Duke's RPI and SOS

[sporthenry]

I know some don't love the idea of RPI and SOS but it has become fairly important in terms of seeding. I also created this thread so we could have a place to talk about Duke's opponents from earlier in the year. Minnesota is going to have a big victory at home versus MSU and as of today gives Duke the 3 top 10 wins and UM looks like a force. Those Hollins kids are going to be tough to cover.

As it stands today, on the 31st day of the 2012 year, Duke has the #1 RPI and the #1 SOS (but the #2 SOS in non-confernence games).

With UK seemingly getting their act together a bit versus Louisville and Temple beating Syracuse a bit earlier, Duke's resume is getting better.

[Class of '94]

I know some don't love the idea of RPI and SOS but it has become fairly important in terms of seeding. I also created this thread so we could have a place to talk about Duke's opponents from earlier in the year. Minnesota is going to have a big victory at home versus MSU and as of today gives Duke the 3 top 10 wins and UM looks like a force. Those Hollins kids are going to be tough to cover.

As it stands today, on the 31st day of the 2012 year, Duke has the #1 RPI and the #1 SOS (but the #2 SOS in non-confernence games).

With UK seemingly getting their act together a bit versus Louisville and Temple beating Syracuse a bit earlier, Duke's resume is getting better.

Just out out of curiosity, who has the #1 SOS in non-conference games? It's hard for me to believe (as biased as I am) to believe someone else had a tougher non-conference schedule than Duke.

[sporthenry]

Just out out of curiosity, who has the #1 SOS in non-conference games? It's hard for me to believe (as biased as I am) to believe someone else had a tougher non-conference schedule than Duke.

A nice little trivia question (and this is according to CBS), it would be the Santa Barbara Gauchos. They have only played 1 team above an 80 out of conference and that was #255 San Diego. Meanwhile Duke has played #164 Delaware, #232 Cornell, and #249 Georgia State.

Now of course this brings about the problem that the difference between 200 and 250 or even 150 and 250 wouldn't seem big whereas the difference between 10 and 60 would seem fairly big and I don't think the formula does much if anything to account for this.

[vick]

Just out out of curiosity, who has the #1 SOS in non-conference games? It's hard for me to believe (as biased as I am) to believe someone else had a tougher non-conference schedule than Duke.

At least according to Statsheet (http://statsheet.com/mcb/rankings/RPI), UC-Santa Barbara is #1, and you can see why looking at their schedule (http://statsheet.com/mcb/teams/california-santa-barbara/schedule?season=2012-2013) and knowing how RPI schedule difficulty is calculated--only two teams below #80. RPI doesn't really record a huge gap between a team ranked around, say, #10 and #50 in terms of strength of schedule, even though for a top team like Duke, that difference matters a lot in terms of how many games the team might win. So Duke is 'underrated' by RPI strength of schedule, although in fairness, there have definitely been years where I felt like Duke's SOS was overrated when the nonconference cupcake schedule was filled with teams which were decent enough (say ranked 80-150) rather than actually bad teams (below 200). RPI isn't really meant to discriminate among top teams anyway, I don't think.

Or, what sporthenry said.

[CDu]

A nice little trivia question (and this is according to CBS), it would be the Santa Barbara Gauchos. They have only played 1 team above an 80 out of conference and that was #255 San Diego. Meanwhile Duke has played #164 Delaware, #232 Cornell, and #249 Georgia State.

Now of course this brings about the problem that the difference between 200 and 250 or even 150 and 250 wouldn't seem big whereas the difference between 10 and 60 would seem fairly big and I don't think the formula does much if anything to account for this.

Kenpom has Santa Barbara at #2 in SOS. Long Beach State is #1, Duke is #9. But these rankings tend to be dominated by the lower-tier teams early. For reference Duke is the only major conference team in the top 30 in SOS (Colorado is #31; Butler, an elite team from a minor conference, is #21). Kansas is the closest good, major conference team at #51.

[cptnflash]

By the end of the season there might be a few Big 10 teams that have comparable overall SOS to ours (Minnesota in particular). But no major conference team will be able to touch our non-conference SOS. As long as we avoid injury and take care of business in the ACC (i.e., no worse than 15-3), we should wind up with a #1 seed.

Interesting tidbit - the 8 teams that have played a tougher schedule than us so far (according to Pomeroy) are a combined 25-67. So it's one thing to schedule tough... it's another thing to schedule tough and still win!

[Olympic Fan]

The last day of 2012 was a good day for Duke's RPI.

Minnesota beat Michigan State to improve to 13-1.

Temple beat Bowling Green to improve to 10-2. Elon (8-5) and Florida Gulf Coast (9-5) also won, although over weak oppoents.

And a secondary boost -- St. Louis (a victim of Santa Clara) beat No. 20 New Mexico.

The way the RPI works (simplifed, it's 25 percent your own wnning pct.; 50 percent is your opponents' winning pct.; 25 percent your opponent's opponents winning pct.) all of those results helped Duke's RPI.

Right now, Duke is not just No. 1 in the RPI, the gap between No. 1 Duke and No. 2 Michigan is wider than the gap between No. 2 Michigan and No. 13 New Mexico. Duke has five top 50 RPI wins (No. 7 Minnesota is the best) and three more top 100 wins. It would help for Kentucky -- currently No. 52 on the list -- to win a couple and get back in the top 50.

[TruBlu]

The last day of 2012 was a good day for Duke's RPI.

Minnesota beat Michigan State to improve to 13-1.

Temple beat Bowling Green to improve to 10-2. Elon (8-5) and Florida Gulf Coast (9-5) also won, although over weak oppoents.

And a secondary boost -- St. Louis (a victim of Santa Clara) beat No. 20 New Mexico.

The way the RPI works (simplifed, it's 25 percent your own wnning pct.; 50 percent is your opponents' winning pct.; 25 percent your opponent's opponents winning pct.) all of those results helped Duke's RPI.

Right now, Duke is not just No. 1 in the RPI, the gap between No. 1 Duke and No. 2 Michigan is wider than the gap between No. 2 Michigan and No. 13 New Mexico. Duke has five top 50 RPI wins (No. 7 Minnesota is the best) and three more top 100 wins. It would help for Kentucky -- currently No. 52 on the list -- to win a couple and get back in the top 50.

I see your logic about the bolded part, but I just cannot bring myself to pull for Kentucky in order for our RPI to improve. Jeez, by that logic, I would have to pull for UNC to win . . . Not gonna happen!

[gumbomoop]

Duke is not just No. 1 in the RPI, the gap between No. 1 Duke and No. 2 Michigan is wider than the gap between No. 2 Michigan and No. 13 New Mexico.

The site I just checked, RealTimeRPI.com ...

http://www.realtimerpi.com/rpi_Men.html

... seems to have been updated today, and now shows #1 Duke's gap over #2 Mich (= .0562) as wider than the gap between #2 Mich and #16 Indiana (= .0561).

Seems statistically significant. Is it? Or just an early season quirk?

[toooskies]

The site I just checked, RealTimeRPI.com ...

http://www.realtimerpi.com/rpi_Men.html

... seems to have been updated today, and now shows #1 Duke's gap over #2 Mich (= .0562) as wider than the gap between #2 Mich and #16 Indiana (= .0561).

Seems statistically significant. Is it? Or just an early season quirk?

Minnesota's win helped our SOS.

[gumbomoop]

Minnesota's win helped our SOS.

I wasn't clear enough. I didn't mean to ask about the change from yesterday. Rather, I meant to ask the more general question, whether the fact that the gap between #1 Duke and #2 Michigan (.0562) is just now larger than the gap between #2 Mich and #16 Ind (.0561) is significant or just a quirk.

That is, the fact seems at first glance to suggest that Duke's RPI is really way out in front of everyone else, since #2 Mich's RPI is as close to #16 Ind as to #1 Duke. And if really way out there, is it significant, or not really?

[Kedsy]

That is, the fact seems at first glance to suggest that Duke's RPI is really way out in front of everyone else, since #2 Mich's RPI is as close to #16 Ind as to #1 Duke. And if really way out there, is it significant, or not really?

Significant in what way? The RPI is a flawed metric and even to the extent to which it works it is not meant to be a predictor of future games or scores. We already knew that Duke's schedule is by far the best of all the top shelf teams, and we're undefeated against that schedule. That history should give us a very strong RPI and, as you point out, it does. I can't imagine it tells us much if anything more than that.

But you knew that.

[gumbomoop]

Significant in what way? The RPI is a flawed metric and even to the extent to which it works it is not meant to be a predictor of future games or scores. We already knew that Duke's schedule is by far the best of all the top shelf teams, and we're undefeated against that schedule. That history should give us a very strong RPI and, as you point out, it does. I can't imagine it tells us much if anything more than that.

But you knew that.

I think your response means Duke's way-ahead RPI signifies only that Duke is undefeated against a very tough schedule. The perfectly straightforward, matter-of-fact substance of your response says, either sort of or exactly, "Your question didn't need to be asked." Alas, that I asked it gives the lie to your punchline.

I suspect you already knew that.

[Any sarcasm in my post is entirely unintentional, and directed, unintentionally, at me, not you, nor any other poster, on any thread, nor at the mods, nor at the American political system, nor at any political party, nor at anyone's ethnic heritage, religious beliefs, views on global warming, or food preferences.]

[Olympic Fan]

I think we all can agree that the RPI is a flawed metric. I much prefer Pomeroy (which has it's flaws too) and even Sagarin.

But so what?

The RPI is important because it's the most important metric used by the NCAA selection committee. Their view the RPI the same way that most baseball writers view batting average and RBIs (two more very flawed metrics).

Actually, the raw RPI numbers and even the RPI ranking is not that decisive when the committee makes its picks. But the RPI is vital when they start talking about "top 50 wins" or "top 100" wins. THOSE figures are based on the RPI rankings and are VERY important to he committee. That's why I saw I wanted to see Kentucky win enough to make it back in the top 50. The difference between being 49 and 51 is very small, but it could mean the difference between seven and eight top 50 victories.

Duke has a few more opponents on the edge -- Davidson is 104 ... playing Duke is likely to boost them into the top 100. Florida Guld Coast is barely outside te top 100, although with their upcomign schedule, it's going to be hard for them to move up.

[sagegrouse]

I think we all can agree that the RPI is a flawed metric. I much prefer Pomeroy (which has it's flaws too) and even Sagarin.

But so what?

The RPI is important because it's the most important metric used by the NCAA selection committee. Their view the RPI the same way that most baseball writers view batting average and RBIs (two more very flawed metrics).



The RPI, while flawed in a technical sense, and a superior KenPom still cannot capture the strength of conferences and the rankings of teams between conferences. The problem is that, after January 2 or so, there are no interconference games. Therefore, the rankings between conferences can only be based on games in November and December, when many teams (hullo, UNC!) are still trying to find themselves. Therefore, lacking meaningful data, it really doesn't matter all that much which algorithms are used.

If there were a prescribed interconference period in early February, for example, when conference schedules stood down and there were highly publicized games among conferences, there would at least be some data, evidence really, to inform the debate about computer algorithms.

sagegrouse

[Kedsy]

[Any sarcasm in my post is entirely unintentional, and directed, unintentionally, at me, not you, nor any other poster, on any thread, nor at the mods, nor at the American political system, nor at any political party, nor at anyone's ethnic heritage, religious beliefs, views on global warming, or food preferences.]

Sorry, I didn't mean my post to come off as snippy. Well, maybe a little. But only a teensy, weensy bit, I swear. ;)

[CDu]

If there were a prescribed interconference period in early February, for example, when conference schedules stood down and there were highly publicized games among conferences, there would at least be some data, evidence really, to inform the debate about computer algorithms.

Of course, to do that, you'd have to do one of a few things:
1. Increase the number of games played (and either add conference games in December or extend the duration of the season)
2. Reduce the number of conference games played (to allow for these late-season interconference matchups)
3. Move conference play to start in December and delete some December interconference games

#3 would basically just result in an offset of the number of interconference games (just moving them from December to February. #2 isn't going to happen. #1 might happen (colleges love making money), but it'd be a tough sell from the "student athlete" perspective.

[sagegrouse]

Of course, to do that, you'd have to do one of a few things:
1. Increase the number of games played (and either add conference games in December or extend the duration of the season)
2. Reduce the number of conference games played (to allow for these late-season interconference matchups)
3. Move conference play to start in December and delete some December interconference games

#3 would basically just result in an offset of the number of interconference games (just moving them from December to February. #2 isn't going to happen. #1 might happen (colleges love making money), but it'd be a tough sell from the "student athlete" perspective.

Sorry. I thought it was obvious: play 2-3 conference games in December and play 2-3 interconference games in February. BTW, this only works if the major conference teams schedule each other and not just the notional St. Leo's.

sage

[CDu]

Sorry. I thought it was obvious: play 2-3 conference games in December and play 2-3 interconference games in February. BTW, this only works if the major conference teams schedule each other and not just the notional St. Leo's.

sage

But what does that change? You play 2-3 conference games in December (instead of February) and 2-3 interconference games in February (instead of December). Doesn't really give us any more data with which to estimate SOS. Just gives the same amount of data shuffled around.

Basically, by the end of the season, you have a slightly more up-to-date estimate of non-conference SOS, but a slightly less up-to-date conference record. It's six of one, half dozen of the other.

[sagegrouse]

But what does that change? You play 2-3 conference games in December (instead of February) and 2-3 interconference games in February (instead of December). Doesn't really give us any more data with which to estimate SOS. Just gives the same amount of data shuffled around.

Basically, by the end of the season, you have a slightly more up-to-date estimate of non-conference SOS, but a slightly less up-to-date conference record. It's six of one, half dozen of the other.

There are three problems with any of the results-driven analyses.

(1) All games from January through March are conference games, providing an excellent if unnecessary ranking of teams within conferences. They are "unnecessary" because there are regular season standings and conference tournaments to give a proper ranking of the teams.

(2) Results and team performances in November and December are often quite different from capabilities the last month of the regular season. Yet all inter-conference rankings -- whether by RPI or KenPom or AP, for that matter -- can only rely on early season results to calibrate across conferences.

(3) Moreover, November and December, to channel Dickie V., are often "cupcake city," when many teams play only much less capable opponents.

The important point is #2: When I looked at the data in detail two years ago, it was clear that beyond the top ten teams, there was very little correlation with ratings and performance measures between December and March. The 2012 data was not quite as strong in this regard.

But if the only ways of calibrating teams across conferences are flawed results from November and December, we are conducting a really lousy experiment and perfecting the various measures is like putting "perfume on a pig." November and December results are "flawed" because of the time interval and degrees of improvement over 3-4 months and because not many teams schedule tough matchups in this part of the season.

If we want to be rigorous about ranking teams for NCAA seeds, we need a different schedule of games. Or, if not -- here's my incendiary remark -- just allow each of the 6-7 major conferences to select one-half (or some other percent) of their teams for the NCAAs and just forget about using RPI etc. for selection.

sagegrouse

[UrinalCake]

Well it would solve the problem of teams still "figuring things out" during the OOC portion of the season.

Of course in UNC's case, they may still be trying to figure things out in February 8-)

[CDu]

Well it would solve the problem of teams still "figuring things out" during the OOC portion of the season.

Right, but at the expense of playing conference games before teams have "figured things out." Currently, we have non-conference SOS based almost entirely on data before teams have figured things out, but the conference schedule is predominantly in the "once things are figured out state." If you move 2-3 conference games to December and replace them with 2-3 non-conference games, you get a non-conference SOS that is partially (only a small part, mind you) based on games after which teams have "figured things out," but the conference results now include games in which teams presumably were still "figuring things out."

So it doesn't really solve anything. It just redistributes the problem.

[Jderf]

Right, but at the expense of playing conference games before teams have "figured things out." Currently, we have non-conference SOS based almost entirely on data before teams have figured things out, but the conference schedule is predominantly in the "once things are figured out state." If you move 2-3 conference games to December and replace them with 2-3 non-conference games, you get a non-conference SOS that is partially (only a small part, mind you) based on games after which teams have "figured things out," but the conference results now include games in which teams presumably were still "figuring things out."

So it doesn't really solve anything. It just redistributes the problem.

Well it's certainly not an absolute solution, but it could still be an improvement. You know what? I actually believe in redistribution. :D

If the problem is that we have a lot of good data for ranking teams in-conference, but no good data for ranking teams between conferences, maybe it makes sense to strengthen our out-of-conference data at the expense of softening our in-conference data. Presumably, our in-conference data would still be pretty good.

Sure, the overall fuzziness of the information stays around the same, but at least then we would have a slightly better idea of where the conferences stand in relation to each other.

Either way, there's no way it's ever going to change. But it is interesting to think about.

[sporthenry]

Duke now has the #1 SOS and #1 RPI in all games and non-conference games.

So far, Duke's most impressive win according to RPI is Minnesota at #7. Duke has 2 top 11 wins but their next two wins are sitting at 27 and 28. Following that is #50 and #52. I mention these because they are usually bracketed together at top 25 and top 50 RPI wins so it will be intriguing to see if Temple/UK can stay in the top 50 and if OSU/VCU can get into the top 25. Duke's sole loss is to #14 NC State.

But so far, most of these teams are trending down. Louisville and UK are coming off head scratching losses to Bama and Nova. Temple just lost to St. Bonnies after squeaking one out at GW. OSU laid an egg at Illinois but did get a big win against Michigan. Minnesota doesn't have any terrible losses but they were down 23 and 19 to Indiana and Michigan at points during the game. VCU has been fairly consistent.

Not playing chicken little, but these guys might and probably will become important in Duke getting the #1 seed.