How Accurate Can The RPI Really Be???

[UcMiami]

I think any SOS or RPI system that does not have a sliding scale for competition in WCBB is not going to be terribly good. The top ten teams any year should never lose to a team ranked 50 or higher so to award a better 'score' for beating team 51 (by any margin) than beating team #251 (by any margin) is just silly. And I would refine it even more - for a top ten team to lose to a team ranked 35 - 50 is terrible whether the ranking of tat team is 36 or 49. And to lose to a team ranked 25 - 34 is pretty bad whether the team is ranked 26 or 34.
AND for a team ranked 40th to have a win over a team ranked #15 or better is great whether that team is #14 or #3, and to beat a team ranked 16 - 25 is very good whether they are #16 or 24.
So ... pooling those teams ranked far above or below a teams ranking should happen while keeping specific values for the teams within say 20+/- ranking spots. I believe Massey is doing something like this with their rankings though I am not sure. But the idea that if Uconn scheduled teams ranked from 75 - 100 in place of the teams that they have played ranked above 125 would actually improve their 'Strength of Schedule' by a significant amount is ludicrous - against any of those teams Uconn can pretty much chose whatever their point spread will be. But that is what the linear SOS does - play the #2 team in the country and the #198 team in the country and it is the same as playing the #99 and #101 team in the country. For the team ranked #100 you might be able to argue that the expectation is the same but the reality is they have a fair chance of winning or losing both games in scenario 2 and an almost 100% chance of ending at 1-1 in scenario 1, for the team ranked #1 there is an almost 100% chance of 2-0 in scenario 2, and a fair chance of 1-1 in scenario 1.

 

[DobbsRover2]

Massey does use a system that has relative comparisons so that a top team playing a top team gets more weight in the team's score than when they play bottom teams, and likewise a bottom team's score is more weighted toward how it does against bottom teams and does not get a big boost for losing by 60 to a top team.

Sagarin uses much more linear rating factors, but it has a couple of different scales and two of them use MOV in some way (the formula is a secret), and the composite score is pretty accurate for predicting game margins.

RPI can give you maybe a 50% chance of predicting which of two monkeys will be the first to hit the dart board.