Why A 14-17 Record Is Not Cause For Alarm

Today the Sports page of the WSJ had some nice statistics on spring training records, and why the Dodgers' 14-17 record is not a meaningful predictor of 2009 season performance (no link):

At Spring Training this year, the typically horrendous Kansas City Royals have a 16-11 record -- one of the best marks in the American League. Does this mean the Royals will defy all odds this year and make an heroic run at the World Series?

The baseball establishment would tell you not to bet on it. Spring Training, they say, is nothing more than a place to evaluate players, regain your timing and work off the lovehandles. Half the players on the field are nobodies with jersey numbers better suited to a football team. So is it possible they've missed something?

James Cochran, a statistician at Louisiana Tech, calculates that over the last six years the average correlation coefficient between a team's winning percentage in the spring and its winning percentage in the real season is .18.

Simply put, that number is a lot closer to 0.0, which suggests there's no correlation at all, than to 1.0, which it would be if the spring standings and final season standings were always identical. In 2005, the number was slightly negative, meaning teams that did poorly in the spring were actually more likely to do well when it counted.

Not so fast, though: Last season, Tampa Bay went 18-8 in the spring before shocking everyone by making the World Series. The correlation coefficient for that season, overall, was a respectable .32, which wasn't even as high as the .43 recorded in 2003. By way of comparison: a recent College Board study of students found a .53 correlation between their SAT scores and their freshman-year grade point averages.