Are the RPI Rankings An Accurate Measure of Basketball Team Strength?

Submitted by mGrowOld on February 19th, 2011 at 11:07 PM

Everybody knows that the sacred RPI rating plays a huge role in determining who is tournament selection worthy and who is not. As of now Michigan is ranked 61st in the most current RPI rankings.  The Belmont Bruins are ranked 53rd with their impressive 24-4 record built by playing THIS schedule:


M …
Date Opponent Opp. RPI Rk Results
  11-16   at  Tennessee  16-10 (6-5) 26  76-85  L
  11-17   vs  Arkansas St.  12-14 (8-5) 203  93-60  W
  11-22   vs  Marist  4-24 (3-13) 313  102-74  W
  11-23   vs  Winthrop  10-14 (8-8) 198  71-44  W
  11-29   at  Tennessee St.  9-15 (8-8) 222  87-72  W
  12-02     Mercer  12-15 (10-7) 246  89-67  W
  12-04   at  Vanderbilt  19-6 (7-4) 14  76-85  L
  12-07     Middle Tenn. St.  13-13 (9-5) 197  88-87  W
  12-16     Kennesaw St.  7-19 (6-11) 324  87-60  W
  12-18   at  Troy  7-17 (6-7) 249  98-63  W
  12-20   at  Alabama St.  7-16 (6-6) 320  66-53  W
  12-23   at  Tennessee  16-10 (6-5) 26  65-66  L
$  12-30     Miami (OH)  13-12 (9-3) 97  83-72  W
  01-03   at  Fla Gulf Coast  6-19 (3-12) 298  83-51  W
  01-05   at  Stetson  7-20 (5-12) 301  70-53  W
  01-08     North Florida  10-18 (7-10) 172  91-59  W
  01-10     Jacksonville  17-8 (12-5) 116  81-50  W
  01-13     Lipscomb  15-10 (11-6) 110  88-52  W
  01-15     Campbell  11-15 (6-11) 252  90-55  W
  01-21   at  South Carolina-Upstate  4-23 (3-14) Mm 317  67-62  W
$  01-23   at  E. Tennessee St.  18-10 (14-4) 100  72-62  W
?  01-25   at  Lipscomb  15-10 (11-6) 110  64-73  L
  01-28     Fla Gulf Coast  6-19 (3-12) 298  89-56  W
  01-30     Stetson  7-20 (5-12) 301  82-64  W
  02-03   at  Jacksonville  17-8 (12-5) 116  76-70  W
  02-05   at  North Florida  10-18 (7-10) 172  69-67  W
  02-12   at  Campbell  11-15 (6-11) 252  78-57  W
$  02-17     E. Tennessee St.  18-10 (14-4) 100  68-58  W

Look at the RPI rankings of their vanquished foes.  How is it possible that beating this collection of cupcakes garners a team an RPI ranking higher than ours?  What is wrong with this picture?????

And did they really play Tennessee twice?



February 19th, 2011 at 11:45 PM ^

albeit with a lot of data behind it.

RPI = (WP * 0.25) + (OWP * 0.50) + (OOWP * 0.25)

where WP is Winning Percentage, OWP is Opponents' Winning Percentage and OOWP is Opponents' Opponents' Winning Percentage.

Or if you prefer, see Ken Pomeroy's summary

I am almost certain one could find a measure that provides a better predictor of future wins. Nonetheless, Ken Pomeroy defended it here (scroll way down to see beginning of the post long ago in comparison to the early BCS methodology, which kept USC out of the title game in 2003

And considering its simplicity, the RPI is pretty good at what it's supposed to do. Winning games is good, playing a tough schedule is good. But more fundamentally, just being a good team is good. There’s really no easy way to cheat the system. And even if you could, there is a bunch of humans waiting at the end of the season to try to smooth out the problems.

Posted on 12/26 at 01:20 AM


February 19th, 2011 at 11:51 PM ^

A home win now counts as 0.6 win, while a road win counts as 1.4 wins. Inversely, a home loss equals 1.4 losses, while a road loss counts as 0.6 loss. A neutral game counts as 1 win or 1 loss. Note that this location adjustment applies only to the WP factor and not the OWP and OOWP factors. Only games against Division 1 teams are included for all RPI factors.


February 20th, 2011 at 12:11 AM ^

Because winning percentage varies much more than opponents' winning percentage.

For example, the best teams have a winning percentage of about .900, the worst have winning percentages of about .100.  Opponents' winning percentage only varies from about .400 to about .600.  So the best teams get (.900 x .25) = .225 RPI points for their winning percentage and the worst teams get (.100 x .25) = .025 RPI points for their winning percentage.  About a .200 difference between the best and worst teams.

But in opponents' winning percentage (OWP), the teams with the strongest schedules get (.600 x .5) = .300 RPI points for their OWP, and the teams with the weakest schedules get (.400 x .5) = .200 RPI points for their OWP.

So even though the factor for winning percentage is only .25 and the factor for opponents' winning percentage is .50, really the difference between the best and worst teams is about twice as big for winning percentage compared with OWP.  You  could say that winning percentage is weighted about twice as heavily as opponents' winning percentage, when variance is taken into account.


February 19th, 2011 at 11:52 PM ^

The RPI is actually a really dumb formula that is intentially dumbed down so that most people who care to learn the formula can easily understand it.

What they should actually shoot for is a ratings index that is the best predictor of future results based on wins and losses alone. This would require lots of fancy math and statistics, so it's not going to happen. The NCAA prefers a system that slaps randomly chosen factors to opponents and opponents-opponents win-loss record because it's easy to compute.


February 20th, 2011 at 12:15 AM ^

Exactly--the big problem is that RPI was created in the late 1970s, about 5-10 years before home computers became ubiquitous.  It was therefore determined that a linear calculation was preferable, rather than a matrix-based or iterated calculation, even though the math makes much more sense with a matrix-based formula.

So the real problem is that they had to use a suboptimal approach to rating, in order to appease the athletic department people who wanted to be able to check the NCAA's math without using a computer.


February 20th, 2011 at 12:51 PM ^

The best predictive systems often rate teams higher than they deserve to be based on their actual accomplishments. For instance, Sagarin's post-national championship game predictive system ranks Auburn 5th, even though they won the national titile. Alabama, by contrast, is ranked 3rd. Oregon is ranked 2nd, and Stanford is ranked 1st. In college basketball, Kenpom ranks Clemson 38th, despite their middling schedule, a few poor losses, and their best win coming over #40 FSU. The reason for both of these oddities is that Clemson and the teams ranked above Auburn lost relatively close games and won a number of games by wide margins. This is important because margin of victory or defeat is generally a better predictive measure than who actually wins or loses a game.

This is a problem area when deciding who is actually more deserving of a tournament spot. On one hand, it is obviously better to beat a team by 20 than to beat them by 2. On the other hand, if Team A wins beats two teams by a margin of 5 points apiece and Team B beats one of them by 15 and loses the other by 1, I doubt anyone would say that Team B is more deserving of a bid than Team A as the ultimate goal in team sports is to win, and the margin of a victory is a secondary concern.


February 19th, 2011 at 11:29 PM ^

It's certainly not everything but it works as a basic indicator. The key with RPI is that you really need to look at SOS at the same time.

Michigan- RPI: 61 SOS: 18

Belmont- RPI: 53 SOS: 230

So yes, Belmont has a slightly better RPI but the SOS has to be considered. If resumes were based on this alone, Michigan would far and away be more impressive.


February 20th, 2011 at 10:09 AM ^

this is why the wisky game is so important.  a W would mean a decent jump in rpi.  our strength of schedule will also jump after the final 3 games.

that said, the rpi is too simplistic a formula.  at least the BCS considers several formulas to attempt to eliminate any strange bias of a single formula.


February 20th, 2011 at 12:17 AM ^

For college basketball, here is how you calculate RPI:

RPI = Winning Percentage * 0.25 + Opponents' Winning Percentage * 0.5 + Opponents' Opponents' Winning Percentage * 0.25.

Winning percentage is adjusted for road and home advantages. Home wins count as 0.6 wins, home losses count as 1.4 losses, road wins count for 1.4 wins, road losses count for 0.6 losses. Neither opponents' winning percentage nor opponents' opponents' winning percentage is calculated with this adjustment.

Opponents' winning percentage is also adjusted to remove any games you play. So if you beat a team, you do not reduce your opponents winning percentage for the purpose of the RPI. Opponents' opponnents' winning percentage, on the other hand, does not remove any games from consideration.

The reason that Belmont's RPI is higher than Michigan's (albeit very slightly) is simple. While Belmont has faced a weak schedule, their winning percentage as calculated by the RPI is very high as they have played a lot of games on the road and won a lot of them too. Meanwhile, they have home losses to drag down their winning percentage.

The biggest issue with the RPI, in my opinion, is that it does not care whether you have any quality wins. That is why the NCAA considers quality wins an additional item to consider alongside RPI when deciding who goes to the tournament.


February 20th, 2011 at 4:42 PM ^

Hey Fuckwad.....I mean Larry:

I've sat quietly and taken your shit long enough.  I'm not quite sure which one of the politically correct edicts I broke to so incur your rightious indignation but I've had enough of it.  So here's a summary of the score to date:

I've got three kids who love me (the most important thing I have).  I would be shocked if you had any clue what fatherhood was like

I've got a far better job than you (own my own company with 250+ employees and are moving forward with an IPO in spring 2012)

I've got way cooler toys than you (just bought a CTS-V's awesome)

I've got a WAY hotter wife than you who also loves me (see my avatar for proof)

I've got lots of friends

I am actively involved in my church, serve on three charitable boards and go to inner city Cleveland Sunday nights to feed the homeless.

So while I cant speak for the entire board I guess I do think i'm better than you.  Go fuck yourself Larry or Busey or whatever the hell you call yourself.

 Truth is a bitch asshole.


February 20th, 2011 at 6:55 PM ^

Hey Faggots, My name is John, and I hate every single one of you. All of you are fat, retarded, no-lifes who spend every second of their day looking at stupid ass pictures. You are everything bad in the world. Honestly, have any of you ever gotten any pussy? I mean, I guess it's fun making fun of people because of your own insecurities, but you all take to a whole new level. This is even worse than jerking off to pictures on facebook.

Don't be a stranger. Just hit me with your best shot. I'm pretty much perfect. I was captain of the football team, and starter on my basketball team. What sports do you play, other than "jack off to naked drawn Japanese people"? I also get straight A's, and have a banging hot girlfriend (She just blew me; Shit was SO cash). You are all faggots who should just kill yourselves. Thanks for listening.

Mitch Cumstein

February 20th, 2011 at 7:34 AM ^

At the end of the day, if you're record is 24-4, you're always going to have a solid computer ranking.  Think of the BCS, undefeated teams are usually very high in the computers regardless of who they've been.   A teams record is always a substantial part of these rankings.


February 20th, 2011 at 9:51 AM ^

I guess that's my point.  Playing cupcake U....and winning....seems to give teams a favorable RPI ranking.  And given that the RPI. not SOS, is the metric most commonly quoted when evaluating a teams tourny chances I'm questioning its validity.  As someone accurately pointed out, our SOS is one of the nation's best (thanks Big 10) yet we are barely mentioned as a bubble team, much less in, because of our poor RPI.

Doesnt seem right to me.


February 20th, 2011 at 1:35 PM ^

As an at-large, Belmont's tourney chances aren't better than Michigan's. The committee likes to use RPI, but not simply as 'team with RPI 57' > 'team with RPI 58'. Record vs top 25 and top 50 teams by RPI matter a lot, and Belmont's 0-3 record vs the top 50 almost certainly keeps them from getting an at-large spot


February 20th, 2011 at 8:46 PM ^

 And given that the RPI. not SOS, is the metric most commonly quoted when evaluating a teams tourny chances I'm questioning its validity.

SOS is part of the RPI formula.  Our SOS is the reason our RPI is as high as it is.  17-11 probably isn't even one of the 100 best records in the country.