FanLax Forum Poll

[Gobigred]

If you look at the rankings in your posting that I quoted, Penn is #7 and Yale is #9. Or do you have multiple rsnkings?

There needs to be an Anchor Ranking or the cat will chase it's tail endlessly. Based on Penn being #7 and Yale being #9, Penn becomes #whatever and Yale become #whatever and... It never stops. There needs to be an anchor. For D1 Men's FanLax Computer Ranking, the Anchor Ranking is the FanLax Forum Poll.

If you're using that as an anchor, then Yale is #9 and Penn #7, not what you stated above. Otherwise, it's you going in circles.

[laxfan1313]

Polls are fun to analyze, but as one who remembers the screw job Cornell received in 1970, I'm happy the final "poll" is decided on the field. Then there's the issue of how the selection committee decides who gets to compete for the NC. It appears we are edging closer to a tourney with fewer unfair exclusions. With two play-in games, there are now 20 teams. Raising it to 24 would perhaps eliminate bad feelings. I know the NCAA doesn't think there are enough D1 teams to justify a 24 team field but as more teams are added to D1, this may come to fruition. It's hard to believe the playoff in 1971 was 8 teams!

[Matnum PI]

If you look at the rankings in your posting that I quoted, Penn is #7 and Yale is #9. Or do you have multiple rsnkings?

There needs to be an Anchor Ranking or the cat will chase it's tail endlessly. Based on Penn being #7 and Yale being #9, Penn becomes #whatever and Yale become #whatever and... It never stops. There needs to be an anchor. For D1 Men's FanLax Computer Ranking, the Anchor Ranking is the FanLax Forum Poll.

If you're using that as an anchor, then Yale is #9 and Penn #7, not what you stated above. Otherwise, it's you going in circles.

7 Penn 50% 7 11
9 Yale 78% 4 18

In the Anchor Ranking (i.e. The Forum Poll, below), Penn is #12 and Yale is #4. Based on using this as an Anchor Ranking, The FanLax Computer ranks Penn as #7 and Yale as #9 (above).

[Gobigred]

Using Quint's rankings (because they're conveniently handy above), top ten wins for his top ten teams:

Maryland 4 (#3, #4, #6, #10)
Georgetown 1 (#10)
Princeton 2 (#2, #4)
Rutgers 1 (#8)
Yale 1 (#3)
Virginia 1 (#10)
Cornell 2 (#5, #8)
Ohio State 1 (#10)
Jacksonville 0
Notre Dame 0

[Gobigred]

Quoting FanLax Computer:

"In the Anchor Ranking (i.e. The Forum Poll, below), Penn is #12 and Yale is #4. Based on using this as an Anchor Ranking, The FanLax Computer ranks Penn as #7 and Yale as #9 (above)."

Then Cornell's loss to Penn is a "better" loss than Princeton's to Yale, as I wrote above..

[Matnum PI]

Quoting FanLax Computer:

"In the Anchor Ranking (i.e. The Forum Poll, below), Penn is #12 and Yale is #4. Based on using this as an Anchor Ranking, The FanLax Computer ranks Penn as #7 and Yale as #9 (above)."

Then Cornell's loss to Penn is a "better" loss than Princeton's to Yale, as I wrote above..

No. The better or worse loss is based on the Anchor Rankings. Cornell's loss to #12 Penn is worse than Princeton's loss to #4 Yale.

P'ton's Three Worst losses are: Yale (4), Maryland(1), and 0. Total=5. Cornell's Three Worst Losses are: Penn (12), 0, and 0. Total=12. 5 is better than 12.

[Wheels]

So with Duke's big win, they'll jump to #2 this week.

[CU77]

So with all this discussion, I decided to dust off my old KRACH program and see what it says.

KRACH uses wins and losses only. If team A has rating RA and team B has rating RB, the probability that A beats B is predicted to be RA/(RA+RB). The ratings are determined by maximizing the probability that all games played to date were won by the favored team. For normalization, each team is assumed to have a win and a loss against a fictitious team with a rating of 100.

Top 25 as of now (including Duke's win over UVA); last number is the KRACH rating rounded to the nearest whole number.

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100
4. Cornell 10-1 1050
5. Rutgers 10-2 827
6. Yale 7-2 702
7. Virginia 8-3 501
8. OSU 8-3 450
9. Harvard 7-2 440
10. Jacksonville 10-2 399
11. Army 9-2 377
12. Boston U 8-3 344
13. Penn 4-4 315
14. Duke 10-5 315
15. UNC 7-4 309
16. St Josephs 9-2 269
17. Lehigh 7-3 266
18. Denver 6-5 254
19. Notre Dame 5-4 251
20. Richmond 6-4 243
21. Brown 6-4 233
22. Utah 6-3 202
23. UMass 6-4 191
24. Villanova 6-4 182
25. St Bonaventure 8-2 176

[rolldodge]

So with all this discussion, I decided to dust off my old KRACH program and see what it says.

KRACH uses wins and losses only. If team A has rating RA and team B has rating RB, the probability that A beats B is predicted to be RA/(RA+RB). The ratings are determined by maximizing the probability that all games played to date were won by the favored team. For normalization, each team is assumed to have a win and a loss against a fictitious team with a rating of 100.

Top 25 as of now (including Duke's win over UVA); last number is the KRACH rating rounded to the nearest whole number.

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100
4. Cornell 10-1 1050
5. Rutgers 10-2 827
6. Yale 7-2 702
7. Virginia 8-3 501
8. OSU 8-3 450
9. Harvard 7-2 440
10. Jacksonville 10-2 399
11. Army 9-2 377
12. Boston U 8-3 344
13. Penn 4-4 315
14. Duke 10-5 315
15. UNC 7-4 309
16. St Josephs 9-2 269
17. Lehigh 7-3 266
18. Denver 6-5 254
19. Notre Dame 5-4 251
20. Richmond 6-4 243
21. Brown 6-4 233
22. Utah 6-3 202
23. UMass 6-4 191
24. Villanova 6-4 182
25. St Bonaventure 8-2 176

Makes a lot of sense.

[Gobigred]

Quoting FanLax Computer:

"In the Anchor Ranking (i.e. The Forum Poll, below), Penn is #12 and Yale is #4. Based on using this as an Anchor Ranking, The FanLax Computer ranks Penn as #7 and Yale as #9 (above)."

Then Cornell's loss to Penn is a "better" loss than Princeton's to Yale, as I wrote above..

No. The better or worse loss is based on the Anchor Rankings. Cornell's loss to #12 Penn is worse than Princeton's loss to #4 Yale.

P'ton's Three Worst losses are: Yale (4), Maryland(1), and 0. Total=5. Cornell's Three Worst Losses are: Penn (12), 0, and 0. Total=12. 5 is better than 12.

Tha Anchor Ranking you posted in this thread at 10:42 am yesterday has Penn at 7 and Yale at 9. Why don't you look at it.

[nms]

So with all this discussion, I decided to dust off my old KRACH program and see what it says.

KRACH uses wins and losses only. If team A has rating RA and team B has rating RB, the probability that A beats B is predicted to be RA/(RA+RB). The ratings are determined by maximizing the probability that all games played to date were won by the favored team. For normalization, each team is assumed to have a win and a loss against a fictitious team with a rating of 100.

Top 25 as of now (including Duke's win over UVA); last number is the KRACH rating rounded to the nearest whole number.

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100
4. Cornell 10-1 1050
5. Rutgers 10-2 827
6. Yale 7-2 702
7. Virginia 8-3 501
8. OSU 8-3 450
9. Harvard 7-2 440
10. Jacksonville 10-2 399
11. Army 9-2 377
12. Boston U 8-3 344
13. Penn 4-4 315
14. Duke 10-5 315
15. UNC 7-4 309
16. St Josephs 9-2 269
17. Lehigh 7-3 266
18. Denver 6-5 254
19. Notre Dame 5-4 251
20. Richmond 6-4 243
21. Brown 6-4 233
22. Utah 6-3 202
23. UMass 6-4 191
24. Villanova 6-4 182
25. St Bonaventure 8-2 176

Thank you for this. By my eye test, I think that one team is seriously overvalued -
St. Joe's - 6 schools beneath them have wins over top15 teams, and St. Joes does not. I expect that the Win % is pushing St. Joe's up.
I also think a case can be made to rank OSU over UVA, since the OSU has one loss to a lower ranked team (Denver), vs. two losses by UVA to lower ranked teams (Richmond and Duke). I know that OSU has one win vs Harvard, and UVA has two over UNC, so this is more debatable.

[Gobigred]

So with all this discussion, I decided to dust off my old KRACH program and see what it says.

KRACH uses wins and losses only. If team A has rating RA and team B has rating RB, the probability that A beats B is predicted to be RA/(RA+RB). The ratings are determined by maximizing the probability that all games played to date were won by the favored team. For normalization, each team is assumed to have a win and a loss against a fictitious team with a rating of 100.

Top 25 as of now (including Duke's win over UVA); last number is the KRACH rating rounded to the nearest whole number.

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100

I like this. Princeton has wins over Rutgers and Georgetown, both much better than Georgetown's best.

[wgdsr]

Polls are fun to analyze, but as one who remembers the screw job Cornell received in 1970, I'm happy the final "poll" is decided on the field. Then there's the issue of how the selection committee decides who gets to compete for the NC. It appears we are edging closer to a tourney with fewer unfair exclusions. With two play-in games, there are now 20 teams. Raising it to 24 would perhaps eliminate bad feelings. I know the NCAA doesn't think there are enough D1 teams to justify a 24 team field but as more teams are added to D1, this may come to fruition. It's hard to believe the playoff in 1971 was 8 teams!

with the 2 play-in games, there are "unofficially" 18 teams in the field. 10 auto qualifier and 8 at large.

officially, it'll be 8 and 8 with the 2 play-in winners.

[Matnum PI]

FanLax Forum Polls due on Monday at noon EST. Enjoy the game this weekend.

[CU77]

By my eye test, I think that one team is seriously overvalued - St. Joe's

KRACH does tend to overrate (compared to an eye test) teams with strong records against weak schedules.

[LynahFan]

By my eye test, I think that one team is seriously overvalued - St. Joe's

KRACH does tend to overrate (compared to an eye test) teams with strong records against weak schedules.

Well, that’s subjective at best. KRACH recognizes the reality that even when team A is “better” than team B, team B will win some percentage of the games between them. If team A is only a “little bit better” than teams B through K, then it would be really unlikely for team A to win all 10 games against them. If all 11 teams are roughly equivalent, that would be like flipping 10 heads in a row.

If team A does, in fact, win all 10 of those games, then it’s far more likely that team A is *a lot* better than the other 10 teams, so that the results are more like not rolling a 1 in ten throws of a 10-sided die. Although A has not beaten “anybody,” they’ve beaten *everybody*, which is unlikely unless team A is pretty darn good.

However, a team in that situation does have a rather “fragile” rating - one bad loss will demolish their KRACH rating, so trying to game the system by scheduling patsies is a double edged sword - if you’re going to take that approach, then you’d better go very close to undefeated. Beating a bad team does not raise your KRACH very much, but losing to a bad team drops your score precipitously. Playing the next bad team (in a marginal sense) is nothing but downside - can’t help you, but can hurt you.

[CU77]

Well, that’s subjective at best.

Well of course. But (as you know) KRACH treats all teams as equal a priori. For KRACH to figure out that one group of teams is overall less good than another requires a significant number of games between those two groups. And typically there just aren't enough.

For example, in 2009, Notre Dame was playing in the Great Western Lacrosse League, and was undefeated. Had one big out-of-conference win, over UNC. KRACH rated Notre Dame #1.

ND got the #7 seed from the committee (UNC was #6). Then ND lost their first-round game to unseeded Maryland, 7-3.

That's just one anecdotal case, but I think a thorough analysis of NCAA tournament results would support the notion that a strong record against weak teams is not a good predictor of tournament success.

[Chousnake]

So with all this discussion, I decided to dust off my old KRACH program and see what it says.

KRACH uses wins and losses only. If team A has rating RA and team B has rating RB, the probability that A beats B is predicted to be RA/(RA+RB). The ratings are determined by maximizing the probability that all games played to date were won by the favored team. For normalization, each team is assumed to have a win and a loss against a fictitious team with a rating of 100.

Top 25 as of now (including Duke's win over UVA); last number is the KRACH rating rounded to the nearest whole number.

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100
4. Cornell 10-1 1050
5. Rutgers 10-2 827
6. Yale 7-2 702
7. Virginia 8-3 501
8. OSU 8-3 450
9. Harvard 7-2 440
10. Jacksonville 10-2 399
11. Army 9-2 377
12. Boston U 8-3 344
13. Penn 4-4 315
14. Duke 10-5 315
15. UNC 7-4 309
16. St Josephs 9-2 269
17. Lehigh 7-3 266
18. Denver 6-5 254
19. Notre Dame 5-4 251
20. Richmond 6-4 243
21. Brown 6-4 233
22. Utah 6-3 202
23. UMass 6-4 191
24. Villanova 6-4 182
25. St Bonaventure 8-2 176

This looks pretty fair and accurate compared to some of the other polls. It doesn't overate the ACC based on reputation. For example, ND at 19 makes more sense than at 11 or even 7(!)

[Matnum PI]

1. Maryland 10-0 2947
2. Princeton 8-2 1105
3. Georgetown 9-1 1100
4. Cornell 10-1 1050
5. Rutgers 10-2 827
6. Yale 7-2 702
7. Virginia 8-3 501
8. OSU 8-3 450
9. Harvard 7-2 440
10. Jacksonville 10-2 399
11. Army 9-2 377
12. Boston U 8-3 344
13. Penn 4-4 315
14. Duke 10-5 315
15. UNC 7-4 309
16. St Josephs 9-2 269
17. Lehigh 7-3 266
18. Denver 6-5 254
19. Notre Dame 5-4 251
20. Richmond 6-4 243
21. Brown 6-4 233
22. Utah 6-3 202
23. UMass 6-4 191
24. Villanova 6-4 182
25. St Bonaventure 8-2 176

This looks pretty fair and accurate compared to some of the other polls. It doesn't overate the ACC based on reputation. For example, ND at 19 makes more sense than at 11 or even 7(!)

From this list (above), Notre Dame beat #14, #NR SU, and #NR UMich and they lost to #3, #7, and #8 and... You think #19 seems about right? i dunno. Logically, based on their Ws and Ls, I'd say they should fall somewhere between #9 and #13.

[CU77]

Richmond has a better win, over #7. Worst loss is to #15.

A method like KRACH looks at ALL wins and losses, and tries to make sense of them.

Here's an interesting question: there are many "closed loops" in the results where A > B > ... > C > A. What should happen to the ratings (any ratings, by any system) if we reverse the loop? That is, suppose that instead of A > B > ... > C > A, the results had been A < B < ... < C < A ? What should happen to the ratings?

KRACH's answer is nothing: reversing a loop does not change KRACH ratings at all.

The idea is that knowing A > B > C > A (the shortest loop) conveys no more or less information that A < B < C < A. We can't order these 3 teams based on these results; all we can do is say that their ratings should be closer to each other than they would be if those games had not been played at all.

(This assumes that we are not taking timing into account; many systems put more weight on more recent results.)