RPI

[Frodo]

Anyone who pays any attention at all probably knows that I think the RPI is a worthless sack of something...but this really is priceless. UConn has moved down to 4th in their power ratings and 3rd in the RPI rating since they lost today to Temple 66-58. You can't make this stuff up.

 

[hussybulldog]

link? that is a good laugh..

 

[DobbsRover2]

Anyone who pays any attention at all probably knows that I think the RPI is a worthless sack of something...but this really is priceless. UConn has moved down to 4th in their power ratings and 3rd in the RPI rating since they lost today to Temple 66-58. You can't make this stuff up.

RPI is indeed brain-dead though it has its fervent supporters on the BY who love to cite it, but the reason that UConn dropped from 1st to 4th in RPI on Sunday was simply that they played SMU for that second time. Even if they had beaten SMU by 300 points, they still would have dropped to #4. Whether UConn will drop out of the RPI top 10 in the coming weeks depends on how the teams in the 5-12 range do, but likely it will happen.

And note that the loss to Stanford is only a blip on the RPI rating effect. In 2002 the Huskies were the 33-0 for the pre-Tournament season and a crunching win over a 28-4 UTenn team at TBA and were still way behind the Vols in the RPI. In fact, UTenn could have suffered a bunch more losses and still have been ahead of UConn.

Totally mindless system whose idiocy can be shown with a few simple case examples, but there are those here who swear by it.

 

[Frodo]

link? that is a good laugh..

http://realtimerpi.com/rpi_Women.html

 

[Vowelguy]

Lost to Temple? Was this for Men's board?

And because I'm a glutton for punishment, I'll repeat my usual comments about the RPI:

1) It is not designed to be used mid-season. It is an end-of-season measure.
2) It is NOT a power rating. It is measure of wins and losses against quality of competition. If the San Antonio Spurs played in the MEAC conference, they would not have a great RPI, even though they are clearly better than any college team.
3) A myth has developed that RPI is the only factor used by the committee and it's power is absolute. That is patently false. It is one of many tools.

 

[Frodo]

Lost to Temple? Was this for Men's board?

And because I'm a glutton for punishment, I'll repeat my usual comments about the RPI:

1) It is not designed to be used mid-season. It is an end-of-season measure.
2) It is NOT a power rating. It is measure of wins and losses against quality of competition. If the San Antonio Spurs played in the MEAC conference, they would not have a great RPI, even though they are clearly better than any college team.
3) A myth has developed that RPI is the only factor used by the committee and it's power is absolute. That is patently false. It is one of many tools.

Can't speak for everyone, but I've never argued that it's the only factor used by the committee...my problem is that it's a factor at all. It's a moronic formula that, frankly, puts a lot of weight on who you play, even if you lose by 50. I don't want to belabor the point, but a few years ago I did some rough calculations and if a team played only those teams ranked between 40-60 and won about 3/4 of them, they would have the #1 RPI...at the end of the season. That's just stupid on it's face.

 

[Frodo]

Wonder if someone is watching...they've updated the site and removed the loss to Temple!!!

 

[DobbsRover2]

Again the RPI apologists are out with the usual garbage about how it's an end of season tool, it is useful for anything, and that there are people spreading some kind of myth that it's the only system the selection committee uses. And even though they say it's just an "end-of-season" tool, the RPI apologists start using it early in January. It is bad for rating the top teams, it's bad for rating bubble teams, but it's great for boosting UTenn above even undefeated UConn teams. Mindless.

RPI should not be used for anything, and if you want to read why its useless for even the things it was supposedly designed to measure, here's a link and another link.

And yes, the resident champion of RPI will tell you that if UConn slides further down the index that we should just ignore that because it's meant to be used for other teams, not UConn. Teams like maybe that #2 RPI Kentucky team. Again, mindless. But hey it's got USCar at #8 right now, so at least it's brainlessness is directed at many different teams.

 

[Vowelguy]

Wonder if someone is watching...they've updated the site and removed the loss to Temple!!!

Note that site is not run the NCAA. It's just some person who does it, thus providing us with a great service. Mistakes can happen...

 

[Vowelguy]

puts a lot of weight on who you play

That's part of the point. The committee has a strong interest in encouraging teams to play quality opponents -- it makes the seeding process much easier, because you have real results to measure everyone up. In contrast, it becomes very difficult to evaluate teams that play weak schedules.

I've always maintained that the big value of the RPI is not the ranks, but the comparatives (ie, W-L vs RPI top 25, top 50, etc). I think it's useful, so as it's interpreted correctly.

 

[DobbsRover2]

Wonder if someone is watching...they've updated the site and removed the loss to Temple!!!

Not sure what you saw before, though something is definitely going screwy with RealTimeRPI, which just switched UConn up to #3 in RPI, which is different from what the Warren Nolan version has with UConn at #4. Maybe you were looking at RealTimeRPI's Power Rating which as I type does have UConn listed at 14-2. Weird.

RealTimeRPI Power Rating

1 South Carolina 16-0 89.69 80 53.54 Sec 4-0
2 Notre Dame 15-2 89.15 8 60.36 Acc 3-1
3 Texas 14-1 86.03 68 54.30 Big12 3-1
4 Connecticut 14-2 85.53 40 55.86 Aac 4-1

RealTimeRPI RPI

1 Notre Dame 15-2 0.6895 5 0.6317 Acc 3-1
2 Kentucky 14-3 0.6850 4 0.6331 Sec 3-1
3 Connecticut 14-1 0.6848 19 0.5934 Aac 4-0
4 Tennessee 14-2 0.6838 9 0.6097 Sec 4-0

 

[Orangutan]

I've never understood why this conversation has to happen every year. Yes, RPI is a terrible way to rank teams. Not even the so-called "apologists" dispute that.

It's basically a way of answering this type of question "UConn is 14-1 against the #23 SOS and South Carolina is 16-0 against the #58 SOS, which is more impressive?" That's it. It doesn't do a very good job of that because it doesn't take scores into account, but the NCAA doesn't want to promote blowouts, which using a score-based method (like Massey or Sagarin) in the selection process would do, even if those would be more accurate.

It is what it is. I think it would be very hard to design a non-score-based ranking formula with better predictive value than RPI.

 

[Intpseeker]

RPI. SchmaRPI.

UConn is brought to you by the letters N and C

 

[Vowelguy]

It's basically a way of answering this type of question "UConn is 14-1 against the #23 SOS and South Carolina is 16-0 against the #58 SOS, which is more impressive?" That's it.

Sorta.
Comparing UConn and SC will be easy, because they will face each other. And also it's generally easier to separate teams at the top. The top 16 seeds are usually more strongly correlated with the AP poll than they are the RPI.

But suppose you want to compare Arkansas and Oklahoma St? That's what these ratings systems are used for -- the middle of the pack.

 

[trisailin]

But what does RPI rank?
It might as well rank the % of the team which ties their shoelaces left over right vs right over left. It certainly has ZERO predictive ability regarding two teams playing each other. Since it can't do that, it doesn't get the teams in the right order, then it certainly can't tell you which team played a tougher schedule.
And I ask again - if a team is ranked #50 and plays #49 and #51 is that a tougher schedule than if they played #1 and #99?
Strength of schedule is a meaningless measure. Both Massey and Sagarin calculate it but it's done AFTER the fact, not as part of the computation for ranking.
So Mr RPI apologist guy - WHAT DOES RPI RANK?

 

[DobbsRover2]

And it's also a lie that RPI encourages teams to play quality schedules, it simple encourages a team to game the system by scheduling a lot of home games against teams that it is confident it can beat, a few away games against teams it is very confident it will beat, and maybe one away game against a very tough opponent just so it can say it played someone. Comparatives of top-25 and top-50 records like what Sagarin shows is very nice, but only if the underlying data for the teams is of any use, but for RPI it isn't. So is Baylor the team that is 3-1 against the top 50 in Sagarin or the one that is 1-1 in RPI? And is UNC at 3-2 or 1-2? Makes for different arguments for the teams.

Overall, just makes sense to use a system that makes sense, not one that's useless.

 

[DobbsRover2]

It is what it is. I think it would be very hard to design a non-score-based ranking formula with better predictive value than RPI.

Um, no. Not even the RPI apologists here claim that it has any type of predictive value for an upcoming game, they just say it has some type of comparative value for teams. As it was set up, RPI was simply a system that uses a three-tier evaluation based on W-L record of games that have been played, and it rates how well it thinks a team has done against that competition, though it does it very badly because of the way it is set up.

Now the RPI also does have a separate component called Predicted Results that does kind of a brute application of a team's remaining schedule, but if anything that tool is even more ridiculous than the main RPI rating system. The RPI Predicted Results computes that USCar will beat UConn at Gampel even though UConn currently has a higher RPI, and that the Huskies should also lose at USF.

Does UConn going 27-3 during the regular season work for you as a better prediction system?

 

[Orangutan]

But what does RPI rank?
It might as well rank the % of the team which ties their shoelaces left over right vs right over left. It certainly has ZERO predictive ability regarding two teams playing each other. Since it can't do that, it doesn't get the teams in the right order, then it certainly can't tell you which team played a tougher schedule.
And I ask again - if a team is ranked #50 and plays #49 and #51 is that a tougher schedule than if they played #1 and #99?
Strength of schedule is a meaningless measure. Both Massey and Sagarin calculate it but it's done AFTER the fact, not as part of the computation for ranking.
So Mr RPI apologist guy - WHAT DOES RPI RANK?

Take it wikipedia...

"the index comprises a team's winning percentage (25%), its opponents' winning percentage (50%), and the winning percentage of those opponents' opponents (25%)"

 

[Orangutan]

It is what it is. I think it would be very hard to design a non-score-based ranking formula with better predictive value than RPI.

Um, no. Not even the RPI apologists here claim that it has any type of predictive value for an upcoming game

Evaluating the predictive value of a ranking system is the best way to evaluate it's accuracy. If it says that Team A is the #1 team and Team B is the #100 team and Team B beats Team A by 30, that's not much of a ranking, right?

Anyhow, feel free to read my statement as "I think it would be very hard to design a non-score-based ranking formula with better comparative value than RPI."

 

[Vowelguy]

Evaluating the predictive value of a ranking system

RPI is not predictive. It only measures performance against the teams played.
As I noted above, if the San Antonio Spurs played only the worst 10 women's college teams, they would likely not have the #1 RPI.

 

[DobbsRover2]

Evaluating the predictive value of a ranking system is the best way to evaluate it's accuracy. If it says that Team A is the #1 team and Team B is the #100 team and Team B beats Team A by 30, that's not much of a ranking, right?

Anyhow, feel free to read my statement as "I think it would be very hard to design a non-score-based ranking formula with better comparative value than RPI."

"Comparative" would be a better word to use than "predictive," but the question would still be why use a "comparative" system that does such a freakishly bad job of comparing, no matter what the resident RPI apologist says? Even if you are hugely concerned about teams being encouraged to "run up the score," there are many ways of limiting the giant margin results and providing a rating system that isn't on crack. Sagarin is built on a number of modules averaged together, and the ELO_Score factor has a means of keeping the blow-outs in check. Massey has likewise has a diminishing-returns feature on the margin along with other comparatives that weight teams more on games with teams that are similar to them and also on more recent games.

There are nuanced systems that do a pretty good job, and there's no need to use a brain-dead RPI system unless you're the resident apologist who wants to push the same old garbage about how well the system does in comparing #2 KY to #4 UConn. That works for him. And note that at the end of the regular season, the RPI Predicted Results has KY as the RPI champion.

Crown those Wildcat women.

 

[DobbsRover2]

And with the dominating win over Temple, UConn has now faded back to #5 in RPI. I'm assuming they need to be a #2 seed in the local guy's next bracketology, as ASU claims a #1 seed along with UTenn, ND and KY.

Just magical.

 

[iri10]

By the end of the season, there's going to be a very large discrepancy between UConn's RPI rank and their "eye test rank." RPI simply wasn't designed to deal with outliers like UConn in this conference.

You can only do so much when you a) don't take margin of victory into account, and b) run a strength of schedule calculator that ignores the reality of WCBB - namely that Division I is top-heavy in terms of team quality.

Basically, RPI tells you that the difference in quality between the number 1 team and the number 25 team is pretty small, and the difference between the 150th ranked team and the 300th ranked team is massive. But if you're an elite team, the difference between the number 1 team and the number 25 team is pretty damn big, and the difference between the 150th team and the 300th team is irrelevant (you'll crush them both).

 

[DobbsRover2]

By the end of the season, there's going to be a very large discrepancy between UConn's RPI rank and their "eye test rank." RPI simply wasn't designed to deal with outliers like UConn in this conference.

You can only do so much when you a) don't take margin of victory into account, and b) run a strength of schedule calculator that ignores the reality of WCBB - namely that Division I is top-heavy in terms of team quality.

Basically, RPI tells you that the difference in quality between the number 1 team and the number 25 team is pretty small, and the difference between the 150th ranked team and the 300th ranked team is massive. But if you're an elite team, the difference between the number 1 team and the number 25 team is pretty damn big, and the difference between the 150th team and the 300th team is irrelevant (you'll crush them both).

And that is why Massey focuses the majority of analysis on performance against the opponents that are similar to a team and gives little value to how UConn does against say an SMU or Houston, and it is also why Massey always has the best SOS index. Whereas RPI will "reward" the worst team in the country for games against top 10 teams, Massey just pretty much discounts them. If you have a comparative rating system that is as badly broke as RPI, the data you get out of it is useless unless your a Vol fan who wants to argue that UConn is only the 5th best team.