PDA

View Full Version : ESPN's New College Basketball Power Index (BPI)



Billy Dat
02-12-2012, 09:55 AM
ESPN has thrown their hat into the RPI, KenPom, Sagarin field by having their analytics department create a new system, the Basketball Power Index
http://espn.go.com/mens-college-basketball/story/_/id/7561413/bpi-college-basketball-power-index-explained
http://espn.go.com/mens-college-basketball/story/_/id/7559170/bpi-feb-11-college-basketball-power-index-rankings

System comparison

How is BPI different than RPI or other advanced rating systems like Kenpom.com and the Sagarin ratings? Here is how the included elements compare to other systems.
Includes RPI BPI Sagarin Kenpom
Scoring margin No Yes Yes Yes
Diminishing returns for blowouts No Yes Yes No
Pace of game matters No Yes No Yes
Home/Neutral/Road Yes Yes Yes Yes
SOS beyond Opponent's opponents' W-L No Yes Yes Yes
All wins are better than losses (before Opp Adj) Yes Yes No No
De-weighting games with missing key players No Yes No No

Olympic Fan
02-12-2012, 11:09 AM
SOS beyond Opponent's opponents' W-L No Yes Yes Yes

Maybe I'm confused, but if you're measuring what I think you're measuring, RPI DOES measure SOS beyond opponent's W-L percentage. The W-L percentage of your opponents' opponents is one-fourth of the RPI.

doctorhook
02-12-2012, 11:15 AM
SOS is number one just in case you want a response to the UNC whining about our schedule.

El_Diablo
02-12-2012, 11:35 AM
Maybe I'm confused, but if you're measuring what I think you're measuring, RPI DOES measure SOS beyond opponent's W-L percentage. The W-L percentage of your opponents' opponents is one-fourth of the RPI.

Yes, but it says "beyond Opponent's opponents' W-L." RPI is accounts for opponents and opponents' opponents, but not beyond that.

Jderf
02-12-2012, 11:42 AM
Interesting stuff. I'm both surprised and impressed that ESPN not only decided to put something like this out, but also seem to have done it well. At this point, it might be hard to tell if the BPI is in any way "superior" to the other ranking systems (except for the horribly flawed RPI), but it is definitely nice to have another evaluative tool. (Here is the table above, hopefully a little clearer.)



Includes
RPI
BPI
Sagarin
Kenpom


Scoring margin
No
Yes
Yes
Yes


Diminishing returns for blowouts
No
Yes
Yes
No


Pace of game
No
Yes
No
Yes


Home/Neutral/Road
Yes
Yes
Yes
Yes


SOS beyond opponent's opponent W-L
No
Yes
Yes
Yes


All Ws are better than Ls
Yes
Yes
No
No


De-weights games with missing players
No
Yes
No
No

Dr. Rosenrosen
02-12-2012, 12:01 PM
So I wonder how much the fellas on espn will be pushing BPI when talking about "the bubble" over the next few weeks. Surely they can't ignore RPI since the selection committee still seems to rely on it, right? But it is annoying to listen to analysts continually talking about RPI when so many people recognize its shortcomings. Will be interesting to see if it spurs a few more insightful conclusions... or maybe I'm just expecting too much from most of them. :rolleyes:

theAlaskanBear
02-12-2012, 12:46 PM
At first glance the BPI seems to do a good job.

It does have Duke higher than either KenPom or Sagarin, but on second glance their calculation of Net Points is the determining factor -- net points directly corresponds with the BPI rankings...there are no exceptions that I saw. It would be nice to know how they determine Net Points vs Avg since it has so much weight.

It also seems as though W-L take a back seat. Michigan St and Kansas both have more losses than Duke, with Duke beating each one head-to-head AND worse schedule strengths overall but are listed in front of Duke. Again...the correlation is that they both have higher Net Points. I'm thinking they may be over ranking margin of victory, but without the formula there is no way to know.

Anyone know if Mich St/Kansas have home losses? It could be the two home losses Duke took drag them down.

TexHawk
02-12-2012, 12:50 PM
Anyone know if Mich St/Kansas have home losses? It could be the two home losses Duke took drag them down.
They do not. KU lost to Davidson in KC, it's close to Lawrence, but it's still not a home game.

Wander
02-12-2012, 01:10 PM
[COLOR=#000000]I'm both surprised and impressed that ESPN not only decided to put something like this out, but also seem to have done it well.


Ha! Ken Pomeroy is the Ron Paul of the college basketball world. He'll never be on ESPN himself, but he's really gotten his views into the mainstream after years of working at it. The weighting games for lost players is an interesting idea, that I would imagine has to increase predictive power if done correctly.

By the way, I thought kenpom rankings did account for diminishing returns in blowouts. Could be mistaken, though.

tbyers11
02-12-2012, 02:22 PM
By the way, I thought kenpom rankings did account for diminishing returns in blowouts. Could be mistaken, though.

Kenpom does not account for diminishing returns in blowouts. It's the main reason Wisconsin is so high in his system. They demolished a bunch of weak teams by 30-40 early on. The diminishing returns and accounting for key players being injured seem to be the main distinguishing factors between the BPI and KenPom.

loran16
02-12-2012, 02:41 PM
For the record, someone checked and found a 98.9% correlation between this and Pomeroy.

I'm curious if the 1.1% difference is actually an improvement.

toooskies
02-12-2012, 03:20 PM
The question is whether it has predictive power. I.e., was it created purposely to fit all existing data? I mean, the article claims minor predictive advantages for past data, but you can tweak a number or two and make that happen in any system. It doesn't mean it's right, though.

So, give the BPI a year or two to "breathe" and let's see how predictive it gets. The tweaks for injured players and blowouts should theoretically help, but their choice of corrective factors may be limited.

And as always, the best question is: does it beat gambling lines in Vegas? Those typically have the throne in terms of predictive power.