Beginning this week we are adding a couple of adjustments to our game predictions. Digging through historical performance we find that teams perform better in the week after a Thursday night game, and after a bye. In both cases, the effect is about 2 points. That is, teams coming off bye weeks or Thursday games outperform the predictions implied by our rankings by about 2 points. These considerations are beyond the scope of our Power Rankings, so to use them we need to adjust our game predictions explicitly.
We prefer to be conservative so are going to “shrink” these effects before adjusting for them. We have much more evidence on bye weeks (20+ years) so won’t shrink that effect much – we’re going to use 1.5 points. We have much less evidence from Thursday games, which means we should shrink the effect more. Moreover, theoretically the benefits of a Thursday game should be muted compared to a full bye. So, we’re going to make an adjustment half the size of the bye-week effect – .75 points.
This kind of post-ranking adjustment is a first for us. It is the product of the expanded Thursday schedule, as well as from our trying to understand a seeming tendency toward a mid-season swoon in prediction performance. We may adopt a few other adjustments in the future. None of them affect our Power Rankings, which we stand by as the best overall look at the relative strength of all teams. We are just trying to incorporate any other advantages we can when using those rankings to predict game outcomes.