March 14, 2018
Sportstar Live
For more than 100 years, tennis, unlike team sports, used statistics sparingly. Basketball, baseball and football needed a plethora of stats, such as shooting percentages, batting averages and touchdowns scored, to measure the performances of their athletes and teams. But tennis players were measured chiefly by their wins, losses, titles and rankings. After all, few cared if the Wimbledon champion made 64% of his first serves or the No. 1 player averaged 77 miles per hour on her backhand.
All that changed in the Computer Age. With more information than they ever dreamed possible, tennis coaches, players, media and fans suddenly craved all sorts of revealing match data, not to mention astute analysis of it. No longer was it just whether you won or lost that mattered, but how and why you won or lost — points, games, sets and matches. Training methods, stroke production, tactics and equipment were also dissected and analysed in much greater depth and detail than ever before.
As the demand for data burgeoned, new technologies, such as sophisticated virtual graphics, tracking technology, statistical applications and telestration, have provided yet more valuable services and information to give athletes that “extra edge.”
Like any prescient, enterprising pioneer, Leo Levin seized the opportunity by developing the first computerised stats system for tennis in 1982. Levin’s seminal work was highlighted by creating the concept of and coining “unforced error,” a term now used in most sports and even by pundits to describe a politician’s self-inflicted blunder.
Since then, the genial 59-year-old, based in Jacksonville, Florida, has covered more than 120 Grand Slam events and countless other tournaments to provide the Association of Tennis Professionals (ATP) and other businesses with match statistics. Levin, dubbed “The Doctor” by broadcaster Mary Carillo for his incisive diagnoses of players’ games, is currently director of sports analytics at SportsMEDIA Technology (SMT), a company that provides custom technology solutions for sporting events.
In this wide-ranging interview, Levin explains his many roles in the exciting, fast-growing field of analytics and how it has changed tennis for the better.
What is sports data analytics?
Sports data analytics is a combination of gathering and analysing data that focuses on performance. The difference between analysis and analytics is that analysis is just gathering the basic data and looking at what happened. Analytics is trying to figure out why and how the basic performance analysis works with other factors to determine the overall performance of the athlete or the team.
When and how did this field start changing amateur and pro tennis? And who were the pioneers?
Honestly, I was. At the end of 1981, the first IBM personal computer hit the market for general consumer use. By the middle of 1982, I was working with a company in California to develop the very first computerised stats system for tennis. The key factor was the way we decided to describe the results of a tennis point in three basic areas. The point had to end with a winner, a forced error, or an unforced error. That created the foundation for how we look at tennis today.
How and when did you become interested in tennis analytics?
I was playing on the tennis team at Foothill College in Los Altos, California, about five miles from Stanford University. When I wasn’t playing matches, I was actually charting matches for my team-mates and then providing that information to the coach and the players to try to help them improve their games.
Brad Gilbert, a former world No. 4 and later the coach of Andre Agassi and Andy Murray, played on your Foothill team. Did you help him?
Brad was on that team, and it was interesting because in his first year, he played No. 2. The player who played No. 1 came to me before the state finals where he had to play Brad in the final, and asked me, ‘How do I beat Brad?’ I was able to give him specific information on strategy and tactics that helped him win the state title.
That was the year Brad took his runner-up trophy and smashed it against a tree and vowed never to lose a match the following year. And the following year, Brad didn’t lose a match.
SportsMEDIA Technology’s (SMT) products and services have evolved from a clock-and-score graphic in 1994 to innovative and sophisticated virtual graphics, tracking technology, statistical applications, and telestration. How do you and your team at SMT use these four methods to analyse statistical data at tennis’ four Grand Slams to provide valuable insight that helps players, coaches, broadcasters and the print media determine how and why a match was won or lost?
One of the challenges with tennis, more so than with any other major sport, is the lack of data. When we started doing this, there really wasn’t any consistent gathering of data from matches. So the first piece we developed was simply a system now known as Match Facts. It pulled factual statistical data directly from the chair umpire. That started with the ATP back in the early 1990s. We were then able to create a base for year-round information on the players. It allowed for the next level of analysis. It has expanded from there. We developed the very first serve speed system to start adding additional data and how players were winning or losing based on the serve speeds. As the technology improved, we’ve been able to harness the new generation — tracking video technology and then on the presentation side, using virtual graphics as a way to be able to place data directly into the field of play to help illuminate what is actually going on. Telestration is a tool that allows the broadcasters to get inside the points and help the fans understand the combinations of shots and strategies the players are using.
Your website (www.smt.com) has a section titled “Visual Data Intelligence” with the subtitle, “SMT delivers the world’s most innovative solutions for live sports and entertainment events across the globe.” What is Visual Data Intelligence? And what are its most important, innovative solutions for live sports and entertainment events?
Visual Data Intelligence goes to the heart of what we try to do as a company. In a lot of different sports, there is a lot of information available. But making it useful to the broadcasters, and specifically to the fans, to help them understand the game is a huge part of what we’re providing. That entails simple things like the first-and-10 line in football. That provides the visual set of information for the commentators and fans that really helps them understand where the teams are and how much yardage they need (to get a first down). It’s gotten to the point where fans in the football stadium are yelling, “Where’s the yellow line?” So we’re expanding that to provide the service to the large screens displayed inside the stadium so teams have their own system to be able to show that to the fans.
How does Visual Data Intelligence apply to tennis?
In tennis where you have a lot of data, the challenge is: how do you provide all that data to the fans and the commentators? We do that through a series of different systems. We have what we call our “open vision system,” which is an IPTV solution that has real-time scoring, stats and video as well as historical data. And it ties it all together and puts it in one place so it provides a true research tool for the commentators and the (print and online) media. Along with that, we have a product we call our “television interface,” which is really a system which drives graphics on air for the broadcasters. This tool allows them to look at the data and see where the trends are. Hit the button and have that information directly on the screen.
Please tell me about the new technology service partnership between Infosys and the ATP, and the analytics and metrics this partnership brings to the tennis world.
I’m not really that aware of what Infosys and the ATP are doing. But I do know that a lot of that hinges on the technology we created for Match Facts. One of the unique things about tennis is the scoring system. Unlike other sports, the player or team that wins the most points doesn’t necessarily win the match. That’s not how our scoring system works. I think they are trying to take a deeper look into the individual points, and how winning or losing specific points in key situations impacts a player’s ability to win or lose matches. The same is true for total games. That’s one of the challenges when you’re trying to do analysis of tennis. In a lot of other sports, you’re just looking at the raw numbers and saying how many points did he score or how many rebounds did she get or how many yards did they gain. But in tennis, it has to be compartmentalised into specific performances in specific situations.
How do insights from game and training data analytics improve coaching?
The key to coaching and player improvement is first to understand what is going on out on the court. It’s a matter of gathering data. One of the challenges tennis has faced because of its late start in the world of statistics and data analysis has been a reluctance by a lot of coaches and players to rely on anything other than what they see and feel. So the real challenge and the real key is to be able to relate the data to what coaches see and what players feel out on the court. When you can make that connection, you have a real chance for improvement.
What are one or two insights that have improved coaching?
The challenge is that every player is different. What the data analysis allows you to do is to customise those things and focus not on what a player does, but what your player does, and how you can get the most out of your player’s game. A simple example of this was when we first started doing detailed statistics and analysis, we worked with the Stanford University tennis programme. Their No. 1 woman player, Linda Gates, was struggling, and the coaches couldn’t figure out where or why. We did an analysis of her game, and we found out that she was dominating her service games on her service points in the deuce court, but she was struggling in the ad court. It wasn’t visually obvious. The coaches couldn’t put their finger on what the problem was. But once we started looking at the numbers and the data, it allowed them to focus in practices on her ad-court shot patterns. Linda went on to win the NCAA Championships that year, 1985, in singles and doubles (with Leigh Anne Eldredge).
An Infosys ATP “Beyond The Numbers” analysis of Rafael Nadal’s resurgence to No. 1 in the Emirates ATP Rankings showed that Nadal ranked No. 1 on tour in 2017 for winning return points against first serves, at 35.2 percent (971/2761). That metric shoots up to an astounding 43.4 percent (454/1045) for his clay-court matches. Which other stunning statistics help explain why other players have had outstanding years this decade?
This goes to the basics of looking at players’ strengths and weaknesses. One stat I always look at is serve and return performance because I still split the game up that way. It’s interesting that when you look at a player like Nadal, you see that he is not only dominant on return of serve. He’s also dominant on his own second serve.
Even with all the analytics we have, an old maxim still holds true: “You’re only as good as your second serve.” You’ll find the players at the top of the rankings for the last four or five years were also at the top of both second serve points won and return of second serve points. Despite all the focus on power and big serves, second serve performance is really a huge key to understanding a player’s overall strengths and weaknesses.
How much do the Women’s Tennis Association tour and its players take advantage of analytics?
Although the WTA was a little behind the ATP curve in terms of gathering and storing match data, the good news is that now they’ve caught up. Their association with SAP and that they’re also now using a Match Facts system to provide data for the players on a match-by-match basis has moved them up the curve.
Which pro players have benefited most from tennis analytics so far? And in what specific ways?
That’s a tough question. Because I don’t work directly with the players and coaches as I used to, I don’t know who is utilising the data more so than others. You can tell just by looking at Roger Federer’s improvement over the last year that his team used analytics to determine that he needed to be more aggressive on his backhand. He’s now hitting a much higher percentage of topspin backhands than he did in previous years and that change has made his game more balanced and puts a lot more pressure on his opponents. Playing to Roger’s backhand used to be the safe play — it’s not any more.
Another area of Federer’s game that came to light using analytics was the difference between his winning and losing matches at Wimbledon. When you compare his final match wins to his matches lost since he won his first Wimbledon in 2003 — 8 titles, 7 matches lost — the numbers that jump out are all about his return of serve, and specifically, his performance on break points. Federer’s serving performance barely changed, but his return game fell dramatically in his losses. In his Wimbledon final wins, Federer converted 30 of 69 break points for 44%. In his losses, he converted only 9 of 53 for 17%. In both cases, he averaged around 8 break points per match. In his wins, he converted almost 4 per match, but in his losses he converted just over once per match. His team looked at that crucial data and added in that nearly all his opponents served and volleyed 2% or less of their service points and concluded that Roger needed to work on hitting his returns deep and not worry about his opponents coming in behind their serves.
Younger players are taking most advantage of the information because they’ve grown up in that world. They’re used to the electronics and the digital experience and having all that information available to them.
How do these insights enhance the fan experience?
I credit (renowned former NFL analyst) John Madden for being one of the very first TV commentators who would take fans inside the game to explain to them things they didn’t necessarily see. Madden would explain to women football fans what the centre or guard was doing on a particular play and why that back ran for 50 yards was all because of this really good block.
What we’re trying to do in tennis and what these insights have provided is to do the same kind of things for tennis fans. Help get them inside the game so they understand the nuances of what’s happening on the court, and they’re not just watching two guys running around hitting the ball.
What is radar-based tracking, which is now used by the United States Olympic Committee (USOC) for every throw an Olympic athlete makes? Is it being used in tennis?
Radar-based tracking is simply tracking the speed and location of the ball or object that is being thrown or hit. Radar-based tracking has been typically used for service speeds in tennis. That is something we pioneered in the late 1980s. The tracking used in tennis has been video-based, as opposed to radar. The advantage of that is that you can track movement of the players as well as the movement of the ball and from a variety of positions and angles.
Can analytics predict which junior players will someday become world-class players or even champions? And if so, can it guide their coaches and national federations to increase the odds that will happen?
Not yet. The challenge is that prediction is different from analysis. You’re trying to draw conclusions from the data, and we don’t have a complete set of data. If you wanted to predict which junior players will become world-class players, sure you can do that if we have genetics, biomechanics, all the physical characteristics measured as well as using analytics to measure the player’s overall performance on the court. We can see whether or not they have specific markers that indicate they will make that jump. But the bottom line is that there are so many factors involved. And a lot of it has to do with the physical side that you can’t necessarily determine from data.
What is bioanalytics? And why is measuring and analysing an elite athlete’s perspiration important?
We’re pioneering bioanalytics in football now. We’re taking biometric readings from players at the university level. The players are equipped with motion sensors and full biometric readers, which are reading things like heart rate, body temperature and respiration. And they’re combining that with the movement data from the tracking information. With that, we’re able to measure the physical output of the players. The sensors in the helmet measure impacts (from collisions).
We’ve been working on this project for a few years. It’s been used for the football programme at Duke University. We’re in the process of adding a couple more universities to this project. At this stage it’s being used for medical purposes. So when a player is on the practice field, they can know immediately if his heart rate starts racing or if his body temperature goes up too high, they can immediately pull him out of practice and get him more electrolytes and hydration. They also weigh the players before and after every practice so they know how much fluid the player has lost during their practice times.
How is bioanalytics used in tennis?
Unlike a team sport where a team can outfit all its players with this equipment, tennis players are all independent contractors. So it’s going to take more of a nationalistic approach — something like what the USTA is doing — to step in and say, “For our junior players, we’re going to outfit some courts and we’re going to provide this level of analysis on the physical side.”
Does analytics apply to tennis equipment and court surfaces? And if so, how?
Sure, it can. Analytics can identify how well players perform using different types of equipment and on different surfaces. For instance, if you’re using some tracking technology to determine what racquet and string combination allows a player to have the most amount of power, that’s a relatively simple exercise. You run a player through a set of drills, hitting particular shots, and measuring the speed of the ball coming off the racquet.
For surfaces, analytics can really help with identifying the type of shots that have an effect on particular surfaces or areas where players’ games break down. For example, you have players who have a long backswing, and that works really well on a slower surface where they have time to take a big backswing. But when you put them on a faster court, where the ball bounces lower and faster, it upsets their timing, and it makes it more difficult for them to adjust. Analytics measures the court’s bounce speed and bounce trajectory. So you can take a player and modify his game on a particular surface taking into account how the ball reacts to it.
You’ve analysed thousands of matches. Which factors influence the outcome of matches the most in men’s tennis and women’s tennis? And why?
The No. 1 factor typically is unforced errors. If you’re making mistakes, you’re basically giving the match to your opponent. Being able to measure and quantify that is a huge factor for player improvement. That entails understanding where you’re making your mistakes — which shots and what situations. The caveat to that is that there are certain players whose games are based on absolutely controlling the pace and tempo of the match. And they have the tools to do that. Two of the best players ever to do that are Steffi Graf and Serena Williams.
What are the disadvantages of and dangers involved with analytics? Will some number crunchers and coaches go overboard with analytics and be guilty of Occam’s razor?
The simple danger is to rely on data alone. The challenge is that you have to make the data relatable to what the player is doing physically and mentally on the court. Analytics doesn’t necessarily measure the mental side of the game, at least not yet. If you’re focusing so much on the analytics of certain shots and not looking at the big picture of their mental focus and how they’re preparing for matches, you can get into trouble.
Since tennis players vary greatly in temperament, talent, current form and other variables, do practitioners of analytics risk over-concluding from their numbers? And what mistakes have you and others made in this regard?
There is always a risk. Data can provide you with valuable information. Then you make that next leap that says, “This information says this, and therefore we have to do this, or therefore we have an issue.” I’ll give you a simple story from a few years ago. Jim Grabb, who was the No. 1 doubles player in the world then, came up to me at a tournament before the US Open and said, “I’m struggling with my first volley in singles. I can’t make a first volley.” And I told him, “You’re the No. 1 doubles player in the world. You have great volleys. And you’re saying you can’t make a first volley in singles.” He says, “Yeah.”
A lot of coaches would say, “How are you hitting it? Let’s analyse the stroke.” I asked, “When you step to the baseline to hit the serve, where is your first volley going?” Jim looked at me like I was speaking a foreign language. So I asked again, “Before you hit your first serve, where are you going to hit your first volley?” He said, “I just react to the ball. I don’t know what you’re talking about.”
So I suggested, “Do this. Every first volley goes to the open court. You serve wide in the deuce court and you volley wide into the ad court. You serve wide in the ad court and volley wide into the deuce court. Just for your first volleys.”
Jim goes out to play and comes back and says, “I didn’t miss a first volley.” The next week he got to the fourth round of the US Open, his best result at a Grand Slam (event) ever in singles. That had to do with the fact that all it really required was a little bit of focus by the player. It didn’t require a level of analysis and stroke production changes. It was simply eliminating decision-making.
What is the connection between analytics and the established field of biomechanics?
Analytics can tell you how a player is performing or how a stroke is performing in key situations. That can then identify that we need to examine the biomechanics of the stroke, particularly if it is breaking down under pressure. Or we can determine that the errors are occurring when the ball is bouncing four feet in the air versus three feet in the air, so their contact point is a foot higher. Now we can look at the biomechanics and see what the player is doing when the ball is a foot higher.
What are player rating systems? And what is the connection between analytics and player rating systems? How valid is the Universal Tennis Ratings system?
I don’t think there is any now. But that’s a direction we can take in the future.
Which match statistic or statistics do you foresee becoming increasingly important as a result of analytics?
I think you’ll see more focus on key point performance as we do more and more analysis of players’ games in key pressure situations. Because you’re serving half of the time and receiving serve half of the time, analytics will look increasingly at each half of the game. We talk a lot about unforced errors, but are they occurring on your serve game or return game? We talk about aggressive play and taking control of the points, but when is that happening? And the serve or return games? On the first serve or second serve?
Data analytics is undeniably changing tennis. Do you think it will revolutionise tennis?
Absolutely! Because the game is always changing. The technology around tennis and all sports keeps changing. Analytics is going to make the athletes better. It’s going to provide them with insights about how they can be at their peak for the key matches. It will help them train better, prepare better, execute shots better under pressure. All those pieces and parts will be available for athletes. And all of their nutritional, sleep, and training regimens will also help tennis players to perform better.