The cost of ending poverty in America has fallen by seventy-one percent since 1967. The reason is not good news.
August 19, 2026
In 1963 an economist at the Social Security Administration named Mollie Orshansky needed a number. She wanted to count poor children, and the United States had no official definition of poor.
So she built one out of groceries. She took the cheapest food plan the Department of Agriculture published, priced it for families of different sizes, and multiplied the result by three. The multiplier came from a 1955 survey: for households of three or more people, food took about a third of after-tax income.
That was the method. A subsistence grocery list, times three. In August 1969 the Bureau of the Budget made it the official statistical definition of poverty in the United States, and it holds that title today. Each year the dollar figure is raised to match the price of goods. Nothing else about it has changed.
A family of four with two children counted as poor in 1967 if it lived on less than $3,386. The same family counts as poor in 2024 below $31,812. By the method's own design, those are the same amount of money.
I did not go looking for any of that. Last month I wrote to a professor at Georgetown University in Qatar to ask whether I had represented his work fairly. I had used a figure of his in two chapters, and I would rather be corrected before printing than after.
He wrote back to say that the paper I had cited was nine years old, that its data came from 2015, that he and his co-author had since corrected it upward, and that a replacement was on its way.
The correction turned out to be the smaller part. Attached to his reply was a column of numbers running back to 1967, and as far as I can tell nobody has said out loud what it shows.
His name is Karl Widerquist. The 2017 paper I had cited was his, and so is the one that now replaces it, written this time with a colleague named Jack Rossbach. Between them they have worked out what it would cost the United States to have nobody poor.
Not to reduce poverty. To end it — to lift every household above the official threshold and leave none below.
The figure is $783.7 billion a year. That is 2.67 percent of gross domestic product (GDP). It is 11.6 percent of federal spending, 7.5 percent of federal, state and local spending combined, and 23 percent of what the country currently spends on entitlements. It would take the official poverty rate from 10.6 percent to zero, which is 35.9 million people, 10.5 million of them children.
The plan they price is a basic income of $16,000 per adult and $8,000 per child, with a flat fifty percent tax on all other income. In all, 130.1 million Americans — nearly two fifths of the country, spread across 50.4 million households — come out ahead. Most households earning up to $70,000 a year are net receivers. For tens of millions of working families it operates less like welfare than like a tax cut.
The number comes from Census Bureau microdata, the same survey that produces the official poverty statistics. It forecasts nothing and simulates nothing. It is an arithmetic operation performed on the actual distribution of American household income in 2024.
Before that figure gets carried anywhere, it needs its limits attached, because it is smaller than it looks and it is easy to misuse.
It is a net figure. The gross cost of writing those checks runs into the trillions. What brings it down to $783.7 billion is that most of the money comes straight back through the fifty percent tax — every household above the break-even point returns more than it received. The net cost is what remains after the clawback. It is not a budget item, and it should never be set against a gross number as though the two marked out a range. They measure different things.
It assumes no existing program is replaced. Nothing is cut to pay for it.
The fifty percent rate is a modeling device, chosen because it makes the arithmetic tractable, not because anyone proposes it.
And "ending poverty" means ending official poverty. The American threshold is widely held to be too low; the authors say so themselves, and note that some researchers put the real requirement at 150 percent of the line or higher. A better-designed alternative has existed since 2009 — the Census Bureau publishes its Supplemental Poverty Measure every year, and the official count still does not use it.
The authors are unusually blunt about all of this. Their paper, they write, is not about the politics, not about which programs could be replaced, not about how to integrate a basic income into the existing tax and benefit system, and not even about how to pay for it. It asks one question: what does this cost in isolation — "in a vacuum, so to speak."
Here is the series.
| Year | Cost of ending poverty, as a share of GDP |
|---|---|
| 1967 | 9.35% |
| 1995 | 4.95% |
| 2015 | 3.70% |
| 2024 | 2.67% |
In 1967, when a guaranteed income was under serious discussion in Washington, ending American poverty would have cost nearly a tenth of everything the country produced. Today it costs a fraction over one fortieth — 28.6 percent of the 1967 share.
It has not fallen in fits and starts. It has fallen more or less continuously for fifty-seven years, through recessions and booms, under both parties, across every administration from Johnson to the present.
And it is still falling. Between 2015 and 2024 the real cost dropped 12.6 percent — from $896.7 billion to $783.7 billion in constant dollars. Not the share. The money.
There is nothing mysterious about the mechanism. But the obvious explanation is the wrong one.
The cost is not falling because America has been winning against poverty.
The United States measures poverty in absolute terms. The threshold tracks prices alone. It does not rise when the country gets richer. Most of Europe measures poverty relatively, as a fraction of median income, so the European line climbs as national income climbs. The American line does not climb. It stays where it was and waits.
The line has not only stayed still. Its premise expired underneath it. Orshansky set the multiplier at three on the strength of that one-third share. In 2024 food is 12.9 percent of what the average American household spends. The two figures do not measure quite the same thing — income in one case, outlays in the other — but no careful accounting closes a gap that wide. The ratio is gone. The line built on it is still in use.
Against a fixed line, a growing economy makes the line cheaper to reach every year, by arithmetic alone.
Now put the two numbers side by side. Between 1970 and 2024 — the window the authors use for their growth comparison, three years shorter than the cost series — American GDP grew 338 percent after inflation. Over those fifty-four years, the real cost of lifting everyone over the poverty line grew 13 percent.
The economy more than tripled. The cost of the floor barely moved.
That is the whole curve. Not a victory. A divergence.
If growth had reached the bottom of the distribution, the cost of a floor would have fallen for a different reason — because there were fewer people underneath it.
That is not what happened, and the Census Bureau's own series is the cleanest place to observe it.
The poverty rate did fall, and that deserves saying plainly before anything else. It was 12.6 percent in 1970. It is 10.6 percent in 2024. Two percentage points of progress in fifty-four years.
But a rate is a share, and the population grew by two-thirds. Count people instead. In 1970, 25.4 million Americans lived below the poverty line. In 2024 it is 35.9 million — the same 35.9 million that $783.7 billion would lift.
Every one of them is counted against Orshansky's grocery list. Two parents and two children are poor in 2024 on $31,811 and not poor on $31,812, at which point they leave the statistic and nothing else about their week changes.
Set the four numbers beside each other. They all cover the same fifty-four years.
The economy grew 338 percent. The cost of lifting everyone over the line grew 13 percent. The poverty rate fell two points. The number of people below the line rose 41 percent.
The floor became cheap to build not because the people beneath it went away but because everybody else moved further above it.
Rossbach and Widerquist put it about as plainly as it can be put:
Despite a well more than tripling of GDP, a substantial number of American households remain in poverty — indicating that we cannot expect the benefits of economic growth to trickle down to common people without active government intervention.
There is a cheap version of this observation, and it is not the one the data supports.
The cheap version is a complaint about fairness. The data says something narrower and harder: the mechanism does not converge. Fifty-seven years is not a business cycle. It covers the strongest sustained productivity growth in recorded history, decades of near-full employment, the entire computing revolution, and the largest expansion of global trade ever recorded. If distribution were going to self-correct under favorable conditions, those were the favorable conditions. It did not correct. It ran in one direction for fifty-seven years, and the gap between what the country produces and what it guarantees widened every year, in good times and bad, under every government.
That is a description of a system with no tendency to reach equilibrium on its own.
Measured against 1963 groceries, the country is winning. Measured against itself, it is losing.
Here is the strongest case against reading anything into that curve.
The cost falls because America measures poverty against a line that does not move. Switch to the European method — a line indexed to median income, climbing as the country gets richer — and the fall should be far smaller, because the target rises with the economy instead of standing still while it passes. On that reading the decline is not a fact about poverty at all. It is a fact about American statistical convention, and a different convention would blunt it.
That is right. It is also not a hostile reading: it is Rossbach and Widerquist's own account of why their numbers fall, stated plainly in the paper.
But follow it through. If the cost fell because the line stayed still, then the line stayed still — for fifty-seven years, while everything it was built to measure moved. A line that does not move is not an accident of arithmetic. It is a decision, renewed by default every year that nobody revisits it. "Not poor" in the United States still means what it meant to a Department of Agriculture food plan in 1963, and the gap between that and what the country now produces is exactly what the curve has been drawing all along.
The objection holds. What it identifies is the finding's cause.
There is one more thing the number cannot do, and it is the thing it gets asked to do most often.
$783.7 billion is what a basic income costs. It is not a plan.
The moment you specify how to fund it and what it replaces, you are no longer measuring a policy. You are measuring a bargain — this program kept, that deduction scrapped, this rate adjusted — and the figure you end up with describes the bargain rather than the basic income. That is why the authors held everything else still. It is also what makes a fifty-seven-year series possible at all: the same assumption runs through every year, so the slope is not an accident of a welfare state that looked nothing in 1967 like it does now.
Rossbach and Widerquist go on to argue that the money should come from taxes aimed at the wealthiest — land value taxes, resource and pollution taxes, a wealth tax, higher capital gains. That is their argument, not mine, and I am not going to borrow it.
What I take from their series is narrower. What a floor costs and how a country pays for it are two different questions, and only the second one is genuinely hard. The first has had an answer for decades, and the answer has been getting easier every year since 1967.
Which is what makes the last fifty-seven years peculiar.
Cost has been the standing objection to a guaranteed income for the whole of that period. It was the objection in 1967, when the same floor would have taken three and a half times as large a share of the economy as it does today. It is the objection now. In between, the figure fell by seventy-one percent and the objection did not move — not restated more carefully, not conceded in part, not adjusted in light of the arithmetic. Repeated at full strength, in every decade, while the ground underneath it gave way.
An objection that outlives its own premise was never really that objection.
One column of figures. Two questions, and they have opposite answers.
Read it as a budget question — can this be afforded — and the answer improves every year. 9.35, 4.95, 3.70, 2.67. By 2024, a floor under every American is one of the cheaper things a rich country could choose to buy.
Read it as a distance — how far below the country's output the floor now sits — and the answer worsens every year, by the same arithmetic, for the same reason. The line stayed still. The economy left.
The number that makes a floor affordable and the number that makes it necessary are not two findings. They are one measurement, read twice.
Both readings come off the same line, and the line came from one desk. An economist at the Social Security Administration priced the cheapest food plan the government published and multiplied it by three. The country has been keeping score against her grocery list ever since, raising it each year for prices and leaving it alone in every other respect.
Widerquist and Rossbach end their paper with a question I have not been able to shake: How rich do we have to get before we decide that no one's income should be so low that they are forced to sleep on the street?
Fifty-seven years of data suggest that getting richer was never the part we were waiting on.