Monday, March 12, 2012

St Louis bracket: log5 prediction

As I did for the Big East tournament in 2011, the NCAA tournament in 2010, and various times in the past, I will be running log5 analysis based on the Pomeroy ratings for at least as long as the Hoyas are in the field.

Ken Pomeroy has run the official log5 odds on his website. Rather than duplicate his, I'm trying something a little bit different.

One major problem with the Pomeroy ratings, especially early in the season, is that it includes all games. Some teams, particularly Wisconsin, are quite proficient at grinding inferior teams into the ground and much less so at doing the same against more closely-aligned competition. As a crude fix for that, I am this year instead preparing predictions based on teams' efficiency margins against opponents ranked in the top 100 of the Pomeroy ratings. Is this a perfect fix? No, absolutely not, but it does "fix" Wisconsin and some other teams.

Here are the percentage odds for advancing to the next round for each team:
Seed  Team           2nd Round  Sweet 16  Elite 8  Final 4
1     North Carolina    87.5      57.2     40.6     25.2
16    Lamar/Vermont     12.5       2.7      0.7      0.1
8     Creighton         60.3      26.4     16.0      8.2
9     Alabama           39.7      13.8      7.0      2.9

5     Temple            53.3      30.7     11.9      5.3
12    Cal/South Florida 46.7      25.4      9.1      3.8
4     Michigan          68.5      34.0     12.6      5.4
13    Ohio              31.5       9.9      2.2      0.6

6     San Diego St.     63.7      31.8     11.8      4.6
11    NC State          36.3      11.1      2.9      0.8
3     Georgetown        63.4      40.1     16.7      7.3
14    Belmont           36.6      17.0      5.0      1.6

7     Saint Mary's      62.4      26.0     15.2      7.1
10    Purdue            37.6      11.7      5.4      1.9
2     Kansas            88.6      59.9     42.5     25.3
15    Detroit           11.4       2.5      0.6      0.1

Note I've listed a couple teams in bold in the brackets. The Hoyas were bolded to highlight their odds. I also have not yet re-written my spreadsheet to account for the two play-in games, between Lamar and Vermont in the 16 seed game and Cal and USF in the 12 seed game. For simplicity's sake, I've assumed the superior team (Lamar and Cal, respectively) will win that game. The listed odds thus somewhat overstate the chance the winner of each play-in game continues on and somewhat understate the odds of all teams not involved in a play-in game.

The Problems Inherent in the Method

More importantly, judging teams based on results against only KenPom top-100 teams results in judging off smaller sample sizes. Most of the time this isn't an issue. For instance, Georgetown played 17 games against the KenPom top 100. Even a non-BCS team like Temple played 15 teams. Unfortunately, there are a couple exceptions. In the Hoyas' region, both Belmont and Ohio only played six games against the KenPom top 100. I prepared the above calculation using Belmont's actual results in those six games. The Bobcats, however, presented a trickier challenge.

Ohio played six games against the KenPom top 100: one game against #20 Louisville, one game against #74 Marshall, three games against #79 Akron, and one game against #95 Northern Iowa. They went 4-2, losing by five points at Louisville and losing once against Akron. Because they blew out Northern Illinois and Akron once, their average efficiency in those games was 109.2 on offense and 100.6 on defense. Based on those numbers, they're the third-best team in the region and roughly the eleventh-best in the field. Considering Ohio is the 72nd-ranked team in the overall Pomeroy ratings, I strongly doubt that accurately reflects how good the Bobcats are.

I therefore manually adjusted the Bobcats' ratings to something more along the lines of what I think they deserve based on where teams in their general vicinity in the Pomeroy ratings came out. I fully admit this adjustment is highly subjective and deeply unscientific, but I prefer it to breaking the system. Looking outside the Hoyas' region, Murray State poses the same problem, unless you believe they really are as good as Kentucky.

Thoughts Outside the St. Louis Region

Looking at the bracket as a whole, Kentucky is clearly the best team, though the margin is less overwhelming than it was before they lost the SEC championship game. There's then a second tier of roughly ten teams, including Kansas, Memphis, Michigan State, Missouri, New Mexico, North Carolina, Ohio State, Syracuse, and Wichita State. I've listed them in alphabetical order, but by the numbers, Wichita State is at the top of the tier, closer to Kentucky than they are to Kansas at the bottom of that tier. The Hoyas thus appear somewhat fortunate in that they avoided Kentucky and only have two of the second tier teams in their region.

You may have noticed that neither the Hoyas nor any other of the #3 seeds are on that list. All the #1 seeds are included, as are all the #2 seeds save Duke, but this year's #3 seeds are mostly unimpressive. Baylor is the best, Marquette second, and the Hoyas come up just barely ahead of Florida State. Kansas is closer to Kentucky, though, than they are to Baylor, let alone the Hoyas, and thus the St. Louis bracket looks the way it does.

For the curious, the teams with the best chances at coming out of the other regions are Kentucky (39%) followed by Wichita State (20%); toss-up between Syracuse (27.2%) and Ohio State (27.4%); and Missouri (26%) over Michigan State (17.2%) and New Mexico (16.9%).

Notes: log5 predictions are made using Pomeroy ratings based on games through Sunday, March 11.
No adjustment has been made for any home court advantage - all games are assumed neutral court.

Finally, a gentle reminder: efficiency ratings and the predictions derived therefrom are not destiny.


  1. Hey Tom,

    I edited your odds for Belmont beating the Hoyas from 11.5 to 36.6, based on what you had for Georgetown in that game.

    Hope that's correct.