ABC Tools

Badminton Match Probability, from One Number

After reading this you will be able to turn a single number, the chance a player wins a rally on their own serve, into exact game odds, best-of-3 match odds and a full game-score distribution for 21-point rally badminton.

What the calculator does

Badminton uses rally scoring: every rally scores a point, no matter who served. A game runs to 21, but you must win by 2, and the cap sits at 30. If the score reaches 29-29, the next rally decides it (the golden point). A match is best of three games.

The calculator asks for one number per player: the probability of winning a rally when that player is serving. From those two numbers it builds the exact Markov chain of a game and reports the chance each player wins a single game, wins the match, and the probability of each final score line (2-0, 2-1, and so on).

Here is the hook. Give Player A a 52% rally-win rate on serve and Player B 49%. Those look almost even. The tool returns a game win probability near 55% for A and a match probability near 58%. Small per-rally edges grow when you play to 21 and then again across a best-of-three. This article shows exactly why.

When to use it, and when not

Use it when you have a defensible estimate of each player's rally-win rate on serve and you want the downstream odds. That estimate can come from match statistics, from a rating model, or from your own scouting. The tool answers a clean question: given these two rally rates, what happens?

Do not use it to predict a specific real match with confidence. Rally rates drift with fatigue, tactics and the shuttle speed on the day. The model assumes each rally is independent and that the two rates are fixed for the whole match. That is a simplification. Treat the output as the logic of the scoring system applied to your inputs, not a forecast with a guarantee.

Because rally scoring gives the server only a small edge, badminton serve-rally rates cluster near 50%. A gap of 3 percentage points (52 versus 49) is already a clear favourite. Rates like 60 versus 40 describe a mismatch, not a close contest.

The rally model and the serve rotation

Two probabilities drive everything. Let s_A be the chance Player A wins a rally on A's own serve, and s_B the chance B wins a rally on B's own serve. The winner of each rally serves the next one. So the probability that A wins the next rally depends on who is serving now.

P(\text{A wins rally} \mid \text{A serves}) = s_A, \quad P(\text{A wins rally} \mid \text{B serves}) = 1 - s_B

Read the second part carefully. When B serves, B wins that rally with probability s_B, so A wins it with probability 1 - s_B. With s_A = 0.52 and s_B = 0.49, A wins 0.52 of rallies A serves and 0.51 of rallies B serves. A is slightly better in both situations, which is what makes A the favourite.

From rally to game with a Markov chain

A game state is a triple: A's points, B's points, and who serves next. From any state, one rally moves you to one of two new states. That is a Markov chain. The tool solves it exactly by working backward from the terminal states (someone reaches 21 with a 2-point lead, or the cap resolves the game).

The win-by-2 rule creates a repeating pattern once the score reaches 20-20. From there each pair of rallies either ends the game or returns to a tied score, so the deuce phase is a small geometric loop. Let p be A's chance of winning a single rally averaged over the alternating serve during deuce. The probability A wins from 20-20 is:

P(\text{A wins deuce}) = \frac{p^2}{p^2 + (1-p)^2}

The two-point swing squares the edge. If A wins 0.51 of rallies during deuce, then A wins the deuce phase with probability 0.51^2 / (0.51^2 + 0.49^2) \approx 0.520. The golden point at 29-29 replaces this loop with a single rally, so the cap slightly compresses the favourite's edge at the very end. The tool handles the cap exactly rather than letting the geometric loop run forever.

Notice the squaring. A per-rally edge of 1 point (0.51 versus 0.49) becomes roughly a 2-point edge at deuce (0.520 versus 0.480). Every layer of the scoring system, deuce then game then match, magnifies the same small advantage.

Worked example: 52 versus 49

These are the demo defaults. Reproduce them by loading the sample data and pressing Calculate with the iid point model.

  1. Set s_A = 0.52 and s_B = 0.49. A wins 0.52 of own-serve rallies and 1 - 0.49 = 0.51 of B's-serve rallies.
  2. Solve the game chain. Averaging over the opening toss, A's chance of winning one game to 21 comes out near 0.549.
  3. Apply best-of-three. A wins the match by taking 2 games first. With a per-game win probability g = 0.549, the match probability is g^2 + 2 g^2 (1-g), which equals 0.577.
  4. Split the game-score distribution. A 2-0 win for A has probability g^2 \approx 0.301, a 2-1 win has probability 2 g^2 (1-g) \approx 0.271. The same formulas with 1-g give B's outcomes.

So a 3-point rally gap turns into roughly 55% per game and 58% per match. The best-of-three barely changes the story here because the games are already correlated through the same rally rate: winning more likely games more often does not add much beyond what one game already tells you.

How A's game win probability climbs as A's serve-rally rate rises, with B held at 0.49. The steep middle shows how small rally edges become large game edges; the marker sits at the demo value 0.52.

Reading and interpreting the results

The three outputs answer different questions. Game probability is the cleanest single number: it isolates the scoring system from the match format. Match probability folds in the best-of-three, which always pushes the favourite a little further from 50%. The game-score distribution tells you how the match is likely to end, which matters if you care about margins or set betting.

A useful sanity check: if the two rally rates are equal, every output should read 50%, and each game score (2-0 either way, 2-1 either way) should be equally likely at 25% each. If your inputs are symmetric but the numbers are not, you have entered them wrong.

The momentum model is a separate view. It runs 20,000 full matches point by point, letting a won rally nudge the winner's next-rally chance up by the momentum strength. At strength 0 it matches the exact iid result within Monte Carlo noise (expect the match probability to wobble by about \pm 0.007 around 0.577). Higher strength fattens the tails: blowouts and comebacks both get more common, while the average match probability barely moves.

A slider sets Player A's serve-rally rate from 0.40 to 0.60 with Player B fixed at 0.49. The display shows three numbers side by side: A's per-rally edge over B, A's game win probability, and A's best-of-three match probability. At 0.52 they read about +0.01 per rally, 0.549 per game and 0.577 per match, illustrating how each stage magnifies the edge.

Common mistakes

The first mistake is entering point-win rate instead of serve-rally rate. In rally scoring these are close but not identical, and mixing them shifts the odds. Use the rate of winning rallies on own serve, which is what the model rotates on.

The second is treating 60% as "a good player." A serve-rally rate of 60% describes a severe mismatch: from the chart above, it gives near 99% per game. Real elite matches between close players sit inside 48 to 52.

The third is over-reading the momentum output. Streaks make individual matches look wilder, but they do not turn an underdog into a favourite. If the momentum match probability differs from iid by more than about 0.02, suspect Monte Carlo noise or an extreme strength setting rather than a real effect.

Do not average two players' serve rates into one number and feed it twice. The rotation only works because A's serving rallies and B's serving rallies have different win probabilities. Collapsing them removes the exact mechanic the chain models.

Related tools

For the same idea in other racket sports, see the Tennis Match Probability Calculator, the Table Tennis Match Probability Calculator and the Squash Match Probability Calculator. The Volleyball Match Probability Calculator uses the same rally-scoring logic.

To turn a single-game probability into series odds directly, use the Playoff Series Probability Calculator. If you would rather estimate each player's strength from ratings first, start with the Elo Rating Calculator.

Frequently asked questions

What number do I enter for serve-rally win rate?

The fraction of rallies a player wins when that player is serving. If a player served 40 rallies in a match and won 22 of them, that is 0.55. Both players usually sit near 0.50 because the serve gives only a small edge.

Why does the match probability differ so much from 50% when the rates are close?

The scoring system squares the edge at deuce, then a game to 21 accumulates it, then best-of-three accumulates it again. A 52 versus 49 rally gap becomes about 55% per game and 58% per match through this chaining.

How accurate is the momentum mode?

It is a Monte Carlo estimate from 20,000 matches, so its match probability carries a standard error near \pm 0.007. Run it twice and the two answers should agree to within about 1.5 percentage points. The exact iid mode has no such noise.

Does it matter who serves first?

Only slightly, and the model averages over the opening toss. The winner of each game serves first in the next, and the game chain already accounts for that within a game.

Everything here is an educational estimate for sport and training purposes. It is not professional coaching or medical advice, and it makes no promise about any real match.

Try it yourself in the tool

Related tools

Elo Rating Calculator

Expected score and rating change for any two Elo ratings and K-factor.

Playoff Series Probability Calculator

Chance of winning a best-of-3/5/7 series from your single-game win probability.

Soccer Score Simulator (Poisson)

Win/draw/loss and most likely scorelines from two expected-goals numbers.

League Season Simulator

Monte Carlo the rest of the season: title, top-N and relegation odds from the current table.

Squash Match Probability Calculator

From serve-rally win % to game and match probability with PAR-11 scoring.

Table Tennis Match Probability Calculator

From serve-point win % to game, deuce and match probability for best-of-3/5/7.