The Sports PageThe Crowd Splits Every Game in Half. You Still Lose a Thousand Dollars.
A betting line is remarkably good: across 4,360 games, every simple rule we tried landed within a point of 50%. The catch is the price. Two sides at −110 add up to 104.76%, and that surplus never plays a down.
The question: if a betting line is a fair guess at who wins, why does almost everybody lose?
Because the line is fair and the price is not. Both sides are offered at odds implying 52.38%, and 52.38 twice is 104.76.
Why that is surprising. People assume they lose because the house knows more. The house does not need to know more. The arithmetic does the work.
What you get: the ability to turn any price into a probability in your head, and to see what it costs you.
First, the good news about the crowd
A point spread really is a fine estimate. We checked it the hard way, over 4,360 college football games across three seasons, graded against the closing number.
| Rule | Record against the spread | 95% interval |
|---|---|---|
| Always back the home team | 50.3% | 48.8–51.8 |
| Always back the favorite | 49.1% | 47.6–50.6 |
| Our own rating model | 51.0% | 48.9–53.1 |
Every one of them is a coin flip. Thousands of strangers, pushing money around, land on a number that splits the outcome almost exactly in half. That is a genuinely impressive piece of collective machinery and we should say so.
So if the guess is that good, where does the money go?
Into the price
American odds of −110 mean you risk $110 to win $100. Turn that into a probability: 110 divided by 210 is 52.38%. That is how often the bet must win to be worth making.
Now look at the other side of the same game. Also −110. Also 52.38%.
52.38 plus 52.38 is 104.76.
The two outcomes of one football game are being sold to you as though they were 104.76% likely between them. They are not; they are 100%. The extra 4.76 points do not belong to either team. They belong to the house, and the house never plays a down.
The line is an honest guess. The price is where the money is made, and it is printed in a notation most people cannot read.
The ProfessorWhat it costs
Read the crossing point. You do not break even at 50%; you break even at 52.4%.
Be right exactly half the time — which is what the crowd’s own number predicts for you — and every bet loses $5.00 on average. You risked $110 to make it, so that is 4.5% of your money, gone. Not occasionally. On average, forever.
Put that on a football season. Fifteen games a week, thirteen weeks, $110 a time. Being right half the time costs you $975.
You will not feel it happening. You will have winning weeks. The money leaves in a thin, even trickle that looks exactly like bad luck.
Why we call it a tax on innumeracy
Not an insult, and not a word about anyone’s intelligence. A tax is a transfer, and this one has an eligibility rule: it is paid by people who cannot convert a price into a probability.
If you can do that conversion, the 4.76 is visible. You may still play, knowing the entertainment costs about four and a half cents on the dollar — that is a legitimate choice and plenty of people make it.
If you cannot do the conversion, the surcharge is invisible. You experience a fair-looking contest at a fair-looking price, and lose steadily anyway. That is the whole mechanism, and it works precisely because the number is never printed.
Notice how much is hidden by notation alone. −110 does not look like a fee. Write the same thing as “we are charging you 4.76%” and the business would have to explain itself.
The same skill, without the tax
There is a version of all this that costs nothing, and you have probably played it.
A pool among friends uses the same lines and the same judgement. The difference is that the money stays in the room. Nobody skims 4.76 points off the top, so somebody in your family actually wins what everybody else loses.
The picking is identical. Only the arithmetic underneath it changed.
What to take home: read the price, not the pick
Two things to carry.
Convert before you consider. A minus price is the stake over the stake plus the payout: 110 over 210 is 52.4%. A plus price is 100 over the payout plus 100: +150 is 100/250, or 40%. Do that before you think about the teams at all.
Then add the two sides together. Whatever they exceed 100 by is what the house takes. If it is 4.76, that is ordinary. If it is 8, you are being charged double and should say no on arithmetic alone, without an opinion about the game.
You are not being asked to predict football better than thousands of people with money at stake. You are being asked to read a price. That one is winnable.
Baseball is ranked by championship leverage: we simulate the rest of the season 60,000 times, then split those seasons by who won each of today's games. College football cannot be simulated that way, so it is ranked by how close to a coin flip the ratings make it — a lopsided game teaches you nothing. Tomorrow this block reports what happened to today's pick.
Notes & sources
The implied probability of American odds of −110 is 110/(110+100) = 52.381%. Both sides of a standard spread are offered at that price, so the two implied probabilities sum to 104.76%; the excess over 100% is the overround, commonly called the vigorish or juice. Normalized back to 100%, each side is exactly 50.00%.
Expected profit on one bet at −110 is 100p − 110(1−p), where p is how often you win. That is zero at p = 110/210 = 52.38% and −$5.00 at p = 0.50. Note the stake is $110, not $100, so −$5.00 is 4.545% of the money risked. The season figure is 15 bets a week for 13 weeks: 195 × −$5.00 = −$975, against $21,450 risked, which is the same 4.55%.
Two numbers here are close together and are not the same thing. The overround is 4.76 percentage points — the amount by which the two implied probabilities exceed 100. The hold, what the house keeps as a share of money wagered, is 4.76/104.76 = 4.545%. The headline rounds $975 to a thousand; the exact figure is here, as it should be.
Rule records are from 4,360 FBS games, 2023–2025, graded against the mean closing spread across books via the CollegeFootballData API, pushes excluded. Our own model’s 51.0% is from Issue #154: 2,140 games, walk-forward, ratings rebuilt each week from earlier weeks only.
Two limits. This piece is about the PRICE, not about how a line is set — the common account, that books simply balance the money on each side, is more complicated than it sounds and we are checking it properly before writing that piece. And −110 is the standard price, not a universal one; reduced-juice books and some markets differ, which is exactly why the habit of summing the two sides is worth more than memorising 4.76.
Related: Issue #118 on the invention of the point spread, and Issue #150 on why probabilities are handled in log-odds.
