I Don't Understand the Latest Sagarin Ratings
I'm focusing on two teams. Both have 24-6 records. Team A has the 32nd ranked schedule; Team B the 10th ranked schedule. Team A is 3-3 against Sagarin's Top 25 and 5-5 against Sagarin's Top 50. Team B is 4-5 against the Top 25 and 9-5 against the Top 50. Team A, Florida, is ranked 3rd. Team B, Michigan, is ranked 9th. If this was 3rd and 4th, I wouldn't be writing this, but can someone explain how Florida is 6 positions higher than Michigan?
Margin of victory.
I'm not sure that's the whole story. When you go to pure points, Michigan actually drops to 10 and Florida stays 3.
It's entirely possible for this to happen with Sagarin, as far as I understand. Some of the major variables in the algorithm are wins, losses and venue. Strength of schedule factors into it as well, as I believe his system compounds the records of opponents as well as opponent's opponents into the overall rating as well. It's similar to what he does with college football rankings. Road wins are weighted too, I believe.
The Predictor number, incidentally, does take victory margin into account. The ELO rating does not.
I think margin of victory is counted. For Florida, that means a lot since they've been beating up on a weak SEC
Seems to me like they should balance out.
High margin of victory but low SOS (or some indicator of weak conference) and low margin of victory but high sos (or some indicator of a strong conference) should keep Michigan and Florida closer, I would think.
Simple terms:
Michigan has outscored opponents by an average of 13 points per game. Florida has outscored opponents by an average of 19 points per game. So Michigan is about 6 points worse to start.
Then add in strength of schedule: Michigan's average opponent has been about 1 point per game stronger than Florida's per Sagarin (that's what the 79.70 vs. 78.56 is - strength of schedule scaled to a points per game metric).
Florida has been 6 points per game better in Margin of Victory, but 1 point worse in strength of schedule. To wit, they should rate about 5 points better than Michigan.
Michigan's predictor ranking is 89.38; Florida's is 94.26. In other words, Florida would be favored on a neutral court (per Sagarin) by 5 points.
It adds up. The reason it seems fishy is just that Sagarin thinks the strength of schedule is much closer (only 1 point) than the margin of victory (6 points).
Margin of victory does play a good deal when look at the losses as well. If it weren't for the East Lansing debacle, things would probably be a little different.
Florida has losses by 1, 3, 4, 6, 6, and 11 points.
Michigan has losses by 1, 3, 3, 6, 8, and 23 points.
Also, Michigan had eight games when they won by eight points or less. Florida's closest win was a twelve point victory. Which means two things aside from strength of competition: Florida knows how to put teams away better than Michigan, and Michigan is better apt to winning close games.
All the computer rankings do this (well at least the ones that don't suck and include margian of victory). Florida has been #1 all year in Kenpom by a wide margain. Brian talked about this early in the season when it was inflating teams that just massacred D-II schools by like a 100 instead of 50 (actually his examples, IIRC were Pitt and Wisconsin, who ended up being better than anyone expected, so maybe there is a lesson there).
There should probably be some diminishing-returns-to-blowouts feature, but I don't think most of the rankings have them.
March 13th, 2013 at 11:40 AM ^
...is a feature at Massey, and the primary reason I like his ratings. But even there, Florida is about 2 1/2 points ahead of Michigan.
Well, not sure if this has anything to do with it, but one team (Florida) actually knows how to box out and rebound, while the other team (Michigan) can't quite figure it out. If you can't rebound, you definitely do not deserve to be ranked in the top five of any rating system since the inability to rebound negates everything that a team does well. A team that can't rebound is susceptible to losing to any given team at any given time. Sorry for the negativity. I am proud of this team, but they simply do not deserve a top five ranking from anybody.
No, I don't think that has anything to do with it, especially when you consider UM is actually a better rebounding team than Florida.
March 13th, 2013 at 12:24 AM ^
His formula has always skewed towards rewarding teams that blow people out. He has always placed a huge amount of emphasis on that. It's one of my biggest complaints with his entire formula. I've never understood why his system gives value to teams that run up the score. It's even more ridiculous in football, where he considers winning by 28 to be somehow super-significantly greater than winning by 21 even though any reasonable person realizes that the difference between the two is simply a garbage-time TD.
The other item here and might be a factor in this case (I'm not claiming it is or it isn't, because I don't know) is the fact that Sagarin discounts earlier results. Something like 2 percent a week, give or take. So last week's games count 100 percent, the week before counts 98 percent, the week before counts (.98^2) = 96.04 percent, all of the way back to mid-November, which counts about (.98^16) = 72.4 percent.
I forget if it's 1 or 2 percent per week, but I do know that he does this. If a team's best results are earlier in the season, it will hurt them in the overall rating. I see the logic here--it's the same as the NCAA counting "record in last 10 games" as a factor--but I don't know if this makes a better prediction or not.
I believe Sagarin himself once revealed that he uses a Bayesian network weighted by starting rankings (preseason, I assume) and if you check his site regularly, the moment you see that the graph is "well-connected", the weighting is gone and the ratings are supposedly unbiased, but like someone else mentioned, teams that routinely crush their opponents do get some favor in his model historically.