Dork NCAA Discussion

[JasonEvans]

Yup, the Dorks own the post-season too!!

Ken Pomeroy has put up seeds next to his rankings on his main page (http://kenpom.com/). A few things jump out when looking at it.

1) The highest rated team to not make the tourney -- Va Tech at #30. Then Maryland at #36 and New Mexico at #43. The ACC got screwed.

2) The lowest rated team to get an at-large bid -- you guessed it, Virginia Commonwealth at #84. The committee has got some 'splaining to do about that one.

3) Utah State and Belmont got the shaft from the committee with their seeds. Utah State is Ken's #16 team and got a #12 seed. Belmont is his #18 team and got a #13 seed. The only other team in Ken's top 25 that got a double-digit seed was #24 Clemson. They only got in enough to make the play-in game as a #12 seed. Did I mention that the ACC got screwed?

4) There is only one ACC team that got seeded higher than Ken's rankings indicate. Guess who? UNC was given a #2 seed despite being the #14 team in Ken's rankings. Ken's rankings thought the Heels deserved a #4 seed. Of course, Ken's rankings include games played while DrewII was still on the TarHeels, which skews things JUST A LITTLE BIT!

5) It is said that the best bet at the National Championship is a team that is ranked in the top ten in both Ken's offensive and defensive ratings. This year, there are only two teams like that -- Ohio St (#1 in O, #10 in D) and Duke (#5 in O, #3 in D). That is not a bad thing ;)

6) Alabama St of the SWAC made the dance (automatic bid, of course) despite being Ken's 295th ranked team. That puts them among the 50 worst teams in all of college basketball. That said, don't sleep on these guys. They were 6-16 at one point but won 11 of their final 12 games to make it to the Big Dance. Of course, they haven't played a team ranked in the top 250 in college basketball since mid-December so they are stepping up in class for the tournament... even when they play UT-San Antonio (Ken's #219 team).

That's all for now. I'll be eager to see how his log5 computations come out as he was really, really good at predicting conference tournament winners (http://kenpom.com/blog/index.php/weblog/log5_vs._seeds/) with them.

-Jason "there is also a great race going on between Jimmer, Nolan, and Sullinger for Ken's POY (http://kenpom.com/blog/index.php/weblog/the_kpoy_the_almost_final_countdown/)-- if Duke goes a lot further than BYU, it could really tilt things for Nolan!" Evans

[loran16]

Yup, the Dorks own the post-season too!!

Ken Pomeroy has put up seeds next to his rankings on his main page (http://kenpom.com/). A few things jump out when looking at it.

1) The highest rated team to not make the tourney -- Va Tech at #30. Then Maryland at #36 and New Mexico at #43. The ACC got screwed.

2) The lowest rated team to get an at-large bid -- you guessed it, Virginia Commonwealth at #84. The committee has got some 'splaining to do about that one.

3) Utah State and Belmont got the shaft from the committee with their seeds. Utah State is Ken's #16 team and got a #12 seed. Belmont is his #18 team and got a #13 seed. The only other team in Ken's top 25 that got a double-digit seed was #24 Clemson. They only got in enough to make the play-in game as a #12 seed. Did I mention that the ACC got screwed?

4) There is only one ACC team that got seeded higher than Ken's rankings indicate. Guess who? UNC was given a #2 seed despite being the #14 team in Ken's rankings. Ken's rankings thought the Heels deserved a #4 seed. Of course, Ken's rankings include games played while DrewII was still on the TarHeels, which skews things JUST A LITTLE BIT!

5) It is said that the best bet at the National Championship is a team that is ranked in the top ten in both Ken's offensive and defensive ratings. This year, there are only two teams like that -- Ohio St (#1 in O, #10 in D) and Duke (#5 in O, #3 in D). That is not a bad thing ;)

6) Alabama St of the SWAC made the dance (automatic bid, of course) despite being Ken's 295th ranked team. That puts them among the 50 worst teams in all of college basketball. That said, don't sleep on these guys. They were 6-16 at one point but won 11 of their final 12 games to make it to the Big Dance. Of course, they haven't played a team ranked in the top 250 in college basketball since mid-December so they are stepping up in class for the tournament... even when they play UT-San Antonio (Ken's #219 team).

That's all for now. I'll be eager to see how his log5 computations come out as he was really, really good at predicting conference tournament winners (http://kenpom.com/blog/index.php/weblog/log5_vs._seeds/) with them.

-Jason "there is also a great race going on between Jimmer, Nolan, and Sullinger for Ken's POY (http://kenpom.com/blog/index.php/weblog/the_kpoy_the_almost_final_countdown/)-- if Duke goes a lot further than BYU, it could really tilt things for Nolan!" Evans

Log5 is out. http://www.basketballprospectus.com/unfiltered/?p=673

[tieguy]

Interesting that Ken gives us a better chance of winning the Elite Eight game than Sweet 16. Not surprising, given how highly his numbers think of Texas. I think that game will be a war (and a trendy underdog pick), though I'll take K over Barnes any day of the week and twice on Sunday.

[ice-9]

Interesting that Ken gives us a better chance of winning the Elite Eight game than Sweet 16. Not surprising, given how highly his numbers think of Texas. I think that game will be a war (and a trendy underdog pick), though I'll take K over Barnes any day of the week and twice on Sunday.

Don't forget there were many posts like yours regarding Lavin and St. John's!

[jdk]

KenPom had us about .0075 points behind OSU before updating yesterday. I was hoping that blowing out the #13 team carried enough weight to cover the gap between us and OSU (who played the 39th ranked team). Right now we are still a quarter of a percentage point behind them, setting up a possible epic matchup in the national semis. Would have been an even more epic final...OSU riding their star freshman center to try and stop the defending national champion from repeating...for the second time in FIVE years.

HOWEVER, I think UNC takes a matchup against OSU should both teams advance.

[MChambers]

In addition to Pomeroy and Sagarin, this time of year I sometimes look at a third ranking system, one created by some Georgia Tech professors. It's supposedly pretty accurate, better than the seeding system (insert your own joke here). Here's some explanation:

http://www2.isye.gatech.edu/people/faculty/Joel_Sokol/lrmc/about_lrmc.html

So this year's Cindarella? Looks like our old friends at Belmont, ranked, wait for it, #4.

http://www2.isye.gatech.edu/people/faculty/Joel_Sokol/lrmc/lrmc.sort0.html

Bad news is that San Diego State is ranked #5, and Texas is #7. The good news is that neither Michigan nor Tennessee is very high. UNC has some very tough matchups

[4decadedukie]

The highest rated team to not make the tourney -- Va Tech at #30. Then Maryland at #36 and New Mexico at #43. The ACC got screwed.

The ACC was screwed; however, the fact that it was Gary/Maryland and Seth/VPI makes this MUCH easier for me to assimilate.

[JasonEvans]

The ACC was screwed...

I find Clemson and FSU's seeds to be the most distressing. How is UGA a #10, UCLA a #7, and Tennessee a #9 seed but Clemson is a #12 that has to play its way into the regular field? Really?

-Jason "the Dorks know a lot better than the committee does" Evans

[TNDukeFan]

Interesting that Ken gives us a better chance of winning the Elite Eight game than Sweet 16. Not surprising, given how highly his numbers think of Texas. I think that game will be a war (and a trendy underdog pick), though I'll take K over Barnes any day of the week and twice on Sunday.

Gottlieb on ESPN picks TX to beat Duke. Then again, he picks UNC as the first 1 or 2 to go down.

[ns7]

I find Clemson and FSU's seeds to be the most distressing. How is UGA a #10, UCLA a #7, and Tennessee a #9 seed but Clemson is a #12 that has to play its way into the regular field? Really?

-Jason "the Dorks know a lot better than the committee does" Evans

It really looks like they used RPI to seed the teams. Or they consulted with the CareerBuilder chimps.

[CDu]

It really looks like they used RPI to seed the teams. Or they consulted with the CareerBuilder chimps.

They have historically seemed to continually rely heavily on the RPI, even though they say they're de-emphasizing the RPI.

One thing I find amusing is that the "resumes" posted by the media for discussion includes RPI and record vs top-50 (which is also based on RPI). So even the resumes are heavily influenced by RPI.

There always appear to be strange decisions made, but it seems clear that RPI continues to guide a lot of the decision-making process.

[Udaman]

Some dorky tidbits from spending way too much time on this stuff today...

1) A 1, 2, 3, or 4 seed is much more likely to win it all (over the last 15 years, only 1 4 seed has won - Arizona, the rest were 1,2's or 3).

2) UNC has never won the national championship without first being a 1 seed. That trend will continue this year.

3) Since 1980 no team has won the national championship after losing in the first round of their conference tournament game. This means you, Pittsburgh. A loss in the first round, I believe, just leaves a team wondering what went wrong. Note that since 2000 (at least) only 1 team that lost in the opening round of its tournament even made the Final Four. That was Michigan State last year...and the ONLY reason that happened is because their bracket was filled with huge upsets. Top seed Kansas lost, # 2 seed Ohio State lost and #3 seed Georgetown lost, so Michigan State played Maryland, Northern Iowa, and then Tennessee to advance. Total joke. They then played Butler (a 5 seed) before losing.

4) Only one team with a freshman point guard has won the National Championship since at least 1990. That was Syracuse in 2003. Their point guard was Gerry McNamara, who was really playing out of position. They won that year because Carmelo Anthony and Hakim Warrick. Two teams fit the bill this year for a true freshman point guard - Kentucky and Texas. Neither will win it all.

5) When all else fails, go with teams that are heavy on Seniors and Juniors, with tournament experience. Those teams are Kansas, Duke, Ohio State, Notre Dame and...um...Pittsburgh. Clearly Pitt is a good team, and they got a joke bracket (with the worst #2, and #3 seeds out there)....but I still think their Big East flameout will hurt them.

Cub

[MChambers]

4) Only one team with a freshman point guard has won the National Championship since at least 1990. That was Syracuse in 2003. Their point guard was Gerry McNamara, who was really playing out of position. They won that year because Carmelo Anthony and Hakim Warrick. Two teams fit the bill this year for a true freshman point guard - Kentucky and Texas. Neither will win it all.

1. Isn't Ohio State's point guard, Craft, a true freshman?

2. Does this mean that I should be hoping Kyrie doesn't return? I'm so confused.

[Lennies]

4) Only one team with a freshman point guard has won the National Championship since at least 1990. That was Syracuse in 2003. Their point guard was Gerry McNamara, who was really playing out of position. They won that year because Carmelo Anthony and Hakim Warrick. Two teams fit the bill this year for a true freshman point guard - Kentucky and Texas. Neither will win it all.


Texas has a freshman backup point guard.

[jdk]

Gottlieb on ESPN picks TX to beat Duke.

Not a big leap, the simulator at Whatifsports has the same thing happening a majority of the time.

And of course they are also the no. 4 on Pomeroy.

[crimsonandblue]

Texas has a freshman backup point guard.

Who? Joseph? He's not really a point. They're basically Balbay and Lucas at point. Joseph's a really good player though. Smart, smooth, poised, but doesn't try and do too much. Basically the anti-freshman Jordan Hamilton.

[davekay1971]

Gottlieb on ESPN picks TX to beat Duke. Then again, he picks UNC as the first 1 or 2 to go down.

This may be a case of a fool making enough guesses eventually guessing correctly. Gottleib picks Texas to beat Duke because he hates Duke. KenPom indicates it may be a good pick, but I bet anything that (1) Gottleib didn't know that; and (2) Gottleib wouldn't understand KenPom's methods if they were explained to him in very...small...words.

[loldevilz]

Duke got completely screwed with Texas as their 4 seed. Texas is the best defensive team in the country and among the most talented. They don't have a great coach, but they have more than enough talent to get to a final four.

Also, San Diego State in Anaheim is also way unfair. Duke "the one seed" has to play "the two seed" at their place. Totally ridiculous.

The sad thing is that I'd rather see SDSU than Uconn in the elite eight. Its safe to say that Duke will have to play some very good games to get to the final four, where they will probably have to hit OSU.

This certainly won't be as easy as last year.

[wilson]

Duke got completely screwed with Texas as their 4 seed. Texas is the best defensive team in the country and among the most talented. They don't have a great coach, but they have more than enough talent to get to a final four.

Also, San Diego State in Anaheim is also way unfair. Duke "the one seed" has to play "the two seed" at their place. Totally ridiculous.

The sad thing is that I'd rather see SDSU than Uconn in the elite eight. Its safe to say that Duke will have to play some very good games to get to the final four, where they will probably have to hit OSU.

This certainly won't be as easy as last year.I think the above is phrasing things a bit too extremely. Texas indeed has a very good profile, probably better than befits a 4 seed, but they don't exactly come blazing into the tournament.
As for SDSU in Anaheim, I view that matchup pretty much exactly the same way as I did last year's regional final vs. Baylor in Houston. It's certainly in the neighborhood, but it's not "their place," as you say, and I'd comfortably pick Duke to beat them, though of course it's no shoo-in (for the record, people, it's not "shoe-in"!).
With regard to UConn, I'd like another crack at them. I think if you can more or less bottle up Kemba Walker (meaning lots of points on lots of shots...I think there's virtually no chance of keeping him from getting lots of points), they're eminently beatable. Some revenge for past NCAAT matchups would be really tasty.

Moving on to Ohio St., my thoughts are twofold:
1. That potential matchup is so far away that I refuse to invest a scintilla of energy worrying about it right now. Furthermore, it's likely that one or both teams will fail to hold up their end of the bargain in bringing it to fruition.
2. They've been looming all season. It's not like they all of a sudden showed up and got hot. To regretfully use a cliché which is nevertheless appropriate, "to be the best, you've got to beat the best." If we should see them on April 2, I know we'll give them everything we've got, and I'll be comfortable with our chances.
You say above that "it's safe to say Duke will have to play some very good games to get to the Final Four."
Um, in other news, the sun came up this morning. Of course we'll have to play thusly to get there, and if we should make it to Houston, we'll have to play two more very good-to-great games if the boys want to win the whole thing. This team is perfectly capable of doing just that.

[hurleyfor3]

Some dorky tidbits from spending way too much time on this stuff today...

1) A 1, 2, 3, or 4 seed is much more likely to win it all (over the last 15 years, only 1 4 seed has won - Arizona, the rest were 1,2's or 3).

No 5 has ever won, either. Not that Butler didn't come close.

Since Kansas in 1988 only a single 4 has won (Arizona), and a handful of 3's (Michigan '89, Syracuse '03 and Florida '06). Every other champion has been either a 1 or a 2.



3) Since 1980 no team has won the national championship after losing in the first round of their conference tournament game. This means you, Pittsburgh.

I wouldn't count out Pitt for that reason. The Big East was positively loaded, UConn is good, and Kemba was possessed. He misses that buzzer-beater, Pitt advances. It wasn't like they lost by 20.

That said, I tend to believe the inverse -- a team that wildly outperforms in its conference tournament will probably come down to earth (revert to the median) in the NCAAs and very likely lose early. Yeah, NC State went on a nice little run once, 28 years ago. I have no hard evidence here, but it seems teams fall flat far more often.

This will be Pitt's best chance to make a Final Four, at least in terms of draw. They'd better not screw this one up.

[mgtr]

Some dorky tidbits from spending way too much time on this stuff today...

That was Michigan State last year...and the ONLY reason that happened is because their bracket was filled with huge upsets. Top seed Kansas lost, # 2 seed Ohio State lost and #3 seed Georgetown lost, so Michigan State played Maryland, Northern Iowa, and then Tennessee to advance. Total joke. They then played Butler (a 5 seed) before losing.


Cub

I am totally confused. I thought we played Northern Iowa last year (and that they were pretty good). I am sure you folks can set me straight.

[pfrduke]

I am totally confused. I thought we played Northern Iowa last year (and that they were pretty good). I am sure you folks can set me straight.

Northern Iowa beat Kansas.

Our road was Arkansas-Pine Bluff - California - Purdue - Baylor - West Virginia - Butler.

[gam7]

I have no idea what the "log5" probabilities mean in layman's terms, but does Pomeroy typically update them after each round?

[Greg_Newton]

Does anyone know for sure whether Pomeroy incorporates recency of results into his ratings?

I've searched the internet far and wide, including his website, and can't find a concrete answer.

[loran16]

I have no idea what the "log5" probabilities mean in layman's terms, but does Pomeroy typically update them after each round?

log5 is essentially a formula used to predict the probability of teams winning games against a given other team. (http://www.tangotiger.net/wiki/index.php?title=Log5, though i don't know if that formula is the same for basketball, but I think it is, using the pomeroy pythag numbers).

[loran16]

1. Isn't Ohio State's point guard, Craft, a true freshman?

2. Does this mean that I should be hoping Kyrie doesn't return? I'm so confused.

IIRC, Craft is actually the backup point guard, though he gets a ton of minutes.

[loldevilz]

Does anyone know for sure whether Pomeroy incorporates recency of results into his ratings?

I've searched the internet far and wide, including his website, and can't find a concrete answer.

I'm pretty sure he does. In just the span of a few games teams jump a lot. For instance SDSU made a huge jump after beating BYU and Notre Dame made a huge jump after crushing Cincinnati.

[Wander]

1) The highest rated team to not make the tourney -- Va Tech at #30. Then Maryland at #36 and New Mexico at #43. The ACC got screwed.

2) The lowest rated team to get an at-large bid -- you guessed it, Virginia Commonwealth at #84. The committee has got some 'splaining to do about that one.

3) Utah State and Belmont got the shaft from the committee with their seeds. Utah State is Ken's #16 team and got a #12 seed. Belmont is his #18 team and got a #13 seed. The only other team in Ken's top 25 that got a double-digit seed was #24 Clemson. They only got in enough to make the play-in game as a #12 seed. Did I mention that the ACC got screwed?


Without getting too zen-like, let's not forget that the entire point of playing a game is to win it.

Pomeroy freely admits that his system makes no attempt to rate how good of a season a team has had - but that's the only criterion, vague as it is, that should be taken into account by the selection committee. Because kenpom ratings never take into account who wins a game, they have no place on selection Sunday.

We all already know this intuitively, I just think some of us forget the game's actual objective when talking efficiency statistics. How many of us were thinking "You know, I really don't care if Gordon Hayward's shot goes in or not, because a score of 61-59 is basically the same result as a score of 61-62."

That may be an extreme example, but I contend it's not fundamentally different than how we should feel during any regular season game. You mentioned Maryland at #36. They were 8-10 in the ACC and had a win at Penn State as their best non-conference win. They didn't deserve to be anywhere near the bubble, and the fact that they were the 2nd highest team left out is an indictment of using these stats during the selection process, not a piece of evidence that the ACC was screwed.

I should note that I look at kenpom stuff all the time, think they're a great tool for breaking down teams, and shamelessly use them as a tool for bracket picking. They're an awesome piece of work. I also am not necessarily defending, for example, VCU over Virginia Tech (though the gap is no where near as large as Jay Bilas thinks it is). I just think we shouldn't lose sight of what makes the sport orders of magnitude better than college football - if you complete the stated objective of scoring more points than your opponent, you'll get a fair shake at winning championships.

[El_Diablo]

3) Since 1980 no team has won the national championship after losing in the first round of their conference tournament game. This means you, Pittsburgh.

Although with the double-bye, Pitt lost in the third round of the Big East tournament, not the first round.

[throatybeard]

Although with the double-bye, Pitt lost in the third round of the Big East tournament, not the first round.

OK, quarterfinals, then.

I'm already seeing silliness in brackets that the NCAA is calling the erstwhile 1R the second round because of the "first four." Um, no. Or, fine, I'm not a prescriptivist. But c'mon, don't lets's act like the first set of games in a BE tourney with a double bye is the same as a QF in an 8-team conference. Did we call the NCAAT 1R "the second round" during the last few years when East Salami State was playing-in with UT-Alice in one "first round" game? No.

Yes I made those schools up. Most people are ignoring the play-in games as well they should. Brackets get finalized before TH noon East, as ever.

[snowdenscold]

Yes I made those schools up. Most people are ignoring the play-in games as well they should. Brackets get finalized before TH noon East, as ever.

I wish more people would. The bracket I've been in for a couple years wants everything in by 6 PM Tuesday. Since this week is crazy busy I haven't had much time to look at the brackets at all, and I don't know if I'll get my selections in in time.

I wondered when they first announced the change if it would backfire on them. People feel rushed to make their selections and may give up on it as a result. Also, you lose the excitement that builds up between Sunday and Thursday due to a half-round not many will get invested in and kills momentum.

If ratings stay consistent, I'll withdraw my concerns. But I do see the potential for backfire. They should follow the old Hollywood maxim of "always leave them wanting more".

[tieguy]

Not to go all Fire Joe Morgan on you, Wander, but...


Without getting too zen-like, let's not forget that the entire point of playing a game is to win it.

Pomeroy freely admits that his system makes no attempt to rate how good of a season a team has had - but that's the only criterion, vague as it is, that should be taken into account by the selection committee. Because kenpom ratings never take into account who wins a game, they have no place on selection Sunday.

Saying Ken doesn't track wins is like saying that the Department of Elections doesn't track election winners, they just count votes.

What Ken takes into account is whether (and by how much) you outscore your opponents. Now, maybe he doesn't call those "wins", but last time I checked, the way you win is by... let me see if I can remember... outscoring your opponents. If you outscore your opponents by a lot, odds are pretty good you're going to win a lot. And vice-versa.

It's true that he doesn't give teams that win close games an extra bonus, but that's because if you win a close game against a team, and then play them again tomorrow, you're likely to lose. (In fact, if you win a close game at home, you're likely to not just lose the road game, you're likely to lose badly.) And of course, all other things being equal, there is some truth to the transitive property- just to pick one example, if Duke beats Miami by 10 at Miami, and UNC beats Miami by 2 in a virtual home game, they are both Ws (how zen!) and yet it also suggests that Duke is a better team than Carolina.

Of course it is also true that outscoring your opponents by a lot doesn't necessarily mean much- Maryland outscored its opponents on the season by a very large margin, which translated into respect from Pomeroy, but it didn't translate into nearly as many wins as you would have expected. Ditto Clemson. Part of that is luck; part of that is mental toughness or intangibles or what have you. But it's not a very big part. Much bigger is shooting the ball well, rebounding, not turning it over, and getting to the line. Those things translate real well to winning.

Too tired to really finish this off with much punch, so I'll just say that if you don't understand Pomeroy you probably shouldn't write so much about him :)

[Reilly]

Wasn't there an article or thread recently about the RPI as an inferior tool? I can't seem to find it looking quickly.

Can anyone explain somewhat succintly what the RPI values, as compared to what Kenpom values, to account for the differences in ranking.

ODU
RPI: 21
Ken: 52

GMU
RPI: 26
Ken: 26

Richmond
RPI: 39
Ken: 46

VCU
RPI: 50
Ken: 84

VT
RPI: 61
Ken: 30

Mason gets the same rating from both (26) and gets an #8 seed (29-32) that seems consistent or close to both the RPI and Kenpom rank.

Richmond seems to exhibit the same sort of consistency: 39 RPI and 46 Kenpom, with a 12 seed (45-48). All in the same neighborhood.

ODU, VCU and VT are wildly different. ODU's RPI is 31 places higher than its Kenpom ranking. VCU's RPI is 34 places higher than its Kenpom ranking. So, the RPI is really favoring ODU an VCU. But the RPI does not favor VT: RPI ranks VT 31 places lower than Kenpom ranks VT.

What does the RPI value so much about ODU's and VCU's resumes?

What does Kenpom value so much about VT's resume?

[Dar95]

The fundamental difference between the RPI and kenpom ratings is that they are designed for different purposes. The RPI is designed to be descriptive, and at a very basic level - whether you won or lost, and how hard it was to win or lose those games. Ken's ratings are designed to be predictive - that is, to give the best estimate of who would win a future game between two teams. So comparing them is, in essence, an apples-to-oranges type of thing, since they aren't attempting to measure the same thing.

[Reilly]

The fundamental difference between the RPI and kenpom ratings is that they are designed for different purposes. The RPI is designed to be descriptive, and at a very basic level - whether you won or lost, and how hard it was to win or lose those games. Ken's ratings are designed to be predictive - that is, to give the best estimate of who would win a future game between two teams. So comparing them is, in essence, an apples-to-oranges type of thing, since they aren't attempting to measure the same thing.

And yet ... both are ranking the same 345 apples. It's not that one is ranking apples and one oranges. And they are both ranking these apples in order from best to worst. So isn't it a closer analysis and shouldn't they track closer?

One person -- let's call him RPI -- takes a small bite out of all 345 apples, and ranks them from 1 to 345 based on taste.

The other person -- let's call him Ken -- measures each apple's circumference, tests its texture for crispness and its skin for tautness, and measures the percentage of sugar or sweetness in a lab, and then ranks the apples from 1 to 345 based on which he believes will taste better.

They are all apples, and they are all being put in order from best to worst. I guess I still don't understand.

[sagegrouse]

And yet ... both are ranking the same 345 apples. It's not that one is ranking apples and one oranges. And they are both ranking these apples in order from best to worst. So isn't it a closer analysis and shouldn't they track closer?

One person -- let's call him RPI -- takes a small bite out of all 345 apples, and ranks them from 1 to 345 based on taste.

The other person -- let's call him Ken -- measures each apple's circumference, tests its texture for crispness and its skin for tautness, and measures the percentage of sugar or sweetness in a lab, and then ranks the apples from 1 to 345 based on which he believes will taste better.

They are all apples, and they are all being put in order from best to worst. I guess I still don't understand.

The RPI harkens to Al Davis ("Just win, baby!") or the NCAA- tournament mantra ("Survive and advance"). All that matters is the W, not the score.

KenPom measures how well a team plays in terms of points made per possession and points allowed per possession. It is far more robust as a measure of how well a team plays. Howsomever, it harkens to the adage: if you want a higher KenPom, make sure you beat the stuffing out of every team you play. Forget your bench; forget your walk-ons; forget easing up on the accelerator with a 30-point lead. And sportsmanship? What is that?

And, let's face it: if coaches' jobs depend on making the tournament, they will do anything reasonable to preserve their extremely high salaries, including winning games by 50 points. And you and I would do the same thing.

sagegrouse
'A reporter asked former Oklahoma coach Billy Tubbs, "What do you say to a coach whom you've just beaten by 80 points?" (He really did beat a team 112-32.) Tubbs: "Can you come back next year?"'

[JasonEvans]

RPI looks at only the following things-- your win percentage, your opponent's win percentage, your opponents opponents win percentage. It does not care how much you won by or where the game was played. So, if Clemson beats Maryland by 1 at home it is the same as Duke beating Maryland by 30 points on the road.

KenPom measures your offensive and defensive efficiency -- how many points you score and how many points you give up. He factors in weightings for where the game was played, the quality of the opponent, and the margin of victory or defeat. There are other things he puts into his ratings too (I believe recentry is a factor). In Ken's rankings, a home win by 1 (which often means the road team is actually the better team) is worth a lot less than a road victory by a lot.

Does that help?

--Jason "other than the lunatics on the selection committee-- most other basketball scholars feel Ken's rankings are vastly superior to RPI" Evans

[CrazieDUMB]

Your analogy is basically right - it's just two different ways to measure how good a basketball team is. RPI is 25% your record, 50% your opponents record, and 25% your opponents' opponents record. Basically, RPI says "who did you beat, and how good are they?"

KenPom uses efficiency, as in "how well does your team put the ball in the basket"? It's scaled by the strength of your opponents, but not nearly to the degree of RPI. I think most people would agree that KenPom is a more complete ranking system because it is more based on how well you play against other teams, and not just in terms of win or lose.

-woops, looks like you guys beat me to it!

[Reilly]

Jason, per wikipedia, home/away is factored in a bit nowadays:

http://en.wikipedia.org/wiki/Ratings_Percentage_Index

Thanks for the replies.

I guess what I'd really like is for some knowledgeable person to go inside the numbers and isolate the numbers, so as to be able to say "VCU is #50 in the RPI b/c they played the following four teams that happened to have tremendously hard schedules themselves .... and VCU is #84 in kenpom because, egad, look at their rebounding efficiency or whatever .... and VT is #30 in kenpom b/c look at their great [insert whatever stat gets them most of the way to #30] ...."

[TNDukeFan]

The bracket I've been in for a couple years wants everything in by 6 PM Tuesday.

cbs online is sticking with Thursday noon -- much to my group's relief.

[Wander]

Too tired to really finish this off with much punch, so I'll just say that if you don't understand Pomeroy you probably shouldn't write so much about him :)

I understand the system very well. It doesn't take into account whether a team wins games. The difference between a 61-60 win and a 63-60 win is the same as the difference between a 61-60 win and a 60-59 loss. That's smart for predictive models like kenpom ratings, but it's absolutely idiotic for selecting teams for a playoff.

Since it was mentioned as a piece of evidence as to why the ACC was screwed: Maryland won less than FSU and Boston College, finished below FSU and Boston College in conference standings, and played what the Pomeroy ratings admit was an easier schedule than either FSU or Boston College. Yet they're ranked higher than either of those teams.

Again, I'd like to repeat that Pomeroy explicitly states that his ratings do not quantify how good a season a team has had.

[Reilly]

... That's smart for predictive models like kenpom ratings, but it's absolutely idiotic for selecting teams for a playoff. ...

Is "absolutely idiotic" maybe overstating things?

The NCAA rewards 31 champions. For the next 37 teams, it seems sane to pick the 37 teams you believe play the best b'ball (kenpom).

Another way to do it is to reward the next 37 teams with the highest WINNING percentage (let's call this the Charlie Sheen model). Who cares who they played or how close the games were -- just win.

The selection committee seems to try to do a little of both above - reward the winning teams, while also looking at who they played (thereby trying to get at the "who plays the best b'ball" model)).

[Wander]

It just seems really obvious philosophically to me that we should be selecting teams based solely on how good of seasons they had. I don't (as you know) advocate the Charlie Sheen model. I absolutely want to look at who teams played and where they played them.

OK, you're right Reilly, I maybe overstated things a bit. As I've said, it's an awesome tool, but I can't tell you how stupid I find it when someone refers to these systems as "reality-based" or when someone says "Team X was screwed because it had a high kenpom ranking and was left out." I mean - Maryland?! How can anyone possibly argue an 8-10 team in the worst ACC in years with one decent non-conference win deserves to be anywhere near the tournament?

[Reilly]

"Based solely on how good of seasons they had" is a loaded phrase and just begs the question of what that phrase means and how is it defined.

Kenpom shows that Maryland has played basketball at such a high level this year that of the 345 Division I teams, Maryland would be predicted to beat 309 of those teams on a neutral court. That's a pretty good season and a model that includes such a team can be said to be "based soley on how good of a season" the team had.

A Charlie Sheen model would have the advantage of being objective. At least it lets folks know the critieria. Either win your conference, or have one of the 37 highest winning percentages of the remaining teams.

What I'd prefer is that we reward the 31 conference champions, then have kenpom pick the next 37 teams. Maryland and VT get in; VCU and somebody else left out. Maybe some other swapping done in the middling ranks.

The current criteria, as it is, is confused. And, leading to improper results, according to some (Bilas et al).

[COYS]

4) Only one team with a freshman point guard has won the National Championship since at least 1990. That was Syracuse in 2003. Their point guard was Gerry McNamara, who was really playing out of position. They won that year because Carmelo Anthony and Hakim Warrick. Two teams fit the bill this year for a true freshman point guard - Kentucky and Texas. Neither will win it all.



Would Duke's 2001 run to the title not count as including a freshman point guard? I recognize that the team essentially had a double point guard attack with Williams and Duhon, and Williams could handle the ball handling duties in crunch time, etc. Nevertheless, that team definitely started a freshman point guard during the entire tournament and relied on him for significant contributions. Duhon might not fit the bill as a freshman starting point guard leading a team to the title, but I think that in the case of Texas and Ohio State, they have enough veteran help that their freshman point guards aren't necessarily relied on to carry the heavy lifting.

[A-Tex Devil]

Back on point. I like to gamble a little and have found Pomeroy helpful, although not consistently so, in finding edges. I would like to see his formula for determining his predictive scores, and if anyone has it, plese post it. But I usually look at that as well as the general formula where a teams score is:

(Off Eff*Opp Def Eff*Average Tempo)/(Average Eff*100)....

.... with adjustments (0.024 I believe) for home/away. I always get the same spread as the Kenpom prediction, but his predictions often have higher or lower totals and I'm not sure why. I think he may be using a variable number besides 100 in the denominator. If anyone knows, please let me know. That would help a lot!

Anyway, one shouldn't use Kenpom as a betting guide, but the model can point out edges. If Vegas and Kenpom are way different, always do your homework because Vegas is smart and there is generally a reason. His spread and Vegas' spread are probably within 2 points 90% of the time, though.

Using an example, with stats through Sunday's games, Kenpom predicts Duke to beat Hampton by 27, and is adjusting his raw score since it's a "Semi-Home" game for Duke. I really like this game for Duke, and would have expected something more along the line of 25. Another one I really like is BYU only giving 8 to Wofford. However, this is one where you shouldn't look at Kenpom as his model probably hasn't fully adjusted to Davies being out.

[Bluedog]

Not sure if this is the right place for it, but Luke Winn has the best analysis out there:

http://tourney.si.com/2011/03/14/an-all-out-bracket-attack/

He always backs up his assertions with facts and numbers (imagine that?!), including relying heavily on efficiency numbers. Has some nice analysis and graphics of various teams/strategies.

He concludes unequivocally that Duke is in the hardest bracket, with the average of the top 4 teams having a ranking of 7.3. Ohio State and the East is second at 8.3, Southwest is at 9.3, and the Southeast brings up the rear at 11.5, but that doesn't take into consideration how BYU has played since Davies went out, so the SE is even worse than that. I really don't know how Pitt was given such an easy route while Ohio State was hosed.

Winn's Final Four: Ohio State vs. Duke, Notre Dame vs. Pittsburgh

And chooses OSU over Pitt in the finals.

[Kedsy]

I guess what I'd really like is for some knowledgeable person to go inside the numbers and isolate the numbers, so as to be able to say "VCU is #50 in the RPI b/c they played the following four teams that happened to have tremendously hard schedules themselves .... and VCU is #84 in kenpom because, egad, look at their rebounding efficiency or whatever .... and VT is #30 in kenpom b/c look at their great [insert whatever stat gets them most of the way to #30] ...."

Well, I don't have time to get that specific, but one huge hole in the RPI is that even after applying your opponents' winning percentage and your opponents' opponents' winning percentage, it still more or less treats all wins as equivalent, and that's not the case in real life.

For example, if you played twenty games against Utah Valley State (Pomeroy's #258 team and RPI's #251) this year and won every one, it would be much better for your RPI than playing and beating Seton Hall (Pomeroy's #58 team and RPI's #102) twenty times this season. This is because Utah Valley State was 19-11 and Seton Hall was 13-18, and 50% of your RPI comes from your opponent's winning percentage (I am assuming all games were in the same venue for the purposes of this explanation).

True, opponents' opponents' schedule also counts, but only 25%, so if you went 20-0 against Utah Valley, your RPI would be 0.662, but if you went 20-0 against Seton Hall your RPI would be only 0.611. In the current RPI that would be the difference between being #4 in the country in the RPI (just ahead of Duke) vs. being #22 (just ahead of Xavier).

Big difference, right? It's even bigger if you consider that if Seton Hall played Utah Valley State 20 times I assume Seton Hall would win every time. The 20-0 record against Seton Hall is much more impressive than 20-0 against Utah Valley State, but the RPI thinks exactly the opposite.

The RPI is a severely flawed system. This, more than anything else, explains why Pomeroy's ratings diverge so much from the RPI.

[Kedsy]

It just seems really obvious philosophically to me that we should be selecting teams based solely on how good of seasons they had.

The problem is the RPI does not properly judge "how good of seasons" anyone had (see my above post). Maybe Pomeroy doesn't, either, but in my mind it's probably closer.

[Wander]

The problem is the RPI does not properly judge "how good of seasons" anyone had (see my above post). Maybe Pomeroy doesn't, either, but in my mind it's probably closer.

Let me make it explicitly clear that I'm not defending the RPI. I agree that it's completely flawed, for the reasons you listed.

[Reilly]

... The RPI is a severely flawed system. This, more than anything else, explains why Pomeroy's ratings diverge so much from the RPI.

Thanks. So where is the movement coming from to overthrow the use of RPI? If it is severely flawed, *somebody* with power must understand that, right?

My sense is that the "sabermetrics" movement in b'ball is gaining more and more ground, so there should be some sort of building movement to get things right, esp. when coupled with Bilas using his soapbox to say how stupid the committee is.

It's not like the committee is evil (Seth Greenberg's maybe theory) ... rather, part (a large part?) of the problem is that it is making decisions based on a severely flawed ranking system.

[Dar95]

"Kenpom shows that Maryland has played basketball at such a high level this year that of the 345 Division I teams, Maryland would be predicted to beat 309 of those teams on a neutral court. That's a pretty good season and a model that includes such a team can be said to be "based soley on how good of a season" the team had.



Wait, do you actually think Maryland should be in the tournament?

While I love kenpom, and rely on his stats and information significantly, that doesn't mean that his numbers are empirically correct. In fact, a deeper look at Maryland suggests that their efficiency ratings are propped up by a large number of major blowouts against crappy teams (they had 6 nc wins by at least 29 points, including two 50+). Their overall kenpom rating is .8836, but if you look at only ACC games, it is .5028. Compare that to Duke, who overall is at .9750, but in ACC games only falls to .8836.

While I think it may make sense to have efficiency ratings such as kenpom as part of the evaluation process, having a pure objective measure for who gets in and who doesn't (or, even, where teams are seeded) has a lot of significant holes.

[Kedsy]

Thanks. So where is the movement coming from to overthrow the use of RPI? If it is severely flawed, *somebody* with power must understand that, right?

One would think so. But these are the same people who decided to use the BCS system to decide the college football champion, so you never know.

[wilson]

While I think it may make sense to have efficiency ratings such as kenpom as part of the evaluation process, having a pure objective measure for who gets in and who doesn't (or, even, where teams are seeded) has a lot of significant holes.This, I think, is an important point. It's good to use quantitative measures as part of the selection process, but there are always means to twist numbers to suit the point you want to make. I strongly believe that there always needs to be a human element involved in the process. Look at the BCS...after several years of purely number-driven selection, they admitted the flaw in their system and threw considerable weight back to the "human polls."
I think people's complaint stems in part from the fact that the human element seems to be erratic. The selection factors they emphasize seem to change year by year, and this year, with regard to several curious selections and non-selections, they have been recalcitrant in identifying any specific factor as to why they chose the teams they did. It's not hard to see why a guy like Greenberg would claim that the process is political. I think his protests are over-zealous, but I also don't think he's totally wrong. By refusing to at least further clarify selection criteria (I don't think it's possible, or even prudent, to determine explicit, "official" requirements for inclusion in the NCAAT), the committee makes it too easy for themselves to choose erratically and erroneously, and to open themselves up to a host of critics, many of whom have legitimate beefs.

[Kedsy]

If it is severely flawed, *somebody* with power must understand that, right?

On the other hand, it's hard to blame the NCAA people for not understanding when the talking heads at ESPN (and elsewhere) don't really seem to get it, either. It drives me crazy when I hear Lunardi or whoever talking about a team's RPI and then mentioning the team's "schedule strength" as if that's a supporting piece of data, when the latter is 50% of the former. Not only that, in my above example, if you played Utah Valley State every game this year, your schedule strength of 0.6333 (UVS's winning percentage) would be third best in the nation. Which truly shows how ridiculous it is to talk about "schedule strength."

Note also that I realize that if you played Utah Valley 20 times and won every time then their record wouldn't be 19-11 anymore (I suppose it would be 19-31, so all of a sudden your RPI would plummet -- oh no!). But for the purposes of my examples, let's all assume there are 20 Utah Valley's out there and you beat each once.

[-jk]

I'm certainly not a numbers guy, so take this with a grain of salt.

Take Maryland, for example. It's axiomatic that different systems will generate different results. And Maryland had some different results. The three most widely cited systems, RPI, Sagarin, and KenPom, all had them with similar strength of schedule, but very differing rankings.

Realtime RPI (http://realtimerpi.com/rpi_Men.html) has them at 95 with SOS of 73
Sagarin (http://www.usatoday.com/sports/sagarin/bkt1011.htm) has them at 50 with SOS of 66
KenPom (http://kenpom.com/) has them at 36 with SOS of 64

Note that KenPom also has them as one of the unluckiest teams (328); for whatever reason, his system couldn't account for their performance.

I also find interesting that the Massey (http://masseyratings.com/cb/compare.htm) ratings composite (40 systems and two human polls) has Maryland at 63 but with a standard deviation of +/-19 places. Out of the top 100 teams, only 11 are more scattered and all but one are ranked lower. Maryland simply defied prediction this year.

On the other hand, VaTech has a Massey composite of 42, +/- 9. They simply got punked. I suppose turnabout's fair play. :D

-jk

[mgtr]

Since much of the discussion boils down to RPI vs. KenPom, so I wonder if someone has taken all the teams from last year's tournament and compared their RPI and KenPom (as of selection Sunday) to the actual tournament result? Not scientific or statistically significant, but I would find it interesting.

[Reilly]

Wait, do you actually think Maryland should be in the tournament?

While I love kenpom, and rely on his stats and information significantly, that doesn't mean that his numbers are empirically correct. In fact, a deeper look at Maryland suggests that their efficiency ratings are propped up by a large number of major blowouts against crappy teams (they had 6 nc wins by at least 29 points, including two 50+). Their overall kenpom rating is .8836, but if you look at only ACC games, it is .5028. Compare that to Duke, who overall is at .9750, but in ACC games only falls to .8836.

While I think it may make sense to have efficiency ratings such as kenpom as part of the evaluation process, having a pure objective measure for who gets in and who doesn't (or, even, where teams are seeded) has a lot of significant holes.

I'm not advocating for or against Maryland. My portion of this discussion started with trying to isolate what exactly about VCU's resume the RPI loved so much, and what about VT's resume the RPI did not love so much, and what about VT Kenpom loved so much, what about VCU Kenpom did not love so much. You have helped that part of the discussion by showing why Kenpom seems to show Maryland so much love: blowouts against non-conference teams.

I agree with you it makes sense to have efficiency ratings as part of the evaluation process. From what I gather, currently the committee is using a rating (RPI) that is pretty whacked and that's not really tethered to b'ball success in ways that a lot of b'ball folks would like to see.

I'm not sure that having a purely objective measure would have a lot of significant holes as compared to an ideal or as compared to the current human element. Doesn't it all depend on what that objective measure is made up of, and what the humans are using?

If the purely objective measure is the RPI, then that looks like it has a lot of holes. But if the purely objective measure is kenpom -- just print his list, scratch out the 31 conference champs, and take the top 37 that remain -- what sort of field does that yield? It might be one that leads you to believe there were not a lot of holes in the selection process.

When you have 31 conferences and 345 teams, that leads me to believe we need more objective criteria, not less. We're trying to compare the 4th place ACC team with the 4th place CAA team with the 3d place A10 team with the 3d place WCC team .... I don't know that humans are real good at that, and I want more objective criteria, not less, and want that objective criteria to be better than what is seemingly being used nowadays.

[Lennies]

Who? Joseph? He's not really a point. They're basically Balbay and Lucas at point. Joseph's a really good player though. Smart, smooth, poised, but doesn't try and do too much. Basically the anti-freshman Jordan Hamilton.

No, I was thinking of J'Covan Brown. For some reason, I thought he was a freshman. In any case, Texas is not reliant on a freshman point guard.

[Kedsy]

Since much of the discussion boils down to RPI vs. KenPom, so I wonder if someone has taken all the teams from last year's tournament and compared their RPI and KenPom (as of selection Sunday) to the actual tournament result? Not scientific or statistically significant, but I would find it interesting.

I agree it's not statistically significant with all the upsets in a one-and-done tournament. Also, for this analysis you can't use the Pomeroy ratings on his website because they are post-tournament ratings. However, I happen to have the pre-tournament ratings and this is what I can tell you:

GAMES IN WHICH POMEROY DISAGREED WITH THE SEEDING IN 2010
-----------------------------------------------------------------
Northern Iowa (9 seed; .9090) over UNLV (8 seed; .8969) -- WIN FOR POMEROY
Georgia Tech (10; .9187) over Oklahoma State (7; .8845) -- WIN
Florida State (9; .8736) over Gonzaga (8; .8637) -- LOSS
Xavier (6; .9284) over Pittsburgh (3; .9147) -- WIN
BYU (7; .9573) over Kansas State (2; .9546) -- LOSS
Washington (11; .9157) over New Mexico (3; .8821) -- WIN
Utah State (12; .9301) over Texas A&M (5; .9253) -- LOSS
Old Dominion (11; .9070) over Notre Dame (6; .8957) -- WIN
St. Mary's (10; .8914) over Richmond (7; .8792) -- WIN


So, when Pomeroy disagreed with the seeding, Pomeroy called the upsets correctly 6 times and incorrectly 3 times. There were a whole bunch more upsets that Pomeroy's ratings didn't envision.

[superdave]

The point has been made about how the RPI does not care if you win by 1 or 30, but another issue along those lines regarding KenPom is that a team can have a great January followed by a lousy February followed by a conference tournament championship. Which teams shows up in the NCAAs?

That's where you use KenPom as a part of how you fill out your brackets, not as the absolute chooser.

I read through Nate Silver's model and got excited, then looked at the brackets his model produced and it was all higher seeds advancing. I understand why the model produces that sort of outcome, but you have to look at the 55%-45% odds games and pick some upsets.

Applying that logic to Texas, they lost some bad games this year (Nebraska, SoCal, Colorado) and won some nice ones (UConn, at Kansas (!), Unc). Does Texas have a bad loss to Oakland in them? By all accounts, Oakland sounds good but are #69 in the Pom. So I say Texas wins.

Does Texas have loss to a good Arizona team in them? Zona is #25 in the Pom...I say yes and picked Arizona. But anyone picking Texas to the Final Four should consider this - can they beat the #69, #25 and #2 (Duke) teams in the KenPom in a row? I can see Texas winning a big game or two, but they have to get there first. And I would not bet on Rick Barnes to get there based on some of these bad losses he's had, much less to beat increasingly good teams each outing. Their inconsistency tells me that KenPom underestimates their downside somehow. My guess is some of their wins are over-shadowing some of their letdowns and screwups more so than it should.

[mgtr]

I agree it's not statistically significant with all the upsets in a one-and-done tournament. Also, for this analysis you can't use the Pomeroy ratings on his website because they are post-tournament ratings. However, I happen to have the pre-tournament ratings and this is what I can tell you:

GAMES IN WHICH POMEROY DISAGREED WITH THE SEEDING IN 2010
-----------------------------------------------------------------
Northern Iowa (9 seed; .9090) over UNLV (8 seed; .8969) -- WIN FOR POMEROY
Georgia Tech (10; .9187) over Oklahoma State (7; .8845) -- WIN
Florida State (9; .8736) over Gonzaga (8; .8637) -- LOSS
Xavier (6; .9284) over Pittsburgh (3; .9147) -- WIN
BYU (7; .9573) over Kansas State (2; .9546) -- LOSS
Washington (11; .9157) over New Mexico (3; .8821) -- WIN
Utah State (12; .9301) over Texas A&M (5; .9253) -- LOSS
Old Dominion (11; .9070) over Notre Dame (6; .8957) -- WIN
St. Mary's (10; .8914) over Richmond (7; .8792) -- WIN


So, when Pomeroy disagreed with the seeding, Pomeroy called the upsets correctly 6 times and incorrectly 3 times. There were a whole bunch more upsets that Pomeroy's ratings didn't envision.

Thanks. There is a great wealth of info stored on members' computers, apparently. To paraphrase Arte Johnson - interesting, very interesting. But not conclusive.

[CDu]

RPI looks at only the following things-- your win percentage, your opponent's win percentage, your opponents opponents win percentage. It does not care how much you won by or where the game was played. So, if Clemson beats Maryland by 1 at home it is the same as Duke beating Maryland by 30 points on the road.

KenPom measures your offensive and defensive efficiency -- how many points you score and how many points you give up. He factors in weightings for where the game was played, the quality of the opponent, and the margin of victory or defeat. There are other things he puts into his ratings too (I believe recentry is a factor). In Ken's rankings, a home win by 1 (which often means the road team is actually the better team) is worth a lot less than a road victory by a lot.

Does that help?

--Jason "other than the lunatics on the selection committee-- most other basketball scholars feel Ken's rankings are vastly superior to RPI" Evans

You are basically correct, but I think the bolded part is not quite accurate. RPI used to not care about home/road, but now there's a weighting for results at home versus on the road. Home wins boost your value less than road wins, and home losses hurt your value more than road losses. There were reasons for this, but it appears that the mid-majors have benefited more from this because they play more OOC road games, and they play weaker in-conference road games (meaning more chances for bonus points for road wins).

So the tweak I'd say is that a 1pt win at home is no different than a 30pt win at home.

[CDu]

Jason, per wikipedia, home/away is factored in a bit nowadays:

http://en.wikipedia.org/wiki/Ratings_Percentage_Index

Thanks for the replies.

I guess what I'd really like is for some knowledgeable person to go inside the numbers and isolate the numbers, so as to be able to say "VCU is #50 in the RPI b/c they played the following four teams that happened to have tremendously hard schedules themselves .... and VCU is #84 in kenpom because, egad, look at their rebounding efficiency or whatever .... and VT is #30 in kenpom b/c look at their great [insert whatever stat gets them most of the way to #30] ...."

VCU is #84 in KenPom because they have a very mediocre (#143) defensive efficiency (adjusted for level of competition) and not that good an offense (#59). VT has a similarly efficient offense (#61) but a MUCH better defense (#25). Hence, they are rated much higher by Pomeroy.

It's harder to explain the RPI stuff because people don't break out the detail. VT played a slightly tougher schedule (according to the RPI), and didn't win quite as many games. I suspect that VCU probably won a few more road games (since they played inferior competition on the road), which is valued more by the RPI.

[tieguy]

Since much of the discussion boils down to RPI vs. KenPom, so I wonder if someone has taken all the teams from last year's tournament and compared their RPI and KenPom (as of selection Sunday) to the actual tournament result? Not scientific or statistically significant, but I would find it interesting.

Nate Silver's gigantic "how we picked" article said that with his data set (don't recall how many years back it went) Pomeroy got it right 74% of the time, which was more than anyone else, but not by a huge margin (e.g., higher seeds win 71% of the time, so that is almost as good a predictor as Pomeroy.)

Exact numbers are approximate, since I haven't looked up the article, but it was something like that- Pomeroy better, but only by a small margin. (This doesn't answer your exact question, because Silver took it for granted that RPI was useless and ignored it, but gives you some sense of where things stand.)

[Dar95]

But if the purely objective measure is kenpom -- just print his list, scratch out the 31 conference champs, and take the top 37 that remain -- what sort of field does that yield?


For the record, if we did this, we would add to the current field:
Virginia Tech
Maryland
New Mexico
St. Mary's
Nebraska

and lose:
VCU
Georgia
UAB
Tennessee
UCLA

My biggest concern with going purely objective - even assuming we had the "perfect" metric that told us exactly which team was better over the course of the season - is that it doesn't take into consideration factors like injuries, suspensions, who's playing worse or better at the end of the year, etc.

[Reilly]

... VT has a ... MUCH better defense (#25). Hence, they are rated much higher by Pomeroy. ... I suspect that VCU probably won a few more road games (since they played inferior competition on the road), which is valued more by the RPI.

Thanks for the breakdown. So, do we want a team that plays tough defense, or one that beat some weak sisters on the road? This is where the human element gets me: what if Duke had taken care of business in 2007 in Buffalo against VCU ... then this year's VCU and this year's VT have the same resumes. Does VCU get in? In other words, is Greg Paulus's defense on Eric Maynor the source of Seth Greenberg's heartburn?

[pfrduke]

Thanks for the breakdown. So, do we want a team that plays tough defense, or one that beat some weak sisters on the road? This is where the human element gets me: what if Duke had taken care of business in 2007 in Buffalo against VCU ... then this year's VCU and this year's VT have the same resumes. Does VCU get in? In other words, is Greg Paulus's defense on Eric Maynor the source of Seth Greenberg's heartburn?

I would be willing to wager a large amount of money that VCU's 2007 NCAA tournament performance had absolutely nothing to do with their selection this season. A very large amount of money.

[Reilly]

For the record, if we did this, we would add to the current field:
Virginia Tech
Maryland
New Mexico
St. Mary's
Nebraska

and lose:
VCU
Georgia
UAB
Tennessee
UCLA
....

Interesting that Colorado, who folks were clamoring for, are not one of the "Kenpom 37".

[pfrduke]

In other words, is Greg Paulus's defense on Eric Maynor the source of Seth Greenberg's heartburn?

Also, if you're referring to the last play of the game, it was Scheyer who was on Maynor. (If you're referring to the game as a whole, Paulus is probably right).

[toooskies]

The biggest argument against the RPI is not that it only rewards winning and losing-- it's that even at that task, it's bad. For instance, if you win against a bad team, your RPI goes down. One could argue that the largest contributor to the ACC's weak RPIs this year was that everyone had to play Wake Forest (and GT/UVA/NC State also produced historically bad RPI ranks for the bottom teams in-conference). Virginia Tech had to play Wake twice, GT three times, and UVA twice. That's 7 games that are usually around RPI 80 that all ended up being well over 100. Shame on VT for losing two of those games, but still-- the ACC schedule wasn't very RPI-friendly to them this year, nor has it ever been.

In addition, you can game the system, and Coach K schedules to do exactly that. A tournament-caliber team is pretty much guaranteed to beat teams in the 100 to 150 range. Scheduling more of those games (instead of bottom-100 cream puffs) are barely more risky when you talk about winning or losing, but they reward you significantly in the RPI.

KenPom isn't measuring the same things, as has been beaten to death-- it's purely a margin-of-victory, "how much should you have won" ranker, versus a "how much did you win" ranker. But, they do show the corollary "Luck" calculation, which is how much your wins and losses deviate from what was expected with your schedule. This year, VT has the least-negative Luck of any worst-efficiency tournament team. What that means is, they did about as good as they were supposed to (slightly worse, but not to a significant degree). Luck isn't so much what the traditional definition of the word means, as much as it means "win when it's close, don't have many blowout wins/losses."

But we need a single composite number to make "real" rankings. I think Sagarin's ELO_CHESS model is significantly better, which is based on the rankings system used for international chess competition (although it has been modified greatly). It's purely a system to reward winning and penalize losing, without margins. In Sagarin's ratings, VPI gets a 44 rating while VCU drops to 61. Which is about where most people perceive them to be, instead of the VPI 62 / VCU 49 RPIs which led to the committee's terrible decision with those two. UAB falls from 31 to 46, Colorado jumps from 65 to 53, Clemson climbs from being very much on the bubble to safely in, and so on. You get some different inflated teams (Harvard and Princeton stick out) but you can still argue that those teams had very successful seasons (despite talent), which you can't say about the VCUs of the world.

That doesn't mean ELO can't be gamed-- SDSU seems to be overrated, but on the other hand, they've only lost to a single, highly-rated team all year. Duke seems to be underrated, but we're very much bunched with teams of similar quality (ND, BYU, Pitt, etc). And, by looking at KenPom, Duke didn't "overachieve" nearly to the degree that OSU/Kansas/SDSU may have.

In summary:
- The RPI is bad, and is a bad guideline for perceived strength of win/losses
- VPI can yell at WF, GT, and their losses to UVA for their terrible RPI
- ELO is better, but not perfect
- KenPom is better at prediction, roughly equivalent to "eye test" tie-breaking factors

[Reilly]

I would be willing to wager a large amount of money that VCU's 2007 NCAA tournament performance had absolutely nothing to do with their selection this season. A very large amount of money.

OK, we have a bet. Let's say $38,000. Payable in small bills. Of course we'll never know for sure to be able to prove it one way or the other to collect. (I refuse to accept the chairman of the committee's non-answers as to what they considered as proof, thank you.) I would think it plays at least a smidgen role. As in, "see, these mid-majors can hang, look what VCU did to Duke" or "look what Northern Iowa did to Kansas." Everytime a non-BCS team wins, that gets filed in the collective unconscious, I would think. Further, it might be OK to get filed there. If in two years Northern Iowa has a middling resume, and is up against a BCS conference 5th place team, you don't think it enters into the equation at all?

[patentgeek]

The biggest argument against the RPI is not that it only rewards winning and losing-- it's that even at that task, it's bad. For instance, if you win against a bad team, your RPI goes down. One could argue that the largest contributor to the ACC's weak RPIs this year was that everyone had to play Wake Forest (and GT/UVA/NC State also produced historically bad RPI ranks for the bottom teams in-conference). Virginia Tech had to play Wake twice, GT three times, and UVA twice. That's 7 games that are usually around RPI 80 that all ended up being well over 100. Shame on VT for losing two of those games, but still-- the ACC schedule wasn't very RPI-friendly to them this year, nor has it ever been.

In addition, you can game the system, and Coach K schedules to do exactly that. A tournament-caliber team is pretty much guaranteed to beat teams in the 100 to 150 range. Scheduling more of those games (instead of bottom-100 cream puffs) are barely more risky when you talk about winning or losing, but they reward you significantly in the RPI.

KenPom isn't measuring the same things, as has been beaten to death-- it's purely a margin-of-victory, "how much should you have won" ranker, versus a "how much did you win" ranker. But, they do show the corollary "Luck" calculation, which is how much your wins and losses deviate from what was expected with your schedule. This year, VT has the least-negative Luck of any worst-efficiency tournament team. What that means is, they did about as good as they were supposed to (slightly worse, but not to a significant degree). Luck isn't so much what the traditional definition of the word means, as much as it means "win when it's close, don't have many blowout wins/losses."

But we need a single composite number to make "real" rankings. I think Sagarin's ELO_CHESS model is significantly better, which is based on the rankings system used for international chess competition (although it has been modified greatly). It's purely a system to reward winning and penalize losing, without margins. In Sagarin's ratings, VPI gets a 44 rating while VCU drops to 61. Which is about where most people perceive them to be, instead of the VPI 62 / VCU 49 RPIs which led to the committee's terrible decision with those two. UAB falls from 31 to 46, Colorado jumps from 65 to 53, Clemson climbs from being very much on the bubble to safely in, and so on. You get some different inflated teams (Harvard and Princeton stick out) but you can still argue that those teams had very successful seasons (despite talent), which you can't say about the VCUs of the world.

That doesn't mean ELO can't be gamed-- SDSU seems to be overrated, but on the other hand, they've only lost to a single, highly-rated team all year. Duke seems to be underrated, but we're very much bunched with teams of similar quality (ND, BYU, Pitt, etc). And, by looking at KenPom, Duke didn't "overachieve" nearly to the degree that OSU/Kansas/SDSU may have.

In summary:
- The RPI is bad, and is a bad guideline for perceived strength of win/losses
- VPI can yell at WF, GT, and their losses to UVA for their terrible RPI
- ELO is better, but not perfect
- KenPom is better at prediction, roughly equivalent to "eye test" tie-breaking factors

There was an interesting article posted on College Basketball Prospectus last summer that introduced another concept into ranking teams. Some of you may be familiar with Win Probabilities, which are essentially graphs that measure a team's probability of winning a particular game as the game is going on. This article proposed using Win Probabilities as a measurement tool for rankings, reasoning that it could eliminate some of the oddities that can occur with pure point-based systems.

http://www.basketballprospectus.com/article.php?articleid=1183

The author concedes that he doesn't know exactly how to implement this system, but it's interesting food for thought for the dork poll fans.

[pfrduke]

OK, we have a bet. Let's say $38,000. Payable in small bills. Of course we'll never know for sure to be able to prove it one way or the other to collect. (I refuse to accept the chairman of the committee's non-answers as to what they considered as proof, thank you.) I would think it plays at least a smidgen role. As in, "see, these mid-majors can hang, look what VCU did to Duke" or "look what Northern Iowa did to Kansas." Everytime a non-BCS team wins, that gets filed in the collective unconscious, I would think. Further, it might be OK to get filed there. If in two years Northern Iowa has a middling resume, and is up against a BCS conference 5th place team, you don't think it enters into the equation at all?

Well, as you say, we'll never be able to know for sure, but I have two general responses. First, VCU in 2007 had a) a different coach, and b) an entirely different roster of players. Quite literally the only similarity between the two teams is the name on the front of the jersey. I could see a one-year effect being considered - something along the lines of "hey, that team with those same players was really good last year, and maybe that haven't quite performed up to the same level this season, but we'll give them the benefit of the doubt." But I really don't think anyone in the room was thinking "VCU beat Duke 4 years ago so we should let them in this year."

Second, there is a lot of evidence, from a lot of different schools, that mid-majors can hang, enough so that I don't think one individual school's performance has carry over effects many years later for tourney admission/seeding purposes.

That said, if it did play a factor, that's a serious abdication of duty by the committee.

[Reilly]

.. That said, if it did play a factor, that's a serious abdication of duty by the committee.

I agree a one-year carryover would be stronger. Also agree other evidence out there. All evidence helps. What gets me is that there is near universal consensus that VCU does not belong. So, some sort of serious abdication of duty by the committee happened, even if that duty is simply to do your job somewhat intelligently. The VCU team did not gather to watch the selection show. Their own coach was surprised. So, what happened? Maybe several errors happened to get them in. And all of this is not to say that VCU won't acquit itself well once in. They may win. Simply that it's near unanimous that other teams would have been stronger based on their resumes.

So what did VCU have going for it - that is, besides its 4th place CAA finish and losing 5 of its last 8 games and its #84 Kenpom placement? It had an RPI of 50, and some name recognition for beating the best college b'ball program in the land in dramatic fashion a couple years ago.

[Udaman]

The biggest problem with Kenpom (and others like it) is the reliance on margin of victory/expected margin of victory. Yes...you should reward a team that wins big when it is supposed to. But when you skew for this with a team that really plays bad teams, it can create a wacky scenario.

Belmont is the perfect example. This is a good team, for sure. They almost won at Tennessee (though it was while the Vols were self-destructing, and the 2nd time they had played them). Their only other big game was at Vanderbilt, where they lost. They also lost to Lipscomb. They almost lost at S. Carolina Upstate (a terrible, terrible team).

Wisconsin on the other hand, beat Purdue, Ohio State, Michigan State, Penn State and Michigan. Still, they enter the tournament after getting blown out and only scoring 33.

This is a huge contrast in styles. Belmont plays 13 guys and runs and runs and runs. Wisconsin slows it down. I think they'll take the air out of the tires, and Belmont will have a brutal time adjusting to that. Then again, the Vegas odds makers have it Wisconsin - 4 1/2, which is by far the smallest line for a 4-13 game.

Still, I think Belmont is getting way too much hype here, and that Wisconsin will win by 9 or more.

[III]

I agree it's not statistically significant with all the upsets in a one-and-done tournament. Also, for this analysis you can't use the Pomeroy ratings on his website because they are post-tournament ratings. However, I happen to have the pre-tournament ratings and this is what I can tell you:

GAMES IN WHICH POMEROY DISAGREED WITH THE SEEDING IN 2010
-----------------------------------------------------------------
Northern Iowa (9 seed; .9090) over UNLV (8 seed; .8969) -- WIN FOR POMEROY
Georgia Tech (10; .9187) over Oklahoma State (7; .8845) -- WIN
Florida State (9; .8736) over Gonzaga (8; .8637) -- LOSS
Xavier (6; .9284) over Pittsburgh (3; .9147) -- WIN
BYU (7; .9573) over Kansas State (2; .9546) -- LOSS
Washington (11; .9157) over New Mexico (3; .8821) -- WIN
Utah State (12; .9301) over Texas A&M (5; .9253) -- LOSS
Old Dominion (11; .9070) over Notre Dame (6; .8957) -- WIN
St. Mary's (10; .8914) over Richmond (7; .8792) -- WIN


So, when Pomeroy disagreed with the seeding, Pomeroy called the upsets correctly 6 times and incorrectly 3 times. There were a whole bunch more upsets that Pomeroy's ratings didn't envision.

Sorry, but this is just not the right way to interpret this. It’s similar to the conference record predictions. At the beginning of the season, Pomeroy calculated that Duke had a better than 50% chance of winning each conference game, but he did not predict that Duke would finish 16-0 in conference.

Similarly, Pomeroy did not correctly or incorrectly predict any of those 9 upsets; he simply calculated which games had a better than 50% chance of being upsets. The fact that 67% turned out to be upsets provides a small sample size of evidence to those calculations having been good predictors. Note, it would have been just as bad an indicator for Pomeroy’s predictions if all 9 of the games had been upsets or if 2-3 had been. 9 “correct” upset calls would have actually been evidence that Pomeroy’s numbers were wrong, not right. Further, the upsets Pomeroy “didn’t predict” were in fact predicted, just not specifically. If there were (for simple math) 12 games that had a 25% probability of upsets, and 3 of the 12 occurred, than Pomeroy predicted the upset probabilities well.

[toooskies]

There was an interesting article posted on College Basketball Prospectus last summer that introduced another concept into ranking teams. Some of you may be familiar with Win Probabilities, which are essentially graphs that measure a team's probability of winning a particular game as the game is going on. This article proposed using Win Probabilities as a measurement tool for rankings, reasoning that it could eliminate some of the oddities that can occur with pure point-based systems.

http://www.basketballprospectus.com/article.php?articleid=1183

The author concedes that he doesn't know exactly how to implement this system, but it's interesting food for thought for the dork poll fans.

I don't think that taking the per-possession margin makes the most sense, for precisely the reservations he expressed. However, KenPom does already calculate "leverage", the percent of change a successful possession would have on the chance to win the game. KP made a defense of keeping players out of the game with a high number of fouls, in that the last few possessions of a close game are the most important; they affect the outcome of the game more directly than other possessions.

Now, KP doesn't put a firm number on this, possibly because he's doing the math to figure out exactly what's going on. What is clear, however, is that the results of a basketball game do have a degree of statistical randomness, i.e. teams can play each other multiple times and have unpredictable results.

There's a lot of equality in the middle, though, and it obscures what is really being picked here. In either RPI or ELO, there is less difference between the 40th and 70th ranked teams in terms of the absolute RPI/ELO "score" than there is between the 1st and 5th. What that tells you is that, ultimately, the committee made more of an error putting Florida at a 2 seed than any of the bubble choices.

[tbyers11]

The biggest problem with Kenpom (and others like it) is the reliance on margin of victory/expected margin of victory. Yes...you should reward a team that wins big when it is supposed to. But when you skew for this with a team that really plays bad teams, it can create a wacky scenario.

Belmont is the perfect example. This is a good team, for sure. They almost won at Tennessee (though it was while the Vols were self-destructing, and the 2nd time they had played them). Their only other big game was at Vanderbilt, where they lost. They also lost to Lipscomb. They almost lost at S. Carolina Upstate (a terrible, terrible team).

Wisconsin on the other hand, beat Purdue, Ohio State, Michigan State, Penn State and Michigan. Still, they enter the tournament after getting blown out and only scoring 33.

This is a huge contrast in styles. Belmont plays 13 guys and runs and runs and runs. Wisconsin slows it down. I think they'll take the air out of the tires, and Belmont will have a brutal time adjusting to that. Then again, the Vegas odds makers have it Wisconsin - 4 1/2, which is by far the smallest line for a 4-13 game.

Still, I think Belmont is getting way too much hype here, and that Wisconsin will win by 9 or more.

Agree with you that margin of victory against really bad teams should be taken with a grain of salt. However, Pomeroy does this to an extent by adjusting his raw offensive and defensive efficiencies to strength of schedule.

You can see here (http://kenpom.com/summary.php?s=RankOE) that Belmont has the #9 raw OE and #2 raw DE in the country. When adjusted for strength of schedule, they drop to #32 Adj OE and #19 Adj DE, respectively. So their number 18 overall Pomeroy ranking has been adjusted for Belmont's weak schedule.

[loran16]

The biggest problem with Kenpom (and others like it) is the reliance on margin of victory/expected margin of victory. Yes...you should reward a team that wins big when it is supposed to. But when you skew for this with a team that really plays bad teams, it can create a wacky scenario.

Belmont is the perfect example. This is a good team, for sure. They almost won at Tennessee (though it was while the Vols were self-destructing, and the 2nd time they had played them). Their only other big game was at Vanderbilt, where they lost. They also lost to Lipscomb. They almost lost at S. Carolina Upstate (a terrible, terrible team).

Wisconsin on the other hand, beat Purdue, Ohio State, Michigan State, Penn State and Michigan. Still, they enter the tournament after getting blown out and only scoring 33.

This is a huge contrast in styles. Belmont plays 13 guys and runs and runs and runs. Wisconsin slows it down. I think they'll take the air out of the tires, and Belmont will have a brutal time adjusting to that. Then again, the Vegas odds makers have it Wisconsin - 4 1/2, which is by far the smallest line for a 4-13 game.

Still, I think Belmont is getting way too much hype here, and that Wisconsin will win by 9 or more.

The reason why Belmont is hyped is because scoring margin DOES matter if you want to best predict future performance. Put it this way: through non-conference play, you can get a decent measure of how well teams compare to each other, since you have essentially a web of connections (Team A plays B, who played C, D and E, who each played....etc.). That gives you how good certain teams are going into conference play.

Now if you put a hypothetical team into a bad conference, lets call them the 2010 Cutler Culldogs, of course that team is going to tend to win either every game or almost every game (upsets will happen by flukes). So how can you tell if they're really good...since you know that the rest of the conference is bad? The answer is scoring margin. The system (Pomeroy) knows that the ASun is a bad conference, but a good real team would thus beat up on the conference...and that's what Belmont did.

The system is not foolproof. Pomeroy admits that in such cases (one team kills rest of conference), it's hard to tell if the rankings are overrating that team. But as Butler proved last year, scoring margin can show that a team is in fact a really good contender (I don't recall Butler's pre-tourney ranking last year, but I imagine it was similar to Belmont's this year).

[snowdenscold]

The RPI harkens to Al Davis ("Just win, baby!") or the NCAA- tournament mantra ("Survive and advance"). All that matters is the W, not the score.

KenPom measures how well a team plays in terms of points made per possession and points allowed per possession. It is far more robust as a measure of how well a team plays. Howsomever, it harkens to the adage: if you want a higher KenPom, make sure you beat the stuffing out of every team you play. Forget your bench; forget your walk-ons; forget easing up on the accelerator with a 30-point lead. And sportsmanship? What is that?

And, let's face it: if coaches' jobs depend on making the tournament, they will do anything reasonable to preserve their extremely high salaries, including winning games by 50 points. And you and I would do the same thing.

sagegrouse'

Sagregrouse hits on a great point. Do we really want to see the behavior changes which would naturally occur if KenPom ratings were used? It's a double-edged sword. The results would be skewed by feedback if used.

I'm torn because obviously it seems like a better system go use. However, I'd rather the behavior adjustments that go w/ RPI (e.g. don't play teams who are absolutely terrible) than w/ KemPom (e.g. throw sportsmanship out the window to inflate margins).

[pfrduke]

IWhat gets me is that there is near universal consensus that VCU does not belong. So, some sort of serious abdication of duty by the committee happened, even if that duty is simply to do your job somewhat intelligently. The VCU team did not gather to watch the selection show. Their own coach was surprised. So, what happened? Maybe several errors happened to get them in. And all of this is not to say that VCU won't acquit itself well once in. They may win. Simply that it's near unanimous that other teams would have been stronger based on their resumes.

So what did VCU have going for it - that is, besides its 4th place CAA finish and losing 5 of its last 8 games and its #84 Kenpom placement? It had an RPI of 50, and some name recognition for beating the best college b'ball program in the land in dramatic fashion a couple years ago.

I think there's a difference between making a bad decision/wrong decision and abdicating duty. I think the decision to choose Virginia Commonwealth was bad, and wrong. That said, there are things that VCU had going for it that the committee does consider, and has said it considered: RPI - better than VT or Colorado; Road record - 8-6, being more wins and a better percentage than VT or Colorado; RPI top 25 wins, 2; RPI top 50 wins, 3. You can debate the merits of these things being considered - I think several of them are not worthwhile as an analytical tool, but one could offer an explanation based on the committee's factors and understand why VCU got included, even if one disagreed with the decision. A lengthy discussion sort of in favor of VCU can be found here (http://joesheehanbaseball.blogspot.com/2011/03/from-345-to-68-epilogue.html).

If, however, they based their decision on what another VCU team did in the NCAA tournament 4 years ago, that would be a wholesale departure from any stated criteria. That, to me, would be an abdication of duty - considering things you don't say you consider and that no rational person could defend considering because they have precisely zero to do with the team's performance this season - in order to pick one team over another.

[Kedsy]

Sorry, but this is just not the right way to interpret this. It’s similar to the conference record predictions. At the beginning of the season, Pomeroy calculated that Duke had a better than 50% chance of winning each conference game, but he did not predict that Duke would finish 16-0 in conference.

Similarly, Pomeroy did not correctly or incorrectly predict any of those 9 upsets; he simply calculated which games had a better than 50% chance of being upsets. The fact that 67% turned out to be upsets provides a small sample size of evidence to those calculations having been good predictors. Note, it would have been just as bad an indicator for Pomeroy’s predictions if all 9 of the games had been upsets or if 2-3 had been. 9 “correct” upset calls would have actually been evidence that Pomeroy’s numbers were wrong, not right. Further, the upsets Pomeroy “didn’t predict” were in fact predicted, just not specifically. If there were (for simple math) 12 games that had a 25% probability of upsets, and 3 of the 12 occurred, than Pomeroy predicted the upset probabilities well.

I suppose. I apologize for speaking in shorthand. What I guess would be a better way to put it was that Pomeroy's ratings disagreed with the seeding in 9 matchups, and in six of those matchups, the team Pomeroy thought was better in fact won the game.

Incidentally, you are referring to Pomeroy's log5 analysis, and I was referring to his ratings. My post was in response to a question asking how Pomeroy's ratings compared with the seeding (really the RPI, but we made the assumption that the seeds were based on the RPI -- that assumption isn't quite correct, either, if you want to get technical about it). That's why I worded my response the way I did.

Finally, if you really want to get mathematical, 12 is still too small a sample size for you to expect Pomeroy to get 3 of 12 in your example. Just sayin'.

[MChambers]

The reason why Belmont is hyped is because scoring margin DOES matter if you want to best predict future performance. Put it this way: through non-conference play, you can get a decent measure of how well teams compare to each other, since you have essentially a web of connections (Team A plays B, who played C, D and E, who each played....etc.). That gives you how good certain teams are going into conference play.

Now if you put a hypothetical team into a bad conference, lets call them the 2010 Cutler Culldogs, of course that team is going to tend to win either every game or almost every game (upsets will happen by flukes). So how can you tell if they're really good...since you know that the rest of the conference is bad? The answer is scoring margin. The system (Pomeroy) knows that the ASun is a bad conference, but a good real team would thus beat up on the conference...and that's what Belmont did.

The system is not foolproof. Pomeroy admits that in such cases (one team kills rest of conference), it's hard to tell if the rankings are overrating that team. But as Butler proved last year, scoring margin can show that a team is in fact a really good contender (I don't recall Butler's pre-tourney ranking last year, but I imagine it was similar to Belmont's this year).
Don't know whether Belmont is really that good, but they are #2 in defensive turnover percentage, #4 in bench minutes, and #304 in FTA/FGA (which is bad). Makes me think they thrive on putting defensive pressure on the other team. That could work well against the right team (think UNC, for example) but very poorly against a team with a talented and seasoned backcourt.

Anyway, I agree that Pomeroy's method should not be ignored by the selection committee.

[loran16]

Don't know whether Belmont is really that good, but they are #2 in defensive turnover percentage, #4 in bench minutes, and #304 in FTA/FGA (which is bad). Makes me think they thrive on putting defensive pressure on the other team. That could work well against the right team (think UNC, for example) but very poorly against a team with a talented and seasoned backcourt.

Anyway, I agree that Pomeroy's method should not be ignored by the selection committee.

Oddly, they're #1 in having their own shots blocked. Alas, Wisconsin is merely average at blocking shots (though in an odd paralel, Wisconsin is #2 at NOT having their own shots blocked).

[Kedsy]

(I don't recall Butler's pre-tourney ranking last year, but I imagine it was similar to Belmont's this year).

Butler was 26th in pre-tourney Pomeroy last year.

[jipops]

Interesting link here (http://www.wired.com/playbook/2011/03/wired-guide-march-madness/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+wired%2Findex+%28Wired%3A+Ind ex+3+%28Top+Stories+2%29%29) indicates how some teams in certains rounds of the espn picks are being over-valued and some are under-valued based on Pomeroy and Sagarin.

[throatybeard]

Butler was 26th in pre-tourney Pomeroy last year.

They were like 12 in Sagarin IIRC. I think I looked that up at the time because I was arguing they were underseeded at a 5.

[Kedsy]

They were like 12 in Sagarin IIRC. I think I looked that up at the time because I was arguing they were underseeded at a 5.

Butler was #31 in pre-tournament Sagarin ("Predictor"). They were #18 in Sagarin's ELO-CHESS rating, and #22 overall.

Post-tournament they jumped up to #17 Predictor and #4 ELO-CHESS (#8 overall).

Butler was #12 in the pre-tournament RPI, however. Maybe that's what you looked up?

[throatybeard]

s #12 in the pre-tournament RPI, however. Maybe that's what you looked up?

Crap, yeah, prolly.

[toooskies]

You have to understand, though, that tournaments are effectively a method of inflating ratings beyond their actual value. Imagine a perfectly even field in a 64-team tournament: every team is exactly equal in performance. However, due to randomness (shots going in, the particular bounce of a rebound, referee calls/no-calls, etc.), some teams will necessarily win. In the next round, again with every team equally matched, there still has to be a winner, and same with the next round, all the way to the final. At the end of the tournament, the winner is still equally matched with everyone else, and yet has a 6-0 record on their resume as opposed to first round losers with uninspiring 0-1 records in the tournament.

Possession-based ratings will skew less because there will be many close games between equally matched teams. Only-winning-matters ratings will skew more, because you're essentially assigning a random positive and negative to differences in quality that don't exist.

That's not to take anything away from the value of winning (or coming close to winning) a tournament. But tournaments have negative effects overall on the accuracy of the ratings of their winners.

Which is why you shouldn't pick UConn in the tournament based on their BE Tournament results: the BE has proven that they're a bunch of more-or-less evenly matched teams through their regular season schedule. Someone had to win, and that may just mean that UConn had luck on their side. It's not like they blew out anyone in their last three games.

[tele]

You have to understand, though, that tournaments are effectively a method of inflating ratings beyond their actual value. ... But tournaments have negative effects overall on the accuracy of the ratings of their winners.

.

Well, if you assume the teams are equally matched, why wouldn't the inherent randomness you point to have the same effect during the regular season as for a tourney?

[toooskies]

Well, if you assume the teams are equally matched, why wouldn't the inherent randomness you point to have the same effect during the regular season as for a tourney?

In a round-robin format (similar to a regular season), you aren't enforcing the rule that one team wins all its games. So, you're likely to get a normal distribution (bell-curve) of teams with wins and losses. You don't expect every team to go 3-3, but you don't expect teams to go winless either.

Which is why the per-possession efficiency numbers are better predictors: instead of taking the win/loss evaluation (with less samples it is less predictive), it effectively counts every possession as a win or loss. With roughly 70 more samples than wins and losses, you can see how effective the team is in a general sense.

You aren't measuring the actual objective anymore when you do that-- you play to win the game, not to create the largest likely margin of victory. Winning the close ones is a trait of a good team that doesn't get picked up by possession-based analysis.

[94duke]

In a round-robin format (similar to a regular season), you aren't enforcing the rule that one team wins all its games. So, you're likely to get a normal distribution (bell-curve) of teams with wins and losses. You don't expect every team to go 3-3, but you don't expect teams to go winless either.

Which is why the per-possession efficiency numbers are better predictors: instead of taking the win/loss evaluation (with less samples it is less predictive), it effectively counts every possession as a win or loss. With roughly 70 more samples than wins and losses, you can see how effective the team is in a general sense.

You aren't measuring the actual objective anymore when you do that-- you play to win the game, not to create the largest likely margin of victory. Winning the close ones is a trait of a good team that doesn't get picked up by possession-based analysis.

I thought the "luck" factored it in. Is that not the case?

[Duke: A Dynasty]

WVU beats Clemson 84-76 so congrats to CLemson for making a fight out of it.

Butler beats ODU 60-58

[toooskies]

I thought the "luck" factored it in. Is that not the case?

I think it's the opposite of what you think it is. "Luck" in KenPom isn't added into your rating; it's simply another calculated value. Luck is a calculation of whether you strung your "random" efficiencies together at the right times to beat teams with better efficiencies, and hold off teams with lower efficiencies. In terms of wins and losses, this directly correlates with winning close games, whether it's against good or bad teams.

You'll find that teams with strongly negative luck have done exceptionally poorly in close games. Texas was 2-5 in games decided by 5 points or less. Kentucky was 1-6 in <5 point games. Clemson didn't lose a game by more than 11 all year.

Whether you believe that those "bad in close games" teams are good teams that simply were unlucky, or they have a definite problem in closing out a win is up to you. But if you're worried about Texas, it may be comforting to think that they'll have to beat Duke by 10 to win!

[tieguy]

Surprised not to see this here yet, but Nate Silver's super-nerd probabilities now have us as a very slight favorite to win it all, presumably because of a slight Kyrie bounce.

(BTW, tooskies, excellent discussion of how tournaments skew per-possession numbers.)

[Pacer]

That lead is widening with each completed game. Now 26% to OSU's 18.4%.

[MChambers]

http://www.nytimes.com/2011/03/24/sports/ncaabasketball/24ncaa.html?adxnnl=1&adxnnlx=1300976514-canbxG4D6OGuAmTokBjb8g

Nice article on Pomeroy. Coach K apparently had him on his radio show.

[loran16]

http://www.nytimes.com/2011/03/24/sports/ncaabasketball/24ncaa.html?adxnnl=1&adxnnlx=1300976514-canbxG4D6OGuAmTokBjb8g

Nice article on Pomeroy. Coach K apparently had him on his radio show.

Yep, in 2009. You can find that here: http://xmsatelliteradio.edgeboss.net/download/xmsatelliteradio/talk_content/coach_k/2009/complete_shows/mp3/coach_k_show_02_04_2009_complete_show.mp3?rss_feed id=588

[JasonEvans]

Pomeroy predicts tonight's games as--

BYU 74-73 over Fla
SDSU 66-62 over UConn
Wisconsin 66-59 over Butler
Duke 79-69 over Arizona

There is now no reason to watch the games tonight. These predictions will come true.

--Jason "other than rooting for underdog Butler, I would really enjoy all of these other results" Evans

[toooskies]

Some more dork notes on my analysis of the analysis:

- A big weakness of most of the public analysis tools is there isn't a lot of recency compensation involved. Nate Silver at least tried to compensate for injury, but he's probably overrating Kyrie's potential contributions to our team (not by over-valuing Kyrie per se, but undervaluing what he's replacing.)

- Similarly, Texas hasn't played great lately, which the NCAA's seeding may have reflected. After consistently being among the top 5 defensive teams which teams played, the Texas defense was mediocre for an entire month. Getting "upset" by Arizona wasn't that big of a deal.

- I find it interesting that you have to go all the way back to Syracuse in 2003 to find an NCAA tournament winner with a "luck" score that wasn't very close to zero. I don't have a great explanation for this, although it might be something to effect of, neutral luck is the mark of consistency. Or, it could be because of tournament factors-- a team that wins the tourney is usually a great team, so 6-0 isn't really out of character. Either way, if this stat persuades you to pick favorites for the championship, you have to like the chances of Duke and Wisconsin this year.

[diveonthefloor]

Pomeroy predicts tonight's games as--

BYU 74-73 over Fla
SDSU 66-62 over UConn
Wisconsin 66-59 over Butler
Duke 79-69 over Arizona

There is now no reason to watch the games tonight. These predictions will come true.

--Jason "other than rooting for underdog Butler, I would really enjoy all of these other results" Evans

Well, it's 951 EDT....Pomeroy is 0-2 so far.

[toooskies]

you have to like the chances of Duke and Wisconsin this year.

So much for that prediction.

[crimsonandblue]

Pomeroy predicts tonight's games as--

BYU 74-73 over Fla
SDSU 66-62 over UConn
Wisconsin 66-59 over Butler
Duke 79-69 over Arizona

There is now no reason to watch the games tonight. These predictions will come true.

--Jason "other than rooting for underdog Butler, I would really enjoy all of these other results" Evans

A nerd 0-fer. Ouch babe. I had Flawrida and UConn, unfortunately, I had them losing to Wisky and Duke. Tough way for Duke and Wisky to end the season. Wisconsin uncharacteristically turns the ball over left and right and Duke uncharacteristically gets shredded on defense and then even more uncharacteristically, panics. The tournament's great, except that it sucks.