Tuesday, March 21, 2006

The Secret to Not Losing Close Games

March Madness is my favorite time of year. However, I picked Kansas to go to the Final 4, so my bracket is now dust in the wind. So I'd like to switch gears to college football. In particular, I'd like to tackle a topic that is always under debate: Why do some teams lose close games? Is it a lack of senior leadership, good coaching, or something intangible? Let's take a look.

If a team wins close games one year, it stands to reason that the same team would win close games the next year. I conducted a simple regression analysis to determine if this is true. I calculated each Division I-A team's record in close games (games decided by 8 points or less) in 2004. Then I calculated each Division I-A team's record in close games in 2005. I used the 2004 winning percentage as the independent variable and the 2005 winning percentage as the dependent variable. In other words, how does a team's performance in winning close games in 2004 predict their performance in close games the following year. The resulting data may come as a surprise to you. A team's record in close games in 2004, was not a reliable predictor of their record in close games in 2005. In fact it was a terrible predictor. The R squared value is .0019. To those of you who are not familiar with R squared, allow me to explain it. R squared is the percentage of variation in the dependent variable (winning percentage in close games in 2005) that is explained by the independent variable (winning percentage in close games in 2004). Only slightly more than 0% (.19%) of the variation in winning percentage in close games in 2005 is explained by a teams winning percentage in close games in 2004.

So if a team's record in close games one season is not predictive of its record in close games the following season, is it's record in non-close games in one season predictive of its record in non-close games the following season? I conducted another regression analysis, this time using every Division I-A team's record in non-close games (games decided by 9 points or more) in 2004 as the independent variable and every Division I-A's team record in non-close games in 2005 as the dependent variable. So is a team's record in non-close games predictive of their record in non-close games the next year? The answer is sort of. The R squared value for this regression is .338. 33.8% of the variation in a team's record in non-close games in 2005 is explained by their record in non-close games in 2004. While not anywhere near a perfect linear model, a team's record in non-close games in 2004 was a much better predictor of their record in non-close games in 2005 than a team's record in close games in 2004 was a predictor of their record in close games in 2005.

Earlier I promised the secret to not losing close games. And I'm about to deliever. The secret to not losing close games, is to not play them. Close games are analagous to a coin flip. Just because you have flipped several heads in a row, you can be sure you will eventually flip a tail, and perhaps a sequence of tails. Just because a team has won several close games doesn't ensure that they will continue to do so.

1 comment:

____________ said...

I think I just earned a MBA from Carolina after reading and learning all of that! J/K good blog Matt