Posted by Doug on September 11, 2006
Here is a quick statistical look at the relevance of week one.
I looked at all week one games during the last ten years and ran a regression of the margin of victory (or loss) in week one versus the rest-of-year win total for the team. Here is the formula:
RestOfYearWins =~ 7.5 + .035 * Week1Margin
The week one margin was indeed highly significant in the official statistical sense. That means that it definitely tells us something. But the coefficient is small, which means that, for practical purposes, it doesn't tell us much. The formula essentially projects everyone to win half its remaining games, give or take just a little.
If you just woke up from a multi-year coma, read this post (obviously the first thing you'd want to do), and then looked at the week's scores, you would expect that the Rams will win 8.8 games in 2006 --- the one they've already won, plus 7.5+.035*8 more --- and the Broncos will win 7.2.
I also tried including strength of opponent (based on last year's strength) into the regression, but it did not turn out to be significant.