Monday, March 8, 2010

A New Approach to Seeding

I don't usually spend much time writing about what seed teams deserve. My focus tends to be on more of a predictive approach, for obvious reasons, and the current criteria for seeding the tournament is anything but that. They may give a nod to tempo-free stats every now and then, but the committee almost always looks at each game as a "1" or a "0", depending on whether you win or lose.

Even with this somewhat silly approach, I think there's room to take a more structured approach to seeding. What I set out to do was get a baseline for the strength of each team's schedule, and compare their record to that mark. The Pomeroy ratings give a Pythag number for each team's schedule -- that's what percentage of the time their average opponent would win against an average D-1 team.

For a couple reasons, that's not quite what I want. First, I'd rather take a more relevant team -- say, one that's right on the bubble -- and run it through each team's schedule. I'd also prefer to be able to remove certain games, particularly to judge the resumes of certain teams that were missing their player of the year candidate for a significant stretch.

To solve both these issues, I used the Pomeroy ratings to see what the expected record of a bubble team (I used Old Dominion, which isn't really on the bubble, especially after Monday night's win, but is 34th in Pomeroy, which is what I was looking for) would be against each potential tournament team's full schedule. We can get a baseline winning percentage from that; for example, ODU would be expected to win 75% of their games against Kentucky's schedule.

From there, we can compare UK's actual winning percentage, .935, against that baseline. So the Wildcats' resume comes out as +0.186, 5.8 wins above expectation. I repeated this exercise for 63 teams with tournament aspirations, as well as three additional times: one looking at only Ohio St.'s games with Evan Turner, one looking at Wisconsin only when Leuer played, and a third excluding the one full game Kalin Lucas missed from Michigan St.'s schedule.

Here's what I got for the top 10 resumes in the country, though Sunday's games:

Starting from the top, the Jayhawks have the best resume in the country; I don't think there's been much debate about that since Syracuse lost at Louisville this weekend.

The surprising thing here is probably Duke easily deserving a #1 seed. To be clear: Duke's lofty rating in KenPom has nothing to do with the Blue Devils' placement here. All that comes into play is their W-L record, and how ODU would do against their schedule.

Here's a slightly more detailed breakdown of Duke vs. the two non-Kansas #1 seed "locks":

Duke's big advantage is in conference play, which isn't quite what you'd expect since they have the worst conference W% of the three contenders.

It's really not what you'd expect in comparison to Syracuse, since the Big East is thought to be stronger than the ACC this year. And they probably do have more teams that can make a Final Four run, and will likely send more teams to the tournament. But the BE also has more teams that really suck.

Strangely enough, Duke's easiest conference game according to KenPom was their final one, home vs. UNC -- Old Dominion wins that game 78% of the time. Contrast that with Syracuse's schedule, which has five games that Old Dominion would win over 80% of the time: vs. USF, at Rutgers, at DePaul, vs. Providence, vs. St. John's. The ACC may have fewer elite teams than the Big East, but it also has fewer nights that you're almost guaranteed a victory.

The order is surprising, but if we skip over Purdue because of the Hummel injury, my rankings agree with the consensus on the four #1 seeds right now. If you put Purdue on the #2 line, my rankings there are fairly standard as well, with West Virginia, Ohio St. (with ET), and Villanova joining the Boilermakers on the second line.

My three seeds are a bit more interesting: Kansas St. and Wisconsin (with Leuer) as shown above, plus New Mexico and Baylor. I may mock the Lobos' tournament potential on Twitter on an almost hourly basis, but for this exercise "28-3" is all that matters, and that's pretty solid.

Kansas St. is a borderline two or three, which is right where they are in the Matrix right now. Losing to Iowa St., a team that ODU beats at home 82% of the time, really killed them.

Wisconsin is looking at a #4 right now, which I think is a bit low. Their resume with Leuer is very strong; only the one bad loss against their in-state rivals, and I don't know that any team can match wins against Maryland, Duke, Ohio St., and Purdue.

Baylor is a borderline 4-5 in the current matrix, behind teams like Tennessee and Vanderbilt. I do not understand this. The Bears have a better record than either of those SEC teams, and have done it against a more difficult schedule, regardless of whether you look at KenPom or RPI SOS. Baylor also has a very low RPI, at 8 -- I don't care about that, but the people putting the bracket together certainly do. Sometimes I think we give too much credit to "signature wins", particularly if they come at home; that may be happening with Tennessee's victories over Kansas and Kentucky.

My four through 11 seeds are here, but let's finish this up with a look at the bubble. Here's my analysis of Lunardi's last six in and first 10 out:

The biggest differences here are Illinois and Mississippi; I have the Illini well out and the Rebels easily in, while Lunardi has the opposite. Here's the breakdown:

Illinois' back-to-back losses to Utah and Bradley back in late November really kill them here. They do have non-con wins against Clemson and Vandy, but they also lost to UGA, Missouri, and Gonzaga, so they were actually only 2-3 in notable out of conference games. Ole Miss didn't have a great conference showing, but they have an excellent non-con win against K-State, also beat UTEP, and only lost extremely difficult games (vs. Villanova, at WVU).


  1. Really interesting stuff. I love it. Reminds me a little of the Memphis debate last year, about how impressive their undefeated run in conference was.

    I think the handling of injury games could be better, as removing them entirely would probably be viewed as a little extreme, but overall I think this is very informative.

  2. "Reminds me a little of the Memphis debate last year, about how impressive their undefeated run in conference was."

    Indeed, I was thinking of that while I wrote it. Hopefully this one makes me look just as brilliant in the tourney.

    "I think the handling of injury games could be better, as removing them entirely would probably be viewed as a little extreme"

    Reasonable. Looking at their overall resumes, Wisconsin comes in 11th (+.114), Ohio St. 13th (+.107), and Michigan St. 15th (+.095).

  3. This is why it's hard to be an Illinois fan who also puts great faith in Vegas lines and the Pomeroy Ratings. All year, I've been trying to convince myself that they were better than their Pomeroy Rating, but the last few weeks have put an end to that facade. Anyways, excellent post, and I endorse the results, even if they look bad for my team.

  4. how did you obtain the expected winning percentage?

    I have been trying to replicate your results and have not been successful?

    Other than that, great article

  5. If the expected score from KenPom is Duke 71, Kentucky 55:


    would be Duke's W%.

    It's the same as Pythag for baseball but using 11.5 as the exponent.

  6. Yes, I know what the Pythagorean metric is and have read up on Kem Pom's blog.

    But how did you specifically manipulate the SOS Pythags to obtain the expected winning %?

  7. I have a spreadsheet that spits out the expected score of each game according to KenPom. That is what I used. If you email me I can give it to you.

  8. Thanks that would be great. Just sent you an e-mail