Does Best-of-5 Entrench Champions? A Tennis Natural Experiment

Men play best-of-five at the majors; women play best-of-three everywhere. That difference looked like a clean way to test whether a lower-variance format keeps the same players on top. The answer turned out to be more interesting than a yes or no.

Where the hypothesis came from

I started from a simple intuition about compounding. If winning — or even just going deep — in a major makes you more likely to do it again (through seeding, confidence, ranking points, experience closing out big matches), then success should be self-perpetuating. Incumbents get returns that first-timers and early-round losers never accumulate.

The second half of the idea is about format. A best-of-five match is a longer sample than a best-of-three, so the stronger player wins more reliably — fewer flukes, fewer upsets. Lower variance should reinforce incumbency and reduce turnover at the top.

Tennis hands you a natural experiment for this. Men play best-of-five at all four Grand Slams and best-of-three everywhere else; women play best-of-three essentially everywhere, majors included. So the first-draft hypothesis was: the men's tour, with its lower-variance majors, should show less turnover at the top than the women's tour.

There's a trap in that comparison, though — and spotting it is half the work. Sex is perfectly tangled up with format. If men show less turnover, is it the best-of-five, or is it differences in field depth, physiology, or which era's dynasty happened to be playing? A raw men-vs-women gap can't separate those. The fix is to find comparisons where only the format changes:

• Within the men's tour, Indian Wells and Miami are 128-draw best-of-three events — the same draw size as a Slam, the same players, just a shorter format.
• The women's tour, being best-of-three everywhere, becomes a control group: its Slam-vs-smaller-event gap can't be caused by format.
• In 2007 the men's Masters events switched their finals from best-of-five to best-of-three — a real before/after inside the same tournaments.

The data, and how you test a claim like this

Everything here comes from match-by-match tour data originally compiled by Jeff Sackmann (the widely-used tennis_atp / tennis_wta archives, accessed via a mirror). It's roughly 360,000 main-draw matches across both tours from 1968–2025, and for each match it records the two players, their rankings, the round, the tournament tier, and — crucially — whether it was best-of-three or best-of-five.

The method has three moving parts, and keeping them separate is what makes the analysis honest:

• A turnover metric. For each rolling 10-year window I count distinct Slam champions and compute the Herfindahl index (HHI) of titles — a concentration measure. Fewer distinct winners / higher HHI = less turnover.
• A mechanism metric. The upset rate: how often the lower-ranked player wins. This is the lever the whole hypothesis rests on, so it gets tested directly — and always controlling for the ranking gap between the two players, since a "favorite" against a near-equal is a very different thing than a favorite against a qualifier.
• A design that isolates format. Difference-in-differences. You never trust a raw before/after; you difference it against a control that didn't change, so any shared trend (like a dynasty suppressing upsets across the board) cancels out.

First surprise: turnover is about eras, not format

If format drove turnover, the men's line below should sit consistently lower (fewer distinct champions) than the women's. It doesn't. The two tours trade places: women were more concentrated than men through the Graf/Seles years, and men became far more concentrated during the Big-3 era. Same formats the whole time — what moved was whether a generational dynasty existed.

But the mechanism is real

The macro claim fails, yet the engine underneath it works. When you hold the draw size fixed and change only the format, best-of-five clearly suppresses upsets. Men's Slams (best-of-five, 128-draw) run a 27.3% upset rate; Indian Wells and Miami — the same 128-draw size, same players, but best-of-three — sit at 32.3%. Controlling for the ranking gap, best-of-five raises the favorite's win odds by about 20%.

Notice the men's and women's Slam bars are nearly identical (27.3% vs 27.4%) despite the format difference — another sign that at the tournament level the format edge is small and easily swamped. The effect is real, but it's a nudge, not a landslide.

The cleanest experiment — and why it looks like nothing

The 2007 Masters rule change is the tidiest before/after available: the same tournaments switched their finals from best-of-five to best-of-three, while every earlier round stayed best-of-three throughout. The naive read is a trap, though — 2007 onward is peak Big-3, so favorites started winning more for reasons that have nothing to do with sets. You can see it in the control rounds, which drop 3.6 points across the boundary purely from the era.

After differencing out the era trend, removing best-of-five from finals raised their upset rate by a statistically invisible +1.0 points. So does that kill the hypothesis? No — and this is the most interesting part.

The right way to look at the data

Best-of-five doesn't lower upsets uniformly — it lowers them only when there is a clear favorite to protect. Stratify every match by how mismatched the two players are, and the picture snaps into focus. When the players are evenly matched, the two formats are indistinguishable; as the gap widens, best-of-five pulls away, roughly halving the upset rate in lopsided matches. Drag the slider:

This is why the 2007 experiment came up empty, and why it's the perfect teaching case for how to evaluate the hypothesis properly. A final is, by construction, the closest matchup of the whole tournament — two players who each won five or six rounds to get there. Finals live in the flat left zone of this chart, exactly where format has nothing to grip. The 2007 rule toggled the format in the one round where it can't matter. The failure isn't evidence against the mechanism; it's a consequence of it.

The lesson for evaluation: don't test a variance-reduction hypothesis on the lowest-variance matchups. The informative comparisons are the early and middle rounds, where genuine favorites meet genuine underdogs — and there, best-of-five visibly protects the favorite. A truly powerful before/after would need a format change applied to whole draws, not just the final.

Conclusion

The clever core of the hypothesis is correct: best-of-five is a lower-variance format that measurably protects the favorite — about 5 fewer upset points and ~20% higher favorite odds, once you control for the ranking gap. But the leap from there to "so men have more entrenched champions than women" doesn't hold. The format edge is small, it only operates when matches are mismatched, and at the level of whole tournaments and eras it gets swamped by the single biggest factor in turnover: whether a generational dynasty happens to be playing. Incumbency compounds powerfully — but it does so on both tours, roughly equally, regardless of format.