OUT: RPI IN: NET
- Thread starter TIGER-PAUL
- Start date
Ironically, an efficiency-based system probably better suits Dixon's teams than the RPI.
In 2013, Pitt was 11 in Kenpom but 46 in RPI.
In 2014, Pitt was 16 in Kenpom but 39 in RPI.
In 2018, TCU was 24 in Kenpom but 28 in RPI.
This is obviously fake news, because it wasn't all that long ago that steel assured us the the NCAA would never replace the RPI with some other ranking system. NEVER!
The potentially interesting thing here is this:
"The model also caps winning margins at 10 points in accounting for the event of a blowout score."
That might actually stop some of these early season cupcake games that no one but basketball coaches seem to like. If the NCAA model doesn't credit you with anything more than a ten point win against a team as bad as, say, Grambling then why would a good team play them? They had a decent record this past season, so playing them wouldn't have been bad for your RPI, but by advanced metrics they were bad. A ten point win against them, especially at home, would be considered to be a poor result for any NCAA tournament hopeful. Does this mean that teams that think they are tournament contenders are going to avoid the top teams from the really bad leagues in favor of playing bottom teams from much better leagues?
"The model also caps winning margins at 10 points in accounting for the event of a blowout score."
That might actually stop some of these early season cupcake games that no one but basketball coaches seem to like. If the NCAA model doesn't credit you with anything more than a ten point win against a team as bad as, say, Grambling then why would a good team play them? They had a decent record this past season, so playing them wouldn't have been bad for your RPI, but by advanced metrics they were bad. A ten point win against them, especially at home, would be considered to be a poor result for any NCAA tournament hopeful. Does this mean that teams that think they are tournament contenders are going to avoid the top teams from the really bad leagues in favor of playing bottom teams from much better leagues?
I always believed RPI was too inaccurate. One big reason for its inaccuracy, IMO, was that it didn't account at all for blowout scores; but only whether a team won or lost a game. Giving credit for up to 10 points for blowouts is a step in the right direction--although I would have liked the cap to be higher than 10 points--maybe 15-20 points just because there are some cupcake opponents that you should be penalized for not beating by at least 15-20.
The above being said, I doubt the new system will discourage teams from scheduling cupcakes OOC since a cap at a 10 point margin of victory might actually encourage scheduling some cupcakes vs the RPI only caring whether you won the game even by a single point.
In addition, there are other reasons for scheduling cupcakes that will still exist. They are your "exhibition" games you use to develop your players' confidence and to tune up rotations and strategies for the more difficult opponents to come as well as to create the public and fan perception "beauty contest" that your overall results are better than they may actually be.
If what you really want is to create a system where at NCAA-time, that you have an as true as possible field of the best 64 teams make the tourney; then you would, IMHO:
1. Use a system like the one just selected (but with a somewhat higher than 10 point cutoff for blowouts).
2. Use that rating system to put in the tourney the top 64 rated teams without regard for auto bid league winners. This might mean teams with overall losing records who played very difficult schedules would get in--not the case in the past (excluding possibly a few tourney champs in auto-bid leagues or some other anomalies).
3. Have winners of auto-bid leagues not already rated in the top 64 all placed in a play in round with whatever number of very bottom of the top 64 rated team list are needed to make the tourney bracket math workable.
I agree with this take. I think this incentivizes scheduling teams you are confident in beating by double digits, rather than one you expect to beat, but could be a little closer.
It really doesn't. If you are expected to win a game by 4 and you win by 8 that's a good thing. Your ranking will go up. If you are expected to win a game by 20 and you win the game by 20 but you only get credit for winning by 10 that's a bad thing. Your ranking will go down.
Essentially what this does is it tells teams not to play anyone that they would be expected to beat by more than 10 points.
I haven't seen that as a part of the new equation. Instead, I've interpreted that winning by >=10 would all be counted the same and as a "cover" or meeting expectation.
That's the way I interpreted it--and why I don't think it discourages anyone from playing "cupcakes." The weaker the cupcake the less likely that you won't win by less than 10 points and hurt your rating.
Last edited:
You have, you just missed it because it doesn't explicitly spell it out. If you are taking into account strength of schedule, then you have built in a "punishment" for playing a bad team. And when you add in scoring margin and put a limit on it, you have changed the value of beating a bad team by a large margin.
If you use scoring margin to do your rankings and you play a team that you are supposed to beat by 20 you get the SOS hit for playing a team you should beat by 20, but you only get the scoring margin "plus" of winning the game by 10 points instead of 20. It's really not that hard to understand. To use an extreme example, if in every game a team plays they under performed what they would expect to do by 10 points what do you suppose would happen to their ranking? It would drop, and by a large amount.
Under this system, if you play a team that you are supposed to beat by 20 the absolute best that you could do is under performing your ranking by 10 points. Teams that under perform their rankings see their rankings drop. That's the way these things work.
Even if you are correct--I would have to actually see the details of the plan and do the math--I still doubt it will have much if any effect on teams scheduling of "cupcakes" OOC, especially ACC teams whose overall schedule is very strong simply due to the number and strength of conference games. Also, at least a handful of OOC cupcakes are needed as part of these early season tourneys were you tune up at home to be ready for the neutral site games vs better opponents in, say, NYC or Brooklyn.
Yes, it will result in a better rating system--probably closer to the Sagarin ratings numbers (which I have always believed to be more accurate overall) than the RPI, but the only teams likely to attempt to strengthen their OOC schedule will be teams not in the P5 conferences who need stronger overall schedules; and they may have trouble doing so as the P5 teams won't have much incentive to increase the scheduling of dangerous non P5 teams. I strongly doubt we see Pitt make any significant changes to its typical OOC schedule. Some Pitt fans will in the future be able to continue to whine about the OOC but it will be to no avail.
Yes, it will result in a better rating system--probably closer to the Sagarin ratings numbers (which I have always believed to be more accurate overall) than the RPI, but the only teams likely to attempt to strengthen their OOC schedule will be teams not in the P5 conferences who need stronger overall schedules; and they may have trouble doing so as the P5 teams won't have much incentive to increase the scheduling of dangerous non P5 teams. I strongly doubt we see Pitt make any significant changes to its typical OOC schedule. Some Pitt fans will in the future be able to continue to whine about the OOC but it will be to no avail.
Last edited:
Well if you understand the Sagarin ratings then you understand that if you play a team that you are supposed to beat by 20 points and you only beat them by 10 (or only get credit for beating them by 10) your rating will drop. Do that on multiple occasions and your rating will drop by a pretty large margin. Do it three or four times and your rating will drop by a full point. To put a "face" to it based on last season's Sagarin numbers (and obviously there is and will be more to it than this), one point was the difference between UCLA being 45th and squarely on the bubble and them being 55th and clearly being on the outside looking in. It's the difference between 16th Florida in a good spot for a top four seed and them being 24th and having no chance to get one.
It won't get rid of all those games, because if there is one thing that college basketball coaches love it's cupcakes. But the smart coaches, the ones who figured out the correct way to "play" the RPI are going to do the same thing with this new system and they simply are not going to play the really bad teams unless they are forced to.
I do believe the Sagarin system does also cap scoring margin--but, I believe not as low as 10 points.
I am sure coaches will schedule to their advantage under the new system. However, when you play 18 ACC level games and a handful of additional P5 OOC games between the ACC-Big challenge and a tourney or two there won't be any real advantage to playing any additional tougher OOC games. So, I don't see Pitt's OOC schedule changing in any significant way.
I am sure coaches will schedule to their advantage under the new system. However, when you play 18 ACC level games and a handful of additional P5 OOC games between the ACC-Big challenge and a tourney or two there won't be any real advantage to playing any additional tougher OOC games. So, I don't see Pitt's OOC schedule changing in any significant way.
His "predictor" system uses of margin of victory, his basic rating does not. The gap between the two usually isn't very big by the end of the year. Last year Pitt ranked 189 and 192
while Nova, Duke, and Purdue were #1, 2, 3 in both.I do think NCAA basically got it right with a 10 point run up the score limit.
His "predictor" system uses of margin of victory, his basic rating does not. The gap between the two usually isn't very big by the end of the year. Last year Pitt ranked 189 and 192
I do think NCAA basically got it right with a 10 point run up the score limit.
Yes, and there is, or was, a composite rating averaging his 3 or so ratings.
I'm not sure what you are calling his "basic" rating, but he has four different ratings on his page. Each of the four uses margin of victory.
"The PREDICTOR, is such that the score is the only thing that matters." In other words, margin of victory.
"GOLDEN_MEAN also utilizes the actual SCORES of the games in a different way but is also completely SCORE BASED" In other words, margin of victory.
"The RECENT, is score-based and weights RECENT play more heavily than earlier games." In other words, margin of victory.
"The overall RATING is a synthesis of the three different SCORE-BASED methods, PREDICTOR(PURE_POINTS), GOLDEN_MEAN, and RECENT and thus should be a good predictor in its own right." In other words, margin of victory.
Every rating that he has on his web page uses margin of victory as it's basis.
It is, but there is no indication that they are going to do that, because it would be extremely stupid.
But then again, it's the NCAA and they stuck with the RPI for many, many years after it was apparent to anyone with a basic understanding of basketball and math that the system didn't make any sense, so....
By incorporating scoring margin and capping it, that is my indication, but to each their own.
Like I said, they could do that but it would be stupid. It would actually make playing poorly against really bad teams less costly. Why on earth would you want to reward a team that plays poorly, maybe even loses the game, to a really bad team?
I can't find a summary on his site, but I know Pomeroy implemented a margin cap, or at least applied a significant falloff at some point, but certainly his cap was more than 10. I think maybe 20.
I think what is unclear, and not a given as in your explanation, is what effect and expectation is placed on margin before a game is played. Certainly you could make the same argument against Pomeroy, that if Pitt were expected to win by 40 and was capped at 20 that it would be bad for Pitt, but it really isn't. It's hard to guess how that would actually be implemented. If Pitt wins 90-65, would the NCAA enter the adjusted score into their formula as 75-65? That seems rather bizarre.
After reading a bit more about the (unexplained) calculation, I am beginning to think that margin actually won't have much impact at all. It's basically going to be like an enhanced RPI with an added efficiency factor.
My guess is that a team like 2015 Florida, which was very high in KP (34) but with a terrible record (16-17) of very close losses, will have a much lower NET rating than you'd see in a pure efficiency system. I'm not fluent in Sagarin, but he had UF at 29/34 in Predictor/Golden mean, but then 70 in ELO.
My guess is that a team like 2015 Florida, which was very high in KP (34) but with a terrible record (16-17) of very close losses, will have a much lower NET rating than you'd see in a pure efficiency system. I'm not fluent in Sagarin, but he had UF at 29/34 in Predictor/Golden mean, but then 70 in ELO.
ELO doesn't use margin of victory. It's the system that Sagarin devised when the NCAA wanted to use computer rankings for the BCS but didn't want to include margin of victory. You can modify them in some ways, but in a pure ELO system the only things that matter is the rating of each team before the game was played and whether a team won, lost or tied the game. It was originally devised (and is still used) to do rankings of chess players, where there was no such thing as "home field advantage" or "margin of victory", you either beat your opponent or you don't.
Pomeroy (and several others) don't really use a hard cap, they use a sliding scale based on the ratings of the teams before the game. A 25 point win over a team that you were supposed to beat by 20 is handled a little bit differently than a 25 point win over a team that you were supposed to beat by 5. A 25 point win over a team you were supposed to beat by 20 most likely isn't modified at all, a 25 point win over a team you were supposed to beat by 5 probably is somewhat modified, but is still worth a lot more than the win over the team you were supposed to beat by 20.
Right, which is why I used that as the example. I think this new system is still going to heavily reward teams who win a bunch of games moreso than teams that actually perform well on the court. The "biggest" factor the NCAA has cited is basically a reworked RPI.
This is all true. Though white always has the advantage in chess and one of the great debates of the game is how great is that first move advantage. I assume ELO takes that into account?
I could never break the 1200 barrier.
It does not. Since most games that are played for rankings purposes are in tournaments and in tournaments you alternate between playing black and white the theory is that any advantage comes out in the wash.
