methomps;632779; said:
Computers don't factor in margin-of-victory. A win is a win, and a loss is a loss.
You are right that the BCS rankings do not explicitly include a margin of victory component but that does not mean that scores are no longer indirectly considered as part of the models.
The standard computer ratings (as opposed to the BCS ratings) often include margin of victory, either explicitly by including scores or implicitly by including variables closely related to score such as complex indexes of weighted adjusted scores, offensive or defensive yardage, or some other measure. Its not completely clear that the BCS computer ratings do not also include a score component, even if it is indirect.
The American Statistician journal reported that the algorithms used by some of the computer ratings are not even clear to the people who conduct the ratings. I don't remember Massey or any of the others who contributed to that issue rebutting that assertion either. So, this is how I understand the BCS computer ratings and I stand subject to correction.
Power ratings are designed to identify meaningful transitive links between teams: A beats B and B beats C, therefore A should beat C. However, as we all know from the Big Ten last year, sometimes C beats A.
The BCS computer ranking formulae still may include some score element in order to help make models more predictive in such circumstances, even if there is no direct margin of victory component included in any of the rankings by agreement.
How would this happen? Well, even though most of the raters closely guard their formulae, some of the computer ratings indicate that they make adjustments for home field advantage, which would be an explicit adjustment for score some standardized home field advantage score, and in that sense be an adjustment for margin of victory even if a small one.
Another example is that some of the models seem to suggest that they are making an adjustment for quality opposition. But how would you do that? Adjust for offensive points scored for or against quality opponents? Use some other proxy variable that is highly related to score such as and index computed by dividing offensive yards against and opponent by the yards scored against them by all others? It's not clear what they are doing.
So, it is not entirely clear that margin of victory is removed entirely from the BCS ratings because it may be in there implicitly or as part of a more complex index of some kind.
I know this wasn't your point, methomps, but margin of victory is a component of the overall computer rankings and can swing press momentum toward a team and perhaps influence the pollsters in the press and coaches polls, Vegas odds, and public opinion. All of which can affect a team's overall BCS rating eventually.
If you think back to last year, we were getting lots of attention for the BCS bowl invitation even when we were on the outside looking in on the BCS. The press started picking up on our position in the normal computer ratings, which showed that even with two losses we were one of the top two or three teams when all of the information was taken into account.
In that case, our normal computer ratings were leading our BCS (i.e., Sagarin ELO) computer rankings by a few weeks and the press started picking up on the computer ratings of us versus Oregon and Notre Dame, especially on strength of schedule. Florida's claim to a #1 spot if it makes it through the next two weeks will largely be based on a strength of schedule argument that is based at least in part on computer ratings. So, margin of victory can impact on the normal and BCS polls, even if the BCS computer rankings explicitly exclude margin of victory.
To get beyond all of the stats and jargon, I just have a hard time believing that a team can hold on to number one if another team is number 1 in all of the computer ratings. Others may hold the opinion that the polls just don't hold that much sway with the poll voters but I think that public opinion can.
And if the voting is tight and three teams are undefeated, this suggests that nice guys could finish last.
