How to Read a Win Probability
What 65% actually means — and what it doesn't
The Short Version
A win probability is a frequency, not a forecast. When we say a team has a 65% chance to win, we mean: if this exact game were played 100 times under these exact conditions, we'd expect that team to win about 65 of them — and lose about 35. The number is the whole distribution, not a call on the final score.
That distinction is the reason people misread these numbers. 65% feels like “they're going to win.” It isn't. It's closer to “they're the favorite, and you should still expect an upset about one time in three.”
Is 65% Good? Depends What You're Asking
“Good” only means something relative to a baseline. A 65% favorite is a meaningful edge over a coin flip, but it's a modest one against the betting market, which is usually already pricing the same information. The useful question isn't “is 65% high?” — it's “is 65% right?”
Here's a rough map for reading the scale:
- 50–55% — essentially a toss-up. The factors nearly cancel; don't over-read the leader.
- 55–65% — a real but small edge. Favorites here still lose often.
- 65–75% — a clear favorite. Upsets are common but no longer expected.
- 75%+ — a strong favorite. When we say 80%, the other team still wins 1 in 5.
A Confident Number Isn't an Accurate One
The single most important idea here is calibration. A model is well-calibrated when its stated probabilities match reality over the long run: the games it calls 70% are won about 70% of the time, the games it calls 30% are won about 30% of the time, and so on across the whole scale.
Calibration is what separates an honest model from a confident one. Anyone can print big numbers. A site that labels every favorite 85% will look decisive and be wrong constantly. The only way to know a probability is trustworthy is to check, over hundreds of games, whether the percentages held up. We track exactly that and publish it on our accuracy page, and we explain how the numbers are built on our methodology page.
The Number Isn't a Promise
Upsets aren't model failures — they're the model working. If 30% underdogs never won, our 30% would be wrong (too low). Some famous examples: the 2007 Giants beating the undefeated Patriots, or a double-digit seed making a deep March run. Those outcomes live inside the probabilities, not outside them.
So judge a probability the way you'd judge a weather forecast. One rainy day after a “20% chance of rain” doesn't make the forecaster wrong. Twenty rainy days out of a hundred such forecasts makes them exactly right.
Reading Probabilities Across Sports
The interpretation is the same in every sport, but the inputs differ — home ice matters less than home field, a back-to-back hurts a basketball team in a way baseball never sees. For how the number is actually calculated in each sport:
The Bottom Line
Read a win probability as a rate, not a verdict. 65% means “favorite, but bring a jacket — it rains a third of the time.” And before you trust any site's number, ask the only question that matters: when they say 70%, does 70% actually happen?
Related Topics
Frequently Asked Questions
It means the favorite wins about 65 times out of 100 and loses about 35. That is a real edge, but it is a long way from a sure thing — roughly a 1-in-3 chance the other team wins. Whether that is "good" depends on what you are comparing it to: the market, another model, or a coin flip.
No. A confident number is not the same as an accurate one. A model can say 90% on every favorite and be wrong constantly. What matters is calibration: over many games, do the teams we call 70% actually win about 70% of the time? That is the test, and we publish how we do on our accuracy page.
Even "close" games usually have a favorite. A 55/45 split still means one side is more likely — it just means the edge is small. We reserve numbers near 50% for games where the factors genuinely cancel out.
Not necessarily. A 30% underdog is supposed to win about 3 times in 10. If those underdogs never won, our 30% would be too low. A single result cannot tell you whether a probability was right — only the pattern across hundreds of games can.
Sources
On Calibration & Forecasting
- Brier, G. W. (1950). Verification of Forecasts Expressed in Terms of Probability. Monthly Weather Review. Link — the origin of the Brier score, the standard measure of probabilistic accuracy.
- Tetlock, P. & Gardner, D. (2015). Superforecasting: The Art and Science of Prediction. Crown. — on why calibration, not confidence, defines a good forecaster.
- Silver, N. (2012). The Signal and the Noise: Why So Many Predictions Fail—but Some Don't. Penguin Press.
Our Numbers
- PlayDecoded Accuracy — our tracked hit rate and calibration across completed games.
- PlayDecoded Methodology — how each probability is calculated and weighted.