Elo-Forecasting the WTA Tour Finals in Singapore

With the field of eight divided into two round-robin groups for the WTA Tour Finals in Singapore, we can play around with some forecasts for this event. I’ve updated my Elo ratings through last week’s tournaments, and the first thing that jumps out is how different they are from the official rankings.

Here’s the Singapore field:

EloRank  Player                Elo  Group  
2        Maria Sharapova      2296    RED  
4        Simona Halep         2181    RED  
6        Garbine Muguruza     2147  WHITE  
8        Petra Kvitova        2136  WHITE  
9        Angelique Kerber     2129  WHITE  
11       Agnieszka Radwanska  2100    RED  
15       Lucie Safarova       2051  WHITE  
21       Flavia Pennetta      2004    RED

Serena Williams (#1 in just about every imaginable ranking system) chose not to play, but if Elo ruled the day, Belinda Bencic, Venus Williams, and Victoria Azarenka would be playing this week in place of Agnieszka Radwanska, Lucie Safarova, and Flavia Pennetta.

Anyway, we’ll work with what we’ve got. Maria Sharapova is, according to Elo, a huge favorite here. The ratings translate into a forecast that looks like this:

Player                  SF  Final  Title  
Maria Sharapova      83.7%  61.1%  43.6%  
Simona Halep         60.8%  35.4%  15.9%  
Garbine Muguruza     59.4%  25.7%  11.3%  
Petra Kvitova        55.2%  23.0%   9.8%  
Angelique Kerber     53.1%  21.7%   8.8%  
Agnieszka Radwanska  37.4%  17.4%   6.1%  
Lucie Safarova       32.3%   9.7%   3.1%  
Flavia Pennetta      18.1%   6.0%   1.4%

If Sharapova is really that good, the loser in today’s draw was Simona Halep. The top seed would typically benefit from having the second seed in the other group, but because Garbine Muguruza recently took over the third spot in the rankings, Pova entered the draw as a dangerous floater.

However, these ratings don’t reflect the fact that Sharapova hasn’t completed a match since Wimbledon. They don’t decline with inactivity, so Pova’s rating is the same as it was the day after she lost to Serena back in July. (My algorithm also excludes retirements, so her attempted return in Wuhan isn’t considered.)

With as little as we know about Sharapova’s health, it’s tough to know how to tweak her rating. For lack of any better ideas, I revised her Elo rating to 2132, right between Petra Kvitova and Angelique Kerber. At her best, Sharapova is better than that, but consider this a way of factoring in the substantial possibility that she’ll play much, much worse–or that she’ll get injured and her matches will be played by Carla Suarez Navarro instead. The revised forecast:

Player                  SF  Final  Title  
Simona Halep         69.9%  40.9%  24.0%  
Garbine Muguruza     59.4%  31.5%  16.5%  
Maria Sharapova      57.6%  29.5%  14.5%  
Petra Kvitova        55.6%  28.4%  14.4%  
Angelique Kerber     52.5%  26.3%  13.2%  
Agnieszka Radwanska  47.9%  22.3%   9.9%  
Lucie Safarova       32.6%  12.9%   4.9%  
Flavia Pennetta      24.7%   8.3%   2.7%

If this is a reasonably accurate estimate of Sharapova’s current ability, the Red group suddenly looks like the right place to be. Because Elo doesn’t give any particular weight to Grand Slams, it suggests that the official rankings far overestimate the current level of Safarova and Pennetta. The weakness of those two makes Halep a very likely semifinalist and also means that, in this forecast, the winner of the tournament is more likely (54% to 46%) to come from the White group.

Without Serena, and with Sharapova’s health in question, there are simply no dominant players in the field this week. If nothing else, these forecasts illustrate that we’d be foolish to take any Singapore predictions too seriously.

Measuring the Effectiveness of Backhand Returns

Italian translation at settesei.it

One-handed backhands can be beautiful, but they aren’t always the best tools for the return of serve. Some of the players with the best one-handers in the game must often resort to slicing backhand returns–Stanislas Wawrinka, for example, slices 68% of backhand first serve returns and 40% of backhand second serve returns, while Andy Murray uses the slice 41% and 3%, respectively.

Using the 650 men’s matches in the Match Charting Database, I looked at various aspects of backhand serve returns to try to get a better sense of the trade-offs involved in using a one-handed backhand. Because the matches in the MCP aren’t completely representative of the ATP tour, the numbers are approximate. But given the size and breadth of the sample, I believe the results are broadly indicative of men’s tennis as a whole.

At the most general level, players with double-handed backhands are slightly better returners, putting roughly the same number of returns in play (about 56%) and winning a bit more often–46.9% to 45.7%–when they do so. The gap is a bit wider when we look at backhand returns put in play: 46.5% of points won to 44.7%. While the favorable two-hander numbers are influenced by the historically great returning of Novak Djokovic, two-handers still have an edge if we reduce his weight in the sample or remove him entirely.

Unsurprisingly, players realize that two-handed backhands are more effective returns, and they serve accordingly. The MCP divides serves into three zones–down the tee, body, and wide–and I’ve re-classified those as “to the forehand,” “to the body,” and “to the backhand” depending on the returner’s dominant hand and whether the point is in the deuce or ad court. While we can’t identify exactly where servers aimed those to-the-body serves, we can determine some of their intent from serves aimed at the corners.

Against returners with two-handed backhands, servers went for the backhand corner on 44.2% of first serves and 34.8% of second serves. Against one-handers, they aimed for the same spot on 47.3% of first serves and 40.9% of second serves. Looking at the same question from another angle, backhands make up 61.7% of the returns in play hit by one-handers compared to 59.0% for double-handers. It seems likely that one-handers more aggressively run around backhands to hit forehand returns, so this last comparison probably understates the degree to which servers aim for single-handed backhands.

When servers do manage to find the backhand side of a single-hander, they’re often rewarded with a slice return. On average, one-handers (excluding Roger Federer, who is overrepresented in this dataset) use the slice on 53.9% of their backhand first-serve returns and 32.3% of their backhand second-serve returns. Two-handers use the slice 20.5% of the time against firsts and only 2.5% of the time against seconds.

For both types of players, against first and second serves, slice returns are less effective than flat or topspin backhand returns. This isn’t surprising, either–defensive shots are often chosen in defensive situations, so the difference in effectiveness is at least partly due to the difference in the quality of the serves themselves. Still, since one-handers choose to go to the slice so much more frequently, it’s valuable to know how the types of returns compare:

Return Type   BH in play W% SL in play W% 
1HBH vs Firsts        43.3%         37.6% 
1HBH vs Seconds       46.0%         44.1% 
                        
2HBH vs Firsts        46.8%         36.2% 
2HBH vs Seconds       48.6%         41.9%

(Again, I’ve excluded Fed from the 1HBH averages.)

In three of the four rows, there’s a difference of several percentage points between the effectiveness of slice returns and flat or topspin returns, as measured by the ultimate outcome of the point. The one exception–second-serve returns by one-handers–reminds us that the slice can be an offensive weapon, even if it’s rarely used as one in the modern game. Some players–including Federer, Feliciano Lopez, Grigor Dimitrov, and Bernard Tomic–are more effective with slice returns than flat or topspin returns against either first or second serves.

However, these players are the exceptions, and in the theoretical world where we can set all else equal, a slice return is the inferior choice. All players have to hit slice returns sometimes, and many of those seem to be forced by powerful serving, but the fact remains: one-handers hit slices much more than two-handers do, and despite the occasional offensive opportunity, slice returns are more likely to hand the point to the server.

These differences are real, but they are still modest. A good returner with a one-handed backhand is considerably better than a bad returner with a two-hander, and it’s even possible to have a decent return game while hitting mostly slices. All that said, in the aggregate, a one-handed backhand is a bit of a liability on the return. It will take further research to determine whether other benefits–such as the sizzling down-the-line winners we’ve come to expect from the likes of Wawrinka and Richard Gasquet–outweigh the costs.

Benoit Paire and Overqualified Challenger Contenders

Italian translation at settesei.it

With three ATP tour-level events on the slate this week, Benoit Paire considered his options and elected to play none of them. Instead, the world #23 is the top seed at the Brest Challenger, making him the highest ranked player to enter a challenger this year–by a wide margin.

Top-50 players may only enter challengers if they are given a wild card, and top-ten players may not enter them at all. Still, since 1990, a top-50 player has played a challenger just over 500 times, at a rate of about 20 per year. (Some of these players didn’t need a wild card, as entry is determined by ranking several weeks before the tournament, during which time rankings rise and fall.)

Many of the high-ranked wild cards fall into one of two categories: Players who lose early in Slams, Indian Wells, or Miami; and clay-court specialists seeking more matches on dirt. Paire’s decision this week–like the Frenchman himself–doesn’t follow one of these common patterns.

Anyway, here are the top-ranked players to contest challengers since 1990, along with their results. A result of “W” means that the player won the title, while any other result indicates the round in which the player lost.

Year  Event           Player               Rank  Result  
2003  Braunschweig    Rainer Schuettler    8     R16     
1991  Johannesburg    Petr Korda           9     SF      
1994  Barcelona       Alberto Berasategui  10    W       
1994  Graz            Alberto Berasategui  11    R16     
2008  Sunrise         Fernando Gonzalez    12    QF      
2004  Luxembourg      Joachim Johansson    12    W       
2011  Prostejov       Mikhail Youzhny      13    QF      
2008  Prostejov       Tomas Berdych        13    QF      
2003  Prague          Sjeng Schalken       13    W       
2005  Zagreb          Ivan Ljubicic        14    W       
2004  Bratislava      Dominik Hrbaty       14    F       
2004  Prostejov       Jiri Novak           14    QF      
2003  Prostejov       Jiri Novak           14    R32     
2007  Dnepropetrovsk  Guillermo Canas      15    SF      
2002  Prostejov       Jiri Novak           15    F       
1998  Segovia         Alberto Berasategui  15    QF      
1997  Braunschweig    Felix Mantilla       15    F       
1997  Zagreb          Alberto Berasategui  15    W

(Schuettler and Korda were outside the top ten a couple of weeks before their respective challengers.)

A look at this list suggests that Alberto Berasategui entered challengers as a top-fifty player more than anyone else. He’s close–with 12 such entries, he’s tied for second with Jordi Arrese. The player who dropped down a level the most times is Dominik Hrbaty, who played 17 challengers while ranked in the top 50. (The active leaders are Jarkko Nieminen with ten and Andreas Seppi with nine.)

Despite all those attempts, Hrbaty wasn’t particularly successful as a high-ranked challenger player. He won only 2 of those 17 events, reaching only one other final. Top-50 players aren’t guaranteed to win these titles, of course, but in general, they have outperformed Hrbaty, winning 18% of possible titles. Here are top-50 players’ results broken down by round:

Result       Frequency  
Title            18.1%  
Loss in F         9.3%  
Loss in SF       11.3%  
Loss in QF       17.1%  
Loss in R16      22.0%  
Loss in R32      22.2%

Paire is a better player than this sample’s average ranking of 37. Combined with a favorable surface, he gets a much more optimistic forecast from my algorithm, with a slightly better than one-in-three chance of winning the title. With a futures title, an ATP trophy, and a pair of challenger triumphs already in the books this year, it seems fitting that Benoit would add another oddity to his wide-ranging season.

Continue reading Benoit Paire and Overqualified Challenger Contenders

Lucky Losers and Familiar Faces

In the final round of qualifying Monday in Moscow, Darya Kasatkina easily defeated Paula Kania. Thanks to a couple of late withdrawals, both players ended up making the main draw … and tomorrow, they’ll play each other again.

This scenario is rare, but not unheard of. Since the mid-1990s, there have been 30 other instances when two women faced each other in qualifying and then again in the main draw. Most recently, Lauren Davis defeated Svetlana Kuznetsova twice at the 2013 Canadian Open. One year earlier, in Sydney, Alexandra Dulgheru beat Sofia Arvidsson in the first round of the main draw despite losing to her in the final round of qualifying.

Tomorrow’s Kasatkina-Kania rematch is far from a sure thing. In those 30 prior matches, barely more than half of the qualifiers–17 of 30–have managed to win both matches.

This sort of rematch is similarly uncommon on the ATP tour. Since 2007 (the earliest year for which I have qualifying results), this has happened a dozen times. Most recently, Albert Ramos-Vinolas defeated Robin Haase in back-to-back rounds in Monte Carlo. Ramos was on the opposite side of things five years ago, when Pablo Cuevas beat him twice in Valencia.

Earlier this year, in a variation on the theme in Auckland, Kenny de Schepper beat Alejandro Falla to qualify, and after both players won their first-round matches, Falla triumphed in the second-round rematch.

—

Programming note: After watching this sort of ad hoc research disappear into the barely-searchable void that is the Twitter archive, it occurred to me to post occasional brief notes such as this one. It’s not groundbreaking stuff, but at least it’ll be easier to find in the future. These curiosities won’t interfere with or replace my longer, more analytical posts.

The Dreaded Deficit at the Tiebreak Change of Ends

Italian translation at settesei.it

Some of tennis’s conventional wisdom manages to be both blindingly self-evident and obviously wrong. Give pundits a basic fact (winning more points is good), add a dash of perceived momentum, and the results can be toxic.

A great example is the tiebreak change of ends. The typical scenario goes something like this: Serving at 2-3 in a tiebreak, a player loses a point on serve, going down a minibreak to 2-4. As the players change sides, a commentator says, “You really don’t want to go into this change of ends without at least keeping the score even.”

While the full rationale is rarely spelled out, the implication is that losing that one point–going from 2-3 to 2-4–is somehow worse than usual because the point precedes the changeover. Like the belief that the seventh game of the set is particularly important, this has passed, untested, into the canon.

Let’s start with the “blindingly self-evident” part. Yes, it’s better to head into the change of ends at 3-3 than it is at 2-4. In a tiebreak, every point is crucial. Based on a theoretical model and using sample players who each win 65% of service points, here are the odds of winning a tiebreak from various scores at the changeover:

Score  p(Win)  
1*-5     5.4%  
2*-4    21.5%  
3*-3    50.0%  
4*-2    78.5%  
5*-1    94.6%

It’s easy to sum that up: You really want to win that sixth point. (Or, at least, several of the points before the sixth.) On the other hand, compare that to the scenarios after eight points:

Score  p(Win)  
2*-6     2.6%  
3*-5    17.6%  
4*-4    50.0%  
5*-3    82.4%  
6*-2    97.4%

At the risk of belaboring the obvious, when the score is close, points become more important later in the tiebreak. The outcome at 4-4 matters more than at 3-3, which matters more than at 2-2, and so on. If players changed ends after eight points, we’d probably bestow some magical power on that score instead.

Real-life outcomes

So far, I’ve only discussed what the model tells us about win probabilities at various tiebreak scores. If the pundits are right, we should see a gap between the theoretical likelihood of winning a tiebreak from 2-4 and the number of times that players really do win tiebreaks from those scores. The model says that players should win 21.5% of tiebreaks from 2*-4; if the conventional wisdom is correct, we would find that players win even fewer tiebreaks when trying to come back from that deficit.

By analyzing the 20,000-plus tiebreaks in this dataset, we find that the opposite is true. Falling to 2-4 is hugely worse than reaching the change of ends at 3-3, but it isn’t worse than the model predicts–it’s a bit better.

To quantify the effect, I determined the likelihood that the player serving immediately after the changeover would win the tiebreak, based on each player’s service points won throughout the match and the model I’ve referred to above. By aggregating all of those predictions, together with the observed result of each tiebreak, we can see how real life compares to the model.

In this set of tiebreaks, a player serving at 2-4 would be expected to win 20.9% of the time. In fact, these players go to win the tiebreak 22.0% of the time–a small but meaningful difference. We see an even bigger gap for players returning at 2-4. The model predicts that they would win 19.9% of the time, but they end up winning 22.1% of these tiebreaks.

In other words, after six points, the player with more points is heavily favored, but if there’s any momentum–that is, if either player has more of an advantage than the mere score would suggest–the edge belongs the player trailing in the tiebreak.

Sure enough, we see the same effect after eight points. Serving at 3-5, players in this dataset have a 16.3% (theoretical) probability of winning the tiebreak, but they win 19.0% of the time. Returning at 3-5, their paper chance is 17.2%, and they win 19.5%.

There’s nothing special about the first change of ends, and there probably isn’t any other point in a tiebreak that is more crucial than the model suggests. Instead, we’ve discovered that underdogs have a slightly better chance of coming back than their paper probabilities indicate. I suspect we’re seeing the effect of front-runners getting tight and underdogs swinging more freely–an aspect of tennis’s conventional wisdom that has much more to recommend itself than the idea of a magic score after the first six points of a tiebreak.

Does Serving First in a Tiebreak Give You an Edge?

Italian translation at settesei.it

Tiebreaks are so balanced, with frequently alternating servers and sides of the court, that it seems they must be fair. As far as I know, there is no commonly-cited conventional wisdom to the effect that the first server (or second server) in a tiebreak has any kind of advantage.

Let’s check. In a dataset of over 5,200 tiebreaks at ATP tour events, the first server won 50.8% of the time. Calculating each player’s service points won for the entire match and using those numbers to determine the likelihood that the first server would win a tiebreak, we get an estimate that those first servers should have won only 48.8% of them.

Two percentage points is a small gap, but here, it’s a meaningful one. It’s persistent across each of the three years most heavily represented in the dataset (2013-15), and it holds regardless of the set. While there might be some bias in the results of first-set tiebreaks, since better servers often choose to serve first and lesser servers choose to receive, the effect in each set favors the first server, and the impact of serving first is greater in the third set than in the first.

However, this effect–at least in its magnitude–is limited to ATP results. A survey of 2,500 recent WTA tiebreaks shows that first servers have won 49.7% of tiebreaks, compared to 49.4% that they should have won. Women’s ITF matches and men’s futures matches return similar results. Running the same algorithm on 6,200 men’s Challenger-level tiebreaks confuses the issue even further: Here, first servers won 48.1% of tiebreaks, while they should have won 48.7%.

A byproduct of this research is the discovery that, for both genders and at multiple levels of the game, the first server in a tiebreak is, on average, the weaker player. At first glance, that doesn’t make a lot of sense: We think of tiebreaks as deciding sets when the two players are equal. And since the effect is present for the second and third sets as well as the first, this finding isn’t biased by players choosing who will serve first.

As it turns out, this result can be at least partially explained by another byproduct of my recent research. In my attempt to determine whether it’s particularly difficult to hold when serving for the set, I calculated the odds of holding serve at every score throughout a set, compared to how frequently players should have held. At most holds–including those with the set on the line–there aren’t any major discrepancies between actual hold rates and expected hold rates.

But I did find some small effects that are relevant here. In general, it is a bit harder to hold serve as the second server, at scores such as 3-4, 4-5, and 5-6, than as the first, at scores like 3-3, 4-4, and 5-5. For instance, in the ATP data, players hold serve at 4-4 exactly as often as we would expect them to, based on their rate of service points won throughout the match. But at 4-5, their performance drops to 1.4% below expectations. In the WTA data, while players underperform at 5-5 by 1.4%, they are far worse at 5-6, winning 5.2% less often than they should.

In other words, if two players of equal abilities stay on serve for the first several games of a set, the second server is a little more likely to crack, getting broken and losing the set. Thus, if neither player is broken (or the number of breaks is equal), the second server is likely to be just a little bit better.

That explains, at least in part, why second servers are favored on paper going into tiebreaks. What it doesn’t account for is the discovery that on the ATP tour, first servers overcome that paper advantage and win more than half of tiebreaks. For that, I don’t have a good answer.

Digging Out of the Holes of 0-40 and 15-40

In the men’s professional game, serving at 0-40 isn’t a death sentence, but it isn’t a good place to be. An average player wins about 65% of service points, and at that rate, his chance of coming back from 0-40 is just a little better than one in five.

Some players are better than others at executing this sort of comeback. Tommy Robredo, for instance, has come back from 0-40 nearly 60% more often than we’d expect, while Sam Querrey digs out of the 0-40 hole one-third less often than we would predict.

Measuring a player’s success rate in these scenarios isn’t simply a matter of counting up 0-40 games. That’s what we saw on the ATP official site last week, and it’s woefully inadequate. That article marvels at Ivo Karlovic‘s “clutch” accomplishments from 0-40 and 15-40, when we could easily have guessed that Ivo would lead just about any serving category. Big serving isn’t clutch if it’s what you always do.

Statistics are only valuable in context, and that is particularly true in tennis. Simply counting 0-40 games and reporting the results hides a huge amount of potential insight. Whether a player wins or loses (a game, a set, a match, or a stretch of matches) is only the first question. To deliver any kind of meaningful analysis, we need to adjust those results for the competition and consider what we already know about the players we’re studying.

Rather than tear apart that article, though, let’s do the analysis correctly.

The number of times a player comes back from 0-40 or 15-40 isn’t what’s important. As we’ve seen, big servers will dominate those categories. That doesn’t tell us who is particularly effective (or, dare we say, “clutch”) in such a situation, it only identifies the best servers. What matters is how often players come back compared to how often we would expect them to, taking into consideration their serving ability.

Karlovic is an instructive example. Over the last few years–the time span available in this dataset of point-by-point match records–Ivo has gone down 0-40 56 times, holding 17 of those games, a rate of 30.4%. That’s third-best on tour, behind John Isner and Samuel Groth. But compared to how well we would expect Karlovic to serve, he’s only 7% better than neutral, right in the middle of the ATP pack.

Before diving into the results, a few more notes on methodology. For each 0-40 or 15-40 game, I calculated the server’s rate of service points won in that match. Since we would expect 0-40 games to occur more often in matches with good returners, in-match rates seem more accurate than season-long aggregates. Given the in-match rate of serve points won, I then determined the odds that the server would come back from the 0-40 or 15-40 score. For each game, then, we have a result (came back or didn’t come back) and an estimate of the comeback’s likelihood. Combining both numbers for all of a player’s service games tells us how effective he was at these scores.

For 30 of the players best represented in the dataset, here are their results at 0-40, showing the number of games, the number of successful comebacks, the rate of successful comebacks, and the degree to which the player exceeded expectations from 0-40:

Player                  0-40  0-40 W  0-40 W%  W/Exp  
Tommy Robredo            110      30    27.3%   1.59  
Denis Istomin            114      26    22.8%   1.36  
John Isner                87      31    35.6%   1.34  
Guillermo Garcia-Lopez   161      29    18.0%   1.32  
Kevin Anderson           130      38    29.2%   1.28  
Bernard Tomic            110      24    21.8%   1.25  
Fernando Verdasco        141      30    21.3%   1.17  
Rafael Nadal             140      32    22.9%   1.15  
Kei Nishikori            122      23    18.9%   1.15  
Marin Cilic              125      26    20.8%   1.14  
                                                      
Player                  0-40  0-40 W  0-40 W%  W/Exp  
Jo-Wilfried Tsonga       124      29    23.4%   1.14  
Novak Djokovic           124      34    27.4%   1.12  
Andreas Seppi            145      24    16.6%   1.09  
Grigor Dimitrov          115      22    19.1%   1.08  
Philipp Kohlschreiber    146      28    19.2%   1.08  
Roger Federer            107      26    24.3%   1.07  
Ivo Karlovic              56      17    30.4%   1.07  
Santiago Giraldo         113      18    15.9%   1.06  
Alexandr Dolgopolov      141      25    17.7%   1.03  
Milos Raonic              82      23    28.0%   1.01  
                                                      
Player                  0-40  0-40 W  0-40 W%  W/Exp  
Tomas Berdych            149      30    20.1%   1.01  
Jeremy Chardy            122      21    17.2%   0.98  
Feliciano Lopez          136      26    19.1%   0.97  
Fabio Fognini            211      24    11.4%   0.97  
Mikhail Youzhny          155      18    11.6%   0.92  
David Ferrer             203      32    15.8%   0.89  
Richard Gasquet          152      25    16.4%   0.87  
Andy Murray              164      24    14.6%   0.80  
Gilles Simon             158      16    10.1%   0.72  
Sam Querrey               84      12    14.3%   0.68

As I mentioned above, Robredo has been incredibly effective in these situations, coming back from 0-40 30 times instead of the 19 times we would have expected. Some big servers, such as Isner and Kevin Anderson, are even better than their well-known weapons would leads us to expect, while others, such as Karlovic and Milos Raonic, aren’t noticeably more effective at 0-40 than they are in general.

Many of these extremes don’t hold up when we turn to the results from 15-40. Quite a few more games reach 15-40 than 0-40, so the more limited variation at 15-40 suggests that many of the extreme results from 0-40 can be ascribed to an inadequate sample. For instance, Robredo–our 0-40 hero–falls to neutral at 15-40. Here is the complete list:

Player                  15-40  15-40 W  15-40 W%  W/Exp  
John Isner                238      122     51.3%   1.33  
Milos Raonic              215       98     45.6%   1.18  
Feliciano Lopez           304      108     35.5%   1.17  
Jo-Wilfried Tsonga        301      119     39.5%   1.17  
Denis Istomin             304      101     33.2%   1.17  
Rafael Nadal              320      118     36.9%   1.16  
Ivo Karlovic              148       68     45.9%   1.15  
Kevin Anderson            338      132     39.1%   1.15  
Guillermo Garcia-Lopez    405      106     26.2%   1.14  
Andreas Seppi             396      113     28.5%   1.12  
                                                         
Player                  15-40  15-40 W  15-40 W%  W/Exp  
Bernard Tomic             273       86     31.5%   1.12  
Kei Nishikori             298       96     32.2%   1.10  
Novak Djokovic            348      132     37.9%   1.07  
Richard Gasquet           325      106     32.6%   1.07  
Roger Federer             281      109     38.8%   1.07  
Fernando Verdasco         306       94     30.7%   1.06  
Philipp Kohlschreiber     352      110     31.3%   1.06  
Andy Murray               431      135     31.3%   1.06  
Santiago Giraldo          331       86     26.0%   1.05  
Tomas Berdych             398      131     32.9%   1.05  
                                                         
Player                  15-40  15-40 W  15-40 W%  W/Exp  
Marin Cilic               357      109     30.5%   1.05  
Sam Querrey               244       78     32.0%   1.04  
Jeremy Chardy             300       91     30.3%   1.04  
Fabio Fognini             422       98     23.2%   1.03  
Tommy Robredo             285       78     27.4%   0.99  
Grigor Dimitrov           307       89     29.0%   0.99  
David Ferrer              498      138     27.7%   0.98  
Alexandr Dolgopolov       299       77     25.8%   0.95  
Mikhail Youzhny           339       77     22.7%   0.94  
Gilles Simon              426       93     21.8%   0.91

The big servers are better represented at the top of this ranking. Even though Isner is expected to come back from 15-40 nearly 40% of the time–better than almost anyone on tour–he exceeds that expectation by one-third, far more than anyone else considered here.

Finally, let’s look at comebacks from 0-30:

Player                  0-30  0-30 W  0-30 W%  W/Exp  
John Isner               338     229    67.8%   1.19  
Bernard Tomic            299     146    48.8%   1.15  
Grigor Dimitrov          342     166    48.5%   1.11  
Novak Djokovic           409     235    57.5%   1.10  
Santiago Giraldo         344     142    41.3%   1.10  
Fernando Verdasco        373     175    46.9%   1.10  
Rafael Nadal             376     194    51.6%   1.09  
Tomas Berdych            492     262    53.3%   1.09  
Tommy Robredo            296     132    44.6%   1.08  
Roger Federer            344     193    56.1%   1.08  
                                                      
Player                  0-30  0-30 W  0-30 W%  W/Exp  
Feliciano Lopez          326     161    49.4%   1.07  
Alexandr Dolgopolov      347     154    44.4%   1.07  
Marin Cilic              378     179    47.4%   1.06  
Jo-Wilfried Tsonga       357     185    51.8%   1.06  
Guillermo Garcia-Lopez   380     146    38.4%   1.06  
Ivo Karlovic             186     118    63.4%   1.04  
Philipp Kohlschreiber    395     185    46.8%   1.03  
Denis Istomin            314     135    43.0%   1.03  
Kei Nishikori            341     145    42.5%   1.03  
David Ferrer             529     227    42.9%   1.02  
                                                      
Player                  0-30  0-30 W  0-30 W%  W/Exp  
Kevin Anderson           361     181    50.1%   1.02  
Mikhail Youzhny          390     142    36.4%   1.00  
Andy Murray              419     185    44.2%   1.00  
Andreas Seppi            418     164    39.2%   0.99  
Jeremy Chardy            316     132    41.8%   0.99  
Milos Raonic             246     139    56.5%   0.99  
Fabio Fognini            478     153    32.0%   0.99  
Sam Querrey              292     131    44.9%   0.97  
Gilles Simon             442     155    35.1%   0.96  
Richard Gasquet          370     159    43.0%   0.95

Isner still stands at the top of the leaderboard, while Bernard Tomic and Grigor Dimitrov give us a mild surprise by filling out the top three. Again, as the sample size increases, the variation decreases even further, illustrating that, over the long term, players tend to serve about as well at one score as they do at any other.

Forecasting the Effects of Performance Byes in Beijing

To the uninitiated, the WTA draw in Beijing this week looks a little strange. The 64-player draw includes four byes, which were given to the four semifinalists from last week’s event in Wuhan. So instead of empty places in the bracket next to the top four seeds, those free passes go to the 5th, 10th, and 15th seeds, along with one unseeded player, Venus Williams.

“Performance byes”–those given to players based on their results the previous week, rather than their seed–have occasionally featured in WTA draws over the last few years. If you’re interested in their recent history, Victoria Chiesa wrote an excellent overview.

I’m interested in measuring the benefit these byes confer on the recipients–and the negative effect they have on the players who would have received those byes had they been awarded in the usual way. I’ve written about the effects of byes before, but I haven’t contrasted different approaches to awarding them.

This week, the beneficiaries are Garbine Muguruza, Angelique Kerber, Roberta Vinci, and Venus Williams. The top four seeds–the women who were atypically required to play first-round matches, were Simona Halep, Petra Kvitova, Flavia Pennetta, and Agnieszka Radwanska.

To quantify the impact of the various possible formats of a 64-player draw, I used a variety of tools: Elo to rate players and predict match outcomes, Monte Carlo tournament simulations to consider many different permutations of each draw, and a modified version of my code to “reseed” brackets. While this is complicated stuff under the hood, the results aren’t that opaque.

Here are three different types of 64-player draws that Beijing might have employed:

  1. Performance byes to last week’s semifinalists. This gives a substantial boost to the players receiving byes, and compared to any other format, has a negative effect on top players. Not only are the top four seeds required to play a first-round match, they are a bit more likely to play last week’s semifinalists, since the byes give those players a better chance of advancing.
  2. Byes to the top four seeds. The top four seeds get an obvious boost, and everyone else suffers a bit, as they are that much more likely to face the top four.
  3. No byes: 64 players in the draw instead of 60. The clear winners in this scenario are the players who wouldn’t otherwise make it into the main draw. Unseeded players (excluding Venus) also benefit slightly, as the lack of byes mean that top players are less likely to advance.

Let’s crunch the numbers. For each of the three scenarios, I ran simulations based on the field without knowing how the draw turned out. That is, Kvitova is always seeded second, but she doesn’t always play Sara Errani in the first round. This approach eliminates any biases in the actual draw. To simulate the 64-player field, I added the four top-ranked players who lost in the final round of qualifying.

To compare the effects of each draw type on every player, I calculated “expected points” based on their probability of reaching each round. For instance, if Halep entered the tournament with a 20% chance of winning the event with its 1,000 ranking points, she’d have 200 “expected points,” plus her expected points for the higher probabilities (and lower number of points) of reaching every round in between. It’s simply a way of combining a lot of probabilities into a single easier-to-understand number.

Here are the expected points in each draw scenario (plus the actual Beijing draw) for the top four players, the four players who received performance byes, plus a couple of others (Belinda Bencic and Caroline Wozniacki) who rated particularly highly:

Player               Seed  PerfByes  TopByes  NoByes  Actual  
Simona Halep            1       323      364     330     341  
Petra Kvitova           2       276      323     290     291  
Venus Williams                  247      216     218     279  
Belinda Bencic         11       255      249     268     254  
Garbine Muguruza        5       243      202     210     227  
Angelique Kerber       10       260      224     235     227  
Caroline Wozniacki      8       208      203     205     199  
Flavia Pennetta         3       142      177     144     195  
Agnieszka Radwanska     4       185      233     192     188  
Roberta Vinci          15       120       91      94      90

As expected, the top four seeds are expected to reap far more points when given first-round byes. It’s most noticeable for Pennetta and Radwanska, who would enjoy a 20% boost in expected points if given a first-round bye. Oddly, though, the draw worked out very favorably for Flavia–Elo gave her a 95% chance of beating her first-round opponent Xinyun Han, and her draw steered her relatively clear of other dangerous players in subsequent rounds.

Similarly, the performance byes are worth a 15 to 30% advantage in expected points to the players who receive them. Vinci is the biggest winner here, as we would generally expect from the player most likely to suffer an upset without the bye.

Like Pennetta, Venus was treated very well by the way the draw turned out. The bye already gave her an approximately 15% boost compared to her expectations without a bye, and the draw tacked another 13% onto that. Both the structure of the draw and some luck on draw day made her the event’s third most likely champion, while the other scenarios would have left her in fifth.

All byes–conventional or unconventional–work to the advantage of some players and against others. However they are granted, they tend to work in favor of those who are already successful, whether that success is over the course of a year or a single week.

Performance byes are easy enough to defend: They give successful players a bit more rest between two demanding events, and from the tour’s perspective, they make it a little more likely that last week’s best players won’t pull off of this week’s tourney. And if all byes tend to the make the rich a little richer, at least performance byes open the possibility of benefiting different players than usual.

The Slow but Steady Erosion of the Server’s Advantage

After a couple of weeks of data-driven skepticism, I can finally confirm a bit of tennis’s conventional wisdom. Over the course of a typical match, breaks of serve are a little easier to come by.

This result–based on tens of thousands of matches from the last few years–is similar for both men and women. After about twelve games (total, not service games for each player), a hold is roughly 2% less likely than it was in the first few games of the match. By the 25th game, a hold is approximately 5% less likely than at the beginning of the match.

To control for the vagaries of surface, opponent, and other conditions, I’ve compared each service game to the server’s hold percentage within that match. Only the closest matches are likely to go very long, so it’s important to compare the last games of those matches to games with similarly even opponents.

It seems that this effect is the result of one or both of two factors: server fatigue (which may have more of an effect on results than an equivalent amount of returner fatigue), and the returner’s increasing familiarity with the server. It would be difficult to separate these two–and with this dataset, probably impossible–so for today, let’s stick with the nature of the effect, not its causes.

The following graph shows the relative probability of a hold of serve based on how much of the match (in games) has been played:

Relative hold percentage

I’ve set the hold probability of the first game at 100%, so all other numbers are relative to that. I’ve excluded tiebreaks from these calculations, though I considered them when counting games–that is, the first game of the second set after a tiebreak is considered the 14th game, not the 13th.

The results get a lot noisier starting around the women’s 25th game and the men’s 35th game, for the simple reason that most matches don’t get that far. For example, while the WTA calculations are based on 11,000 matches, only one-third reached the 25th game and less than one-tenth made it to the 31st.

The general downward trend indicates that the fatigue and/or familiarity effect dwarfs the effect of new balls. I have found that in men’s matches, the age of balls has a very small effect on hold percentage, and in women’s matches, it has no effect. In any case, the steady ebb of the server’s advantage is a stronger effect.

It is likely that some players suffer more from fatigue or familiarity than others. Due to the smaller size of the per-player samples, especially beyond the 20th game or so, I’m reluctant to draw any strong conclusions. Still, there are some intriguing numbers for the players for whom the dataset contains the most matches.

Here, I’ve calculated the hold percentage for several top players at various stages of the match, relative to their hold percentage in the first ten games. Thus, a number below 100% indicates less frequent holds, while a number above 100% means more frequent holds:

Player                 Matches  11 to 20  21 to 30  31 to 50  
Tomas Berdych              337     98.5%     98.3%    101.5%  
David Ferrer               330     97.0%     99.4%    102.4%  
Novak Djokovic             325    100.1%    101.8%    101.7%  
Roger Federer              325    100.2%     99.6%    100.4%  
Andy Murray                295     97.7%     98.7%     97.9%  
Rafael Nadal               293     99.2%    100.3%     93.7%  
Jo-Wilfried Tsonga         255    100.4%    100.9%     99.6%  
Philipp Kohlschreiber      252    101.4%     97.9%     96.7%  
John Isner                 251    100.4%    100.4%    100.3%  
                                                              
Player                 Matches  11 to 20  21 to 30  31 to 50  
Kevin Anderson             247    100.0%     98.1%     97.5%  
Richard Gasquet            246     99.1%     98.4%    105.1%  
Gilles Simon               245    100.1%    103.7%     95.0%  
Milos Raonic               238     97.1%     96.1%     96.7%  
Marin Cilic                238     95.4%     97.5%     94.5%  
Fabio Fognini              235    100.4%     99.6%     98.2%  
Kei Nishikori              233    101.8%    104.1%    107.2%  
Grigor Dimitrov            224    100.9%    100.3%     94.6%  
Andreas Seppi              221    106.4%    100.4%    103.1%  
Feliciano Lopez            221     99.2%     99.7%     98.4%  
                                                              
Total                    23326     98.1%     96.1%     95.1%

While John Isner is steady throughout the stages of the match, other big servers such as Milos Raonic and Marin Cilic are less dominant as the match progresses. The players whose hold percentage improves through the match–such as Novak Djokovic and David Ferrer–tend to be those without big serves, so we may be looking at more of an overall fatigue effect in those cases.

The most extreme number in the table is Rafael Nadal‘s relative hold percentage after the 30th game. Perhaps after that much time on court, his opponents finally figure out how to defend against the ad-court slider.

Here are the same calculations for top WTA players:

Player                Matches  11 to 15  16 to 20  21 to 40  
Agnieszka Radwanska       299    101.0%    104.9%     98.0%  
Sara Errani               279     97.7%     91.2%     92.7%  
Caroline Wozniacki        279    103.1%    102.3%    104.9%  
Serena Williams           266    102.8%    102.4%    104.9%  
Angelique Kerber          265    101.9%    103.0%    101.5%  
Samantha Stosur           253     99.2%    105.0%     97.6%  
Carla Suarez Navarro      252    102.2%    101.8%     93.7%  
Petra Kvitova             251     93.9%    100.4%     95.9%  
Roberta Vinci             250     94.2%     97.9%     95.4%  
Ana Ivanovic              241    100.8%    106.0%     95.2%  
Jelena Jankovic           241    102.2%    108.7%     96.4%  
                                                             
Player                Matches  11 to 15  16 to 20  21 to 40  
Maria Sharapova           236    100.1%    105.9%    104.9%  
Victoria Azarenka         228    100.6%    103.7%     97.8%  
Lucie Safarova            227    102.7%    100.5%     94.4%  
Simona Halep              224     89.2%     95.3%    101.7%  
Dominika Cibulkova        210     98.7%     89.9%     99.9%  
Alize Cornet              210     96.2%    102.8%     96.4%  
Andrea Petkovic           194    101.5%    104.2%    107.5%  
Sloane Stephens           185     97.5%     90.1%     88.7%  
Sabine Lisicki            185     97.4%     97.5%     96.6%  
Ekaterina Makarova        185     96.6%    102.8%     92.8%  
Flavia Pennetta           180    105.1%     92.9%    103.9%  
                                                             
Total                   22406     98.6%     97.2%     95.0%

Here is some confirmation that Serena Williams–at least on serve–gets better as the match progresses. Many of the other players with the strongest serve results late in matches are those known for fitness (like Caroline Wozniacki) or steeliness (Maria Sharapova).

Whether the root cause is fatigue or familiarity, most players are less effective on serve as the match progresses. With further research, I hope we’ll be able to better understand the cause and determine whether there are advantages to serving particularly well at certain stages of the match.

The Odds of Successfully Serving Out the Set

Italian translation at settesei.it

Serving for the set is hard … or so they say. Like other familiar tennis conceits, this one is ripe for confirmation bias. Every time we see a player struggle to serve out a set, we’re tempted to comment on the particular challenge he faces. If he doesn’t struggle, we ignore it or, even worse, remark on how he achieved such an unusual feat.

My findings–based on point-by-point data from tens of thousands of matches from the last few seasons–follow a familiar refrain: If there’s an effect, it’s very minor. For many players, and for some substantial subsets of matches, breaks of serve appear to be less likely at these purportedly high-pressure service games of 5-4, 5-3 and the like.

In ATP tour-level matches, holds are almost exactly as common when serving for the set as at other stages of the match. For each match in the dataset, I found each player’s hold percentage for the match. If serving for the set were more difficult than serving in other situations, we would find that those “average” hold percentages would be higher than players’ success rates when serving for the set.

That isn’t the case. Considering over 20,000 “serving-for-the-set” games, players held serve only 0.7% less often than expected–a difference that shows up only once every 143 attempts. The result is the same when we limit the sample to “close” situations, where the server has a one-break advantage.

Only a few players have demonstrated any notable success or lack thereof. Andy Murray holds about 6% more often when serving for the set than his average rate, making him one of only four players (in my pool of 99 players with 1,000 or more service games) to outperform his own average by more than 5%.

On the WTA tour, serving for the set appears to be a bit more difficult. On average, players successfully serve out a set 3.4% less often than their average success rate, a difference that would show up about once every 30 attempts. Seven of the 85 players with 1,000 service games in the dataset were at least 10% less successful in serving-for-the-set situations than their own standard.

Maria Sharapova stands out at the other end of the spectrum, holding serve 3% more often than her average when serving for the set, and 7% more frequently than average when serving for the set with a single-break advantage. She’s one of 30 players for whom I was able to analyze at least 100 single-break opportunities, and the only one of them to exceed expectations by more than 5%.

Given the size of the sample–nearly 20,000 serving-for-the-set attempts, with almost 12,000 of them single-break opportunities–it seems likely that this is a real effect, however small. Strangely, though, the overall finding is different at the lower levels of the women’s game.

For women’s ITF main draw matches, I was able to look at another 30,000 serving-for-the-set attempts, and in these, players were 2.4% more successful than their own average in the match. In close sets, where the server held a one-break edge, the server’s advantage was even greater: 3.5% better than in other games.

If anything, I would have expected players at lower levels to exhibit greater effects in line with the conventional wisdom. If it’s difficult to serve in high-pressure situations, it would make sense if lower-ranked players (who, presumably, have less experience with and/or are less adept in these situations) were not as effective. Yet the opposite appears to be true.

Lower-level averages from the men’s tour don’t shed much light, either. In main draw matches at Challengers, players hold 1.4% less often when serving for the set, and 1.8% less often with a single-break advantage. In futures main draws, they are exactly as successful when serving for the set as they are the rest of the time, regardless of their lead. In all of the samples, there are only a handful of players whose record is 10% better or worse when serving for the set, and a small percentage who over- or underperform by even 5%.

The more specific situations I analyze, the more the evidence piles up that games and points are, for the most part, independent–that is, players are roughly as effective at one score as they are at any other, and it doesn’t matter a great deal what sequence of points or games got them there. There are still plenty of situations that haven’t yet been analyzed, but if the ones that we talk about the most don’t exhibit the strong effects that we think they do, that casts quite a bit of doubt on the likelihood that we’ll find notable effects elsewhere.

If there is any truth to claims like those about the difficulty of serving for the set, perhaps it is the case that the pressure affects both players equally. After all, if a server needs to hold at 5-4, it is equally important for the returner to seize the final break opportunity. Maybe the level of both players drops, something we might be able to determine by analyzing how these points are played.

For now, though, we can conclude that players–regardless of gender or level–serve out the set about as often as they successfully hold at 1-2, or 3-3, or any other particular score.