re: A statistics nerd on Reddit plugged all 128 teams into a computer model

quote:

---

I mean, it's a neat idea and this guy did a good job, but it's hardly a good predictor for this year. It would be a much better predictor if he used the first 3-5 games of the season as a basis for the model, in my opinion.

---

It's difficult to really come up with anything substantial. With football even for a whole season you have a sample size of 11-13 games, which is small. You need a sample size of about 6 games before the data is useful enough.

I am a big Ken Pomeroy fan, so I modeled my football preseason ratings similar to how he does his basketball preseason ratings.

One thing that is hurting Auburn is the baseline portion of my system. They were so bad in 2012 that that figure is still dragging them down. The baseline is the rolling 5 year average, with more weight to recent seasons.