Having gotten drawn into so much discussion about the RPI, I decided to build a ranking system myself. Here's what I came up with. Code: Team Coef 1 New Mexico 1.25 2 UCLA 0.91 3 CSU Bakersfield 0.86 4 UC Irvine 0.85 5 UC Davis 0.47 6 CSU Fullerton 0.42 7 Washington 0.39 8 UC Riverside 0.36 9 CSU Northridge 0.35 10 St Mary's 0.35 11 San Diego 0.33 12 San Jose St 0.29 13 UCSB 0.29 14 Sacramento St 0.22 15 San Diego St 0.21 16 Cal Poly 0.15 17 Air Force 0.14 18 Santa Clara 0.11 19 Portland 0.08 20 Oregon St 0.07 21 Denver 0.05 22 San Francisco 0.00 23 Loyola Marymount -0.02 24 Gonzaga -0.21 25 Cal -0.32 26 Stanford -0.33 27 Seattle -0.40 28 UNLV -0.41 1. These rankings make use of all games between western teams, which is to say, they ignore all out-of-region games. Making nationwide rankings would've required a lot more work. 2. The rankings ignore the scoreline of the games, taking each game's outcome as either a win or a loss (or 1/2 win, 1/2 loss in the case of a tie). The rankings do incorporate home field advantage and strength of opposition. 3. The coefficients come from logistic regression, which is a standard technique used to predict binary outcomes. There are other details, but I doubt anyone cares much. My system turns out to like New Mexico a lot and doesn't regard UC Irvine too highly. It also doesn't rate UCSB very highly, but that's because UCSB's best results came against teams from the east. It also looks like there's a lot of parity (not too surprising).
Through 10/31 Code: Team Power 1 New Mexico 1.43 2 UCLA 1.05 3 CSU Bakersfield 0.82 4 UC Irvine 0.70 5 UCSB 0.49 6 UC Riverside 0.48 7 Washington 0.44 8 UC Davis 0.41 9 Air Force 0.39 10 CSU Northridge 0.36 11 St Mary's 0.35 12 San Jose St 0.34 13 CSU Fullerton 0.34 14 Portland 0.28 15 Sacramento St 0.27 16 Santa Clara 0.23 17 San Diego St 0.21 18 Cal Poly 0.17 19 San Diego 0.17 20 Loyola Marymount 0.12 21 Denver 0.09 22 San Francisco 0.01 23 Oregon St -0.03 24 Cal -0.30 25 Stanford -0.31 26 Gonzaga -0.33 27 Seattle -0.49 28 UNLV -0.50 Two big differences between my system's results and the RPI: 1. My system puts Bakersfield a lot higher. This isn't because of out-of-region results because all of Bakersfield's games were in the west. 2. My system puts the Big West lower. This is partially because of out-of-region results (esp UCSB's), but I think it's mostly because RPI's formula skews things too much towards a conference's overall record.
Your system does the opposite of the RPI, it penalizes teams for going out of region, by not counting those games... Still the big west has three of the top six western teams, four of the top 8 and five of the top 10, so the penalty is not that severe. Irvine and Ucsb played a lot of good out of region games that don't count in this.
Do you think GauchoDan would be willing to share his results file? If so, I'd be happy to cook up nationwide ratings. My system is nothing fancy, but I'm pretty sure that it makes sense. It avoids the RPI's most glaring weaknesses, like having a team's rating go up when it loses or down when it wins.
That isn't necessarily a weakness. If the best team in the country were to play the worst team in the country and only eek out a 1 or 2 goal win, I would expect to see that "best" team drop and the "worst" team improve. Any ranking that uses a team's strength of schedule as a factor in its calculation will do this, as far as I can figure out. I think the two worst things about the RPI mechanism are that it doesn't reward teams for playing on the road and that it doesn't take margin of victory into account. A 4-0 victory is about the same as a 5 or 6-0 victory (at some point domination is domination), but it certainly isn't the same as a 1-0 double OT game. There's the whole East Coast bias thing, but I don't really know how I would correct for that, while I can think of easy ways to correct for the issues I just stated (play with the value of a "win" in the system, so that all wins are not equal). The best thing about the RPI is that it's actually very easy to understand in comparison to most computer ranking systems. Your western rating is interesting, and I think reflects a lot of what we left coasters have been saying this year - the Big West is good, the Pac-10 and WCC are in a down year, and the MPSF has a couple of great teams. It does underrate schools who played big games out of the west, but without going national, there's no way around that. I've been tempted to do something similar, but having to manually look up and input the score of every game in the country turns me off every time. I have a lot of respect for those who can pull through that tedium and come up with something useful.
I think the top drawer soccer scoreboard has it one page per week, so it should be too hard to get all the results (unless it's incomplete or inaccurate). http://www.topdrawersoccer.com/college-soccer/college-scoreboard/men 11 weeks so far?
As much as I agree with you, I also think this is very complicated. You don't want one lopsided scoreline to distort a team's entire season, and you also don't want to assign too much importance to a team's playing style. Some teams win with blowouts, others focus more on defending small margins. For those reasons, I decided it was safer just to focus on wins/losses/ties.
That's a good suggestion. I've been following their scoreboard a lot (and also lateralsports). It's a terrific resource. Unfortunately, they don't appear to identify neutral site games, so it's not ideal for my purposes. Perhaps I'll use it anyway.