BYU, Ranking and Lamar

[pelagius]

There is a pretty big difference between BYU's ranking in most of the computer model rankings and the RPI. I guess there is a big difference between the computer models and humans polls now as well. For example, Sagarin currently ranks BYU at #43. Actually, the ranking is even worse if you look at Sagarin's margin of victory based model: #55. I was wondering how much the Lamar loss is affecting these ratings. To answer the question, I computed my own computer model for the 2006-2007 season. It is very similar to the Sagarin margin of victory based rating system. Here is my model's top ten, BYU's ranking, and Lamar's ranking:

----------------------------
rank team            rating
----------------------------
1    North Carolina  28.2682
2    Kansas          23.9379
3    Florida         23.5786
4    Texas A&M       23.1958
5    Memphis         21.3694
6    Ohio State      21.2867
7    UCLA            21.0292
8    Maryland        20.6553
9    Duke            20.5696
10   Wisconsin       20.1581
----------------------------
52   Brigham Young   12.7539
236  Lamar           -3.7958
----------------------------

My point is not that BYU deserves to be rank 52nd or that my model's top 10 is correct (it is basically the same as Sagarin's MOV ranking). What I am interested in is measuring the effect of the Lamar game on BYU's computer rating. So I recomputed the model dropping the BYU vs Lamar game from the data. BYU ranking increases to the following:

----------------------------
rank team            rating
----------------------------
48   Brigham Young   13.6314
----------------------------

Dropping the Lamar game improves BYU's ranking 4 spots. I was actually a bit surprised by this. I thought it would be more influential. Still bad losses early in the season are clearly affecting things a bit.

[Indy Coug]

Lamar is only 3.33% of BYU's season.

[pelagius]

Originally Posted by Indy Coug

Lamar is only 3.33% of BYU's season.

Of course. But that doesn't mean it isn't an influential observation in terms of the computer models. My general question is how much does a really bad loss hurt a team in terms of the computer rating? Clearly, the fact that one single game is 3.33% of the season is relevant (and limits how influential it can be), but it is still possible that a single game could be highly influential in terms of model estimation. Lamar is clearly an outlier; on a neutral court, the model above would predict that BYU would beat Lamar by 17 points. The Lamar game is affecting BYU's ranking in computer polls more than any loss (hardly surprising since it is the worst team BYU lost to, but I wanted to get a sense for the quantitative magnitude of the effect):

How much the rank increases by if we remove a loss:
UCLA          0
Boise St.     2
Michigan St.  2
Lamar         4
UNLV          0
Colorado St.  3
San Diego St. 2

[Indy Coug]

There are 336 teams, so a move of 4 places in the rankings is minimal, relative to 336. This is due in large part because 96.7% of BYU's season performance is fairly indicative of a 40th to 50th place team; which is

1. Undefeated or nearly undefeated at home
2. Beat most of the poor teams they faced on the road
3. Lost to only teams 0.500 and above on the road, other than Lamar.

[pelagius]

Originally Posted by Indy Coug

There are 336 teams, so a move of 4 places in the rankings is minimal, relative to 336. This is due in large part because 96.7% of BYU's season performance is fairly indicative of a 40th to 50th place team; which is

1. Undefeated or nearly undefeated at home
2. Beat most of the poor teams they faced on the road
3. Lost to only teams 0.500 and above on the road, other than Lamar.

I don't agree with your first sentence, but the rest of it seems about right.

[jay santos]

I like the computer avg at http://www.masseyratings.com/cb/compare.htm better than any one ranking, whether it be Sagarin or Massey.

My model has consistently had BYU about ten ranks higher than the computer average, but lately that gap has narrowed and my model has BYU at #31 compared to #32 for the average.

I tested the same test, taking out the Lamar game.

In my model, BYU went from #31 to #23.

In the RPI, BYU went from #20 to #17.

fyi, my top 10

1 UCLA
2 North Carolina
3 Ohio St.
4 Kansas
5 Southern Illinois
6 Wisconsin
7 Arizona
8 Texas A&M
9 Florida
10 Memphis

[pelagius]

Originally Posted by jay santos

I like the computer avg at http://www.masseyratings.com/cb/compare.htm better than any one ranking, whether it be Sagarin or Massey.

My model has consistently had BYU about ten ranks higher than the computer average, but lately that gap has narrowed and my model has BYU at #31 compared to #32 for the average.

I tested the same test, taking out the Lamar game.

In my model, BYU went from #31 to #23.

In the RPI, BYU went from #20 to #17.

fyi, my top 10

1 UCLA
2 North Carolina
3 Ohio St.
4 Kansas
5 Southern Illinois
6 Wisconsin
7 Arizona
8 Texas A&M
9 Florida
10 Memphis

Jay, The win/loss based computer models I think on average are ranking BYU higher than the MOV victory based models. For example, the Sagarin W/L model ranks BYU at #28. Do you know what the split is between these two kinds of models at the Comparison page.

[jay santos]

Originally Posted by pelagius

Jay, The win/loss based computer models I think on average are ranking BYU higher than the MOV victory based models. For example, the Sagarin W/L model ranks BYU at #28. Do you know what the split is between these two kinds of models at the Comparison page.

Good question. Interesting point. I don't know what the split is.

I was wondering earlier in the season why I had BYU so much higher than the computer avg at massey comparison. That's probably the reason. I use both W/L and MOV (with diminishing principle), but I think the W/L might have more weight.

[pelagius]

Originally Posted by jay santos

Good question. Interesting point. I don't know what the split is.

I was wondering earlier in the season why I had BYU so much higher than the computer avg at massey comparison. That's probably the reason. I use both W/L and MOV (with diminishing principle), but I think the W/L might have more weight.

Your top 10 seems to reflect the W/L weight. You have Southern Illinois at #5. Sagarin MOV and my MOV have them at 31 and 27 respectively. The Sagarin W/L has them at #9. Where do you have Nevada-Reno? They look like at W/L darling as well.

[jay santos]

Originally Posted by pelagius

Your top 10 seems to reflect the W/L weight. You have Southern Illinois at #5. Sagarin MOV and my MOV have them at 31 and 27 respectively. The Sagarin W/L has them at #9. Where do you have Nevada-Reno? They look like at W/L darling as well.

I have Nevada at 18. I'm pretty sure Colley is W/L only, and he has BYU #26, as well as a high ranking for S Ill. I checked my model, and I have near equal weights to W/L and my modified MOV.