## Wednesday, January 10, 2007

### Close Game Variance Ratio

One of the resources I use is something I call the "Close Game Variance Ratio" or CGVR. It's an idea I picked up from Phil Steele's preseason magazines, and so far, it's been a pretty strong predictive tool.

The way to find a team's CGVR is to take the difference of games won by 8 points or less from the games lost by 8 points or less. For instance, Tennessee in 2006 beat Air Force by 1, Alabama by 3, South Carolina by 7, and Kentucky by 5, giving Tennessee 4 close wins. UT lost to Florida by 1 and LSU by 4, giving UT 2 close losses. Two from four is +2. So, Tennessee's CGVR is +2.

The idea behind the theory is that teams with higher CGVRs were extremely lucky, or, to sound a little more serious, had positive variance. I believe variance tends to balance itself out, usually from year to year, but sometimes longer. Conversely, teams with lower, negative CGVRs will usually get no worse in record. A 4-7 team with a -3 CGVR will probably improve their record the next season, or at the very least, stay at the same level.

The evidence? I still consider my research to be done over a small sample size, but it's looking pretty good so far. I calculated every team's CGVR for every season from 2001-2005, and so far, the outcomes are pretty telling. First, we'll look at teams with high CGVRs:

+4 and up - Has occurred 34 times, with 91% of the teams having the same or worse record the next season
+3 - Has occurred 24 times, with 75% of the teams having the same or worse record
+2 - Has occurred 49 times, with 71% of the teams having the same or worse record

Solid evidence. Now, for negative CGVRs:

-4 and worse - Has occurred 22 times, with 86% of the teams having the same or better record
-3 - Has occurred 37 times, with 68% of the teams having the same or better record
-2 - Has occurred 50 times, with 71% of the teams having the same or better record

As far as I can tell, this theory can continually be a solid resource in preseason predictions. Obviously, preseason predictions are about as common as the sniffles, but it's still fun to be right. If you used this resource before the 2006 season, you could have predicted Arkansas's improvement (-4 in 2005), Alabama's and Penn State's record drop (both +2 in 2005), Oregon's troubles (+3), Arizona's 3 win jump (-4), and other teams' fortunes.

Later this week, I will outline which teams from 2007 fall into the important CGVR ranges.