Total Pageviews

Wednesday, June 19, 2013

Digitizing Sports to Slake a Thirst for Stats

When Joe Inzerillo started taping baseball games for the Chicago White Sox a quarter-century ago, about all he needed was a video camera, some blank tapes and a bunch of VCRs. He would edit the tapes and send copies to the video staff members for the opposing team â€" job done.

Mr. Inzerillo and baseball scouting have come a long way since then. Now, as chief technology officer at MLB Advanced Media, he helps maintain a vast, fast-growing digital record of the national pastime. Each season, MLB Advanced Media tracks about 17,400 games. Every pitch, hit, out and more can be sorted, analyzed and shared instantaneously with fans and teams.

The undertaking is immense. Consider pitching. MLB Advanced Media uses Pitchf/x, a tracking technology developed by Sportvision, to capture three dozen data points on every pitch of every game, which are then used to calculate and sort pitches according to speed, trajectory, location and type.

With roughly 290 pitches thrown each game, more than 25 millio data points are created each season, information that scouts, coaches, general managers and now fans sift to determine which pitchers have potential and which ones are veering toward injury or irrelevance.

“When I was with the White Sox, we were one of the first teams to match pitch speed to video,” Mr. Inzerillo said in MLB Advanced Media’s office in Manhattan. “Now, we have all these other statistics like arm angles, slots and breaks. What we did 15 years ago for coaches has moved into the psychology of the average fan.”

Spurred by increasingly sophisticated computers, cameras and Internet connections, statistics have seeped into nearly every sport, from auto racing to basketball to tennis. The reliance on data has created an entire job category for programmers, statisticians and videographers that barely existed a decade ago.

Michael Lewis’s book “Moneyball,” which illustrated how the Oakland Athletics used statistics to produce winning teams with lean payrolls, ! is now widely imitated, and it is no longer unusual for a brainy blogger to be hired to scout players.

Leagues and sports associations are also licensing team and player data to companies like Bloomberg Sports, part of Bloomberg L.P., which are repackaging the information into products that predict plays and games.

“It’s not about collecting data, but crunching it,” said Bill Squadron, the president of Bloomberg Sports. “We can look at team performance with greater insight and forecast match results.”

To feed the growing curiosity about the game within the game, leagues have recognized that sharing statistics can enhance fans’ enjoyment while helping sell premium products, not unlike the way the Grateful Dead let fans make bootleg tapes of concerts to popularize the group.

“If you’re a fan and you have an idea about what the substitutions should be for the Miami Heat, you literally had nowhere to turn to support your hypothesis,” said Steve Hellmuth, executive vice resident for technology and operations at NBA Entertainment, which this year opened its entire statistical archive to fans. “Now, if you’re a blogger and you want to explore it, you can find that answer.”

For decades, the majority of statistics in basketball, baseball and other sports were concerned with offensive skills because things like hits, home runs and three-point shots were the easiest to quantify.

It has been harder to track and rate defensive skills. In baseball, teams have long relied on statistics like errors, putouts and passed balls that involve players who are in position to make plays. Evaluating an infielder’s range, however, is more difficult because it requires measuring situations like balls that fell for hits instead of being caught.

To do that, software has been created to track players as they chase balls in a multitude of directions across a vast field, not simply tracking a ball in the predictable path from the pitcher’s mound to home plate, the w! ay Pitchf! /x does. To gather enough data to evaluate a fielder, cameras shoot many more pictures of each play to create a database deep enough to be accurate.

MLB Advanced Media has been testing a system since 2009, when it began installing Fieldf/x, another Sportvision product, in ballparks in Boston, Kansas City, Milwaukee, San Francisco and St. Petersburg, Fla.

Every game, four high-resolution cameras snap thousands of pictures that are analyzed to track balls in play as well as the speed and path fielders take to reach them. While evaluating a fielder’s effectiveness is useful, the real utility to teams is the software that can estimate how efficiently another player might have reached a similar fly ball.

“The baseball guys have gone crazy with this data,” said Hank Adams, the chief of Sportvision, which has developed video-based tracking technology for Nascar and other sports. “We’ve done so much around understanding pitching and hitting, but when it gets to defense, 150 years of baseall have been played and we’re still reacting to intuition.”

Creating a system has taken time. Algorithms have been thrown off by things like sunlight streaming across the field and sea gulls, which can be confused with players. The chaos of a rundown can make tracking difficult.

The digitizing of defensive statistics has created an explosion in the amount of data being collected. Fieldf/x creates about two terabytes of data every baseball game. And because every game is recorded on multiple feeds, MLB Advanced Media generates more than 1.5 petabytes of live and on-demand video each season (a petabyte is 1,000 terabytes). In all, the subsidiary of Major League Baseball stores about six petabytes of content per season, one reason it is opening a new data center in Omaha.

“We’re marrying data with video so you get the stats with the live video in real time,” said Bob Bowman, chief executive of MLB Advanced Media. “We’re trying to give to fans what every manager knows in ! his gut.â! €

Stats L.L.C., which owns a technology called SportsVU, is doing something similar in basketball. Half of the N.B.A.’s 30 teams have six palm-size video cameras installed in the rafters of their arenas. The cameras track the movement of the players and the ball at 25 frames a second. Algorithms produce a time stamp, identify the players and the ball, and create x, y and z coordinates for their positions on the court.

This allows teams, which pay up to $100,000 a year for the service, to track when a rebound was grabbed, who grabbed it and how many players were nearby. “Essentially, we’re adding the context around the play,” said Brian Kopp, vice president for strategy and development at Stats.

By calculating speed, distance and the separation between players, SportsVU can tabulate the number of dribbles, touches and passes to determine a player’s offensive efficiency. It can also explore how a player’s presence on defense can alter an offense. This year, two Harvard researches wrote a paper titled “The Dwight Effect,” which concluded that Dwight Howard of the Los Angeles Lakers was the best at preventing close-range, high-percentage shots from being taken.

The paper provides a glimpse of the many questions about defenses that teams and fans will be able to answer in the coming years, Mr. Kopp said. “If we come up with a new definition, we don’t have to find new data, we can just re-run all the historical data.”