ESPN's BPI vs RPI rating

[paladinhawk]

Iowa is 19th in BPI and 34th in RPI (at least by realtimerpi). It seems like ESPN under-ranked the Iowa football team all season long.

Is ESPN ranking over-ranking the basketball team?

What's your favorite 'metric' to compare teams?

 

[StormHawk42]

ESPN's metrics are crap. Not that I love the RPI, but the fact is that it is still a piece used by the committee, so that's the one we have to pay attention to. Kenpom is the best of all of them IMO.

I believe the BPI had us losing by double digits to Davidson last year. Whoops.

 

[DanL53]

My favorites are KenPom followed closely by Sagarin.

 

[yerpHAWK]

BPI is good cause it rates us high. FPI is trash cause it hates us.

 

[Clifford C. Clavin]

RPI is so archaic. It's pretty much the worst out there, which makes it all the more irritating that it is used at all. BPI is much more dynamic, as is Sagarin, Kenpom, and others. Anyone with a little skill in Excel could replicate RPI.

 

[RocknRollface]

I believe the main reason the rpi is still used is that it generally rates the mid majors higher than the other metrics.

If they went by the more accurate metrics there would be allot less at large mid majors and the mid major upsets are a big part of what draws the casual observer for the first week of the tourney.

 

[skallday]

RPI is so archaic. It's pretty much the worst out there, which makes it all the more irritating that it is used at all. BPI is much more dynamic, as is Sagarin, Kenpom, and others. Anyone with a little skill in Excel could replicate RPI.

Part of the trouble that people have in understanding differences in rating systems is not understanding the purpose of a given system. RPI is not supposed to be predictive like the other systems are. Part of me thinks that the committee continues to use it because they don't want to incentivize blowouts.

While I agree RPI is awfully simplistic, and not necessarily a good predictor of future results (because it only uses wins and losses, your opponents' wins and losses and your opponents' opponents' wins and losses), it is still not a terrible representation of a team's resume. But it is susceptible to being abused (D2 scheduling, etc.).

Predictive models, on the other hand, rely on points scored/allowed rather than solely wins and losses. Over time, it's been shown that points scored/allowed are a better indicator of future wins and losses than past wins and losses.

The main difference between predictive models is how fancy they get. KenPom, for example, is really simple/elegant, and can also be replicated pretty closely by someone with good Excel skills (protip: one way around paying for a membership ). KenPom's inputs are adjusted offensive and defensive efficiency - that is, points scored/allowed per 100 possessions, adjusted for strength of schedule. The system does not care whether the team actually wins or loses.

BPI, on the other hand, makes a few adjustments to this approach. It includes a diminishing return for blowouts, it matters in the model whether the team actually wins or loses, and it de-weights games in which a team was missing a key contributor. Other models weight recent results more heavily or make other adjustments, but the basic principles are the same.

So I guess when evaluating any model, it's really important to understand what the system is trying to show. And also understand that rankings are not necessarily the best way to look at differences between teams.

Anyway, sorry for the rant, and apologies in advance if I forgot anything. Being an actuary, stats are kind of a passion for me