I was about a thousand words into a meandering first draft when I realized that the ultimate goal of tennis analytics could be described simply:
The ability to quantify the impact of each individual shot, probably using camera-based player- and ball-tracking.
The purpose of every shot is to increase the odds of winning the point. (There are exceptions; you might hit a suboptimal shot to make a later shot harder to read or otherwise mislead your opponent.) Serves offer clear-cut examples. Carlos Alcaraz wins about 66% of his service points. If he hits an ace, that one shot increases his chances of winning the point from 66% to 100%, a swing of 34 percentage points. A missed first serve is worth -11 percentage points, as his chance of winning the point drops from 66% to 55%.
Shots that don’t end the point have a more modest effect than an ace, winner, or error. If you respond to a neutral forehand down the middle with a slightly more powerful forehand back down the middle, you might be upping your odds from 50% to 55%. If you run down a strong drop shot and just barely chip the ball back into play, your chances of winning the point might increase from 3% to 5%.
Point being: Each shot has some impact on the likely result of the point. If someone has, say, an above-average backhand, that will show up in these hypothetical numbers. Not every one of his backhands will move his single-point win probability in the right direction, but when we put them all together, we would be able to say that in a given match, his backhand was worth 1 point above average, or 2.5 points, or whatever else the sum of the individual impacts worked out to.
Shot-by-shot stats like these probably require camera-based ball and player tracking. We can come up with rough estimates using Match Charting Project data. (I’ve tried; results are mixed.) To get anything close to accurate measurements of win probability when a point is in progress, though, we need to know where each player is positioned as well as the progress of the ball.
Of course, the “end goal” of analytics differs depending on your own aims. If you are a player or coach, you want to know how to get better, or what tactics to use against your next opponent. These individual shot-impact stats wouldn’t identify mechanical flaws, but they would make it possible to isolate each individual type of shot at a very granular level–for instance, running backhands against left-handed forehands hit harder than 80 miles per hour. In terms of tactics, the benefits should be clear. The more detailed your understanding of an opponent’s strengths and weaknesses, the better your ability to tailor a game plan.
If you are a bettor, you are primarily concerned with predicting the future. A key component of that is to separate luck from skill. That’s the purpose of every sports stat with the word “expected” in it, from baseball’s xwOBA to soccer’s xG. Tennis doesn’t have much in the way of “x” stats because we generally don’t have access to underlying data that would allow an estimate of how many points each player “should” have won. Done correctly, “expected service points won” (call it xSPW) would be a better predictor of future results than the actual SPW we work with now.
Finally, if you’re a fan like me just hoping to better understand the game, these numbers would be a gold mine. Impressed by a “steal” when Andy Murray runs down a lob and hits a winner in return? How much better would it be to know exactly what his chances were of winning the point–and how it compared to his career bests? The next time Novak Djokovic and Daniil Medvedev slug out a 20-stroke rally, wouldn’t it be fascinating to know exactly who had the edge at each stage, and which shots shifted the momentum?
Now imagine those numbers for every steal, every momentum shift, every rally. We would learn so much about each player’s skills and tendencies, far beyond the few examples I’ve given so far.
The possibilities are endless. Having these numbers, especially if they became available in real time, would transform the way we talk about the game. Every time a baseball player hits a home run, we immediately find out the exit velocity and launch angle–measurements that tell you just how well it was hit. The more we can talk about the details of fundamental skills athletes are asked to execute, the better we understand just how well or poorly they are playing. Top-level results like set scores and match wins are lagging indicators, not leading ones.
I don’t know how, when, or even if the tennis-loving public will get stats like these. But I get excited just thinking about it.
* *
Subscribe to the blog to receive each new post by email: