sagarin's eigenvector predictions??

[gojetsgomoxies]

anyone know what this is?

sagarin a few years ago starting adding specific predictions for games.... 3 of his long-time ratings plus a recent rating and an eigenvector rating.

recent is interesting as you can work on whether lines over/under react based on recent play.....

eigenvector, I have no idea. I somewhat know what an eigenvector is.... but no idea how he's using it.

and as presented, some of sagarin's Eigenvector picks are way off his 3 conventional and recent ratings picks.

bowl games I see,

big one is psu-memphis.... conventional (C) = 7-9 psu fave, recent(R) 2 psu fave, Evector = Memphis 6 fave

hawaii, kast, Oregon, Utah, Iowa... all are different EV than C/R

let me double-check on those as I wrote them fast.

[gojetsgomoxies]

psu-memphis is basically correct as I stated.


Hawaii conventional (C) zero.. EV 8 fave
kast C 0 EV 7 fave
wash C 0 to 6 fave... EV boise 8.5 fave... different game winner...
social C zero EV USC 6
Oregon C zero EV Oregon 16 massive difference in spread
Utah C 3.5... EV 14.5 big difference

note that for most of the games I mentioned are basically sagarin basic rating suggesting close game... I didn't filter for spreads that are way different.

[KVB]

Likely using eigenvector to limit noise in the data.

It's possible the eigenvector rating is coming from different data as well.

Not sure, not familiar with Sagarin at all but there are only so many uses and whatever data he's using, he's probably just breaking down principle components then trying to id the highest eigenvalue and then I suppose making a prediction based on that one component with the higher value.

It might differ a great deal in prediction because the other components are used in the other predictions but the eigenvector may just select a principal component to base it on.

That's likely how it would be used, whether or not he is ranking components who knows, but he is probably identifying a principal component anyway.

[stevenash]

I eat, sleep, and shit everything Sagarin.

Short answer he places a premium on teams beating stronger teams.

[KVB]

The more I think about this, though, the less I am sure.

lol.

[gojetsgomoxies]

Likely using eigenvector to limit noise in the data.

It's possible the eigenvector rating is coming from different data as well.

Not sure, not familiar with Sagarin at all but there are only so many uses and whatever data he's using, he's probably just breaking down principle components then trying to id the highest eigenvalue and then I suppose making a prediction based on that one component with the higher value.

It might differ a great deal in prediction because the other components are used in the other predictions but the eigenvector may just select a principal component to base it on.

That's likely how it would be used, whether or not he is ranking components who knows, but he is probably identifying a principal component anyway.

KVB, thanks. the PCA idea is interesting................ the problem with the eigenvector predictions is that in at least one case (psu-memphis) the prediction is so different from his other 4 ratings. so either 1) other info?, 2) error? 3) something else............ actually the reason psu-memphis eigenvector is so noticeable is that it has a different significant favorite. but the spread in oregon-wisconsin is way different although the conventional ratings (recent too?) = close game.

[gojetsgomoxies]

I eat, sleep, and shit everything Sagarin.

Short answer he places a premium on teams beating stronger teams.

historically, i think he had 2 or 3 basic ratings (and then maybe an average of 2 or more of them)... he definitely had the rating (elo from chess) that was only based on winning the game.

more i think about it, my comment doesn't really relate to your comment so i will think about your comment some more.