KEY TAKEAWAYS

• A win probability model estimates from game state; a live price is that estimate plus a margin plus market flow.
• Converting a posted price straight to a percentage overstates what the sportsbook actually thinks.
• The order is always convert, de-vig, then compare — skipping the middle step invents disagreements.
• A -250 / +200 live market implies 71.43% and 33.33%, which de-vig to 68.18% and 31.82%.
• The size of a probability gap means little until you price it as expected value.
• A gap usually reflects something the model cannot see, such as an injury the market has already priced.

During a live game you are often looking at two numbers that claim to describe the same thing. A broadcast or analytics feed shows a win probability — say 68% for the home team — while the sportsbook shows a live moneyline. Both are statements about how likely a result now looks, yet they rarely agree, and bettors frequently treat the difference as a signal when it is often just a difference in construction. A win probability model is an estimate built from game state. A live price is a traded number that carries a margin, reflects what the book knows, and moves with the money coming in. Comparing them is genuinely useful, but only after the price has been converted into the same units, which is a question of implied probability rather than of judgement. This article explains how the two numbers are built, how to convert one into the other, and what a gap between them does and does not mean.

Win Probability and Implied Probability Are Not the Same

A win probability model takes the current state of a game and estimates how often, historically, teams in that situation have gone on to win. The model’s output is a pure probability estimate with no commercial component attached to it, derived from score, time remaining, possession and whatever other state variables it tracks. It is a description of the game, not a price.

A live betting price is a different kind of object. The number a sportsbook posts is a probability estimate plus a margin plus whatever the book has learned from the betting it has taken, and the way those prices move through a game is covered in how live betting odds change during a game. Converting that price straight into a percentage therefore gives you something slightly larger than the book’s actual opinion.

The practical consequence is that a raw comparison is always unfair to the market. Implied probability taken directly from a posted price overstates the chance the sportsbook actually assigns, because the margin has not yet been removed. Any gap you find before de-vigging includes the book’s built-in edge, and mistaking that for a disagreement is the single most common error in this whole exercise.

How to Calculate Win Probability From Live Odds

Working out how to calculate win probability from a sports betting price is a three-step process, and skipping the middle step is what produces most false signals. The sequence is convert, de-vig, then compare — in that order, every time.

Step 1: convert the price to implied probability

Use the standard conversions. For a favourite, implied probability is the absolute value of the odds divided by that value plus 100; for an underdog it is 100 divided by the odds plus 100. A live price of -250 implies 250 / (250 + 100) = 71.43%, and the other side at +200 implies 100 / (200 + 100) = 33.33%.

Step 2: remove the margin

Add the two implied probabilities together and divide each by the total. Those two figures sum to 104.76%, so the margin is 4.76%, and dividing 71.43% by 1.0476 gives a de-vigged estimate of 68.18% for the favourite and 31.82% for the underdog. This is the step covered in detail in how to calculate no-vig odds, and it is what turns a price into something comparable to a model output.

Step 3: compare like with like

Only now is the comparison meaningful. The book’s genuine estimate in that example is 68.18%, not the 71.43% the raw price suggested, so a model showing 68% is in near-perfect agreement with the market even though the unadjusted figures looked three points apart. Most apparent disagreements disappear at this step.

A Worked Comparison

Suppose the live favourite is priced at -250 while an in-game model shows that team at 75%. After de-vigging, the market’s estimate is 68.18%, so the genuine disagreement is about 6.8 percentage points rather than the 3.6 points the raw conversion implied. The gap only becomes a betting question once you ask what that difference is worth at the posted price, which requires the payout rather than the probability.

At -250, a $100 wager returns a profit of $100 x (100 / 250) = $40, for a payout of $140. If the model’s 75% were exactly right, the expected value of that $100 bet would be (0.75 x $40) – (0.25 x $100) = $30 – $25 = +$5, or +5% of the amount risked. That is a real theoretical edge, but it exists only if the model’s 75% is closer to the truth than the market’s 68.18%, and nothing about the arithmetic establishes that. The same calculation with the market being right instead gives (0.6818 x $40) – (0.3182 x $100) = $27.27 – $31.82 = -$4.55. These are illustrative prices rather than current market odds, and the two results show how completely the conclusion depends on which estimate you trust.

How to Read a Gap Between Model and Market

The honest starting assumption is that a gap reflects something the model does not know. Public win probability models generally work from game state alone and are blind to injuries, substitutions, foul trouble, pitch counts and anything else the market has already priced. A model that still has a star player on the field after he has limped off will look confidently wrong for several minutes.

Timing cuts the same way. A live price incorporates new information within seconds, while a model updates on the discrete events its data feed records, so a disagreement during a fast sequence of play is usually a lag rather than an insight. Prices that have been suspended and reopened at a new number have already absorbed something the model may not have seen at all.

There is also the question of what the market as a whole represents. A traded price aggregates many informed opinions along with the book’s own modelling, which is a materially different input set from one model’s output. Treating a single model as the benchmark and the market as the error is a strong claim, and it needs more support than the fact that the two numbers differ.

Common Mistakes

The most frequent mistake is comparing a model’s probability against a raw implied probability. Skipping the de-vig step manufactures a disagreement of roughly half the margin on every single comparison, which is enough to make a perfectly efficient market look consistently mispriced.

A second error is treating a probability gap as if it were a profit. A seven-point disagreement at a long price and the same disagreement at a short price are worth very different amounts, which is why the expected value calculation matters more than the size of the gap itself. Probability differences do not translate linearly into money.

The third mistake is the most damaging in a live context. A high win probability is not a statement that the result is settled, and a team at 90% still loses roughly one time in ten. Chasing a losing position because a model says the outcome is likely is how in-play betting turns from a view into a reaction, a dynamic also at work in whether momentum matters in live betting.

Where the Two Diverge in Practice

Divergence clusters around specific moments rather than appearing at random. The widest and most persistent gaps show up right after a significant event, when the market has repriced and the model has not yet caught up, or in the reverse case where the model reacts mechanically to a scoring play that the market considers less meaningful than the raw state implies.

Late-game situations are the other recurring case. Models that treat time remaining as the dominant variable can be slow to account for a specific tactical position, such as a team holding possession with no intention of scoring, where the game state understates how settled the result actually is. In those spots the price often looks expensive against the model precisely because the market is right. This is also where cash out offers get compared against model numbers, and the same de-vigging discipline applies to judging them.

This topic sits on top of two separate foundations, and it is worth having both before acting on any gap you find. The pricing half rests on turning odds into clean probabilities, which makes no-vig odds the prerequisite rather than an optional extra. The decision half rests on expected value, which converts a probability disagreement into a figure you can actually evaluate, and together they turn an interesting comparison into a usable one.

Frequently Asked Questions

How do you calculate win probability from betting odds?

Convert each side’s price to implied probability, add the two together, then divide each by that total to remove the margin. The resulting pair sums to 100% and represents the sportsbook’s own estimate rather than its posted price.

Why do live odds and win probability models disagree?

Because they are built differently. A model reads game state from a data feed, while a price also carries a margin, the book’s own information about injuries and lineups, and the effect of money already wagered.

What is Elo win probability?

It is a probability derived from a rating system that assigns each team a number and converts the difference between two ratings into a chance of winning. It describes general team strength, so on its own it does not reflect live game state.

Does a bigger gap mean a better bet?

Not by itself. The value of a disagreement depends on the price attached to it, so the same percentage gap is worth different amounts at different odds. It also assumes your model is the more accurate of the two, which is rarely established.

Does a 90% win probability mean the game is over?

No. It means that in similar historical situations the result went that way about nine times in ten, which leaves one in ten going the other way. A high probability describes a likelihood, never a certainty.

Why does the sportsbook’s number look higher than the model’s?

Usually because the margin has not been removed. A raw conversion of a posted price always overstates the book’s real estimate, and on a typical live two-way market that inflation is worth a couple of percentage points on each side.