Women's Hoops Blog

Inane commentary on a game that deserves far better


Monday, November 14, 2005

It's never too early to start talking about the RPI, muse of the all-powerful Selection Committee.

Reflecting coaches' growing frustration with the RPI, the WBCA recently announced that it would "no longer generate the RPI" and that it had "solicited the NCAA to make their RPI public." Later this year, NCAA will apparently do just that — but that's unofficial, it might be that only the men's Committee will disclose.

I like the idea of more transparency. If the Committee wants to rely on the RPI, then it should make it public and defend it in the open. If the RPI can't be defended in the light of day, then it shouldn't be used.

Last year I outlined a few complaints about the RPI. The 25-50-25 weighting system has some quirks. Among other things, it produces a strange mathematical inversion in the Strength of Schedule calculations. Last year that meant, for example, that beating Eastern Michigan was worth about as much to your RPI as beating UConn.

So what would the RPI look like if we used a different weighting system? What would your RPI be if it were 33-33-33 weighted rather than 25-50-25 weighted?

At the very top, there would be almost no difference. But once you get into bubble territory, you start to see some significant moves. Here are a few 2004-05 rankings under the alternate 33-33-33 RPI system, with the actual 25-50-25 RPIs in parentheses. (Thanks to Jerry Palm for providing the underlying data to make the calculations possible.)

22. UWGB (37)
26. Gonzaga (48)
27. Virginia (15)
31. Delaware (41)
32. Penn State (20)
37. Chattanooga (56)
41. Virginia Tech (30)
50. Liberty (67)
54. Eastern Kentucky (64)
55. Auburn (38)

Under the alternate system in the relevant range, small conferences did substantially better and big conferences did substantially worse. Some other possible systems — such 25-37-37 and 30-30-40 — have similar but generally less drastic changes.

It's tempting to conclude that the current 25-50-25 system has a built-in big conference bias. But stated unconditionally, that's not quite right. Some other system — such as 20-60-20 — would make the bigs look even better. Relative to the 20-60-20 system, the current system has a built-in small conference bias.

So what weighting should the RPI use?

In the abstract, there's no non-arbitrary way to answer that question. In the abstract, there's no (neutral, behind-the-veil) reason why 33-33-33 is better or worse than 25-50-25 — or, for that matter, 10-10-90 or 40-5-55.

But science could shed light. You could determine which weighting system has the greatest predictive success. It wouldn't be easy to study, and no system would be perfect or even close to it. But some systems might be better than others.

If some math PhD student can devise a better system, the Committee should switch. If nothing else, the Committee should at least be aware of these quirks and relative biases when it uses the RPI to make selections.