SITE NEWS: We are moving all of our site and company news into a single blog for 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. ยป Sports Reference

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

Do specific matchups matter?

Posted by Doug on January 7, 2009

Two teams are, all things considered, exactly equal. They're playing tomorrow on a neutral field. This game is a 50/50 proposition.

Now suppose I give you two more pieces of information: one of the teams has a very poor pass blocking line, and the other has a very good pass rush.

Is this game still a 50/50 proposition?

The 2001 Falcons and Saints provide a decent concrete example, with the Falcons playing the role of the sieve and the fearsome twosome of Charlie Clemons and Joe Johnson fueling New Orleans' league leading pass rush. Both teams, though, were 7-9 and they had a point differential within 10 points of each other.

Conventional wisdom would, I think, say that this game is not a toss-up. The team with the matchup to exploit --- the Saints in this case --- would have the edge.

I've always been a little suspicious of that line of thinking. If the teams are of equal quality overall, then the team with the bad pass blocking line must have advantages over the good pass rushing team in other areas. The 2001 Saints turned the ball over a lot, for example. Those advantages might be less extreme, but they have to add up to the same thing if the teams' overall strengths are equal.

Or do they? The logic I just outlined assumes that matchup advantages are additive and not multiplicative. But that might not be true. Maybe one big advantage is better than three or four little ones. Or maybe it's not.

This is obviously a complicated question, but I decided to take a very quick stab at it.

I looked at every game since 1990 and I noted the SRS difference between the two teams and the location of the game: overall team quality and home field advantage, the two standard independent variables. Then I looked at the difference between one team's sacks allowed SRS number and the other teams sack SRS number, as described in this post.

To continue the Saints/Falcons example, the "Matchup SRS difference" was about 2.6. Which means that in a Saints/Falcons game, we would expect Saints to have about 2.6 more sacks than we would expect an average 2001 team to get against another average 2001 team. That's a lot of sacks, and that's a pretty serious problem for the Falcons.

Then I ran a logit regression with those three variables --- (1) overall SRS difference, (2) home field, (3) sack SRS difference --- as inputs and win or loss as the output.

Win Prob =~ 1 / (1 + exp(.441 - .143*SRSDiff - .882*HOME - .048*SackDiff))

Since .048/.143 is right about one third, this model would indicate that 1.0 of SackDiff (the units here are sacks) is worth about 1/3 of SRSDiff (the units here are points). So the Saints' 2.6-sack advantage makes them equivalent to a team that's one point better and has no sack mismatch. And 2.6 is extreme. In the vast majority of cases, the SackDiff between two teams will be between -.6 and .6 or thereabouts. Finally, I'll point out that, though it's close, the coefficient on SackDiff in the above regression is not statistically significant (p =~ .15).

But this makes sense. A SackDiff of .2 or .4 probably isn't relevant. What matters --- maybe --- is whether or not there is a big mismatch to exploit. So I created a yes/no variable for whether or not the SackDiff was really big. I defined really big to be 1.0 or greater, and about 12% of the games met that criterion. I re-ran the regression with the SackDiff variable out and the Was-The-SackDiff-Big? variable in. Here are the results:

Win Prob =~ 1 / (1 + exp(.446 - .144*SRSDiff - .882*HOME - .18*BigSackDiff))

This time, the BigSackDiff coefficient was statistically significant (p =~ .04). This says that a big mismatch in that particular phase of the game is worth a little more than a point.


A couple of things I need to be very clear about:

  • As I mentioned above, this is a fairly simple look at one particular kind of matchup. The question in the post's title is obviously a much broader issue.
  • There is a possibility that the regression above is suffering from some cause-and-effect-itis. In particular, I don't know this for sure, but I think it's likely that, all else equal, teams that are ahead tend to accumulate more sacks because they have more opportunities for sacks because their opponents, who are behind, are passing a lot. Thus, much in the same way that high rushing yards are simultaneously a cause of wins and caused by wins, so might be sacks. And this would skew the results of the regression. I can think of a few possible ways to deal with this, but I think I'll end it here and get your feedback before figuring what direction (if any) to go from here.

Finally, because I know you're dying to know, the Falcons and Saints split in 2001. Their first matchup, whose boxscore shows nothing remotely interesting, was a Falcon win in New Orleans. There's a bit of a story to tell with the rematch, though: New Orleans sacked Chris Chandler nine times en route to a 28-10 win.

This entry was posted on Wednesday, January 7th, 2009 at 5:42 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.