The House Is the State

IThe wrong-way tax

Income tax runs uphill. Earn more, pay a higher rate. Whatever you think of how well it works, the direction is deliberate and it is the entire design principle.

The lottery runs the other way, and it does it in a way that's hard to see because the mechanism is voluntary.

Everybody pays the same price. A hedge fund manager and a night-shift cleaner buy the identical two-dollar line. But among people who actually play, the ones with least spend most — a study commissioned by South Carolina's own lottery found players with household incomes under $35,000 spending more than twice what players earning $100,000 to $150,000 spent. Not as a share of income. In raw dollars.

Then the money comes back out as schools, scholarships, senior services — the things used disproportionately by lower-income households.

Collected regressively. Spent progressively. What looks like a transfer is mostly a lateral shuffle, from poor to slightly-less-poor, with the state taking a third off the top for running it.

There's a sharper version of this, and it explains why lotteries exist at all.

You cannot pass a tax rise. You can run a lottery. Forty-five states and the District of Columbia have worked out that the second thing raises money the first thing can't, and does it without a single vote, a single hearing, or a single person describing it as taxation. Lottery proceeds now account for about 2% of all state taxes collected across those states.

The lottery isn't really a game. It's a revenue instrument that lets a state collect from the bottom of the income distribution without ever having to call it a tax.

That's the argument. The rest of this moves through three stages: why anybody buys a ticket, what a ticket actually is, and who ends up paying for it. The third stage is where it gets uncomfortable, including for the argument I've just made — one of the better datasets says the opposite of what I expected, and I've left it where it fell.

IIEveryone gets the same odds

Start with the part that's true.

Every ticket has an identical chance. There is no premium tier, no relationship pricing, no minimum balance, no credit check. A hedge fund manager and a night-shift cleaner buying the same Powerball line have precisely the same 1-in-292,201,338 shot. In a country where almost every other route upward is visibly, structurally unfair, the lottery is the one place where it isn't.

And this is the place to say that everything following applies to the whole category, not just the billboard games. A $1.8 billion Powerball jackpot and a $5 scratch ticket with a $10,000 top prize are the same product at different price points. No skill. Identical odds for every buyer. The same beach-and-palm-tree imagery. The same money going to the same place, taxed the same way on the way out. Instant tickets are the larger business — roughly 65 to 70% of sales in a mature market like Massachusetts — but nothing in the argument turns on which one you buy. The mechanism is the promise of a fast, unearned, better outcome, sold in bright colours.

That promise turns out to matter more than ignorance does. A 2008 experimental study in the Journal of Behavioral Decision Making concluded flatly that it would be naive to attribute disproportionate low-income lottery play to ignorance or cognitive error. What the researchers found instead was that participants primed to feel low-income played more — and the mechanism they proposed was exactly this sense of a uniquely level playing field.

Read that as rational rather than pathetic and it's much harder to dismiss. If the ladder is rigged everywhere else, the one unrigged thing in your week is worth two dollars.

Clotfelter and Cook titled their 1989 book on state lotteries Selling Hope, and the phrase is more precise than it sounds. The product isn't the prize. Almost nobody gets the prize. The product is the interval between buying the ticket and finding out — the few days, or the few seconds under a coin, in which the answer is not yet no. That product is delivered in full, every time, to every buyer. It is one of the few things sold in America that always works.

So far, so benign. Here's where it stops being benign.

Powerball's odds never change. One in 292,201,338 when the jackpot is $20 million, and one in 292,201,338 when it's $1.8 billion. Identical, every draw, forever. But ticket sales don't hold still at all. Fitting a curve to jackpot growth between draws gives roughly 63,000 tickets sold per million dollars of advertised jackpot — near-linear, from ordinary draws right up through the billion-dollar runs.

Sit with what that means. If people were reading the jackpot as how much I'd get, a bigger number would change how much they wanted to win. It wouldn't change whether they bought a ticket. Instead it changes whether they buy, and almost proportionally.

The number on the billboard is not a claim about the prize. It's a lever on perceived probability.

Two pieces of published work say the same thing from different directions. Cook and Clotfelter, in the American Economic Review in 1993, found lotto sales per capita rise with state population even though larger states run games with worse odds — and proposed that players judge their likelihood of winning by how often somebody wins. Frequency of any winner substitutes for probability of this winner. A bigger state, a bigger jackpot, more coverage, more visible winners: the outcome feels more available even as it becomes less likely.

Then in 2024, Allcott, Lockwood, Taubinsky and Sial published a finding in the Review of Economic Studies that is, to my mind, the single most damning number in this entire subject. Lottery sales respond strongly to the size of the jackpot. They are completely unresponsive to the size of the second prize.

Powerball's second prize is one million dollars. That is a life-changing sum for essentially every person who has ever bought a ticket. Move it up, move it down — sales don't notice.

People are not buying a chance at money. They are buying a chance at the number on the sign.

Which raises the obvious question. What is the number on the sign?

IIIWhat the sign says, and what arrives

In December 2025, Powerball advertised a jackpot of $1.817 billion. Second largest in American history.

Nobody has ever received $1.817 billion.

There are two reductions between the billboard and the bank account, and they are completely different in kind. Conflating them is the most common error in writing about this, so it's worth installing the distinction before the numbers arrive.

The first reduction is not a loss. The advertised jackpot is the arithmetic sum of thirty annual payments running out to 2055. It is a nominal total, added up across three decades and printed as though it were a quantity of money that exists somewhere. The cash option — $783.2 million, about 43% of the headline — is what it actually costs to fund those payments today. That's not a haircut. That's the honest present value, and the $1.034 billion difference was never money in the first place.

Every other price a consumer encounters is a present value. A car costs what a car costs. The lottery quietly breaks that convention and displays a nominal thirty-year sum in the slot where a price belongs. It's disclosed. It's also the only number on the billboard.

The second reduction is a real loss. For a New York City winner taking the cash: federal at 37%, New York State at 10.9%, New York City at 3.876%. Combined, about 51.7%. That's $405.3 million.

What lands is $378.0 million — 20.8% of the number on the sign.

A detail that catches people who should know better: the withholding doesn't cover the bill. Federal withholding on lottery prizes is 24%, but the actual top marginal rate is 37%. On a $115 million cash option, that's a $14.7 million balance falling due at filing, months after the cheque cleared.

And the choice between lump sum and annuity, which is presented as a matter of temperament, is really a question about discount rates that most winners have never been asked to think about. Model a $30 million Powerball win and an $18 million cash option over thirty years at a 6% return with capital gains paid annually, and every route converges into a band between $20 and $32 million. The smallest-sounding prize on the board finishes highest.

The number doing the work there isn't the prize. It's that 6% compounds at 3.95% once federal capital gains, the net investment income tax, and New York State and City have taken their share. That single figure — the gap between the return you earn and the return you keep — determines more of the thirty-year outcome than which jackpot you won.

And the largest lever of all isn't in the prize structure. New York taxes non-residents on New York-source income, and lottery prizes over $5,000 are explicitly New York-source, so the prize follows the ticket wherever you go. But New York City tax is residency-only, and New York doesn't tax non-residents on investment gains at all. Move out of the state and your after-tax compounding rate goes from 3.95% to 4.57%, worth 19–30% over thirty years.

Relocating is worth more than winning a bigger prize.

That sentence is the whole essay in miniature. The outcome is determined less by what you win than by what you know — and what you know correlates ferociously with where you started.

IVWhere the dollar goes

Before the numbers, one distinction that almost every article on this subject gets wrong, including my own first draft.

Sales are not losses. When Americans buy $104.7 billion of lottery tickets, they do not lose $104.7 billion. Most of it comes straight back as prizes, and most of that gets wagered again. A scratch player putting $200 a week through the counter in Massachusetts, where the prize payout runs around 74 to 76% of revenue, is getting roughly $150 of it back. Their actual annual loss is closer to $2,600 than to $10,400. The same dollar can be counted as sales three or four times as it cycles.

Hold onto that, because it cuts both ways. It makes the eye-watering spending figures less impossible than they sound. It also means the real extraction is a smaller and much harder number than the one usually quoted.

Here it is. In fiscal 2024, Americans bought $104.7 billion of lottery tickets, up from $52.8 billion in 2008. Prizes paid came to $70.2 billion — about 67 cents of every dollar wagered goes back to players.

Which leaves $34.5 billion that doesn't. That is the number. Not the hundred billion. Thirty-four and a half billion dollars is what American lottery players actually lost in a year, and it covers state proceeds, retailer commissions and administration. In 2008 the same figure was $20.6 billion.

Thirty-three cents on every dollar wagered is the extraction rate, and it's worth holding against something familiar. It is far above the house edge of any casino game. It is roughly seven times the annual expense ratio of an expensive mutual fund — except charged per transaction rather than per year, on money that recirculates. And unlike either of those, it's collected by a government that also writes the rules about what may be advertised and where.

Then, at the other end, the winnings get taxed. And here the tax code does something that deserves more attention than it gets.

All gambling winnings are taxable as gross income. Losses are deductible only as an itemised deduction on Schedule A. The 2026 standard deduction is $16,100 for a single filer and $32,200 for a couple, so anyone who doesn't clear that in total deductions — which is most people, and nearly everyone at the bottom of the income distribution — deducts nothing at all, regardless of how much they lost.

On paper, the scratch player above wins around $7,800 across the year, is down $2,600 net, and is taxable on the full $7,800 with no offset. The ability to net losses against winnings is a benefit reserved for people wealthy enough to itemise.

And it just got tighter. The One Big Beautiful Bill Act rewrote §165(d) effective for tax years beginning after 31 December 2025: only 90% of gambling losses are now deductible, still capped at winnings. Break even on $50,000 and you deduct $45,000, leaving $5,000 of phantom income — tax on money you never made. It was projected to raise $1.1 billion over eight years, and there is a bill pending in Congress to repeal it.

The honest qualification, which matters: this is largely unenforced at the bottom. The W-2G reporting threshold rose from $600 to $2,000 in 2026, so small wins are never reported to the IRS and in practice almost nobody declares scratch winnings. But that is its own indictment. The rule is written so the offset only functions for people who itemise, and it was made stricter this year, while the people doing most of the wagering can't reach it and are simply expected not to comply.

VWho's actually paying

This is where I have to complicate my own argument, because one of the better datasets did.

The claim I expected to find was straightforward: poor people play more. It is repeated everywhere. It is not what the participation data says.

Gallup, surveying 1,025 American adults in 2016, asked whether people had bought a state lottery ticket in the past year. Among households under $36,000: 40%. Households between $36,000 and $89,999: 56%. Households above $90,000: 53%. Overall, 49%.

Participation rises with income. It doesn't fall.

The education breakdown is flatter still — high school or less 47%, some college 53%, college degree 53%, postgraduate 45%. Gallup found the same pattern in 1999, 2004 and 2007, so this is not a fluke, and Gallup is not a marginal outfit.

Anyone arguing that lotteries prey on the poor has to deal with that. Here's how it resolves.

Participation is not intensity. "Did you buy a ticket this year" is a binary. Somebody who bought one ticket during a record jackpot counts identically to somebody spending two hundred dollars a week. And on the question that actually determines where the revenue comes from, the picture inverts.

The Massachusetts Lottery commissioned a study in 2016 that found the top 10% of players account for about 40% of all sales. South Carolina commissioned one in 2014 that found players earning under $35,000 spending more than twice what players earning $100,000 to $150,000 spent — in absolute dollars, before you adjust for income at all. Older academic work by Clotfelter, Cook, Edell and Moore found the poorest households spending around 3% of everything they had.

So: about half of Americans buy a ticket in a given year, and those buyers skew slightly upward in income. But most of them buy one ticket, when a jackpot makes the news, and forget about it. The revenue is generated by a heavy-playing minority spending at a completely different order of magnitude — and among people who play, the poorest spend the most in raw dollars and vastly more as a share of what they have.

A revenue system doesn't need everyone poor to be regressive. It needs its heaviest contributors to be poor.

And the machine is installed where they live. The Howard Center for Investigative Journalism obtained lottery retailer locations for 44 states plus the District of Columbia and matched them against census data. In every single one — all 44 states and DC, no exceptions — neighbourhoods with a lottery retailer have a higher poverty rate than neighbourhoods without one. In Michigan, close to double the poverty rate and $16,000 less in median household income. The industry's standard rebuttal is that people don't buy where they live, so the Howard Center tested it with mobile location data covering roughly three-quarters of US lottery retailers: the average customer lives within 1.1 miles of the store, and in all but two jurisdictions a majority of retailers drew their customers from the surrounding neighbourhood.

That data establishes where the machine is and that the people using it live nearby. It cannot tell you how much any individual spends; for that you need the state studies above. But the placement is not accidental, and it isn't a conspiracy either. States recruit retailers on criteria like store security, compliance, and the ability to hit sales targets. Nobody in that process is asked about neighbourhood income. A system optimising purely for volume finds the heaviest users on its own, and the map it produces is indistinguishable from the one deliberate targeting would draw.

One dispute I'll flag and not resolve. Allcott and colleagues found that worse mathematical reasoning strongly predicts higher lottery spending, and their regression implies Americans would spend roughly 43% less if unbiased. Notably, income predicted spending only weakly while numeracy predicted it strongly, so this is not simply a proxy for being poor. That sits directly against the 2008 level-playing-field finding, which says low-income play is a rational response to unfairness elsewhere rather than a failure of arithmetic. Both are credible. They imply completely different policy responses. I don't think it's settled and I'm not going to pretend otherwise.

What isn't in dispute is the shape of the flow. Money is collected at a rate that rises as income falls, from stores concentrated in poorer neighbourhoods, and spent on services whose users skew poor. It is the exact inverse of how taxation is supposed to work, and it is the mechanism by which the least advantaged people in the country partly fund their own assistance.

The poor funding the poor, with a third skimmed for administration.

VIAnd it doesn't buy what they think it buys

There's a folk story about lottery winners that everyone knows: they blow it, they're broke in five years, the money ruins them. It's told as a warning and it functions as a comfort. It is also mostly false.

Hankins, Hoekstra and Skiba tracked 35,000 Florida lottery winners against bankruptcy court records. Under 6% ever filed — not the 70% of internet lore. Winners of $50,000 to $150,000 were somewhat more likely to file than non-winners, but the finding that matters is subtler and much bleaker: the money postponed bankruptcy by a couple of years rather than preventing it. At five years there was no difference. And the large winners who eventually filed had net assets and unsecured debt indistinguishable from people who'd won almost nothing.

The money passed straight through without changing the underlying position.

Swedish data using the same natural experiment finds winners keeping their wealth over a decade, reducing work only modestly, and reporting significantly higher satisfaction with their lives and finances. Nobody is being destroyed. They're just not being rescued either.

Which brings this back to The Price of Time, and to a study that settles something that piece could only correlate.

That essay laid out the gap in life expectancy across the income distribution — ten to fifteen years, depending how you cut it — and presented it as correlation, breaking out education, race and the other factors that travel alongside income. What it couldn't do was isolate money itself. Nobody can, observationally. Wealth never arrives on its own.

Except in a lottery, where it does.

Cesarini, Lindqvist, Östling and Wallace used Swedish records to ask the question directly: if you make people richer purely by chance — by lottery — do they then live longer? The sample was huge, almost nobody dropped out of the data, and because the winners were picked at random, the winnings are the only thing separating them from everyone else.

They found no evidence that it does. Winning moved how long people lived by an amount you can't tell apart from zero.

It's worth being exact about what that rules out, because it's the whole point. Across the population, richer people really do outlive poorer people — that gap is real and large, and it's the reason wealth and lifespan look linked in the first place. Take the size of that gap as the yardstick. This study was precise enough to say that the part actually caused by the money — what you'd get from the cash itself — is at most a sixth of it, and most likely none of it.

So the gap is real, but the money is almost none of the reason for it. Wealth travels with the things that genuinely lengthen a life — education, the kind of work you do, where you grew up, the accumulated conditions of an entire childhood — without being the ingredient doing the work. Hand someone the cash in adulthood and the extra years don't arrive with it.

The caveat matters and I'll state it plainly: Sweden has universal healthcare. In a country where a medical bill can bankrupt you, a windfall might do more. A US replication would be genuinely valuable and doesn't exist.

But take the finding at face value and the golden ticket is emptier than even the cynical reading assumed. It isn't only that the billion is really $378 million. It's that the thing people are actually buying — the escape, the extra years, the different life — may not be a thing money buys.

VIIThe objection, and my answer to it

The obvious response to everything above is a practical one, and it deserves a straight answer rather than a dodge.

Fine. Scrap the lottery. Where does the thirty-four billion come from? Does it get raised somewhere fairer, or do the services just get cut?

Nobody knows. It's genuinely unresolved, and I can't settle it. There's no natural experiment for abolishing a state lottery, and the states without one differ in too many other ways to serve as a control.

There's also a serious argument on the other side, which I'd rather present than have someone find. Allcott and colleagues built a welfare model from their own estimates and concluded that current multi-state lottery designs increase social welfare — mainly because lotteries raise public funds that would otherwise have to come from somewhere. The result is conditional: if behavioural bias is more than about twice their baseline estimate, lotteries reduce welfare instead. And under a zero-bias assumption, low-income heavy spenders derive the most surplus of anyone.

My response is that the welfare gain in that model runs through government revenue. The state's benefit from the transaction is counted as a plus. Which is precisely the circularity this essay is about: the argument that the lottery is good because the state profits from it is not a defence of the lottery, it's a description of the problem.

But the deeper answer is that I think the question is aimed backwards.

"If we stop taking disproportionately from the least advantaged, what will we do instead?" is not an objection. It's an admission. It concedes that the current arrangement works by extracting from people with the fewest options, and then treats that extraction as a constraint on the alternatives.

If a service can only be funded by selling hope to people who need hope most, the funding model is the thing that's broken, not the proposal to end it. That's a good problem to have. It's a problem about how a wealthy country chooses to raise money, and wealthy countries have many options.

There's a name for an arrangement where need buys entries, and it's older than any state lottery. In The Hunger Games, the poorest children could put their names into the reaping extra times in exchange for grain — so that being hungry raised your odds of being taken. The book made it a horror. Strip away the fiction and the shape is familiar: the less you have, the more the system invites you to enter, and the entering is the cost. The only real differences are direction and honesty. The reaping came once a year and nobody chose it; this comes twice a week, all year, forever, and buying more chances isn't so much permitted as advertised.

Panem, at least, was honest about what the arrangement was. Ours calls it a public good, prints the odds on the back, and sends the money to schools. Every player knows the house always wins — that was never the secret. The secret, the thing the palm trees and the sunsets and the dollar and the dream are all built to keep you from saying out loud, is who the house is.

Sources & method

Prize structures and odds come from Powerball's published prize charts for Powerball, Power Play, Double Play, and Millionaire for Life — the last a new game launched 22 February 2026, replacing Cash4Life and Lucky for Life. New York added Double Play on 2 June 2026, and Mega Millions relaunched in April 2025 at a $5 price with a built-in multiplier. All odds and fixed prize amounts are taken directly from those charts rather than secondary summaries, which are frequently out of date on all three of these recent changes.

The opening. The game names are documented New York instants — Big Money, Fast Money, My Money, Sweet Million, Magnificent Millions, Cashword, 200X — as is the $30 ticket carrying a $10,000,000 top prize. Both slogans, all you need is a dollar and a dream and hey, you never know, were written for the New York Lottery by DDB New York; the agency also produced good things happen in an instant for the instant-game line. The visual details in the first paragraph (palm trees, convertible, sailboat) are archetype rather than a specific verified campaign — see open decisions.

National sales, prizes and net revenue come from the US Census Bureau's Annual Survey of State Government Finances, fiscal 2024: $104.7 billion in sales, $70.2 billion in prizes, $34.5 billion in net revenue, against $52.8 billion and $20.6 billion in fiscal 2008. A trap worth naming: NASPL reports different totals — $113.5 billion for 2024, $109.4 billion for 2025 — because of differences in fiscal-year definition and the treatment of video lottery terminals. The two series should never be mixed. This piece uses Census throughout.

On sales versus losses. Lottery sales figures are handle, not player losses. Prizes recirculate — a scratch player's winnings are largely re-wagered, and the same dollar appears in the sales total repeatedly. The essay therefore treats the $34.5 billion residual, not the $104.7 billion of sales, as the extraction figure. Prize payout ratios vary by game and state; Massachusetts, the highest per-capita lottery state in the country at roughly $856–935 against a national average near $320, reported payouts of 74.6% and 76.2% in April 2024 and April 2025. Instant tickets typically make up 65–70% of sales in that market. Generalising from Massachusetts is risky and the piece says so.

Tax figures model a New York City resident at top marginal rates — 37% federal, 10.9% New York State, 3.876% New York City — computed bracket-by-bracket rather than as a flat top rate, which matters because it makes annuity payments meaningfully cheaper to receive than lump sums. Federal bracket thresholds for 2026 are estimates. The domicile treatment follows New York Tax Law §631, which places lottery prizes over $5,000 in New York-source income for non-residents, and the absence of any New York City non-resident income tax. The SALT deduction is ignored throughout; at these income levels the cap phases to $10,000 and the effect is under half a percentage point. The gambling loss provisions are IRC §165(d) as amended by OBBBA §70114, effective for tax years beginning after 31 December 2025. The Fair Bet Act, which would repeal the 90% cap, was pending at the time of writing. None of this is tax advice.

Ticket volume is the weakest input here and is flagged as such. Lotteries do not publish per-draw sales in usable form. The figure of roughly 63,000 tickets per $1 million of advertised jackpot is fitted from observed jackpot cash-pool growth between draws and cross-checked against Match-5 winner counts in large draws. It is an estimate, and the Figure 2 sales series is marked as modelled rather than observed.

Who plays, and how much. Participation figures are Gallup, June 2016, n=1,025, ±4 points overall — but with roughly 340 respondents per income band the subgroup margin is nearer ±7, and the survey measures only whether a person bought any ticket in twelve months. It cannot measure intensity, which is the question that matters. Spending figures come from two studies commissioned by state lotteries themselves: South Carolina (2014), finding sub-$35,000 players spending more than double $100,000–$150,000 players; and Massachusetts (2016), finding the top decile of players generating about 40% of sales. Clotfelter, Cook, Edell and Moore (1999) provide the share-of-income estimate. Self-reported gambling expenditure is systematically understated across all such instruments.

Retailer placement comes from the Howard Center for Investigative Journalism's 2022 Mega Billions investigation, which matched lottery retailer locations in 44 states and DC against census demographics and used SafeGraph mobile location data covering roughly three-quarters of US retailers to establish customer catchments. It is journalism, not peer-reviewed work, but the methodology is documented and it is the best available evidence on the question. It supports claims about where lottery access is concentrated and who lives nearby; it does not observe individual spending and is not used here as if it did. One forward-looking caveat: as online lottery sales expand, retailer-location analysis will weaken as evidence. It doesn't affect the sources used here — Massachusetts had no online lottery until summer 2026, South Carolina's study predates US iLottery entirely, and online sales were a small share of a minority of states during the Howard Center's study period.

Behavioural evidence comes from Cook and Clotfelter, American Economic Review 83 (1993); Haisley, Mostafa and Loewenstein, Journal of Behavioral Decision Making 21 (2008); and Allcott, Lockwood, Taubinsky and Sial, Review of Economic Studies (2024). The 43% figure is a regression projection, not an experimental result — nobody was taught probability and observed to spend less — and is described that way in the text.

Winner outcomes come from Hankins, Hoekstra and Skiba, Review of Economics and Statistics 93:3 (2011), and Cesarini, Lindqvist, Östling and Wallace, Quarterly Journal of Economics 131:2 (2016).

Two disputes are flagged and not resolved: whether low-income lottery play is driven by innumeracy or by a rational response to unfairness elsewhere; and whether abolishing lotteries would shift revenue to a fairer source or simply cut services.

One figure could not be sourced and has been left out: the share of jackpot winners who elect the cash option rather than the annuity. It's widely described as very high. No verifiable figure was found, so no number appears.