Lunar Eclipse

In a recent post, Ken Pomeroy explains how and why the "adjusted RPI" is not much of an improvement on the home-road neutral RPI.

Kenpom begins by reviewing Joe Lunardi's latest column in which ESPN's "Bracketologist" considers the impact of the NCAA's decision to weigh road games more than home games on the 2003-2004 Richmond Spiders. Running the numbers, Lunardi, a big proponent of the championships committee's adjustment, finds our beloved bubble team safely tournament-bound.

Importantly, the Spiders would go from a No. 49* RPI on Selection Sunday to a ranking most likely in the high-30s. Their at-large bid is a whole lot more secure in that range, as it should have been all along. There is also a much greater incentive for teams to schedule more road games in the first place.

Sounds great, no? The "adjusted RPI" works as planned, rewarding road victories and providing an incentive for greater travel. But after Pomeroy checked Lunardi's math, I learned that "Say it Ain't So" Joe only told us half the story.

Lunardi asserts that Richmond would have vaulted 10-12 spots up the RPI food chain under the new formula. The fault with this is that he only did the calculation for the Spiders. With a road bonus, everyone's RPI value will improve, assuming they won at least one road game. So I went ahead and recalculated RPI data for all teams, just to see how much Richmond actually improved.

Richmond was a +3 from 47* to 44. Not the huge move that Lunardi predicted. Why is that? While Richmond's road record of 8-7 was slightly better than average, when compared to the better teams of college hoops, it's not exceptional.

And neither is the new formula's impact, according to Ken. Putting the chimps to work, Pomeroy found none of the "adjusted RPI's" big winners or losers among the basic RPI's Top 50, rendering the new system "largely irrelevant" in his eyes.

Though I happen to believe dropping Georgia out of the running and cementing Washington's place in the tournament are critical contributions.

Teams that were hurt
1) Rutgers -11 (56 to 67)
2) Michigan -5 (55 to 60)
3) St. Louis -4 (64 to 68)
4) Georgia -3 (48 to 51)
4) Nothern Iowa -3 (59 to 62)

Teams that were helped
1) Villanova +10 (67 to 57)
2) Pacific +6 (65 to 59)
3) Wisconsin Milwaukee +5 (75 to 70)
4) Washington +4 (56 to 52)
4) Colorado +4 (57 to 53)

Importantly, however, Pomeroy also identified a critical side-effect of the "adjusted RPI." His calculations suggest the new formula will benefit major conference teams that have mediocre records. Witness those Wildcats.

An alleged improvement can have unintended consequences. Villanova had only five road wins. However, the effect of the road bonus has more impact on teams with worse records, and Villanova had the worst record of teams in the top 70 at 15-15. Think about it, if a team has a perfect record, their winning percentage cannot be adjusted any higher no matter how many road wins they have.

Wow. I see the boost to average teams from elite conferences as more than an unintended consequence, but as a critical flaw. As it stands, the big boys get enough love from the NCAA on Selection Sunday. If Kenpom's analysis rings true this March, look for (muted) calls to scrap the "adjusted RPI."

*You may have noticed that Lunardi (49) and Pomeroy (47) peg the Spiders at different home-road neutral RPIs. Why? Because Lunardi computes "InsideRPI" numbers using a formula that tries to replicate the secret RPI adjustments the NCAA selection committee is believed to make. As Ken notes, these alleged adjustments would have hurt Richmond in 2003-2004.

**But you may not have noticed the change in Lunardi's column since it was first published. I believe I just caught ESPN.com red-handed.

Interestingly, Lunardi appears to have revised (or fudged, but let's give the worldwide leader's "Bracketologist" the benefit of the doubt) his most important assumption, the value he employed as a multiplier. In my original post on Lunardi's work, I excerpted from his piece as follows, suggesting Joe used a multiplier of 1.5.

How would this translate numerically into an RPI rating? Since we know that home teams win roughly two-thirds of the time in Division I, a road victory should be worth about twice as much as winning at home (because, by the numbers, it is twice as hard to achieve).

I'm guessing the new NCAA formula won't go that far (essentially saying that road wins are worth 1.50 victories and home wins are only 0.75), as this would be too dramatic a departure from the "known" without sufficient study of the unknown effects. But we can at least split the difference and reprogram Factor I (winning percentage) to, say, 1.2 "equivalent victories" for road wins and 0.8 equivalent victories for those at home. Similarly, road losses would count as only 0.8 while losing at home would count 1.2.

But in the version currently posted online, Joe seems to have used a multiplier of 1.25.

How would this translate numerically into an RPI rating? Since we know that home teams win roughly two-thirds of the time in Division I, a road victory should be worth about twice as much as winning at home (because, by the numbers, it is twice as hard to achieve).

I'm guessing the new NCAA formula won't go that far (essentially saying that road wins are worth 1.50 victories), as this would be too dramatic a departure from the "known" without sufficient study of the unknown effects. But we can at least split the difference and reprogram Factor I (winning percentage) to, say, 1.25 "equivalent victories" for road wins.

Given that Ken, Joe and I agree that the size of the road game bonus has a significant affect on the impact (or lack there)of the "adjusted RPI," changing multipliers is more than risky business. It's scandalous.

***For even more fun with numbers, read Ken's thoughts on why (1) while home teams win roughly two-thirds of the time in Division I, a road victory should not be worth about twice as much as winning at home and (2) winning percentage has much more influence on the RPI than does strength of schedule.