Most interesting patterns in data die the same death. Someone points out that the groups were different sizes, or that the categories were chosen after the fact, and the finding quietly evaporates.
Occasionally a dataset hands you a comparison with none of those escape routes available. Professional tennis has one, and it is worth walking through slowly because the structure is rarer than the subject.
The setup
Across 116,382 completed professional singles matches, a player who loses the opening set goes on to win 16.52% of the time. One figure for every situation, which is where most discussions stop.
Split those matches by the score of the first set and seven different answers appear:
– lost it 7-6: wins 21.92% of the time, from 14,265 matches
– lost it 6-4: 19.41%, from 25,546
– lost it 7-5: 18.89%, from 10,252
– lost it 6-3: 17.39%, from 25,970
– lost it 6-2: 13.30%, from 19,739
– lost it 6-1: 11.06%, from 14,269
– lost it 6-0: 7.18%, from 5,833
The comparison with no escape route
Look only at the first row and the sixth.
A first set lost 7-6 appears 14,265 times. A first set lost 6-1 appears 14,269 times. The two groups differ in size by four matches out of fourteen thousand. That is as close to a matched comparison as observational data gets.
One group recovers 21.92% of the time, the other 11.06%. Twice as often.
Now try the usual objections. Unequal sample sizes cannot explain it, because the samples are the same size. Cherry-picked categories cannot explain it, because the categories are the seven possible set scores and all seven are shown. A lucky run cannot explain it, because the rate falls at every single step from 7-6 down to 6-0 without one reversal.
What is left is the thing itself. How close the first set was predicts the rest of the match, and it predicts it strongly.
Why the ordering matters more than the gap
The headline difference is the eye-catching part, but the monotonic sequence is the stronger evidence.
Seven categories, ordered by margin, and the outcome moves the same direction at every step. If the pattern were noise you would expect at least one row out of place. If some other variable were driving it, the sequence would usually buckle somewhere. It does not.
That is what separates a result you can rely on from one that happens to look good.
The same discipline, applied to the next set
Follow the losing player into the second set and the data keeps its shape.
Holding level to one game all in the second is worth 17.66%, from 63,405 matches, against 16.52% at the start of the match. All that holding on buys about one percentage point.
Going two games up early in the second is worth 36.80%, from 16,660 matches. Three games to one up, 39.02% from 18,412. A break of serve roughly doubles the chance.
Being a break down instead, at 1-2, leaves 10.42% from 47,102 matches. The match is decided inside eight games of the second set.
The habit worth stealing
When you meet a claim built on grouped data, look for the group sizes before the percentages. If they are wildly uneven, the comparison is doing less work than it appears. If they are close, and the pattern holds its order across every category, you are looking at something real.
The figures come from the Live Tennis API, a JSON service for developers. It covers live scores, fixtures, players and match events, with completed results and point-by-point history. They are published at blog.livetennisapi.com/studies.json under CC BY 4.0, free to reuse with attribution.
Method: completed best-of-three professional singles matches, 1 January 2023 to 14 September 2026. Retirements and walkovers excluded. Each match state counted once per match.
