Thursday, March 16, 2006

RPI Project revisited

Couldn't post last night this stuff, but I wanted to get it down before the games started.

Last year I did a long, drawn out post series on using the RPI ratings to predict the NCAA games. It didn't work all that well at predicting winners, and in fact it may have proved that the RPI isn't the best indicator. And that's really my point. The NCAA uses the RPI in terms of at-large selections and seeding. I always wonder about using an instrument like the RPI. If it is a proper indicator of relative strength between teams, why not solely use it (it is bias-free). But if it "doesn't tell the whole story", then why use it at all? If it's a good tool, then use it. If it isn't, don't.

Anyway, the series last year was a little too focused on rankings and too little on matchups, and I'm tweaking it some.

The basic idea was to rank the 64/5 teams based on the RPI ratings. Then look to see if each team is "over-seeded", meaning their seed in the actual tournament is higher than it would've been if only the RPI ratings were used, or "under-seeded", meaning their seed in the actual tournament is lower than it would've been if only the RPI ratings were used. Then look to matchups. I ran the system twice, once by slotting teams based on their RPI overall (meaning if a team is ranked 12th in the RPI, they're a 3 seed. and if they're ranked 80th in the RPI, they'd be a hypothetical 20th seed) and once by slotting teams based on their RPI only compared to the other 64 teams (meaning nobody can be worse than a 16th seed). In the first method, naturally, about half the teams were either appropriately seeded or underseeded and the other half overseeded by some amount. In the first method, it was evenly split three ways - 1/3 correct, 1/3 underseeded, and 1/3 overseeded. Then I looked at both to get an idea of who fits where.

Then I looked to matchups.

First round matchups between teams appropriately seeded (RPI tells us little):
All of the 1/16 games.
Each of the 2/15 games except UCLA-Belmont
Washington-Utah State

First round matchups between an appropriately seeded favorite and an overseeded underdog (RPI suggests the favorite got a "good draw" in having to play a weaker than seed opponent):
Gonzaga-Xavier
Nevada-Montana
Wichita State-Seton Hall

First round matchups between an underseeded favorite and an underseeded underdog (RPI tells us little, but in effect both teams got a "bad draw" in having to play a tougher than seed opponent):
George Washington - UNC-Wilmington
Illinois - Air Force (yes, Air Force was actually underseeded)
Michigan State - George Mason
UNC - Murray State
Arizona - Wisconsin

First round matchups between an overseeded favorite and an overseeded underdog (RPI tells us little, but in effect both teams got a "good draw" in having to play a weaker than seed opponent):
Arkansas - Bucknell
Indiana - San Diego State
Kentucky - UAB
Boston College - Pacific

First round matchups between an overseeded favorite and an appropriately seeded underdog (RPI suggests the underdog got a "good draw"):
UCLA - Belmont
Iowa - Northwestern State
Florida - South Alabama

First round matchups between an underseeded favorite and an appropriately seeded underdog (RPI suggests the underdog got a "bad draw"):
LSU - Iona
Pittsburgh - Kent State

And now the most interesting, mixed-up ones...

First round matchups between an overseeded favorite and an underseeded underdog (RPI may suggest that the favorite got a "bad draw" and the underdog got a "good draw", also, suggests the teams are closer than the seeds suggest):
Syracuse - Texas A&M
West Virginia - Southern Illinois
Cal - NC State
Kansas - Bradley
Georgetown - Northern Iowa

First round matchups between an underseeded favorite and an overseeded underdog (RPI suggests that the favorite got a "good draw" even though underseeded, the underdog got a "bad draw" even though overseeded, and suggests the teams are further apart than the seeds suggest):
Marquette - Alabama
Oklahoma - UW-Milwaukee

I know it's late in terms of affecting your own brackets, and I also know that this "system" isn't really a system, since last year it didn't particularly work perfectly (I think it did pick Vermont and Pacific to have upsets though). Use at own risk. I just thought it was interesting. I also think it's interesting that Southern Illinois and Wisconsin were underseeded both this year and last (and both won their opening round games last year) while Iowa and West Virginia were overseeded both this year and last (Iowa lost in the opener, while West Virginia had a long run in the tournament).

To do things more similarly to last year, the most underseeded teams in the tournament are Bradley, Wisconsin, Northern Iowa, George Mason, Oklahoma, and Air Force. The most overseeded teams in the tournament are Syracuse, California, Xavier, Kansas, Boston College, Pacific, Montana, Florida, Georgetown, Bucknell, Arkansas and San Diego State.

Just some interesting stuff to follow as the tournament progresses, I suppose.

Personally, I'm pulling for Villanova to win it all. I did two brackets this year, identical except for the national semifinal on the right side (in one I pick 'Nova and the other I pick UConn, and the winner of that game wins it all). Final Four: Villanova, UConn, Texas and Gonzaga.

0 comments: