Embed from Getty Images
Italian translation at settesei.it
This year’s US Open was the first grand slam to use a countdown clock before each serve. The time between points was set at 25 seconds, up from the official grand slam time limit of 20 seconds, partly to acknowledge the reality that 20 seconds was never going to happen, and to compromise with the ATP and WTA, whose limits have long been 25 seconds. The clock was tested at several North American events this summer, and I’ve already measured the effect of the clock on match times: once at The Economist’s Game Theory blog, and a second time here at Heavy Topspin.
In those two articles, I found that the serve clock seemed to make the sport slower. Using the limited data at hand–the number of points in each match and its overall time–it turned out that at every event using the clock in 2018, matches were slower by between 0.3 and 2.0 seconds per point. That doesn’t sound like much, but it adds up to a few minutes per match, and this is an innovation that was designed to hurry up play, not hold it back.
The US Open gives us a larger set of matches to study as well as more detailed data to work with. Before we attempt a less ham-handed approach to the problem, let’s see how the matches in New York measured up by the simple standard of seconds per point. Here is that calculation for all main draw singles matches in 2017 (without the clock, and a nominal 20-second time limit) and 2018 (with a 25-second clock):
Draw 2017 2018
Men 40.0 43.4
Women 40.7 42.3
Those are some awfully slow matches. Of the other summer events I analyzed, only the 2018 men’s draw in Washington exceeded 42 seconds per point.
However, the excessive heat probably played a part in some of the glacial play. The US Open heat policy certainly slowed down matches, as it allowed for a 10-minute break after the first two sets of women’s matches and the first three sets of men’s matches when the conditions were particularly bad. Those breaks are included in the official match times, so we need to account for them somehow.
Let’s skip some extra work and avoid the heat policy entirely by comparing only straight-set matches from 2017 and 2018, none of which were eligible for a heat break. That still leaves us with half of the original data points:
Draw 2017 Straight-Sets 2018 Straight-Sets
Men 39.2 43.4
Women 39.8 41.3
That was not what I expected. The straight-set matches this year were almost the same speed as the longer ones, even without the possibility of a 10-minute heat break. Maybe players don’t dally as much during straight-set matches because so many of them are lopsided. Or perhaps the mix of players is a bit different. Whatever the reason, this apples-to-apples comparison shows that this year’s apples were quite a bit slower than last year’s.
Again, with better data
The heat policy issue illustrated the problem with using overall match time: It includes set breaks, changeovers, challenges, lets, and every other random type of delay you can imagine. In the long run, all the delays will even out, but in the long run, we’ll all be dead. So far, we’ve seen only a few hundred matches on each tour using the serve clock.
The US Open Slamtracker includes timestamps for the beginning of every point of most singles matches. That’s still not perfect–it doesn’t tell us when points end, for one thing–but with a bit of care and handling, it’s something we can work with. First, I took the Slamtracker data and identified every first-serve point that didn’t end the service game. I filtered out second serves because players use such wildly differing times between first and second serves, and that’s not something addressed by the serve clock. And I filtered out game-ending points because the pause after those points would be longer, involving switching servers and often changing sides.
That left about 16,000 points, a healthy amount of data to work with. From there, I tried to figure out how time was spent actually playing tennis. You know, serving, returning, hitting a bunch of slices, that sort of thing. It turns out that each additional shot adds roughly two seconds to the time between the start of that point and the start of the next. A portion of that might be additional fatigue, resulting in a longer between-points break, but I’ll give the players the benefit of the doubt and assume it’s all time spent playing tennis. I’ll also be generous and say that the first shot–the length of an ace or unreturned serve–is five seconds, to allow for some of the more elaborate service motions.
Put it all together, and we have 16,000 points for which we can estimate the length of the break after the point. If the timestamps for point 1 and point 2 are 35 seconds apart and point 1 was a five-stroke rally–5 seconds for the first shot, 8 seconds for the ensuing shots, for a total of 13 seconds–we can conclude that it took 22 seconds for the server to towel off, choose between various amounts of tennis-ball fuzz, and get ready to serve again.
One last step, again in the spirit of generosity: I eliminated the longest 5% of between-point breaks in each match. Some of those are probably challenges, or let serves, or other disruptions not reflected in the data. I’ve probably filtered out some legimate cases in which the server was really, really slow, but I want to do what I can to give us results that are uncontaminated by too many external issues.
Enough methodology, here are the results. The table shows the number of between-point pauses that were under 20 seconds, under 25 seconds, over 25 seconds, and over 30 seconds. Remember that these times, and the resulting rates, are built on a series of player- and official-friendly assumptions. I’m fairly confident that if we took a stopwatch to 16,000 points and audited the process in person, we would be much more likely to come up with equal or longer times between points than shorter ones.
Time Between Points 2017 2018 Change (%)
Under 20s 86.5% 78.6% -9.2%
Under 25s 97.0% 95.1% -2.0%
Over 25s 3.0% 4.9% 63.1%
Over 30s 0.4% 0.8% 91.0%
The number of excessively long breaks was not very high–less than one point in 20 this year–but the figures skyrocketed in comparison with last year. We could attribute this to the rule change from 20 seconds between points in 2017 to 25 this year, but as we’ve seen, matches with the 20 second limit last year were about as fast (on a match-time per point basis) as those with the 25 second time limit. So I think that’s a non-starter.
The heat, of course, remains a factor, even when heat policy breaks are taken out of the equation. Hotter, more humid conditions will tire players out more quickly, and that will show up in the amount of time they spend recovering between points. Maybe that accounts for the near-doubling of 30-second-or-longer pauses since last year.
Still, there are plenty of questions left to be answered about the serve clock and the way umpires are using it. The rate of 30-second or longer breaks, 0.8%, sounds tiny, but across 16,000 points, it’s over 100 cases. My study was able to include only about half of the points in Slamtracker-covered matches, which itself represents perhaps three-quarters of singles rubbers. Thus, we could be talking about over 300 instances of a player taking more than 30 seconds before serving over the course of the tournament. (And remember, we excluded the longest 5% of between-point pauses.) The number of 25-second-or-longer breaks is even more damning: By the same reasoning, there may have been nearly 2,000 times when a player exceeded the 25-second limit. A few time violations were called, sure, but only a tiny fraction of these probable offenses.
As I noted in my previous article here, a big part of the problem stems from officials waiting until after the crowd has settled down to start the clock. Thus, in an exciting, well-attended match, the time limit effectively becomes 35 seconds or more. This may be what umpires are instructed to do, but it is a sure-fire way to slow matches down. There’s no reason not to start the clock immediately and pause it later for the rare instances when the crowd is making too much noise 25 seconds later.
The simple approach to evaluating the effect of the serve clock, outlined at the beginning of this article, continues to suggest that the serve clock has made matches slower. The more sophisticated tack, made possible by the more detailed data available for most grand slams, supports the same argument, and shows us just how often players are still able to take extra time between points. Let’s hope the serve clock is a work in progress, because changes are necessary if it’s going to contribute to a speedier sport.
