Understanding Pomeroy's Ratings... - Basketball Forum : Professional and College Basketball Forums

Title_BU · 12-31-2012, 08:47 AM

Preseason ratings: Why weight?

----------------------

Simply, they aren't "rankings", they are a statistical attempt at projecting incremental games. The ranking mechanism is biased to preseason rankings until nearly February because of the rules of statistical significance. In short, they are the polar opposite of RPI (or ELO Chess) which rank what has happened. They attempt to "rank" what is to come.

It'll be interesting to see what happens when Ken gets around to implementing his MOV cap (ie beating Alcorn State by 30 and 60 is the same thing). He said previously he was looking at that and understands the problem no cap presents.

uz2b-len · 12-31-2012, 08:55 AM

If that's your simple version, send me an aspirin with the comprehensive version.

BM-GW · 01-02-2013, 06:02 AM

KenPom does conference title predictions. Number of times each team comes out on top out of 10,000 simulations...

Code:

Atlantic 10
VCU               6364
St. Louis         1263
Butler             775
Saint Joseph's     565
La Salle           381
Temple             267
Dayton             171
Xavier              94
Richmond            43
St. Bonaventure     33
Charlotte           15
Massachusetts       15
George Washington   13

**agoo101284** · 01-02-2013, 07:00 AM

A .33% chance? Lets go Bona's!!!!

I think there's a total of 9,999, so I'll go ahead and give Bona's a .34% chance!

uz2b-len · 01-02-2013, 07:36 AM

Looks like his simulation doesn't include the luck factor. Or includes it the wrong way.

res · 01-02-2013, 10:00 AM

Quote:

Originally Posted by uz2b-len

Looks like his simulation doesn't include the luck factor.

It shouldn't. As Pomeroy notes on his blog, the luck factor has nothing to do with his rating calculation. It is merely the deviation in winning percentage between a team’s actual record and its expected record.

Bona84 · 01-02-2013, 10:54 AM

With the New Year upon us, and non-conference schedules nearing completion for most, here are Pomeroy's current "rankings" and projected conference records based on data through 1/1/13:

13 VCU 13-3
30 St. Louis 11-5
38 Butler 11-5
52 La Salle 10-6
54 St. Joe's 10-6
62 Temple 9-7
67 Dayton 9-7
84 Richmond 8-8
94 Xavier 7-9
99 St. Bona 8-8
103 Charlotte 7-9
113 GW 7-9
114 UMass 7-9
200 URI 4-12
205 Duquesne 4-12
249 Fordham 3-13

Didn't add them up, but the records may not sum to .500 due to rounding.

Free Quebec · 01-02-2013, 11:24 AM

Quote:

Originally Posted by Bona84

With the New Year upon us, and non-conference schedules nearing completion for most, here are Pomeroy's current "rankings" and projected conference records based on data through 1/1/13:

13 VCU 13-3
30 St. Louis 11-5
38 Butler 11-5
52 La Salle 10-6
54 St. Joe's 10-6
62 Temple 9-7
67 Dayton 9-7
84 Richmond 8-8
94 Xavier 7-9
99 St. Bona 8-8
103 Charlotte 7-9
113 GW 7-9
114 UMass 7-9
200 URI 4-12
205 Duquesne 4-12
249 Fordham 3-13

Didn't add them up, but the records may not sum to .500 due to rounding.

GW has been playing much better the last three weeks,with Dwayne Smith re-discovering his missing game and the Garino/Armwood combination turning GW into a very good defensive ball club. It's nice to see the efficiency ratings picking that up - and to see GW moving ahead of UMASS into 12th.

That said, I would gladly take 7-9 in the A-10 today. I would consider that a successful season for this young team.

jpschmack · 01-02-2013, 12:44 PM

Quote:

Originally Posted by uz2b-len

Looks like his simulation doesn't include the luck factor. Or includes it the wrong way.

It's funny to me that no one has created a system which drops the most unexpected results at each end of the spectrum for each team. We're talking about 18-22 year old college kids here. Hell, everyone has "the flu game" each year, where because you're on a college campus, traveling, living and eating together, it runs through the entire team in the course of a week.

uz2b-len · 01-02-2013, 12:46 PM

I think UMass may have the widest divergence between KenPom and RPI in history, like 100 vs. 40, or something like that.
edit: 114 vs. approximately 40

Title_BU · 01-02-2013, 12:54 PM

Quote:

Originally Posted by jpschmack

It's funny to me that no one has created a system which drops the most unexpected results at each end of the spectrum for each team. We're talking about 18-22 year old college kids here. Hell, everyone has "the flu game" each year, where because you're on a college campus, traveling, living and eating together, it runs through the entire team in the course of a week.

The issue is sample size. The reason that preseason ratings maintain weight until the end of January is the sample is just too small. Eliminating "outlier" games (if they were uniformly identifiable) would create a situation where there would be MORE weight to preseason ranking. That's a problem.

The simplest "haircut" without eliminating sample, so to speak, is the margin of victory/loss cap.

jpschmack · 01-02-2013, 12:56 PM

Quote:

Originally Posted by Bona84

With the New Year upon us, and non-conference schedules nearing completion for most, here are Pomeroy's current "rankings" and projected conference records based on data through 1/1/13:

13 VCU 13-3
30 St. Louis 11-5
38 Butler 11-5
52 La Salle 10-6
54 St. Joe's 10-6
62 Temple 9-7
67 Dayton 9-7
84 Richmond 8-8
94 Xavier 7-9
99 St. Bona 8-8
103 Charlotte 7-9
113 GW 7-9
114 UMass 7-9
200 URI 4-12
205 Duquesne 4-12
249 Fordham 3-13

Didn't add them up, but the records may not sum to .500 due to rounding.

We really need Fordham, GW, URI and Duquesne, Bona and Richmond to only beat each other (no offense to fans of those teams -- and I am a Bona fan).

Our ideal standings would be:
16. FOR: 0-16
15. URI - 1-15
14. GW - 2-14
13. DUQ - 3-13
11. Bona 4-12
11. Rich 5-11
Seven teams tied at 11-5
Three teams with weakest resumes at 12-4

won't happen, but I'm rooting for us getting a ridiculous amount of bids.

jpschmack · 01-02-2013, 01:01 PM

Quote:

Originally Posted by Title_BU

The issue is sample size. The reason that preseason ratings maintain weight until the end of January is the sample is just too small. Eliminating "outlier" games (if they were uniformly identifiable) would create a situation where there would be MORE weight to preseason ranking. That's a problem.

The simplest "haircut" without eliminating sample, so to speak, is the margin of victory/loss cap.

It's all a matter of what you're trying to accomplish with these ratings.

I don't ever bother with Sagarin or PomRoy because... I don't gamble. That's really the only reason for these ratings to exist: gambling purposes.

Rating teams to rate teams and see who's better is relatively worthless. We have a nice big tournament at the end of the year to determine who's the best. Replicating the NCAA's formula for RPI has some value, because that's the metric they use to pick teams and we want to know where we all stand in that now.

Numbers are just numbers. The important thing is to know WHY the numbers are what they are, and what the numbers mean.

Free Quebec · 01-02-2013, 02:46 PM

Quote:

Originally Posted by uz2b-len

I think UMass may have the widest divergence between KenPom and RPI in history, like 100 vs. 40, or something like that.
edit: 114 vs. approximately 40

GW is 207 in the RPI, but 113 in the Pomeroys. That's a difference of 84 (so even larger than UMASS' difference - though GW is hurt by the RPI, whereas UMASS is helped by it).

12-31-2012, 08:47 AM	#1 (permalink)
Title_BU 6th Man Join Date: Oct 2009 Posts: 335 Rep Power: 7011	Understanding Pomeroy's Ratings... Preseason ratings: Why weight? ---------------------- Simply, they aren't "rankings", they are a statistical attempt at projecting incremental games. The ranking mechanism is biased to preseason rankings until nearly February because of the rules of statistical significance. In short, they are the polar opposite of RPI (or ELO Chess) which rank what has happened. They attempt to "rank" what is to come. It'll be interesting to see what happens when Ken gets around to implementing his MOV cap (ie beating Alcorn State by 30 and 60 is the same thing). He said previously he was looking at that and understands the problem no cap presents.

12-31-2012, 08:55 AM	#2 (permalink)
uz2b-len Player Join Date: Oct 2005 Posts: 817 Rep Power: 5048	Re: Understanding Pomeroy's Ratings... If that's your simple version, send me an aspirin with the comprehensive version.

01-02-2013, 06:02 AM	#3 (permalink)
BM-GW Player Join Date: Dec 2002 Location: Arlington, VA Age: 49 Posts: 635 Rep Power: 4813	Re: Understanding Pomeroy's Ratings... KenPom does conference title predictions. Number of times each team comes out on top out of 10,000 simulations... Code: Atlantic 10 VCU 6364 St. Louis 1263 Butler 775 Saint Joseph's 565 La Salle 381 Temple 267 Dayton 171 Xavier 94 Richmond 43 St. Bonaventure 33 Charlotte 15 Massachusetts 15 George Washington 13

01-02-2013, 07:00 AM	#4 (permalink)
agoo101284 -LIFETIME MEMBER- Join Date: Jun 2003 Location: Bronx, NY Age: 28 Posts: 7,760 Rep Power: 1595652	Re: Understanding Pomeroy's Ratings... A .33% chance? Lets go Bona's!!!! I think there's a total of 9,999, so I'll go ahead and give Bona's a .34% chance!

01-02-2013, 07:36 AM	#5 (permalink)
uz2b-len Player Join Date: Oct 2005 Posts: 817 Rep Power: 5048	Re: Understanding Pomeroy's Ratings... Looks like his simulation doesn't include the luck factor. Or includes it the wrong way.

01-02-2013, 10:54 AM	#7 (permalink)
Bona84 Star Join Date: Oct 2003 Location: Westchester County, NY Posts: 2,760 Rep Power: 261039	Re: Understanding Pomeroy's Ratings... With the New Year upon us, and non-conference schedules nearing completion for most, here are Pomeroy's current "rankings" and projected conference records based on data through 1/1/13: 13 VCU 13-3 30 St. Louis 11-5 38 Butler 11-5 52 La Salle 10-6 54 St. Joe's 10-6 62 Temple 9-7 67 Dayton 9-7 84 Richmond 8-8 94 Xavier 7-9 99 St. Bona 8-8 103 Charlotte 7-9 113 GW 7-9 114 UMass 7-9 200 URI 4-12 205 Duquesne 4-12 249 Fordham 3-13 Didn't add them up, but the records may not sum to .500 due to rounding. __________________ Bona84

01-02-2013, 12:46 PM	#10 (permalink)
uz2b-len Player Join Date: Oct 2005 Posts: 817 Rep Power: 5048	Re: Understanding Pomeroy's Ratings... I think UMass may have the widest divergence between KenPom and RPI in history, like 100 vs. 40, or something like that. edit: 114 vs. approximately 40 Last edited by uz2b-len; 01-02-2013 at 01:41 PM.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode