Pages

A Method to This (March) Madness? Dissecting the 2007 NCAA Brackets

3/16/2007
It's two days after the Selection Sunday. For bracketologists (and quasi-bracketologists like me, according to Washington Post writer Eric Prisbell), it's the calm after the storm. And in a few days, the tournament will start and nearly everyone will forget about who got hosed in the selection and seeding process. And rightfully so...the brackets are a distant second to the actual tournament. Unless, of course, your team was directly impacted, in which case you can stew about it while watching your team compete for the NIT (or as I like to call it, the Tommy Amaker Invitational).

For the second year in a row, my projections on The Bracket Project performed the worst of the 30+ bracket projections compiled. However, this is nothing for me to be ashamed of. I fully expected to be the worst. Because I look at it differently and base my projections on where I think the teams SHOULD be seeded, instead of trying to predict where the Committee WILL seed them, and more importantly because my brackets are firmly rooted in logic, clarity and consistency, right off the bat you know it's not going to do very well. Recall, I base my seeds on the following three criteria applied consistently across the board:

1. Body or work (Colton Index) - resume of wins and losses for the entire season
2. Quality Wins - summing up all of a teams wins based on degree of difficulty
3. Performance over last 10 - did team play above or below their body of work rating over the last 10

Two interesting points about my projections. First, I was the only one out of 30 that correctly picked Syracuse to not make the field (maybe the Committee read my recent blog post!) and was one of only a few that had Arkansas in the field. Also, overall I only did slightly worse than the king of Bracketology, Joe Lunardi at ESPN. Pretty amazing considering I spent about 15 minutes a day just updating the numbers and poor ol' Joe probably spent 15 HOURS a day pouring over the numbers. Lunardi pretty much invented the 'science' of bracketology (soon to be a celebrity religion), but he can't be having too many more years like this one before ESPN is telling him to stick to his day job (Assistant VP of marketing communications at St. Joseph's University).

The key thing to note is that bracketology in general is quickly becoming a joke. How does one predict what a group of 10 guys are going to decide to do, especially when there's a different set of guys every year? Their criteria could be strictly subjective or strictly objective, and oh by the way, that criteria can change from team to team. Good luck!

Before 2004, I believe the brackets were much easier to predict. But two factors have changed all of that: the New (and unimproved) RPI and Craig Littlepage. With the New (and unimproved...do I need to keep saying it?) RPI, the numbers are skewing the profiles of teams even more than before, and it's impossible for an outsider to know for which teams the Committee will be able to 'see through' the numbers and for which teams they will rely on the numbers. Depending on the path, the results will be drastically different. Related to the New (you know by now) RPI, Craig Littlepage and Co. turned the whole college basketball world upside down last year with some of the most puzzling brackets of all time. Nobody could make any sense of it (I tried here and here). I don't really blame Littlepage...I blame the NCAA for providing him with bad information. But coming into this year, no really knew what to expect.

I think the general consensus is that Gary Walters and Co. did an okay job (it is definitely a thankless job). Anything would've been an improvement over last season. Not too many people had beef with the teams selected, but there are some head scratchers as it relates to the seeding. The nice thing about my brackets (and the primary reason that I even bother posting them) is that we can evaluate where the Committee deviated from my consistent criteria (of what the Committee states it is doing) and see if there's any consistent pattern to the teams that the Committee is overseeding and underseeding.

First, comparing how well my projected brackets performed year over year really highlights that seeding was the bigger question mark relative to last year:

Number of Teams Picked Correctly: 2006 - 59, 2007 - 62
Exact Seed Picked Correctly: 2006 - 23, 2007 - 18
Seed Picked Within One: 2006 - 40, 2007 - 41

A higher number of teams correct this year, but far less with the exact seed picked correctly. So let's dive in and see if we can determine if there's a method to this madness.

THE OVERSEEDS

VIRGINIA

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
4RPI19-105550 1431766-44-12-35-3
8Colton19-103628 98185-0.414-14-44-3

TEAMTOT CONFNCL10R/N QWQL
Colton Splits362198 50782656

First, let me give you a lay of the land. The top line of the first table shows some of the key RPI-based attributes that the Committee tends to look at. The bottom line shows the same attributes instead using the Colton Index numbers. Any deviation by the RPI from the Colton Index is purely artificial and driven by the flawed nuances of the RPI formulas. So the first thing to ask yourself is: 'Is it likely that the Committee would have looked at this team differently if it had the numbers on the bottom instead of the numbers on the top?' If so, then the RPI likely played a role in the decision making.

For performance over the last 10 games, I look at it a different way. Instead of looking at the absolute number of wins and losses, it helps to look at how that record compares to expectations based on the strength of the opponent and the location of the game. A team may get dinged for finishing 5-5 but the truth is that 5-5 record might be consistent with the team's overall ranking. Conversely, a 8-2 record over the last 10 games might look good on paper, but what if those two losses were to very weak opponents? So you'll see a +/- in my 'Last 10' column that represents the performance versus expectations over the last 10 games for a team with that ranking. In the Virginia example, their 6-4 record was 0.41 less wins than expected for a typical 36th-ranked team.

In addition to the Colton-based attributes, this year I'm also showing some key splits by the Colton Index, representation the teams relative ranking by conference vs. non-conference games (special thanks to avid reader Todd Lucas for helping me clarify my thinking around this issue), performance over the last 10 games (this time in terms of overall ranking instead of +/-), performance in road and neutral games, and the team's ranking in quality wins (summing up all of the team's wins relative to the expectations used for the SOS measure) and quality losses (same thing as QW, for losses instead). So you can see that Virginia performed like the 36th best team over the course of the season, but that can be split between performing better than 36th for the first 19 games of the season and 50th for the last 10 games, or 21st and 98th between conference/non-conference, etc.

So, getting back to Virginia, is there any indication in the numbers that would lead you to believe they should be a 4 seed? Their RPI-based numbers are actually deflated across the board. The only things that sticks out is their Conference Split of 21 and their Quality Wins ranks of 26, however both are below what you'd equate with a 4 seed.

Verdict: Conference Performance. They did tie for the regular season ACC title (is there a regular season ACC title?)

BUTLER

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
5RPI25-627108 15276-41-13-03-2
11Colton25-634 1422473-1.942-12-0 3-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits347224 98364133

I was way off on Butler. I didn't really expect them to fall to 11, but I didn't expect them to be a 5 either. Butler didn't get dinged at all for their lackluster performance over the last 10 (6-4, but nearly 2 games below expected, an on par with the 98th best team over that stretch). Their non-conference performance was very good, and looks even better under the RPI. That was probably what carried the day for them.

Verdict: Non-Conference RPI and SOS. Also, 6-4 record doesn't look so bad on paper

VIRGINIA TECH

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
5RPI20-113423 53765-54-12-24-6
7Colton20-114024 80190-0.394-13-45-4

TEAMTOT CONFNCL10R/N QWQL
Colton Splits402480 52511575

Virginia Tech is another team that didn't get hurt by a less than stellar finish to the season. Their 5-5 record was consistent with the 52nd best team. They do rank well in Quality Wins and Conference Split. Maybe the Committee felt they had to have both Virginia and Virginia Tech ahead of Duke. NC RPI numbers are inflated, especially NC SOS.

Verdict: Non-Conference RPI and SOS, Quality Wins and ignoring 5-5 Finish

NOTRE DAME

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
6RPI24-731106 683097-33-34-13-1
8Colton24-72389 10276-0.171-24-25-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits233410 35504319

Notre Dame's RPI numbers make them look worse than the Colton Index numbers, but I still have them as an overseed. Quality wins is probably the biggest culprit...the RPI gives them 3 top 25 wins and 7 top 50 wins, the Colton Index gives them 1 top 25 win and 5 top 50 wins.

Verdict: RPI Quality Wins. Not a huge gap here...maybe a 7 seed is more appropriate.

LOUISVILLE

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
6RPI22-93736 1341447-32-61-14-2
9Colton22-94273 104279+1.601-51-26-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits4217104 15204041

Louisville's RPI SOS number is vastly different than its Colton Index cousin. This is probably driven by the fact that Louisville didn't play a non-conference road game, which is not captured in the RPI SOS measure (amazing huh?). Louisville performed very well down the stretch as well...it's 7-3 record doesn't do it justice.

Verdict: RPI SOS and Performance over the Last 10

VANDERBILT

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
6RPI19-114522 1011305-54-31-25-3
10Colton19-1143 2192222+0.114-33-3 4-4

TEAMTOT CONFNCL10R/N QWQL
Colton Splits432592 45532962

Florida seemed to be revered by the Committee. Ohio State had a much better profile than Florida other than their head-to-head meeting four months ago, but the Committee rewarded the Gators with the #1 overall seed. Vanderbilt probably got some residual benefit from beating Florida recently. Even with this win, however, Vandy's doesn't seem to fit the profile of a 6 seed: 43rd body-of-work, 29th quality wins, 45th over the last 10.

Verdict: They beat Florida, therefore they must be good.

INDIANA

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
7RPI20-102851 401485-53-42-42-2
11Colton20-1028 273486-0.993-43-5 2-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits283634 64765520

PURDUE

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
9RPI21-114244 37877-31-42-26-3
11Colton21-1146 4642147+0.881-44-4 5-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits464742 24735931

I'm a Big Ten guy, but I can't quite get a handle on the way the Big Ten seeds broke down. Indiana 7, Michigan St 9, Purdue 9, Illinois 12. Michigan State is probably fine at 9, but the others should've been more evenly bunched in the 10/11 range. Indiana played poorly down the stretch, but it didn't seem to hurt their seed.

Verdict: Hard to figure this one out.

XAVIER

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
9RPI24-83283 47449-11-12-19-3
OUTColton24-860138 59130+1.190-03-27-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits605859 20674769

I was one of the few that had Xavier out of the tournament. They were clearly helped by some inflated RPI numbers (32 vs 60...a huge difference for a bubble team. 83 vs 138 SOS) and some bad losses (Cincy, Duquesne, Rhode Island) didn't seem to hurt them much. I'm not surprised that they made the tournament, but I'm still not sure why they were considered a lock.

Verdict: The RPI Numbers

GONZAGA

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
10RPI23-106093 4897-31-41-14-3
12Colton23-1054 834513-0.902-41-3 3-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits546845 86404658

Gonzaga as a 10 seed shows that they probably would've made it as an at-large had they not won their conference tournament. I'm not so sure that they deserved it. Gonzaga seemed to get a lot of credit for simply playing a tough non-conference schedule, although their NC performance wasn't much better than their overall performance (54 vs 45). Unlike other years where I felt Gonzaga wasn't getting the credit they deserved, this is one year where their profile doesn't fit the seed.

Verdict: Played a tough non-conference schedule.

TEXAS TECH

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
10RPI21-125339 46626-43-22-42-3
OUTColton21-125761 56145+0.033-32-43-4

TEAMTOT CONFNCL10R/N QWQL
Colton Splits575456 59543382

Texas Tech was the second of three teams that I didn't have in the tournament. Tech had some very good wins (Kansas, Texas A&M twice) that probably helped their cause tremendously, enough to overshadow losses to Baylor, Oklahoma and Nebraska. RPI SOS numbers are a little inflated. Overall, I can't really raise a stink that they got in. They were certainly more deserving than Drexel or Syracuse.

Verdict: Quality wins. Quality wins. Quality Wins.

OLD DOMINION

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
12RPI24-84096 52359-11-03-22-3
OUTColton24-863136 6478+1.651-10-24-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits635664 10576055

Well, I'm just glad it wasn't Drexel. ODU got some serious help from some inflated RPI numbers. RPI: 40 vs 63. SOS: 96 vs 136. NC SOS: 35 vs 78. Those are the kind of gaps you want if you are one of the last teams in the field. Plus ODU got credit for 4 top 50 wins when I only have them with one (2 thanks to Drexel). Performance over the last 10 is strong, but the rest of their splits are mediocre.

Verdict: Inflated RPI numbers. Favorable conference and head-to-head to Drexel.


THE UNDERSEEDS

SOUTHERN ILLINOIS

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
4RPI26-6731 12429-12-16-35-1
2Colton26-61148 2248+1.310-06-39-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits111222 516817

CREIGHTON

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
10RPI22-102019 72247-31-26-14-4
6Colton22-103555 99117+0.551-23-08-6

TEAMTOT CONFNCL10R/N QWQL
Colton Splits351599 30371666

MISSOURI ST

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
RPI21-103642 41577-31-52-16-3
8Colton21-103954 4896+0.861-30-39-3

TEAMTOT CONFNCL10R/N QWQL
Colton Splits393548 26242359

The Missouri Valley got no love from the Committee this year. Seeds were down across the board, and Missouri St got hosed for the second year in a row. For all of the stink about Syracuse not getting a seed, Missouri St has a much bigger beef. What more do you want from Missouri St? Overall performance? 39th. Quality wins? 23rd. Performance over last 10? 26th. Road/Neutral performance? 24th. And their RPI numbers are even better than their Colton Index numbers.

It's the same deal with Southern Illinois. Their numbers are very strong across the board, at least enough to warrant a 3 seed. Creighton also has some very good numbers.

So what's up? I think this is an example where the Committee simply didn't believe the numbers. The RPI numbers for the MVC are inflated across the board. The MVC did get a lot of negative press (whether justified or not) for cooking the numbers last year. That could have had a role in Committee members being skeptical again here, only they over corrected. Maybe it's backlash for trying to game the system. Or maybe they simply don't believe that the MVC teams are really that good despite what the numbers say.

And although the Committee states that conference affiliation doesn't matter (and for the most part I believe them), conference affiliation DEFINITELY played a role last year...when the Committee was dead set against having five MVC teams. If you don't believe me, check out Craig Littlepage's comments here (under 'Missouri St and Conference Affiliation'). Missouri St lost out in that comparison last year, and they lose out again here. Sorry Bears fans.

Verdict: No love for the MVC. No faith in the MVC RPI numbers.

ARIZONA

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
8RPI20-10142 216-43-72-29-1
4Colton20-10161 1215-0.353-94-16-0

TEAMTOT CONFNCL10R/N QWQL
Colton Splits162812 2571223

Another team notorious for scheduling to make itself look as pretty as possible. Maybe the Committee read my blog post from last year. Arizona didn't get any benefit from their lofty numbers this year. It's surprising too, because their 'true' numbers are still very strong. 12th in Quality Wins. 7th in Road/Neutral games (although I was at the Illinois/Arizona game in Phoenix, that was hardly neutral. At a very minimum, Arizona got some homer-ish calls). 12th in non-conference performance. None of these things scream 8 seed at me. But since they got in the tournament last year when they shouldn't have, I can't really feel sorry for them.

Verdict: What goes around comes around.

UNLV

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
7RPI27-61038 13389-13-21-16-1
4Colton27-61790 1498+0.341-13-28-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits172614 12212118

UNLV had some inflated RPI numbers, it didn't do much for them on Selection Sunday. Even with my lower numbers, UNLV fit the profile of a 4-5 seed. It's Conference Split is the lowest one on the list, and that's still 26th overall.

Verdict: Not a strong schedule, but still a pretty solid resume. What more do you want?

KENTUCKY

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
8RPI21-11131 624-61-64-39-2
5Colton21-11202 1629-1.671-65-49-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits203116 67291825

VILLANOVA

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
9RPI21-10197 22467-31-44-45-2
7Colton21-103136 35181+0.193-41-16-5

TEAMTOT CONFNCL10R/N QWQL
Colton Splits314035 37222746

An interesting match-up of two Wildcats (awww..how cute!) that were both underseeded...a 8-9 matchup that could've easily been a 5-7 match-up. Kentucky played one of the most difficult schedules in the country, performed reasonably well, yet wasn't rewarded for it. A lackluster finish (4-6, 67th best performance over the last 10) and 1-6 against the RPI Top 25 probably hurt its chances. As for Villanova, I expected them to be seeded much higher than the 7th seed I had them since their RPI numbers were much higher across the board (46 vs 181 NC SOS! It's amazing that the RPI is even measuring the same thing.) Villanova's numbers look like a pretty solid 7-8 seed.

Verdict: Last 10 for Kentucky. Not sure about Vilanova.

GEORGIA TECH

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
10RPI20-115241 381577-33-44-23-2
7Colton20-112913 15124+1.043-46-43-2

TEAMTOT CONFNCL10R/N QWQL
Colton Splits294815 21643037

Georgia Tech had deflated RPI numbers across the board. RPI: 52 vs 29. SOS: 41 vs 13. NC RPI: 38 vs 15. 7 Top 50 wins vs 9 top 50 wins. The Colton Index numbers paint a better picture for Georgia Tech.

Verdict: Deflated RPI numbers.

WINTHROP

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
11RPI24-470270 314510-00-42-01-0
6Colton24-410150 203+0.490-41-02-0

TEAMTOT CONFNCL10R/N QWQL
Colton Splits10920 612902

I've said everything there is to say about Winthrop here. By giving them a 11 seed, the Committee is treating Winthrop like a 22-6 team instead of a 24-4 team. Winthrop's 70 RPI and 270 SOS probably didn't help their cause either.

Verdict: Deflated RPI numbers. Lack of quality wins (90th overall).

ARKANSAS

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
12RPI21-13359 23326-41-44-16-6
9Colton21-134716 36112+0.871-54-17-7

TEAMTOT CONFNCL10R/N QWQL
Colton Splits475136 32382273

Arkansas played its way into the tournament in the conference tourney, bolstering its quality wins (22nd) and performance over the last 10 (32nd). Most focused on Arkansas' sub-500 conference record. A tough one to seed with quality wins and some bad losses (73rd QL)

Verdict: Performance down the stretch got em in. Conference performance and bad losses held them back.

FLORIDA ST

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
RPI20-124121 261114-63-51-23-5
10Colton20-1233 1113151-0.623-52-7 3-0

TEAMTOT CONFNCL10R/N QWQL
Colton Splits335213 55344530

Florida St just gets no love. Last year they played a weak non-conference schedule and had some deflated RPI numbers. This year their numbers were deflated as well (33 vs 41 RPI, 11 vs 21 SOS) despite a better schedule this time around. Florida St had a weak 4-6 finish to their season, but this wasn't much lower than expected (55th overall). All 12 of its losses were inside the top 50 of the Colton Index (much different than 3-5 in the RPI 51-100 range). Florida St is the perennial bubble team.

Verdict: Deflated RPI numbers. 4-6 finish looks worse than it really is.

CLEMSON

SEEDTEAM W-LRPISOSNC RPI NC SOSLast101-25 26-5051-100
RPI20-105152 302333-70-55-24-2
10Colton20-1027 2327282-2.650-55-4 5-1

TEAMTOT CONFNCL10R/N QWQL
Colton Splits162812 2571223

I didn't really expect Clemson get in given their 17-0 start and 3-10 finish. If it was the other way around, they would've made the tournament. If it's truly body of work...Clemson gets in: 27th overall. 32nd in quality wins. 27 vs 51 RPI probably didn't help much either.

Verdict: Could say deflated numbers, but in all honesty 3-10 finish just ain't gonna cut it.

Overall, it's hard to really detect a pattern in the Committee's thinking. In some cases, performance over the last 10 killed teams like Clemson, Florida St and Kentucky but didn't do any harm to Indiana, Butler, Vanderbilt or Virginia Tech. RPI likely played a role for Xavier, Georgia Tech, Winthrop, Old Dominion, but didn't in the MVC teams, Arizona, Arkansas or UNLV.

It seemed like Walters & Co. depended more on taking and compiling votes of the 10 members more than I remember, so it was more of a democratic process. However, it's impossible to know what criteria each and every Committee member was using to rank order the teams -- were they looking at body or work, voting for the most deserving teams, or simply ranking what they thought were the best teams? Hard to tell exactly from just looking at these examples.

One other interesting tidbit that came out of the post-bracket teleconference. Somebody asked Walters about a new ranking, called LRNC, that was being used for the first time. This was completely out of left field. Nobody I know has even heard of this LRNC rating. I'm surprised more wasn't made out of it. I can only imagine what the basis of it is. When I hear anything, I'll let you know. Stay tuned.

"Q. How much, if at all, did your committee use the LRNC rankings provided to you this year for the first time?

GARY WALTERS: We had access to all of the quantitative data that is out there, and it's certainly one tool that people could look at. But, you know, the main quantitative model that we use is obviously the RPI. We have enormous respect for what Jeff Sagarin has done over the years, as well. We can also reference that. But our primary quantitative tool is the RPI. That provides the basic framework from which we operate.

Q. Did you look at this at all?

GARY WALTERS: Yeah, I had a chance to look at it. I mean, anybody who's a basketball aficionado would look at it. But I don't know what the thrust of that question is. Is this an advertisement?

Q. I'm just asking about it because it's new.

GARY WALTERS: No, I mean, we looked at it. Again, I think I answered the question."

0 comments:

 
Wegoblogger #31 © 2011 | Designed by Bingo Cash, in collaboration with Modern Warfare 3, VPS Hosting and Compare Web Hosting