Can Anybody Explain Sagarin's Ratings?

[furpop16]

With apparently nothing better to do, I've been keeping track of Furman's relative position as posted by several of the Usual Suspects. Of course, FU's standing improved considerably after the Loyola-Chicago and Villanova games. Since Villanova things have settled into patterns and here are my general observations:

• Massey (today 32) has kept FU in the low 30's, except after Asheville when FU had a Massey ranking of 26
• KenPom (today 115) has rated FU ~ 115-120; again except after Asheville when KenPom rated FU at 112
• ESPN BPI (today 71) has indexed FU in the low 70's, except after So. Wesleyan (79) and WCU (89)
• NCAA NET and AP, though recent, have FU consistently improving, From 46 to 28 and from 25 to 23, respectively

Rather consistent in each case, wouldn't you agree? At least consistent enough from which to draw conclusions relative to those of other teams. Now here are Sagarin's ratings for FU:

• After Villanova: 74
• After So. Wesleyan: 76
• After UNC Asheville: 95
• After WCU: 119
• After Elon: 84
• Today, after SC Upstate: 110

Can anybody make any sense of that? Some on this board have a high regard for Sagarin's ratings. I even heard Dan Scott recently refer to Sagarin on his show. But if there's information contained in Sagarin's ratings, it is completely escaping me.

[youwouldno]

Sagarin is roughly comparable to Pomeroy but with one major difference - Pomeroy places much more weight on prior year results as a starting point early in the season. That has a "smoothing" effect on the ranking unless the results are dramatically different from expectations. Sagarin jumps around early on due to the limited number of matrix data points.

Massey is a little different because there is a benefit to winning games, which seems reasonable but is problematic from a computer rankings standpoint . . . it might have some merit late in the season but it can cause some odd results early on.

[furpop16]

Sagarin jumps around early on due to the limited number of matrix data points.

My goodness; we’re 75% through the OOC schedule. If I were to draw any conclusion from his ratings I’d say the guys were getting worse. Granted the OOC schedule since Villanova hasn’t been grueling, but that only underscores the point that all of the other systems at least have us consistently holding our own.

I wonder if Jeff Sagarin developed his ratings for some purpose different than those of the other systems ... and at what point during the season will his ratings be valid for his purpose?

[Paul C]

I read an exchange on twitter with Pomerory where he said his pre-season rankings are fully "out of the calculation" by mid-Jan.

[youwouldno]

My goodness; we’re 75% through the OOC schedule. If I were to draw any conclusion from his ratings I’d say the guys were getting worse. Granted the OOC schedule since Villanova hasn’t been grueling, but that only underscores the point that all of the other systems at least have us consistently holding our own.

I wonder if Jeff Sagarin developed his ratings for some purpose different than those of the other systems ... and at what point during the season will his ratings be valid for his purpose?

At a high level, Sagarin is similar to Pomeroy in focusing mostly on point margin to determine the ratings. The ratings displayed by Sagarin ("Predictor") and Pomeroy are intended to predict the point differential outcome of a game between any two teams (e.g., Furman is 5 points better than Team X - which then has to be adjusted for home team, unless played at a neutral site).

In a completely pure application of this method, with no pre-season weighting, the results after the first game would be largely meaningless. Say the first 2 games of the season were at a neutral site:

Furman 80, UNC-Asheville 60
Florida 80, UNC 60

The system would rate Furman and Florida as +10, and UNC-Asheville and UNC as -10.

Then, let's say the third game happens at the same neutral site and UNC beats UNC-Asheville 80-60. Even though Furman and Florida didn't play another game, their ratings will change because more information is available about their previous opponent.

The system after that third game would say something like, Florida +15, Furman +5, UNC -5, UNC-Asheville -15. Florida is up and Furman is down because of the additional data point.

There is no exact number of data points that make this method "valid," but after 5 games or so they will generally be reasonable. The process is complicated by the fact that teams can play better or worse as the season progresses (the different systems have various ways of trying to account for this).

The ratings therefore are some combination of how good a team has been overall, assuming all teams were static in their abilities, and how good a team is right now, based on more recent results. Sagarin and Pomeroy vary somewhat in those kinds of areas, but overall are fairly similar and tend to converge as the season progresses.

[furpep]

Don't give a rip about rankings. Todays AP story called Furman the "feel good story" of college basketball. Yep, we are. The heck with these computer guys. Keep winning that is all that matters!

[tya1]

I seem to remember Ken Pomeroy saying that his preseason rating (last year's results adjusted for the roster changes for this year) doesn't leave the system until all D1 teams are "connected." They become connected when you can trace a game results link between every team. That usually happens in January.

[dinhead]

The Sagerin rating system is rather complex but it basically considers strength of schedule and margin of victory. However there is a principal of diminishing returns which limits how much a team can gain by rolling up the score. Since wins and losses are not considered, even though Furman is undefeated the fact that we play one of the weakest non-conference schedules in the country greatly limits how high we can be ranked.

Furman is currently ranked #112
Here are the Sagerin rankings for the teams we've played:
#24 Villanova
#108 Loyola Chicago
#196 Gardner Webb
#250 Charleston Southern
#311 Elon
#312 Western Carolina
#336 USC-Upstate
#344 UNC-Asheville
Games against Bob Jones, North Greenville and Southern Wesleyan don't count but clearly none of these non-D-1 teams would be even considered to be in the top 400 in the country.
Therefore, Furman has only played four teams ranked in the top 300 in the country.
Furman's overall strength of schedule is ranked #335 out of 353 D-1 teams and that doesn't even include the non-D-1 games. Cleary one of the weakest if not the weakest D-1 schedule in the country to date.

UNC-Wilmington is #201 and LSU is #49 so that will surely increase it some.

In the SoCon Furman trails #77 Wofford and #88 UNC-Greensboro and is slightly ahead of #115 ETSU.

So far Furman appears to be a "one hit wonder" with a fantastic win over Villanova, no other top 100 wins and a tendency to often play down to it's level of competition. However it's hard to measure "finding a way to win" and the confidence and belief that comes from winning and so far the Paladins have that much going for them.

[stonemd]

Wofford and uncg ~80 in sagarin and kenpom

Furman 112, 115

Doesn't make any sense to me
At least they will be good wins if we get em