Already a Bloomberg.com user?
Sign in with the same account.
On a Saturday morning in March, some 400 people crowd a conference room at the Boston Convention Center. Mostly men, and mostly paying customers, they are there to listen to six other guys talk about baseball statistics. It's day two of the MIT Sloan Sports Analytics Conference, an annual gathering dubbed "Dorkapalooza" by ESPN's Bill Simmons. The buzz from the panel is about something called Fieldf/x.
"Do you feel like Fieldf/x will essentially make all other fielding stats irrelevant?" asks the moderator, the writer Rob Neyer.
"Yes," says Tom Tippett. Tippett is the director of baseball information services for the Boston Red Sox, which means he is in charge of gathering and crunching numbers to help put together a winning team. "I kind of feel like I have to throw away everything I've done for the last 20 years and start over."
But in a good way. Fieldf/x will create a digital catalog of virtually every movement at every Major League Baseball game in every park. Already in place in San Francisco's AT&T Park, it is coming to four more venues this year. If all goes according to plan, it will be in every major league park by 2012.
Fieldf/x is a motion-capture system created by Chicago-based Sportvision. It uses four cameras perched high above the field to track players and the ball and log their movements, gathering more than 2.5 million records per game. That means you could find out whether Ichiro Suzuki truly gets the best jump on fly balls hit into the right-field gap, or if Derek Jeter really deserved that Gold Glove last year.
The deluge of numbers will send analysts scrambling to answer just such questions. What they find will change the way teams choose and pay their players, and the way fans watch and talk about the game.
Motion capture, or optical tracking, promises to rid sports of the biases of the human eye and quantify the formerly unquantifiable art of being in the right place at the right time. In the National Basketball Assn., teams have begun using it to sort out the conditions that make for a high-percentage shot. In European soccer, clubs are mapping the endless streams of passes and moves to determine which lead to goals. In baseball, it promises to solve the age-old statistical riddle of defense, a feat that could make and break players and ball clubs.
Curiously, Fieldf/x has roots in one of the most unpopular additions to sports of all time, the much-reviled glowing puck, which annoyed hockey fans for two years between 1996 and 1998. In 1998, three Fox Sports executives who had helped to create the effect left to form their own company. Under its original chief executive officer, Bill Squadron (formerly a vice-president of strategic planning at News Corp., now the head of Bloomberg Sports, whose parent company, Bloomberg LP, also owns this magazine), Sportvision went on to build the technology that gave fans the yellow line that marks first downs on football broadcasts and the data boxes that shadow cars during Nascar races.
In 2001, the company worked with ESPN to create "K-Zone," which showed Sunday Night Baseball viewers the location of pitches as they crossed home plate. Creating K-Zone required a batch of data which became useful on its own. After several years of negotiations with Major League Baseball, Sportvision created an upgrade of K-Zone called Pitchf/x that uses a pair of cameras set along the baselines to capture every pitch roughly 40 times on its path between the mound and the plate. In 2007, MLB installed Pitchf/x throughout the league at a cost of roughly $2 million.
The point, at first, was to evaluate home-plate umpires' calls. The league wanted "a thoughtful, objective way that would be the same in every ballpark," says Bob Bowman, CEO of Major League Baseball Advanced Media.
Under a co-licensing agreement, Sportvision went on to sell Pitchf/x to broadcasters while the major league's Advanced Media wing (known as BAM) used it in its online game-tracking service called Gameday. Fans loved it. BAM also provides the raw data for free to all 30 teams, who use it to evaluate pitchers and hitters. "That was a tremendous amount of data," says Mike Chernoff, the 29-year-old assistant general manager of the Cleveland Indians. Digesting it, he says, was a major undertaking. "We continue to find useful pieces," he says. The data gave front offices a taste for motion capture.
Sportvision began cooking up broader applications. Hank Adams, who replaced Squadron as CEO, recalls a 2009 meeting with the Chicago White Sox. "We had shown them data capture on one play, a steal, and they kept asking us questions," he says. "'Do you get the initial lead? Do you get the secondary lead? Do you get the windup time, the pitch time, the pop time, the time it takes to throw down to second base?'" The answer in every case was yes. "You could see it on their faces," says Adams, "'Oh my God, what are we going to do with all this data?' "
The widest frontier is in fielding. No one knows this better than John Dewan, the owner of Baseball Info Solutions. He has spent decades building the current standard in defensive metrics. Dewan estimates that game analysts are collectively about 90 percent along the way to the creation of a complete picture of hitters, and close to 85 percent with pitchers.
With fielders, he says, they started with a severe handicap. For over 100 years, scorekeepers have described a player's work in the field based only on what he does after he gets to the ball. If a shortstop gathers in a grounder and throws it to first in time to make the out, he is credited with an assist. If he bobbles it and can't make the throw in time, he gets an error. According to Dewan, this basic accounting captures only about "5 percent of the information of what a player is all about."
Dewan's life story is largely the story of trying to close this gap. The 56-year-old Chicago native is one of baseball's Moneyball revolutionaries—the statistical analysts chronicled in Michael Lewis's bestselling 2003 book. Before getting into baseball, Dewan spent 10 years as an insurance actuary. In 1982, he came across a copy of The Bill James Baseball Abstract. "I was absolutely mesmerized," Dewan says. "He was doing with baseball data what I was doing with insurance." James, the Moses of Moneyball, is now a senior adviser to the Red Sox.
In 1984, frustrated at the lack of information in the standard box score, James enlisted his Abstract readers in Project Scoresheet. The idea was to standardize and collect the work of amateur scorekeepers in order to create a more complete database of statistics. Dewan signed on early. He wrote a program to log the data and soon became the project's director. In 1985, he decided to try to make a living out of it, investing $30,000 to found STATS Inc. He ran the company from a bedroom of his Chicago home with his wife, Sue, and co-founder, Dick Cramer.
Initially, Dewan, James, and the rest of the so-called sabermetricians (after the Society for American Baseball Research), focused on gathering and analyzing data for a familiar set of events: balls, strikes, hits, walks, stolen bases, home runs, and so on. Moneyball is essentially the story of how the values placed on these events were upended when they began to look at the data carefully. It turned out that, when it comes to producing runs, and therefore wins, batting average and stolen bases had long been overvalued, while the ability to draw walks had been overlooked. Oakland A's General Manager Billy Beane listened to the analysts before other GMs took them seriously and used their insight to build a team of slow-footed, patient hitters who made the playoffs four years running from 2000 to 2003.
Now, Bill James acolytes occupy most major league front offices, and have built an online industry around baseball statistics. The approach's popularity has also eroded its effectiveness. "The statistical analysis game has grown on what I would call stats of outcomes," says Vince Gennaro, the author of Diamond Dollars: The Economics of Winning in Baseball, meaning the Moneyball heroes were simply rethinking the traditional numbers. "We were looking at what happens at the batter-pitcher match-up ... and then analyzing that to death." Now the game is moving, he says, toward looking at "attributes and processes," at not just what happened, but how.
At STATS, Dewan helped to pioneer this transition. Looking for a better way to assess defense, he decided to divide the field into hundreds of zones and then count how often fielders were able to make outs on balls hit into those zones. He then measured each fielder against the average: If a center fielder, for instance, made a catch on a ball that 70 percent of major leaguers would also catch, he was credited with .3 toward what Dewan dubbed his Ultimate Zone Rating. In 2000, News Corp. (NWS) bought STATS and the Ultimate Zone Rating system with it. Dewan left and founded Baseball Info Solutions (BIS) two years later. There he refined his system, increasing the number of zones to more than 3,000 and converting the zone rating into a number of runs saved by a fielder's play. In 2010, little-known Daric Barton of Beane's Oakland A's led all first basemen by (theoretically at least) saving 20 runs.
At BIS, compiling these numbers is a laborious task. The company employs roughly 20 "video scouts" at its offices in Allentown, Pa. They watch video of every game at least three times and tag every batted ball with a direction (from 135 to 225 degrees), distance (0 to about 400 feet, depending on the size of the park), pace (hard, medium, or soft), and type (grounder, fly ball, line drive, or "fliner"). The results are information that major league teams pay to see, though Dewan won't say how much. The BIS data, however, is limited. It has nothing to say, for example, about where a fielder was standing when the ball was hit. And it is liable to human error, with a margin of 15 to 20 feet on some plays. Dewan estimates he's now only 60 percent of the way to fully measuring a fielder's ability.
This is where Sportvision enters. Fieldf/x essentially automates and massively expands the work of Dewan's video scouts. Each game at AT&T Park last season produced files of about 2.5 million results, or 2 terabytes' worth of data. At that rate, eight baseball games would fill the memory bank of IBM's (IBM) Watson supercomputer. "It's almost overwhelming how much data we're creating," says Ryan Zander, the company's general manager of baseball products. Much of it is extraneous: 20 records per second of players warming up between innings or milling around between pitches. Zander says Fieldf/x is accurate to within a foot. At AT&T, it would sometimes grab hold of seagulls and cotton candy vendors who came into the frame and begin tracking their movements. The slice of data that teams are most likely to want amounts to about one million results per game.
Via major league's BAM, front offices are likely to get just that, and without paying a cent. In 2000, every club agreed to kick in $1 million a year for the first four years of running BAM, with the expectation that the project would eventually cost $120 million. The idea was to manage a league-wide website and figure out how to make money from baseball online. BAM became a moneymaker ahead of schedule in 2003. MLB.com now gets 50 million to 60 million unique visitors per month during the season. Its mobile app, At Bat 2010, was the top grosser in Apple's online store last year. BAM's annual revenue is now nearly $500 million, and teams have already been paid dividends totaling three times their initial investment. The data is gravy.
For the first installation in San Francisco last year, Sportvision, whose labs are in nearby Mountain View, worked directly with the Giants, under BAM's oversight, to get permission to test the technology. In return, the Giants received exclusive access to the data. BAM arranged this year's installations—already in place at Yankee Stadium and in San Diego, and under way in Kansas City and Tampa. "Our plan is ultimately to put it in every park," says CEO Bowman. But first, he wants the commissioner and clubs to see how it works. (Bowman plans to install a competing technology, called PlayItOver, in one park "to keep everybody honest.") If teams like what they see, BAM will likely help shoulder the installation cost—about "six figures per," according to Sportvision CEO Hank Adams—and then try to figure out how to make revenue from the technology. "We think it's a project that may deliver value to our fans through the myriad of devices they use," says Bowman. Sportvision, meanwhile, will go to work on broadcast products, such as overlaying the field with concentric circles illustrating a player's range.
Clubs will get a mountain of data and race to make use of it. The coming deluge reminds the Red Sox's Tippett of Project Scoresheet in 1984. Like Dewan, he cut his teeth on that new batch of data. "I happened to be at the right age at the right time and got in on that early, and that's why I have a job with the Red Sox today," the 53-year-old says in the hallway, amid the crowd at the Sloan conference. Fieldf/x, he says, has the potential not only to be a big step forward in how clubs evaluate players but also in opening the door to "anybody who wants to jump in and be among the leaders in figuring out what we can do with this stuff."
There's no shortage of people waiting to take him up on the challenge, as suggested by the 1,500 who showed up at the Sloan conference. Moneyball made analytics sort of sexy—Brad Pitt will play Billy Beane in the movie version to be released this fall—and the market for number crunchers is glutted with talent. The Internet is teeming with bloggers hoping to get noticed for their work at sites such as Baseball Prospectus and the Hardball Times. The Indians' Chernoff rattles off a list of four names the team has recruited from their ranks and says his team's analytics budget continues to grow. Much of the insight gained so far from the Pitchf/x data came from such bloggers, who figured out how to gather the information from MLB's online Gameday service.
As a rule, baseball teams don't talk about their front-office spending, but according to sources familiar with their budgets, they now range from about $100,000 to half a million or more for analytics. The teams at the top end of that range treat work done in public forums as tryouts. At the bottom, they treat it as free labor. And where they get their stats analyzed will influence each team's opinion about whether or not they want the public to see the Fieldf/x data.
"You hear stories from front offices about how much they are relying on free work from the Internet," says Robert "Voros" McCracken, a onetime baseball analyst who famously gave away one of the game's great statistical epiphanies to a small online group in 1999 (the only reliable measures for a pitcher are the outcomes that don't involve fielders: walks, strikeouts, and home runs). Teams, according to Voros, are asking, "'Why do we need to hire this guy? He's giving us what we want for free.'" Voros's work earned him a job devising systems to evaluate college players for the Boston Red Sox for three years. The pay was peanuts, less than $30,000 per year. Sportvision's Adams admits the passive approach to freelancers, is "probably not the best business model or even the most consumer-friendly, frankly." For now, however, the company is focused on its broadcast products and its partnership with the league. The fate of the Fieldf/x data is unclear. "I think we would make it public," says Bowman. "We live in an age where, if you've captured it somewhere, somebody's going to find it, so you're better off making it transparent." But that decision will ultimately come from the commissioner and teams.
If Tippett has any say in the matter, clubs will keep the data to themselves. "I want this to be adopted by Major League Baseball, made available to all the clubs, but kept within the industry," he says. Tippett admits that broad access to Pitchf/x data produced useful ideas and that as a fan he values open access, but he knows who signs his checks. "We don't get paid to advance the state of the art in analysis," he says, "We get paid to put a winning team on the field." If he could, of course, Tippett would keep the information exclusive to the Red Sox, but Fieldf/x only becomes truly useful as league-wide data set, so teams will have to share.
For the heavily analytical teams, the competition is not with other teams, but with bloggers. Those clubs, says Gennaro, will like their chances in a race restricted to a field of 30. "A team at the bottom of the food chain," he says "would benefit grossly from [Fieldf/x] being in the blogosphere." If it gets there, teams will get information potentially worth millions for free.
Tom Tango is the nom de plume of one of the most respected minds in sabermetrics, although he writes and posts anonymously. ("Tommy is pretty much the best out there," says Voros.) Tango, who consults for the Toronto Blue Jays and Seattle Mariners, estimates that getting a jump on the new fielding data could be worth an extra win or two per season for a major league team. "Teams will spend a lot of money to purchase an extra win or two ($5 to $10 million)," he writes by e-mail. That's for players. Spending on analytics has not caught up. While he is fiercely protective of his identity, Tango is an advocate for public access to Fieldf/x. "MLB can hire the 30 best analysts," he writes, "but the next 3,000 best analysts would be able to do better and faster work as a community than any single analyst can do on his own."
Companies like Baseball Info Solutions and Bloomberg Sports are in wait-and-see mode. "I would love to get my hands on it and use it in our stuff," says Dewan, "but how that's going to work in the long run, I don't know." Bloomberg already gets Pitchf/x data as part of a larger partnership with BAM and includes it in the analytics package it sells to 18 major league teams. "We would certainly be interested," Squadron says of Fieldf/x. "We might be best positioned to make sense of that volume of data."
Last summer, Sportvision invited six analysts to see what they could find in 13 games' worth of Fieldf/x data. Even with that small slice, they were able to begin unwinding the difference between the results of a play and its quality. Soon teams will be able to distinguish between plays made because a fielder was already standing in the right place and those made because of exceptional quickness to the ball. There will be new metrics—such as degree-of-difficulty ratings—and more precise coaching. "There are going to be pregame meetings on where the shortstop is going to play exactly for each of the hitters," says Gennaro.
One way or another, the exhaustive data will also seep into the game's vernacular. "We see center fielders make catches or shortstops make diving plays and we go, 'great range, great jump,'" says Bowman. "That's what it looks like to our naked eye." Now fans will begin to see what they've been missing. Range will be described in numbers and not just adjectives. "It's going to make the game more scientific," says Gennaro. More science, Bowman assures, doesn't have to mean less poetry. "The facts will not end the discussion, they will only enlarge it."