Friday, July 28, 2006

Bowl Games and the Spread

Look at my greasy hair.

As promised last week for all you degenerates, here are the results for how certain team characteristics predict bowl winners against the spread. If you don't feel like reading the last post, here's a quick refresher. The sample size is every bowl game from the past 6 seasons (2000-2005). I determined how often the team with the better overall record, point differential, conference record, offense (rated by points scored) and defense (rated by points allowed) won the bowl game against the spread. In the interest of staving off premature blindness and causing confusion, I will eschew year by year results and just hit you with the cumulative data.

Better Record: 62-62-3 .500
Better Point Differential: 69-87-3 .443
Better Conference Record: 47-63-3 .429
Better Offense: 64-93-3 .409
Better Defense: 75-81-3 .481

Nothing. Zilch. Nada. Neither characteristic predicts bowl spread coverers (for lack of a better term) with any accuracy. Looks like Vegas wins again. But wait. When you dig a little deeper and separate each category into underdogs and favorites something does show up.

Better Record
Dog: 28-40 .412
Favorite: 34-22-3 .602

Better Point Differential
Dog: 21-64 .247
Favorite: 48-23-3 .669

Better Conference Record
Dog: 24-36 .400
Favorite: 23-27-3 .462

Better Offense
Dog: 24-62 .279
Favorite: 40-31-3 .561

Better Defense
Dog: 30-55 .353
Favorite: 45-26-3 .628

For the most part, in every category except better conference record, underdogs perform very poorly, but favorites perform quite well. Why is this so? I can't really come up with any ideas off the cuff, but the phenomenon does seem to exist. So if hypothetically, the Sun Bowl matches up two teams, and the team with the better record (or point differential, or offense, or defense) is also the favorite, they might be a good play. Of course, past performance is no guarantee of future performance, but if you do use this knowledge to win a substantial, any donations will be accepted.


Sam said...

My gut reaction to the results you posted about why having a better record, especially a better conference record isn't a very solid metric for predicting the outcome of a bowl game is that any team's opponent in a given bowl game is not representative of the type of opponent that team would face during the regular season. For example, last year's Cotton Bowl put Alabama up against a Texas Tech team with a more "high-powered" offense unlike any that 'bama had faced all year. So if you believe that SEC football is inherently different from Pac-10, Big XII, etc football then it just makes sense. This is compounded by the fact that most team's non-conference schedules tend to be composed of weaker teams (1-AA schools, teams from the mid-majors, etc) than they would normally face in a bowl situation.

A suggestion for future study would be to include combinations of the categories. For example, if the favorite has better defence and better point differential does that increase the likleyhood that they will cover the spread.

Also, I'm a little unclear on what your data actually is saying. When you write:
Better Record
Dog: 28-40 .412
Favorite: 34-22-3 .602

Does that mean that in 68 games the underdog had a better record than the favorite and that 28 of those times the underdog won outright, or just lost close enough to win with the spread?

matt said...

Thanks for your input Sam. The data you reference means that in a bowl game where the team with a better record is the betting underdog, they cover the spread about 41% of the time. However, when the team with the better record is the betting favorite, they cover the spread about 60% of the time. Hope that clears things up.

Sam said...

Thanks for clearing that up. I'm curious as to how you compile these statistics. Do you use a spreadsheet, a database, pen and paper(!), something else? Would you be willing to share the raw data with your readers?

matt said...

For most its good old fashioned pen and paper. For the studies that involve r squared and correlations I use Microsoft Excel. I would be more than happy to share some of my data.

Anonymous said...

You also have to figure in that teams have a month to prepare for an opponent in a bowl game as opposed to one week during the season.