Deconstructing Tennis Goes Up Against Big Blue
Big data is the hot new thing. Potentially its correct use will offer all sorts of insights into how things work. But how do the insights of Big Blue compare to those of Deconstructing Tennis?
IBM has long run the websites at all four of the Grand Slam events. IBM, at least in my mind, has a reputation with data that the NY Times has with journalism. They are top-rated. For each match, “Keys” are presented which, if achieved, will most likely lead to victory for each player. Data is gathered from 8 years of Grand Slam play and crunched through IBM’s AI (Artificial Intelligence) computer. These “Keys” however suffer from two flaws: 1) they are not easily observable by fans or commentators (although the commentators have the luxury of instantaneous calculations at their fingertips); and 2) they are often misleading, if not outright incorrect. Dare I say that they are the equivalent of “fake news” in that they get the audience and the broadcasters alike to take their eyes off of the ball?
Here’s what I mean:
Here are the “Keys to the Match” for the Fed-Anderson quarter-final played on Wednesday of this past week. Each of these metrics is calculated as the odds of winning a set. These “Keys” are part of the pre-match broadcast team’s warmup before each of the matches begins. Fed’s three “Keys” were:
- Win more than 72% of first serve points
- Keep the average speed of return shots below 59.1 mph on second serves
- Keep first serve percentage above 69%
As for my first point about observability: are there any fans who can calculate a player’s return of serve speed … and then create an average … an then have it accurate down to a tenth of a mph? Obviously not! This metric is almost laughable. That being said, IBM has had promising success with using data to help doctors to prescribe the correct meds for patients. In that setting precise measurements are enormously helpful. Tennis is not medicine and the AI platform needs to be radically simplified to make it accessible and meaningful.
Now let’s look at my second point: are these measures working to predict the winner? Let’s take a look.
Fed accomplished #1 in sets 1, 3, 4, and 5. He lost 3 of 4 of those sets and won set 2 where he failed to reach his target.
Fed accomplished #2 in all sets except for the first. This is the set which he won handily. He lost sets 3, 4 and 5 despite achieving his goals (well, not his goals, Big Blue’s goals!)
Fed reached his goals with #3 in all but the final set. So he won two and lost 2 using this criterion. The computer was correct in 3 of the 5 sets.
The results for Anderson are no better:
- Win more than 27% of first serve return points
- Win more than 80% of points when serving at 15-30
- Have a second serve kick height of greater than 4.9 feet
Anderson reached his #1 goal only once and this was in the second set which he lost.
He achieved his #2 targets in sets 1, 3 and 5 winning 2 of 3. He failed to achieve his target in set 4 which he won.
Only in set #4 did Anderson achieve the required kick height on his second serve. This measure correctly predicted his losses in sets 1 and 2, but failed to account for his victory in sets 3 and 5.
Overall mark with Fed: 5 out of 15.
Overall mark with Anderson: 6 out of 15.
So, my second point is proven by the results: The “Keys to the Match” often do a poor job of predicting who will win.
Is there a more reasonable alternative? Yes! Have the expert commentators pick a simple criterion that is easily observable. It can serve as a talking point throughout the match. Here is what I would have chosen: “Anderson is a player who thrives on attack and is far less comfortable when defending. Fed needs to attack his returns in order to immediately get Anderson into defensive mode. This was precisely the way he played in the 6-1, 6-1 demolition of a couple of years ago.”
On a whole other subject, the website now provides a graphic on momentum throughout a match. My guess is that this is supposed to be interesting because it strongly suggests that the possessor of momentum will go on to victory. Otherwise, why would we talk about it?
In the attached document, you will find my simple analysis of Momentum using the IBM graphics. As it turns out, most commentators are often saying things like this:”He had the momentum just 10 minutes ago and then it switched.”
The question of whether or not momentum exists has been tested using 10 years of Wimbledon data (not by IBM though!). The findings were that momentum is such a small factor in tennis that it may as well be ignored. Yet IBM presents Momentum graphics on each match as if they matter.
Early in the 20th century nearly all physicists believed in the existence of ether – a weightless substance which bound the Universe together. This belief held despite the fact that no one had ever seen or measured ether. See any parallels here? It’s time to abandon momentum as a reference point when discussing tennis. (That being said, there might be some valid discussion of the effects of rain delays or postponements due to darkness).
In the attached document, I show the website graphics and my simple interpretation. In the Fed-Anderson match, momentum could be said to have forecast the outcome in only 1 out of the 5 sets – the first set. But in that case, an alternative hypothesis – players in the lead tend to win – serves to explain the graphic just as well.
Want to learn about tennis? Then check in here with Deconstructing Tennis and let Big Data go!