Internet discussion has perked up about a post of mine from last month, The Mirage of Surface Speed Convergence.
Many people don’t like my results, and plenty of people just don’t like having someone challenge their preconceived notions–or those of the players they idolize.
Yet for all the chatter, no one has even attempted to address the question at the end of that post:
If surfaces are converging, why is there a bigger difference in aces now than there was 10, 15, or 20 years ago? Why don’t we see hard-court break rates getting any closer to clay-court break rates?
Unless there is a valid answer to those questions, it really doesn’t matter how you felt after watching the Miami final, or what a top player said in some press conference.
Jeff, I think a problem here is that what to you is self-evident and indisputable is actually an assumption, even if quite a reasonable one. And it may be an assumption some readers don’t share, even if they don’t do a good job of explaining why not.
It is sort of like the labeling problem I described in a comment on the original post. But let me see if I can explain it another way that may be clearer, using the Toulmin model of fair argument.
I have actually taught the Toulmin model in class – anyone unfamiliar with it can check out the Wikipedia article on Stephen Tolumin for more information. Briefly, Toulmin sees an argument as made up of not just evidence and a claim, but an underlying assumption or “warrant” that links the evidence to the claim. The example given in the Wikipedia article is that if a man claims that “I am a British citizen,” with his sole piece of evidence that “I was born in Bermuda,” then the underlying warrant is “Persons born in Bermuda are legally British citizens.”
In your case, you too have a warrant, and it is captured by this paragraph from your original piece:
“When courts play faster, there are more aces and fewer breaks of serve. The slower the court, the more the advantage swings to the returner, limiting free points on serve and increasing the frequency of service breaks.”
This may seem self-evident – so much so that you did not spend much if any time qualifying it. But I think some readers, some pundits, and possibly some players as well wouldn’t agree.
The only way to know if others agree with a warrant is for it to be brought forward as an explicit contention to be discussed. Once the community agrees on the warrant, only then can participants move onto the implications.
So I think the question here about ace rates is not the one you propose – “If surfaces are converging, why is there a bigger difference in aces now than there was 10, 15, or 20 years ago?” – but rather, “Do we agree that ace rates are a sufficiently valid proxy for surface speed?” And ditto the question for break rates – we cannot yet ask, “Why don’t we see hard-court break rates getting any closer to clay-court break rates,” but must first ask, “Do we all agree that break rates are a sufficiently good proxy for surface speed?” Note the word “sufficiently”!
It may seem like a picayune difference, but I don’t think it is. If you really had full agreement on the appropriateness of these two metrics, who on earth could argue with you? And the fact that some people *are* arguging may be less about math and more about their not being persuaded of this particular warrant. I’m only guessing, of course.
As an example, look at how pundits use a player like Roger Federer to build a story around about surface speed. I’ll use the first hit I came upon when Googling for “Federer court fast slow”:
http://espn.go.com/tennis/story/_/id/8625576/roger-federer-wants-faster-courts-encourage-attack
Mid-story, we get this paraphrase: “[Federer] said slower courts are also good for long rallies — which are a big crowd pleaser — but that having more variety in the surfaces would force players to learn to be more aggressive.”
Federer is then quoted about what “more aggressive” on a faster court might actually mean. And what interests me is that the terms he uses seem to have little to do with aces & breaks:
“‘What you don’t want is that you hit 15 great shots and at the end, it ends up in an error,’ he said. ‘So I think sometimes quicker courts do help the cause. I think it would help from time to time to move to something a bit faster. That would help to learn, as well, for many different players, different playing styles, to realize that coming to the net is a good thing, it’s not a bad thing.'”
So for Fed here, not aces & breaks, but rather, slightly shorter rallies (and I’m guessing, better chances of a winner not being retrieved, too) and more net play.
And maybe some fans are “thinking like Federer.” Maybe they too are associating “fast” play with more aggressive point construction, more reward for attempted winners, and ditto for more net play.
If so, how about rally length mixed in as a part of a formula for court speed? And winner rates? Of course, just as with ace rates & break rates, I think these stats wouldn’t work for readers if they were presented as warrants. They’d have to be argued for explicitly. Which gets messy.
So to me, the question of surface speed still seems partly a labeling problem. The verbal discussion of “what’s fast, what’s slow” is something that’s hard to measure not just because of lack of data – and not just because so many of us are bad at math – but also because the verbal categories of “fast” and “slow” in people’s heads may include qualities that can’t be captured by ace and break rates alone. And these qualities, though seemingly ineffable, may be important to fans & players.
Jeff — I really enjoyed your article and I found it very surprising! Your methodology is interesting — in that you “normalize” by taking H2H and also by taking the ratio of the clay-to-hard court breaks (and aces). I have a few comments on your article:
The peaks in the plots do not seem to have an “obvious” explanation. For eg., take the “2005-peak”; what happened around 2005 to produce more breaks (aces) in the clay courts? or conversely fewer breaks in the h/c? Why such an evolution in time?
Because the peaks for these ratios appear “ad-hoc” to me, I am inclined to believe that the way the matches got sampled in your data (because you retain only pairs of players) for each year makes it not representative across time (while remaining a perfectly reasonable measure for that year). i.e, how are you sure that the population of players that you selected for that year is “consistent” with the population of players for the previous year?
In contrast take a look at the plots on this website for the four GS without such a “selection bias” (i stumbled on this page):
http://www.physics.usyd.edu.au/~cross/GrandSlamStatistics.htm
Clearly, the first plot shows that the speeds have converged for four grand slams; the second plot (aces) confirms that too (if we ignore the Isner-Mahut match). Admittedly, this study was done only for the GS, but it still corresponds to what the present and the past players say! More importantly, it does not seem to produce peaks that are “difficult” to explain.
It will be great to know your thoughts on this…
To be honest I think your research about it was too simple for once.
Your articles are really good because you weight the different stats so we can really use comparisons rather than just numbers.
But on this one you are comparing different players, different racquets, different balls, different player entries on each court…and you concluded about surface speed.
If you read for examle Stefan Edberg he said that strings made a major difference and now give more control to the returner and it s why you can t play anymore like he used to do. It s just an example.
If the rallies are longer now it can be because of the surfaces. Max Mirnyi for example would have made the difference between hard court and clay court for ace very big( I m sure you can find out the exact numbers), surely more than a clay courter, but those type of players are gone since the surface have been made slower and they cant enter the draws anymore. If you were comparing same players between the years it would make sense but you are comparing very different players…The fact that you had 20 server volleyer among the top 100 and now just one for example make a big difference in the composition of your main tour list of matches.
Gentlemen – it’s nice that you assume aces are symptomatic of court speed. However, why not use average length of rally? If the courts are getting slower, then rallies should last longer (average racket head size probably hasn’t changed much over the last 10 years). And adjust for taller players (who presumably have longer reaches). Yes, it is assumed today’s athletes are “better” athletes than those of 10-15 years ago. But then should they not be able to hit the ball harder/more easily through the court? Measure average rally length…see what you get.
The second line of research to pursue is average bounce height. Why is that important? It might follow that the shift to higher bounces has produced a shift to more extreme grips, right? In turn, higher bounces hurt under-spin shots, which diminishes the ability to “attack”. And higher bounces provide MORE time to defend. So…how high are bounces on courts today?
Curious minds want to know….
The data just doesn’t exist for rally length, let alone bounce height.
TC was showing the Fed v Safin Aussie Open Semi’s…We know they slowed the courts in moving to Sydney…here’s your chance to collect data from same tourny 10 years apart…