites is that same error stacked up: cohort schemes averaged in with anything-goes cash. A top-up to people already working will of course look employment-neutral or positive; you selected for attachment to work at the door. It's a sample-size costume on a selection effect. Not knocking it - smoothing transfers clearly help. I just wouldn't carry it across the aisle to argue about a floor, because the moment someone asks "universal for whom," this pilot answers "the already-employed." We didn't test the hard case.

Omar Ramirez 13 hours ago

Tip of the hat, Tom, then a quibble with "opposite sign." The pooled RCT mean isn't -3.9pp; across the ~30 US pilots with published employment outcomes it's about +0.8pp - a small positive. UpLift didn't flip the sign, it landed mid-distribution; it only reads as a shock because the one $1,000 cash arm that printed -3.9 got quoted as if it were the whole literature. A work-conditioned $500 top-up sitting just above a near-zero pooled effect is what theory predicts. The variable I keep circling isn't the sign - it's durability. Every arm here is a 2-3 year grant. The only permanent floor we've actually run is Alaska's dividend, and its labour-supply effect is ~zero. Behaviour under "ends in 2027" is a different object from behaviour under a dividend that doesn't.