Best of five cuts upsets less than the Slams show

Upsets are rarer at the Grand Slams than anywhere else on the men's tour. Since 1990 the lower-ranked player has won about three Slam matches in ten, against more than a third at the Masters. The usual explanation is the format: over five sets the better player gets more chances to prove it. That is true, and it is not the whole story.

The majors are the only tour events where the men play the best of five sets. The Grand Slam Rule Book requires it for every main-draw match; the rest of the ATP Tour plays the best of three.

What the format alone does

Five sets alone take about a seventh off the upset in a close match. The model behind that figure plays a match point by point, and every point is decided by a fixed chance that never moves: no nerves, no fatigue, no momentum. Both players start from the tour-typical six serve points in ten, and the better player's small edge is added on his serve and on his return alike.

From those two chances the games, tiebreaks, sets and match follow by exact arithmetic, with a seven-point tiebreak at six games all in every set. A small edge grows with every set played. Take a player who wins 52% of all points. He wins a set roughly two times in three; in a best of three the weaker man still wins three matches in ten, in a best of five about one in four.

The curve runs the same match from an even contest to a rout: the wider the gap between the players, the harder five sets bite in relative terms.

What the record shows

At the Slams the lower-ranked player has won 29.5% of completed matches since 1990. At the Masters he has won 35.7%. The gap is 6.2 points, and on samples of this size it is not chance.

The record is every completed ATP main-draw match from 1990 to 2025 in the TennisMyLife database. An upset is a win by the man ranked lower on the day. Walkovers, retirements and defaults are left out, as is any match missing a ranking. The format is read from the match, not the event, so the Masters finals played over five sets before 2008, 104 played and 100 completed, belong to neither series.

The Slam rate has also moved more over time. In the 1990s the lower-ranked player won 32.5% of Slam matches; in the 2010s he won 26.1%. The Masters rate barely moved.

Where the format matters most

Where the pairing is lopsided, the record moves twice as far as the format can carry it. When a top-10 player meets a man outside the top 50, the lower-ranked man wins about one Masters match in five. At the Slams he wins about one in ten, half the rate. In the model a best-of-three chance of one in five sits on the 53% row, and there five sets take away about a quarter, not half.

Interpretation: how to read the gap

In ordinary pairings the format carries much of the gap. On the model rows that bracket the Masters rate, five sets cut between two and a half and four points, against a recorded gap of 6.2. The model fixes one kind of match and the record mixes them all, so the comparison gives an order of magnitude, not a decomposition. In lopsided pairings the arithmetic explains a quarter of the fall and the record shows a half. Something beyond the number of sets works for the favourite at the majors.

My reading is that the draw does more than the stakes. A 128-player field sends top-10 players against qualifiers and wild cards in the first week; a Masters draw of 56 or 96 seldom does. The seeding keeps the best apart for longer. The Masters carry far more points and money than an ATP 250, yet with the same format they produce upsets at almost the same rate as the rest of the tour. If stakes alone steadied a match, the Masters would sit between the two.

What this does not show

These tables cannot settle whether best of five causes the lower Slam upset rate. The gap bundles the format with draw size, qualifiers and wild cards, surface, scheduling and where players choose to peak. No round in this window was played in both formats, and the pre-2008 Masters finals are too few to serve as the experiment.

The model is also simpler than the game. It assumes independent points and a seven-point tiebreak in every set, while real fifth sets ran to advantage at three of the four majors for most of the window and go to a 10-point tiebreak since 2022. Leaving out retirements drops some upsets in progress and some favourites who withdrew; the direction of that bias is unknown, its size bounded by the excluded rows. And the ranking is the rank on the day, not a measure of how close the match was.