I’ve written several things over the years about players who win more or fewer tiebreaks than expected. (Interested readers should start here.) Fans and commentators tend to think that certain players are particularly good or bad at tiebreaks. For instance, they might explain that a big serve is uncommonly valuable at the end of a set, or that mental weakness is more harmful than ever at such times.
My research has shown that, for the vast majority of players, tiebreak results are indistinguishable from luck. Let me qualify that just a bit: Tiebreak results are dependent on each player’s overall skill, so better players tend to win more tiebreaks. But there’s no additional factor to consider. While players tend to win service points at a slightly lower rate in tiebreaks, the effect is similar for everyone. There’s no magical tiebreak factor.
However, a single season is short enough that some players will always have glittering tiebreak records, tricking us into thinking that they have some special skill. In 2017, John Isner won 42 of his 68 tiebreaks, a 62% success rate. Based on his rate of service points won and return points won against the opponents he faced in tiebreaks, we’d expect him to win only 34–exactly half. Whether by skill or by luck, he exceeded expectations by 8 tiebreaks. Armed with a monster serve and a steady emotional presence on court, Isner is the kind of guy who makes us think that he has hacked the game of tennis, that he has figured out how to win tiebreaks. But while he has beaten expectations several times throughout his career, even Big John can’t sustain such a level. In 2018, he played 73 tiebreaks, and the simple model predicts that he would win 41. He managed only 39.
For additional examples, name whichever player you’d like. Roger Federer has built a career on unshakeable service performances, yet his tiebreak performances have been roughly neutral for the last four years. In other words, he wins tiebreak serve and return points at almost exactly the same rate as he does non-tiebreak points. Robin Haase, infamous for his record streak of 17 consecutive tiebreak losses, has paralleled Federer’s tiebreak performance for the last four years. 2018 was particuarly good for his high-pressure record, as he won two more breakers than expected, putting him in the top quartile of ATP players for the season.
Meaning from randomness
In short, season-by-season tiebreak performance resembles a spreadsheet full of random numbers. A player with a good tiebreak record last year may well sustain it this year, but only if it’s based on good overall play. If there is an additional secret to tiebreak excellence (beyond being good at tennis), no one has told the players about it.
But in sports statistics, every negative result has a silver lining. We might be disappointed if a stat is not predictive of future results. However, the very lack of predictiveness allows us to make a different kind of prediction. If a player has a great tiebreak year, beating expectations in that category, the odds are he just got lucky. Therefore, he’s probably not going to get similarly lucky this year, and his overall record will regress accordingly.
The player to watch in 2019 in this department is Taylor Fritz, who recorded a sterling 20-8 record in tiebreaks last season. Based on his performance in the whole of those matches, we would have expected him to win only 13 of 28. His Tiebreaks Over Expectations (TBOE) of +7 exceeded that of any other tour player last season, even though many of his peers contested far more breakers. It’s always possible that Fritz really does have the magical mix of steely nerves and impeccable tactics that translates into tiebreak wins, but it’s far more likely that he’ll post a neutral tiebreak record in 2019. In 2017, the player after Isner on the TBOE list was Jack Sock, and it’s fair to say that his 2018 campaign didn’t exactly continue in the same vein.
With that regression to the mean in mind, here are the TBOE leaders and laggards from the 2018 ATP season. The TBExp column shows the number of tiebreaks that the simple model would have predicted, and TBOR is a rate-stat version of TBOE, reflecting the percentage of tiebreaks won above or below average. Rate stats like TBOR are usually more valuable than counting stats like TBOE, but in this case the counting stat may have more to tell us, since it takes into account which players contest the most tiebreaks. Sam Querrey’s rate of underperformance isn’t quite as bad as Cameron Norrie’s, but the number of tiebreaks he plays is a result of his game style, justifying his place at the bottom of this list.
Player TBs TBWon TBExp TBOE TBOR
Taylor Fritz 28 20 13.3 6.7 0.24
Bradley Klahn 22 16 10.6 5.4 0.24
Martin Klizan 16 13 8.1 4.9 0.31
Kei Nishikori 22 17 12.5 4.5 0.20
Bernard Tomic 18 14 9.6 4.4 0.24
Alexander Zverev 23 17 13.2 3.8 0.17
Albert Ramos 22 15 11.2 3.8 0.17
Adrian Mannarino 25 16 12.3 3.7 0.15
Stan Wawrinka 21 13 9.6 3.4 0.16
Juan Martin Del Potro 32 22 18.7 3.3 0.10
Borna Coric 21 8 10.8 -2.8 -0.13
Denis Shapovalov 30 12 15.0 -3.0 -0.10
Karen Khachanov 42 20 23.4 -3.4 -0.08
Ivo Karlovic 47 19 22.6 -3.6 -0.08
Denis Istomin 31 13 16.7 -3.7 -0.12
Ricardas Berankis 22 7 10.9 -3.9 -0.18
Pablo Cuevas 21 7 11.3 -4.3 -0.20
Andrey Rublev 18 5 9.6 -4.6 -0.26
Fernando Verdasco 25 8 12.8 -4.8 -0.19
Roberto Bautista Agut 26 10 14.8 -4.8 -0.19
Cameron Norrie 22 5 9.9 -4.9 -0.22
Sam Querrey 36 12 18.5 -6.5 -0.18
The guys at the top of this list can expect to see their tiebreak records drift back to normalcy in 2019, while the guys at the bottom have reason to hope for an improvement in their overall results this year.
Converting tiebreaks to wins
I’m sure we all agree that tiebreaks are really important, but what’s the real impact of the over- and underperformance I’m talking about here? In other words, given that Kei Nishikori won 4.5 more tiebreaks last season than expected (than he “should” have won), how did that effect his overall won-loss record? And by extension, what might it mean for his match record in 2019?
The math gets hairy*, but in the end, two additional tiebreak wins are roughly equal to one additional match win. Nishikori’s 4.5 bonus tiebreaks are equivalent to about 2.25 additional match wins. He was 48-22 last year, so with neutral tiebreak luck, he would’ve gone 46-24. Of course, that still leaves some unanswered questions; translating match record to ranking points and titles is much messier, and I won’t attempt anything of the sort. His lucky tiebreaks might have converted should-have-been-losses into wins, or they might have turned gut-busting three-setters into more routine straight-set victories. But blending all the possibilities together, each player’s TBOE has a concrete value we can convert to wins.
The exact numbers aren’t important here, but the concept is. When you see an extremely good or bad tiebreak record, you don’t need to whip out a spreadsheet and calculate the precise number of breakers the player should have won. Given neutral luck, every ATP regular should have a tiebreak record between 40% and 60%–40% for the guys at the fringe, 60% for the elites. (In 2018, Federer’s expected rate was 60.1%, and Sock’s was 40.9%.) Any number out of that range, like Richard Gasquet’s 13-of-16 in 2016, is bound to come crashing back to earth, though rarely so catastrophically as the Frenchman’s did, falling to a mere 5 wins in 17 tries.
Any given tiebreak might be determined by superlative serving, daring return tactics, or sheer mental fortitude. But over time, those effects even out, meaning that no player is consistently good or bad in breakers. The better player is more likely to win, but luck has a huge say in the outcome. In the long term, that luck usually cancels itself out.
—
* A quick overview of the math: In a best-of-three match, there are three possible times that the tiebreak can take place. Flipping the result of a tiebreak could change the result of the first set, the second set, or the third set. The win probability impact of flipping the first set is 50%–assuming equal players, the winner has a 75% chance of winning the match and the loser has a 25% chance. The win probability effect of reversing the second set is also 50%. Either the winner takes the match (100%) instead of forcing a third set (50%), or the winner forces a third set (50%) instead of losing the match (0%). Changing the result of the third set directly flips the outcome of the match, so the win probability effect is 100%.
Every completed match has a first and second set, but fewer than 40% of ATP matches have third sets. The weighted average of 50%, 50%, and 100% is about 58%, which would be our answer if ATPers played only best-of-three matches. The math for five-setters is more involved, but the most important thing is that best-of-five gives each of the first four sets less leverage, and by extension, it does the same to tiebreaks in the first four sets. Weighing that effect combined with the frequency of best-of-five set matches would give us a precise value to convert TBOE to wins. Rather than going further down that rabbit hole, I’m happy with the user-friendly andapproximately correct figure of 50%.