That old Greek dude was on to something.

I have a few thoughts on why some weeks are better than others. Logically, you’d think that it would improve as the season went along, except for wk 17 as noted. It seems to have roughly done this, which tells me it’s generally a keeper. But why would there be any anomalies aside from Week 17?

The same reason Indy won the SB in 2006 and not other years: matchups. It’s a 53 on 53 game and even a team that averages plus 7 pts a game can lose to a team that is roughly even because of matchups.
1) Was that +7 average beefed up by drubbing somebody 30-0? (the implication is that early season with fewer games could be skewed by a single outlier more than late season).
2) Was the breakeven team playing all SB contenders, and therefore doing quite well breaking even?
3) Did the +7 team face creampuffs for the past few weeks?
4) How do these particular teams match up? Any key injuries? Home/away correlation? If HFA is worth 3 theoretical points, that could also be calculated in, no?

The reason the Steelers beat the Colts in the 2005 playoffs, why the Colts won the 2006 SB (no Chargers), and why the Giants came close to beating in week 17 and did beat the Pats in the SB the next year, all come down to matchups. The Giants had the right D package to get to Brady and nobody else really did all season. They matched up well. If the Colts had WRS who were 6-3/220 in 2003 and 2004, how would those playoff losses changed because NE’s “bump and mug” D technique no longer matched up as well?

So it really comes down to team- and individual matchups. While Pythagoras is a good rough rule of thumb, it’s no substitute for actual knowledge of the game. Or, like Diane Chambers on Cheers used to do when she won the betting pool every week in one episode (Yes, I am dating myself here—I was in HS at the time): “Sam, the Lion’s over the Dolphins. Come on, how is a dolphin going to defend itself against a lion? Eagles eat cardinals, don’t they?” Clearly she had no knowledge of the actual game coming up next Sunday. Neither does Pythagoras—he only has past knowledge, which, like they say in investment prospectii, is no guarantee of future results.

The number clearly improves as the season progresses until you get to weeks 15-17 where I think you might start seeing the affect of both the best teams starting to ease up and coast as the playoffs approach and their seeding is starting to lock in, and players on generally bad teams start playing for their jobs (or to get new jobs with other teams for the following year).

I wonder if some of the weakness/variability in the model in weeks 3,4,5,6,7 might come from the affect of bye weeks. Good teams without rest playing a previously poor performing teams coming off a bye week might have an edge and create a forcing in the numbers. I think you almost have to throw out weeks 1 and 2 as statistically irrelevant to the mission. If you throw the first and last 2 weeks out of the average you improve to 63.3%. If you threw out the weeks affected by bye weeks you improve to almost 66%. You might be able to make money at that rate.