Sign in / Join

Luck vs. Skill

In his recent Moneyball post, Skip alluded to the “Fire Joe Morgan” blog, website dedicated to pointing out silly commentary from sports journalists. In homage to this sadly departed blog, I point out this commentary from Don Banks at Sports Illustrated today.

No way, no how would the Buffalo Bills have ever won this game last season. But because of last season, when the heartbreaking, narrow loss was the Bills' specialty, Buffalo learned the kind of resiliency in the face of adversity that proved to be essential to its corner-turning 34-31 comeback win over New England Sunday at Ralph Wilson Stadium.
Essentially, Banks is saying that experience at playing close games leads to an improved ability to win close games in the future. This type of statement is easy to check empirically. Obviously, a serious test of the proposition requires more data collection that a simple blog post deserves, but a roughly correct way to test this would be to check the correlation between number of close games played and win percentage in those close games. I tested the correlation between OT games played and win percentage in OT games for the 2010 season. The correlation is positive but barely above 0.1 and extremely far from statistical significance. In MLB in 2011, the correlation between the number of 1-run games played and win percentage in those games is actually negative, -0.20, but again not statistically significant.

 

Much more likely is the proposition that close games are basically a crap shoot. Teams that lose a bunch of close games are more likely unlucky than unskilled. One referee's call, funny bounce, or blown coverage, and not the relative talents of Tom Brady vs. Ryan Fitzpatrick, often means the difference between winning and losing. A poor record in close games results in a bad overall win-loss record, but it otherwise masks the true talent level of a team. In future games, it will appear as if the team has improved its ability to win close games when in fact all that is happening is a reversion to the mean (of roughly 50%) in close game winning percentage.