Is it better to bet overs or unders on player props?

Somebody will tell you unders are the smarter side of a player prop. There is a hit rate that appears to prove it, and for a while I believed the version of this argument that I could measure.

Across 1,776 settled MLB prop leans we tracked between 7 and 26 August, unders hit 55.6 percent and overs hit 46.0 percent. Nearly ten points. If a hit rate were the whole story the argument would be finished.

It is not, and the missing step is the one most prop content skips entirely.

What each side needed

A price is a break-even rate wearing different clothes. Convert every price in that sample to the probability it implies, average those, and you get what each side actually had to hit to come out level.

It is worth seeing both sets of clothes. Across this sample the average over was priced at the equivalent of +122, and the average under at roughly -120. A +122 returns $2.22 for every dollar staked, so it pays for itself at 45 percent and every win above that is profit. Something near -120 costs about $1.20 to win a dollar, so it has to land closer to 55 just to stand still. Neither side needed 50 percent, because neither side was priced at even money, and almost no player prop is.

Those two prices are not the two sides of one market. Every lean in the sample is its own prop with one side recorded, so the two figures describe different markets rather than a book’s two-way line.

The unders were not winning more often for free. They were winning more often because they were priced as the likelier outcome, and they cleared that bar by roughly the same margin the overs cleared theirs.

Calibration is simply how often something happened divided by how often the market said it would. It lands within half a point of identical on both sides.

A cleaner check, on a different sport

One dataset can flatter you, especially one built from lines we chose to look at. So here is a test that cannot. Across nine WNBA slates we logged both sides of every prop on the board, whether we liked it or not. That is a census rather than a selection.

Overs hit 47.77 percent of 942 settled props against a devigged 48.73 percent. Bootstrapping by game puts the gap somewhere between minus 4.5 and plus 2.5 points, so the effect is about a fifth the size of the uncertainty around it.

The census also shows where the folklore starts. The market made the over the favourite on only 40.4 percent of props. Unders win more often because unders are more often expected to.

What to do with it

Read no hit rate without its price. That is the entire lesson and it has survived every dataset we have put it through.

• A player who has gone over in 8 of his last 10 is not a bet yet. He is a bet at some numbers and a pass at others, and the number decides it.
• When a tout shows a hit rate with no odds attached, the one variable that could make it meaningful has been removed.
• If you track your own bets, record the price at entry. Without it you can measure whether you were right and never whether you were paid properly for being right.

This is why every line on the Lab board carries its price and its devigged fair number beside the form, and why today’s MLB player props publishes the rate and the price it was taken at together, every morning, before first pitch. A rate on its own decorates a page. If you want the arithmetic behind the fair number, we wrote that up in how to devig odds.

How we measured this

1,776 settled MLB form leans, 7 to 26 August 2026 across 18 slate dates, at the best price available when each lean was recorded. These are reads off the board rather than bets this desk placed, and they are MLB only.

Break-even is the mean of the implied probability of every price in the sample. It is not the implied probability of the mean price, which is not a number that means anything: American odds are not a linear scale, so averaging them produces a figure nobody could have bet.

The prices quoted above are that same mean probability converted back: 45.015 percent expressed as a price is +122, and 54.593 percent is about -120. They are the price equivalent to an average probability, not an average of prices. Rounding to a price a book would actually post costs a few hundredths of a point in each direction, which is the floor of what whole-number American odds can express.

Repeating the whole thing on bettable prices only, with sharp exchanges removed, leaves 1,385 leans and gives calibration of 1.021 for overs and 1.035 for unders. Same conclusion.