Friday, March 12, 2010

Similarity Predictions

Since there were no tickets punched last night, I spent part of the evening developing an Excel workbook that makes game predictions based on how a team has performed in games against similar opponents.  I’ll use today’s game between Wisconsin and Illinois as an example.

First, I found the ten Wisconsin opponents whose offenses were most similar to Illinois, using the same method I’ve been using in the Tickets Punched series.  I then calculated Wisconsin’s individual game adjusted defensive efficiency against each of those opponents, using this equation (derived from the standard adjusted efficiency formula):

Wisc Game Adj Def Eff = [Opp Game Raw Off Eff] * [NCAA Ave Eff] / [Opp Season Adj Off Eff]

[NOTE: I actually treated HFA the same way Pomeroy does, but I didn’t want to make the equation look even messier]

Finally, I took a weighted average of those efficiencies, with the weight being the similarity score.  This gives me an adjusted defensive efficiency for Wisconsin that is specifically for games against offenses similar to Illinois.

I repeat the above process for Wisconsin’s offense, and Illinois's offense and defense, and the results all go into these pretty charts (where, as usual, green=good and red=bad):


As you can see, Illinois’s offense has played better against Wisconsin-ish defenses than they have overall this season, and their defense is roughly the same.  Conversely, Wisconsin has done worse against Illinois-ish offenses and Illinois-ish defenses.  To find out what this means for their chances of beating the Illini, I compared these similarity-derived efficiencies to their regular Pomeroy numbers (again using the standard efficiency equation – Offense*Defense/Average=Expected):


The similarity-derived numbers are way higher on Illinois than the standard Pomeroy ratings are.  Now, the Badgers are still favored, it looks like nearly a tossup, instead of an easy win.  For a lot of games, the two predictions are nearly identical, but there are other cases like Wisconsin-Illinois, where there are large differences.  Here are the other major conference games from today where there was a sizable difference.  Some have already taken place, and for those, I’ll note the result and whether the similarity-derived prediction seemed better or worse than the standard Pomeroy (I’ll update the rest tomorrow).
[NOTE: Updated.  I also removed games where the difference between the 2 predictions was less than 10% (there was one “win” for each system).]


So, there you have it.  I’ve automated all this so that all I have to do is type in the names for two teams, and the “vs. Comps Of” and “Efficiency Prediction” charts are created nearly instantly, so if the predictions for tonight turn out reasonable, I’ll hopefully find time to use this method to look at more games going forward.

Happy Championship Week!

No comments:

Post a Comment