Subject: | |
From: | |
Reply To: | |
Date: | Wed, 17 Nov 1993 14:45:20 PST |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
I imagine my computer rating system is about as meaningless as any poll,
this early in the season, but I thought I'd post anyway. BGSU fans
might thank me!
The W-L records shown include only those games against other teams on the
list, and not exhibitions, Canadian college games, etc. The "rating"
in the last column is on a "log-odds" scale, explained below; a 1-point
difference means that the higher team has a 73% chance of beating the
lower team.
1. Michigan 7- 0- 1 12.718
2. Bowling Green 5- 0- 1 12.577
3. Maine 5- 0 12.488
4. New Hampshire 6- 0 12.166
5. Harvard 3- 0 11.726
6. Lake Superior 6- 2 11.442
7. Colorado Coll 6- 0- 2 11.394
8. Mass-Lowell 4- 0- 1 11.019
9. Northeastern 4- 1 10.996
10. Michigan Tech 4- 2- 2 10.842
11. N Michigan 6- 2 10.751
12. Boston U 3- 2 10.591
13. St Lawrence 3- 2 10.571
14. Brown 2- 1 10.471
15. Merrimack 4- 3 10.408
16. RPI 3- 2 10.340
17. Wisconsin 4- 3- 1 10.222
18. Michigan State 4- 2- 1 10.136
19. Cornell 1- 0- 1 10.118
20. Miami 2- 4 9.943
21. Minn-Duluth 3- 3- 2 9.921
22. Notre Dame 2- 3- 1 9.872
23. Alaska-Anchorage 3- 5 9.778
24. W Michigan 3- 4- 1 9.763
25. Providence 2- 5 9.749
26. Clarkson 2- 2- 1 9.724
27. Colgate 2- 3 9.661
28. Denver 3- 3 9.620
29. Boston College 1- 2- 1 9.575
30. Dartmouth 1- 2 9.505
31. North Dakota 3- 5 9.435
32. Ohio State 1- 4- 1 9.386
33. Alaska-Fairbanks 3- 5 9.376
34. Kent 2- 3- 1 9.286
35. Minnesota 1- 5- 2 9.194
36. Massachusetts 0- 1 9.127
37. St Cloud 2- 5- 1 9.083
38. Ill-Chicago 2- 6 9.053
39. Princeton 0- 2 8.994
40. Yale 0- 2 8.994
41. Union 1- 2 8.945
42. Ferris State 1- 8 8.774
43. Air Force 0- 6 8.498
44. Vermont 0- 4- 1 8.398
45. Army 0- 4 7.559
Harvard's surprisingly high position is based mostly on their defeat of
#14 Brown (otherwise undefeated), though Brown's victories were not
exactly against powerhouses!
The rating system is my attempt to combine win-loss records with strength
of opposition to produce one overall figure representing the strength of
each team. The ratings are related to win probabilities by (non-
mathematicians can ignore this bit) prob=1/(1+exp(-rating diff)), which
converts any (+ or -) rating difference into a probability between
zero and one. The ratings are found by matching the observed wins with
expected wins, given the strength of opposition.
One of these days I'll find a more user-friendly rating scale!
--
Ken Butler
[log in to unmask]
|
|
|