Line shopping: what a better number is actually worth

Every article on line shopping tells you the same thing: books disagree, so open more accounts and always take the best number. The first half is true and measurable, and this page measures it. The conclusion is where they all go wrong — because the better number is not free, and once you pay for it the arithmetic looks very different from the way it is usually sold.

How far apart the books actually are

Not about who wins — about the number. At the close, this is how much the widest and narrowest spread on the same game differ by.

The mean gap is 0.632 points and it is non-zero on 81% of games. Fewer than one game in five closes with the market in agreement. On a tenth of the slate the books are a point and a half or more apart, which is roughly the difference between betting a game at −3 and betting it at −1.5 — the same side, and not remotely the same bet.

Half a point does not sound like much until you price it. A point of NFL spread is worth about 3.1 percentage points of cover probability, and the whole edge in a −110 bet is 2.4 points. The gap between two books on an ordinary Sunday is larger than the house edge you are trying to beat.

What you can actually hold, and what it costs

Dispersion is not the same as opportunity. What matters to a ticket is the best number a bettor could really have taken on that side, and the price sitting next to it.

Read the third column, not the second. The extra half point is available on 55.6% of sides, and where it is available the median price is −115 rather than −110. The full point costs −118. The books are not leaving the better number lying around; they are selling it, and they are pricing it about right.

Taken across every side, betting the best available number pays a mean of 0.8999 per unit staked, against 0.9091 for a flat −110. You gain +0.445 points of line and hand most of it straight back in the price.

The part every other guide leaves out

A bettor who blindly takes the best available number on every bet goes from roughly −4.3% to roughly −2.0%.

That is a genuine, permanent, effort-free improvement, and it is still a losing number. Shopping halves what the house takes. It does not decide whether you win, because it does nothing about which bets you make — and that is the half that has to be positive before any of this matters. Selection is what turns a card profitable; shopping is what stops the vig eating the result once it is.

This is worth being blunt about, because “line shopping is free money” is the single most repeated claim in betting content and it is repeated by people paid per account opened. The honest version is duller and more useful: shopping is a discount on your losses, not a source of profit. Take it every single time, and do not mistake it for an edge.

How many accounts is enough

Sub-linear, as you would expect — each additional book is less likely to be the one holding the outlier.

Two books capture 0.095 points, three 0.130, five 0.210. Eleven reaches 0.316 — so going from five books to eleven buys about as much as going from one book to two. Five is where it stops mattering much. The sixth account is a chore, not a strategy, and the fourth is worth more than the tenth ever will be.

Two looks beat six more accounts

The more useful axis is not how many books you hold but how many times you look. A line is a thing in motion, and dispersion at any one instant is a snapshot of some books moving before others.

Shopping only the closing board is worth +0.316. Shopping only the board a day out is worth +0.317 — the same, within noise. Reading both is worth +0.445, and lifts the share of sides where you get half a point or better from 51% to 69%.

The reason is that 59% of the best numbers first appear a day before kickoff rather than at the close. Two looks catch two different books being slow. Six more accounts catch the same instant six more times.

Which books are worth shopping

Books are not uniformly good or bad; they are good and bad at different things, and a book that is cheapest on spreads can be among the dearest on props. Rung for rung on the same bet, the gap between the best and worst book on this board is 1.49 points of required win rate — about $1.91 per point on a $100 bet.

On game lines, the top book posts the best available price on 71% of the bets it takes, while one book on the same board has the best price on 3% of them — shopping away from that one pays almost every time. On player props the ordering changes completely. The current, regenerated table is in the sportsbook rankings, ranked separately for game lines, props and teasers.

What to actually do with this

• Hold about five accounts, not eleven. Past five the curve flattens and you are managing accounts instead of betting.
• Look twice: a day out, and again before kickoff. Worth more than doubling your number of books, and it costs you two minutes.
• Compare the number and the price together. −3 at −118 is not obviously better than −3.5 at −105, and taking the best line without reading the price is how shopping stops paying.
• Shop props separately from game lines. The book with your best spread is very unlikely to have your best prop.
• Do not treat it as an edge. It is a discount. It makes a winning card win more and a losing card lose slower.

How this was measured

11 books, two snapshots of every board — one a day before kickoff, one half an hour before — over 2024-2025. Books are only ever compared on the same game at the same moment.

Two contaminations were removed first, and both were large enough to have produced a wrong answer. The odds feed carries an incorrect kick-off time for some games, so a “closing” snapshot could actually have been taken after the game started — one game showed a 15.5 spread priced −1000/+600 on a game that closed at 6.5. Lead times were recomputed from the play-by-play kick-off and anything at or past it dropped. Six further games whose consensus sat more than 1.5 points off the recorded closing line were dropped outright.

Before those fixes the board disagreed with the closing line with a standard deviation of 2.095 points; afterwards, 0.451. An early version of this analysis reported half a point converting at 59.85%. It does not, and that number should not be quoted from anywhere.

For what a half point is worth when you buy it outright rather than find it, see buying half points: what they cost and what they are worth. For why some numbers matter far more than others, NFL key numbers: why 3 and 7 decide everything. And for converting any of this into the win rate a bet needs, the odds calculator.

Common questions