SITE NEWS: We are moving all of our site and company news into a single blog for Sports-Reference.com. We'll tag all PFR content, so you can quickly and easily find the content you want.

Also, our existing PFR blog rss feed will be redirected to the new site's feed.

Pro-Football-Reference.com ยป Sports Reference

For more from Chase and Jason, check out their work at Football Perspective and The Big Lead.

Strange seasons

Posted by Doug on July 4, 2007

The 2006 Jaguars had what I'd call a strange season. They were 7-4 against above average (according to the simple rating system) teams, and only 1-4 against below average teams, including a 20-point loss against the Texans. Here are their opponents, sorted from best to worst, along with the result of the Jaguars game against that opponent:

   Opp  rating  result
======================
   nwe  +10.2    -3
   ind   +5.9    -7
   ind   +5.9   +27
   dal   +3.7    +7
   pit   +3.4    +9
   phi   +3.4    +7
   buf   +2.2    -3
   nyj   +2.0   +41
   kan   +1.0    -5
   mia   +0.7   +14
   nyg   +0.1   +16
   ten   -1.3    -7
   ten   -1.3   +30
   was   -4.0    -6
   hou   -4.5    -3
   hou   -4.5   -20

The correlation coefficient between the two columns of data is +.21, which is pretty strange if you think about it. A positive correlation means that the two quantities tend to vary together, which means that the Jags did better --- not just relatively better, but absolutely better --- against stronger teams than weaker ones. The correlation of +.21 is, in fact, the fourth-highest such value since the merger. For reasons I can't explain, the top three all occurred in 2003.

2003 Cardinals
   Opp  rating  result
======================
   gnb   +8.1    +7
   bal   +6.3    -8
   stl   +5.9   -24
   stl   +5.9    -3   
   sea   +4.1   -18   
   sea   +4.1   -38
   sfo   +3.1    +3
   sfo   +3.1   -36
   min   +2.9    +1
   dal   -0.5   -17
   car   -0.9    -3
   pit   -1.1   -13
   cin   -2.4    +3
   cle   -2.9   -38
   chi   -3.5   -25
   det   -5.8   -18

2003 Giants
   Opp  rating  result
======================
   nwe   +6.9   -11
   stl   +5.9   +10
   phi   +4.4   -18
   phi   +4.4    -4
   mia   +3.4   -13
   min   +2.9   +12
   tam   +1.6    -6
   nor   -0.3   -38
   dal   -0.5    -3
   dal   -0.5   -16
   nyj   -0.6    +3
   car   -0.9   -13
   buf   -1.0   -17
   was   -5.7   -13
   was   -5.7    +3
   atl   -7.4   -20

2003 Vikings
   Opp  rating  result
======================
   kan   +8.3   +25
   gnb   +8.1    -3
   gnb   +8.1    +5
   stl   +5.9   -31
   den   +5.5    +8
   sea   +4.1   +27
   sfo   +3.1   +28
   chi   -3.5    -3
   chi   -3.5   +11
   oak   -5.5   -10
   det   -5.8   +10
   det   -5.8   +10   
   sdg   -6.8   -14   
   atl   -7.4   +13
   nyg   -8.6   -12
   ari  -12.6    -1

It occurs to me that the correlation coefficient might not be the best way to measure the strangeness of seasons, since it can be overly sensitive to outliers. I'm open to suggestions for other ways to capture what I'm trying to capture.

For comparison's sake, here is the least strange season since the merger --- correlation coefficient -.87 --- which also happened in 2003.

2003 Saints
   Opp  rating  result
======================
   ind   +7.0   -34
   ten   +6.5   -15
   phi   +4.4   -13
   sea   +4.1   -17
   tam   +1.6    -7
   tam   +1.6    +3
   dal   -0.5    +6
   car   -0.9    -3
   car   -0.9    -6
   jax   -2.4    -1
   chi   -3.5    +7
   was   -5.7    +4
   hou   -6.0   +21
   atl   -7.4   +28
   atl   -7.4    +3   
   nyg   -8.6   +38   

The least strange season of 2006 was by the 49ers, who were destroyed by the best three teams they played. They were 1-5 against above average teams and 6-4 against below average teams.

2006  49ers
   Opp  rating  result
======================
   sdg  +10.2   -29
   chi   +7.9   -31
   nor   +4.0   -24
   phi   +3.4   -14
   den   +1.3    +3
   kan   +1.0   -41
   sea   -3.6    +6
   sea   -3.6   +10
   stl   -4.0    +7
   stl   -4.0    -3
   min   -4.1    +6
   gnb   -4.4   -11
   det   -6.4    +6
   ari   -6.9    -6
   ari   -6.9    -7
   oak   -9.6   +14

Other 2006 teams who had strange seasons were the Dolphins, who routed the Pats and Bears but lost to the Packers and Texans, and the Lions. Aside from San Francisco, the least strange 2006s belonged to the Giants and Bengals.

So does this mean anything, or is it just trivia? Not expecting to find anything, I ran a regression of Year N+1 wins versus Year N wins and Year N "strangeness coefficient." The result:

Year N+1 Wins =~ 4.97 + .432*(Year N Wins) + .932*(Year N Strangeness Coefficient)

The last coefficient is statistically significant, which indicates that there might be something real there, especially since the result agrees with intuition. If you're capable of beating good teams, then losing to bad teams seems like something you can figure out how to correct. On the other hand, if you are beating bad teams and losing to good ones, it's not clear that you have much untapped upside. The Jags and Giants were both 8-8 last year, but this formula projects the Jags with 8.6 wins this year and the Giants with 7.8. That's quite a difference.

This is an indicator that I haven't heard discussed much. Intuitively, it seems like it could be a general indicator of future success in all sports, not just NFL football. There's more work to be done, but I think there is potential here for identifying groups of teams that are likely to improve next season, much in the same way that deviation from projected winning percentages --- whether projected from the SRS, or from a pythagorean type procedure, or from something like footballoutsiders' DVOA --- does.

This entry was posted on Wednesday, July 4th, 2007 at 4:59 am and is filed under General. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.