ADVERTISEMENT

Computers, including Herbie's FPI, weigh all games equally

11NCs

ND Expert
Mar 4, 2011
1,447
123
63
and are heavily skewed by margin of victory. If later games are weighted more than earlier games, then computer rankings look an awful lot like the human polls. Sagarin's "Recent" ratings, below, demonstrate how applying less weight to Notre Dame's first four (very) narrow victories (including Michigan) yields a more accurate rating.

Rank (Recent) Rank (Standard) School Rating (Recent)
1 1 Alabama 105.73
2 2 Clemson 100.38
3 9 Notre Dame 95.5
4 3 Michigan 95.5

5 4 Georgia 94.84
6 5 Oklahoma 92.31
7 11 LSU 90.16
8 6 Ohio State 89.74
9 15 Washington State 86.79
10 7 West Virginia 86.37
11 25 Iowa State 85.92
12 8 Penn State 85.5
13 23 Central Florida(UCF) 84.99
14 12 Mississippi State 84.39
15 20 Missouri 84.33
16 18 Texas 83.9
17 10 Washington 83.81
18 24 North Dakota State 82.72
19 33 Northwestern 82.08
20 14 Utah 81.26
21 29 Utah State 81.2
22 34 Syracuse 80.75
23 48 Pittsburgh 80.64

24 21 Michigan State 80.44
25 16 Texas 80.18
 
Computers need to meet a hammer. Period.

Just like boxing styles make fights. Not points scored or how pretty a team looks.
 
and are heavily skewed by margin of victory. If later games are weighted more than earlier games, then computer rankings look an awful lot like the human polls. Sagarin's "Recent" ratings, below, demonstrate how applying less weight to Notre Dame's first four (very) narrow victories (including Michigan) yields a more accurate rating.

Rank (Recent) Rank (Standard) School Rating (Recent)
1 1 Alabama 105.73
2 2 Clemson 100.38
3 9 Notre Dame 95.5
4 3 Michigan 95.5

5 4 Georgia 94.84
6 5 Oklahoma 92.31
7 11 LSU 90.16
8 6 Ohio State 89.74
9 15 Washington State 86.79
10 7 West Virginia 86.37
11 25 Iowa State 85.92
12 8 Penn State 85.5
13 23 Central Florida(UCF) 84.99
14 12 Mississippi State 84.39
15 20 Missouri 84.33
16 18 Texas 83.9
17 10 Washington 83.81
18 24 North Dakota State 82.72
19 33 Northwestern 82.08
20 14 Utah 81.26
21 29 Utah State 81.2
22 34 Syracuse 80.75
23 48 Pittsburgh 80.64

24 21 Michigan State 80.44
25 16 Texas 80.18

How is that “more accurate”??
 
and are heavily skewed by margin of victory. If later games are weighted more than earlier games, then computer rankings look an awful lot like the human polls. Sagarin's "Recent" ratings, below, demonstrate how applying less weight to Notre Dame's first four (very) narrow victories (including Michigan) yields a more accurate rating.

Computer models are much more sophisticated than what you’re portraying them here. They literally take a look at every play at every gsme situation (tight gsme, blowout situation, opponent, etc.) and give an approximation of how a team has performed counting EVERY play. There is no “eye-test” equivalent of evaluating how a team played on every down against every opponent for all teams considered.

Computers need to meet a hammer. Period.

Just like boxing styles make fights. Not points scored or how pretty a team looks.

What does this even mean? Who’s to determine which “styles” will make for good “fights”? That’s purely subjective.

Margin of victory, strength of opponents, yards and first downs gained/allowed in non blowout situations ARE the best way to fairly evaluate teams.
 
ND is still getting judged pretty harshly in the computer generated systems due to the first 3 games of the season where ND struggled to move the ball efficiently vs Ball St and Vanderbilt. Back when Dexter Williams and Ian Book were on the bench.

Human polls are better able to account for this effect (the Ian Book effect). Which is why ND is ranked in the 3-4 range in human polls. While the computer generated ranking systems have them in the 6-7 range.

ND has been an entirely different team/offense with Dexter Williams and Ian Book since week 4 but the computer systems have no way of discerning the player personnel changes.

EDIT: After re-reading the OP, it looks like "sagarin" 'weights' recent games higher than earlier games in the season, which is why NDs sagarin rank is #3/4.

Other systems like FEI/S&P+ don't weight earlier games in the season any different than recent ones. And i kind of agree with this methodology. The sample sizes are already so small that weighting earlier games in the season less just makes the data even less reliable.
 
Last edited:
How is that “more accurate”??

Very simply, a team’s expected future performance and, more importantly, its current strength relative to other teams, is more accurately determined by how they performed in weeks 8-12 than in weeks 1-4.
 
ADVERTISEMENT