Hold is not the house edge

Open any report in the series and the statewide summary gives every game and every slot denomination a column headed "Win Percent". It reads as one measurement applied consistently. It is two.

For slot machines the board is explicit: "The “Win Percent” for slot devices provides a ratio which represents the reported win amount divided by the total dollar amount played by patrons." That is win divided by handle. It is the same quantity as a house edge, in the same units, and the complement of the return-to-player figure a manufacturer prints. When this study says a denomination kept $7.06 per $100, that is exactly what it means.

For table games the same page says something much shorter: "The “Win Percent” for games provides a ratio which has been adjusted for effects of credit play." What it does not say, because the industry takes it as read, is what the ratio is over. It is drop: the money and markers that went into the box at the table, which is what players bought in for, not what they bet. A player who buys in for $200 and plays it through twenty hands has generated $200 of drop and far more than $200 of wagers.

This is not a subtlety anyone in the industry disputes. The standard public reference on the subject, Robert C. Hannum, University of Denver's A Guide to Casino Mathematics for the UNLV Gaming Studies Research Center, is blunt about it: hold percentage "is equal to win divided by drop", and is "sometimes erroneously called win percentage", and a casino "will not in the long term keep 24% of the money bet on the spins of roulette wheel". The confusion survives anyway, because the two numbers are printed in the same column of the same table under the same heading.

Everything in the next section follows from that table. Everything after it is about slots, where the denominator is the amount wagered and the arithmetic is simple.

Over the twelve reports to 2026-07, Nevada's casinos won $1.3bn at blackjack against $9.9bn of drop, a reported win percent of 12.79%. Our own house edge table puts the cost of six-deck blackjack played to basic strategy at 0.41% of each wager, derived from this site's engine rather than transcribed.

Those two numbers are thirty-one times apart and both are correct. To see why, divide one by the other for every game the board reports separately and this site publishes an edge for. The result is not a house edge and is not a rate. It is the number of times a dollar of buy-in would have to be wagered for optimal play at the main bet to produce the hold the state actually reports.

Read the bottom row first. Roulette lands at 3.6 times, and roulette is the one game on the list where a player has essentially no way to make the price worse: nearly every bet on an American layout costs the same 5.26%, and there is no strategy to misplay. So its multiple is the closest thing here to a clean reading of how many times a dollar of buy-in gets bet before it leaves the table, and the answer is somewhere near three and a half.

Now read the top row. Blackjack's 31.2 times cannot be turnover. Nobody's buy-in goes round thirty-one times. What it is instead is a measurement of the distance between how blackjack can be played and how it is played: side bets priced several times the base game, hands misplayed against the chart, insurance taken. The cheapest game on the floor produces the largest gap in this table, and that is not a coincidence. A game only has room to be played badly if it has a strategy to get wrong.

Keno is on the chart as the control, and it is the reason to believe the rest of it. A keno ticket's drop is the amount written on it, so for keno alone the denominator the board divides by is the amount wagered, and its multiple should be one. It comes out at 0.99: a reported 26.79% against a house edge of 27.00%. The arithmetic behaves exactly as it should where the two denominators agree, which is what licenses reading everything above it as the denominators disagreeing.

One number in this section has visibly moved since it was written down. A Guide to Casino Mathematics uses Nevada roulette at about 24% as its illustration, and it is quoted onward from there constantly. In the board's own series, statewide roulette last held 24% of drop in the years before this archive opens. The trailing twelve months come in at 19.06%, and the UNLV University Libraries, Center for Gaming Research compilation puts the annual average below 20% for most of the past two decades. The guide's argument is unaffected. Its example is out of date.

That Nevada's slot hold has risen is established and not ours. The UNLV University Libraries, Center for Gaming Research publishes it annually, its August 2026 edition puts the statewide rise since 2004 at 29%, and it was covered widely when that edition landed. Our series reproduces it: $5.72 kept per $100 wagered in 2004, $7.13 in 2025, 16 of 20 year-on-year comparisons upward.

What nobody appears to have published is why, and there are two candidate explanations that carry completely different meanings for a player. Either the machines themselves were repriced, or players migrated onto machine types that were always more expensive. The floor changed enormously over these years, so the second is not a straw man: penny machines and multi-denomination cabinets took the floor from the quarter and dollar reels that dominated it in 2004.

The two explanations separate cleanly, because the board reports each denomination's win and win percent separately, and a denomination's share of the money wagered can be derived from both. Split the change into the part explained by shares moving at each denomination's average price, and the part explained by each denomination's own price moving at its average share.

Because win percent for slots is win over handle, and the board publishes both the win and the percent, the handle falls straight out of the arithmetic. It is not a figure anyone publishes, and over the whole series it comes to $2.73 trillion wagered into Nevada slot machines for $179.7bn of casino win.

The year-by-year shape of that number is the part worth sitting with. The most money ever wagered into a Nevada slot floor in a calendar year was $150.3bn, in 2025. Nineteen years before 2025, in 2006, players wagered $138.0bn. In 2025 they wagered $150.3bn: 8.9% more, in dollars that are not adjusted for inflation at all. Over the same nineteen years the casinos' win from slot machines rose 29.0%, from $8.3bn to $10.7bn.

Nearly all of the growth in what Nevada's slot floors earn, then, has come from the price of a dollar wagered rather than from more dollars being wagered. In nominal terms play is roughly flat across two decades; in real terms it is well down. This is a finding about the industry rather than about any player, and it is the single thing in this study we would most like someone with better inflation and visitation data than us to take further.

The denomination ladder in that table is the oldest known fact in this data and it survives every year of the series: the cheaper the machine is to play, the more of each dollar it keeps. It is not a trick and it is not hidden. It follows from fixed costs per machine spread over much smaller average bets. It is worth stating plainly anyway, because it runs exactly opposite to the intuition that a machine advertised at a low price is a low-priced machine.

Monthly gaming figures get reported as news, and the reported win percent in them gets read as though it were a property of the games. Over the last nine complete years in the series, one month of baccarat came in as low as 6.67% and as high as 26.47%. Nothing about the game changed between those two months. A small number of very large players did.

The ordering is the useful part. Table games with big average bets and few decisions swing hardest; slot floors with millions of small spins barely move at all. All slot machines has a standard deviation of 0.41 points across the same period, which is why the slot series in this study can be read year by year and the table series cannot.

This is the law of large numbers doing its ordinary work, and it is the same reason a player's own session tells them nothing about a game's price. You can check what the distribution of a few hundred spins actually looks like on our slot simulator, and what it does to a bankroll on the variance simulator.