UCAFANS.COM | GO UCA BEARS!

A Computer Cracks the Cooperstown Code
wsjonline.com and 7/27/09 paper edition in "The Count'

Now that Rickey Henderson and Jim Rice have been inducted into the National Baseball Hall of Fame, fans the country over can return to the only thing they like as much as watching games: Arguing over who should join them in Cooperstown. Sadly, machine learning may have just taken much of the steam out of such arguments.

According to research by Lloyd Smith, a professor of computer science at Missouri State University, and James Downey, an assistant professor of management information systems at the University of Central Arkansas, whether a player is elected to the Hall of Fame by veteran baseball writers is an entirely predictable outcome based on a few statistics.

Using a radial bias function network, a sort of neural net, Dr. Smith and Dr. Downey were able to identify statistical commonalities among Hall of Famers. As it turns out, hits, home runs and on-base plus slugging percentages are what count for hitters, while wins, saves, earned run average and winning percentage are what count for pitchers. All-Star Game appearances count for both, being especially valuable for hitters as they serve as a useful proxy for position.

Chance of Election
Using a computer model, a player's likelihood of being elected to the Hall of Fame can be predicted.

PLAYER HALL OF FAME PROBABILITY
Vladimir Guerrero 88.8%
Trevor Hoffman 86%
Chipper Jones 81.7%
Edgar Martinez 53.5%
Mike Mussina 47.8%
Curt Schilling 44%
Todd Helton 22.3%
Jeff Kent 14.2%

Armed with these numbers, the researchers are able to assign quite accurate odds that a given player will be elected, even without taking intangible factors into account at all. Of the 1,592 players considered by their study -- anyone who retired between 1950 and 2002 and met several other criteria such as having played 10 years in the majors -- the model was able to accurately identify whether they had been elected 98.7% of the time. Before the 2009 class was voted in, the model gave Mr. Henderson a 97.2% chance of being elected. On the flip side, Mo Vaughn had a 1.3% chance.

There is, though, good news for fans in the upcoming elections. Just as the public does, the model sees such borderline cases as Edgar Martinez and Mike Mussina as having essentially 50/50 chances of making Cooperstown. So let barroom debates commence.

—Tim Marchman
Printed in The Wall Street Journal, page B6