ISU #1 in BPI

[kucyclone]

What's your metric of choice, Don't even come back with "use the voters" as that is even dumber than switching to using the BPI.

At some point in the night, you switched from being fairly good at arguing to being quite awful. Alas:

For predictors, I prefer KenPom with a sprinkle of Sagarin.

For resume, I suppose I'd prefer BPI (if I can use victory margin), or Sagarin PURE_ELO and Massey (if I can't). I also think there's a lot of value in looking up and down a team's schedule and results, and not just relying blindly on these numbers. But, make no mistake, any of these is better than RPI.

 

[TurbulentEddie]

As far as mostly who you play, I still say that you don't understand RPI. That is not true at all. I've already explained this if you don't understand my explanation I don't know how to help you but do some research and take a statistics class.

I understood what you we're trying to say... that opp/opp's opp record part of RPI has low variance (around a mean of .5), so it needs to be "stretched" to actually have an impact on the total RPI. But that does basically set the base of a team's RPI, no matter how the team actually plays against its schedule, and that's kind of crappy.

As for gaming, just pointing out that it's a stupid argument to say MOV or PPP based ratings are bad because they can be gamed. All systems can be gamed (and adjusted for after the fact). I'd rather not have teams game a MOV system by playing 100% against cupcakes; I'd also like to see DI teams playing more DI teams OOC.

Anyway, to actually argue *for* something, I'd like to see the committee use a combo of RPI, Sagarin, and KenPom - not for specific teams' rankings - but for their SOS/record against top n teams. SOS from a predictive system is a much better measure of "who you played" than from a system that only ranks on W/L. I'd like to see a record-based (resume) ranking system developed that weights strength of wins and losses on predictive (MOV, PPP) measures of team strength.

Or, at least, the committee can do a better job of weighting home/away (beyond the shim that's in the RPI formula now): the kenpom.com blog

 

[mikeiastat]

I understood what you we're trying to say... that opp/opp's opp record part of RPI has low variance (around a mean of .5), so it needs to be "stretched" to actually have an impact on the total RPI. But that does basically set the base of a team's RPI, no matter how the team actually plays against its schedule, and that's kind of crappy.

As for gaming, just pointing out that it's a stupid argument to say MOV or PPP based ratings are bad because they can be gamed. All systems can be gamed (and adjusted for after the fact). I'd rather not have teams game a MOV system by playing 100% against cupcakes; I'd also like to see DI teams playing more DI teams OOC.

Anyway, to actually argue *for* something, I'd like to see the committee use a combo of RPI, Sagarin, and KenPom - not for specific teams' rankings - but for their SOS/record against top n teams. SOS from a predictive system is a much better measure of "who you played" than from a system that only ranks on W/L. I'd like to see a record-based (resume) ranking system developed that weights strength of wins and losses on predictive (MOV, PPP) measures of team strength.

Or, at least, the committee can do a better job of weighting home/away (beyond the shim that's in the RPI formula now): the kenpom.com blog

I'm not saying RPI isn't gamable so i don't see why you think any part of that is dumb. I'm saying that when poeple argue that RPI is being gamed so BPI is better, this is a bad premise because there is no incentive to game BPI. Of course its not better on this basis because it currently isn't being gamed. I also concede that it is not any worse. It is a good predictor. Yes, I also Like Sagarin better as a predictor as you stated we're in agreement. But again, we're not just talking predictor here, and also if the NCAA hasn't weighted Sagarin after what, 20 years of existence, what makes you think they'll change their justifiable position now?

I think the committee should and know that they do use a combo of data for seeding purposes. Look at when Colorado State gamed the system. Teams that game the system are fairly easily identified. Did their 29 RPI get them an 8 seed? No they got bumped back down to an 11. I feel even that was better than they could have bumped them, but again they don't just go by who they think should win. They reward making the system work, for making NCAABB more exciting, and for taking chances on the road with the big boys. They have to.

And in a way the Co. States of the world make the system work. They could easily improve RPI as a predictor by allowing D2 teams in the calculation and putting a penalty in for that, but NCAA scheduling wouldn't work without some D2s. The Drakes of the world would never get enough games and certainly not enough home games.

I just believe that for a number of reasons RPI should continue to be used for getting in. And as a predictor sometimes I can take it or leave it when people don't like it. But the poeple that say "who you play is 75% of your score" don't understand a tenth of what they should to speak on the issue.

That like say Mr officer, you are totally wrong you're using the wrong metric. I'm not drunk you have my score at .19. I'm not 19 percent drunk. I only have .00019% alcohol in my system. YOu are using the wrong metric because you are using a multiplier to make the number make more sense. Just stupid. Your RPI is nowhere near varied 75% by who you play. Some people are just dumb enough to state this. Because they listen to some ESPN talking head moron and regurgitate his argument that has zero understanding of variation. Some people are good with x's and o's and should stay there.

 

[Dryburn]

An ESPN explanation why ISU is no. 1.......

BPI Talk: Iowa State is No. 1 - Stats & Info Blog - ESPN