Writers of Pro Football Prospectus 2008

15 Dec 2011

ESPN: Will Defense Hold Contenders Back?

My column for ESPN Insider this week points out that massive imbalance between offense and defense hasn't necessarily kept teams from winning in past postseasons. It's not true that the Saints and Patriots can't compete in the playoffs because of awful defense; what matters is the whole package.

Posted by: Aaron Schatz on 15 Dec 2011

13 comments, Last at 16 Dec 2011, 6:36pm by tuluse


by dmstorm22 :: Thu, 12/15/2011 - 1:33pm

Unbalanced Defensive teams: 12-3 (three Super Bowl titles)

Unbalanced Offensive teams: 8-5 (two Super Bowl titles)

There were 13 teams on the list. Obviously, this is a very small sample size (I'm not sure what the teams that just missed the cut accomplished), so teams that do win the Super Bowl can really skew the results, but it does look at least that great defensive teams do perform better.

by tuluse :: Thu, 12/15/2011 - 1:48pm

I'm guessing one of the unbalanced offensive teams in the 2006 Colts?

Which isn't completely correct as Bob Sanders came back for the playoffs and they played much better defense.

If the Saints or Packers have a defensive player who will play at a HoF level for them during the playoffs who isn't playing for them right now, I think that significantly increases their chances.

by tally :: Thu, 12/15/2011 - 4:39pm

12-3 vs. 8-5 is nowhere near a significant difference in proportion. You'd basically have a 41% chance of seeing that proportion from completely identical populations.

by DisplacedPackerFan :: Thu, 12/15/2011 - 6:16pm

I understand the audience being written for, but I would love to see unbalanced in terms of DVOA, not ordinal rank.

So GB is 35.1% vs 9.1% for a 44.2% "imbalance"
NE would be 35.3% vs 15.0% for a 50.3%
NO 33.7% vs 16.1% for 49.8%

So that 1998 Denver at 30.7% and 1.8% for 32.5% was not nearly as imbalanced as this years Green Bay.

Sure that would still show a team with a historically great O or D and a fairly pedestrian D or O as potentially very imbalanced. The 2007 NE with 45.2 and -5.0% would show as a 40.2% imbalance, but if you take balance to mean that offense and defense are playing roughly the same (be that both very well and both very average) they were an imbalanced team.

Even if you use the rankings to determine imbalance it's still nice to see the offensive and defensive DVOA to get a feel for scale. That 2006 Colts team was 29.0% and 8.8% (37.8% total) which actually seems to make them a closer match to the Packers than the Pats or Saints that they were compared to.

by Perfundle :: Thu, 12/15/2011 - 7:01pm

"So that 1998 Denver at 30.7% and 1.8% for 32.5% was not nearly as imbalanced as this years Green Bay."

I don't have Insiders, so I don't know if this was addressed, but doesn't the league average have to be taken into account? Here is the league imbalance average for all the years in FO's database:
1992 -10.08%
1993 -11.20%
1994 -4.19%
1995 -0.33%
1996 -9.88%
1997 -7.65%
1998 -7.69%
1999 -11.81%
2000 -3.94%
2001 -6.48%
2002 2.48%
2003 -4.37%
2004 1.24%
2005 -3.39%
2006 0.46%
2007 1.36%
2008 7.77%
2009 5.66%
2010 6.25%
2011 7.18%

The rules are slanted more towards offense nowadays. With that in mind, Denver's imbalance was 40.2% above the league average last year, while GB's is 36.0% above.

by DisplacedPackerFan :: Fri, 12/16/2011 - 12:18pm

Yes I would think the league average should be taken into account to adjust for environment but it should be be split still, not just overall, you'd want over/under average for offense and for defense. I was being quick and sloppy and wrong and ordinal rank does capture a bit of the league environment.

My main complaint was that DVOA wasn't shown at all in the article and was barely mentioned, it was all based on the DVOA rankings. So Green Bay right now is #2 offense and #20 defense so the imbalance used was 18. I'd have to do more work but I'd think it's still possible that a difference of 18 in ordinal rankings is less imbalanced than a difference of say 25.

So let me correct my math and support you in proving that I was wrong in my first post.

1998 League average offense was -4.1%, league average D was -3.6% (both rounded). Broncos were 30.7 and 1.8% so yeah I should use 34.8 and 5.4 for a 40.2

2011 O is 3.9 and D is 3.3 so

GB 35.1 and 9.1 adjusts to 31.2 and 5.8 for a 37.0%
NO 33.7 and 16.1 adjusts to 29.8 and 12.8 for 42.6% adjusted
NE 35.3 and 15.0 adjusts to 31.4 and 11.7 for 42.8% adjusted.

So GB is 3.2% less unbalanced than Denver was and NO and NE are 2.4 and 2.6% so they all still compare to that team fairly well by this metric but not by the ordinal ranks used in the article.

Oh and since the 2006 Colts were used, not surprisingly there is very little adjustment for the Colts.

2006 average O 0.5% and D 0.03% (so 0)
So the Colts are 28.5% and 8.8% for 37.3% and yep match the 37.0% Packers pretty well.

So the adjusted numbers of all the teams I mentioned by adjusted DVOA imbalance.

2011 Patriots..31.4..11.7..42.8
2011 Saints....29.8..12.8..42.6
1998 Bronco....34.8...5.4..40.2
2006 Colts.....28.5...8.8..37.3
2011 Packers...31.2...5.8..37.0

I just like that info better than what the article showed. The article said the 06 Colts most resembled the 2011 Pats and Saints, and 1998 Broncos most resembled the Packers of the imbalanced teams that won the SB. Without adjusting I disagreed, but adjusting I think I do again, even if the total imbalance of the Pack and Colts are closer, the way the imbalance happens is closers for the Packers and the Broncos.

Again I understand the ESPN audience issue. I just still think this gives a better picture than ordinal rank. Seeing the Broncos D was only 5.4% worse than an average 1998 D tells me a lot more than it was ranked #20 out of 30.

That was my main point, I muddled it, thanks for pointing it out so I could clarify better.

by DisplacedPackerFan :: Fri, 12/16/2011 - 3:38pm

Just to track it. The average DVOA for OFF and DEF.

1992 -4.68 -5.40
1993 -5.60 -5.60
1994 -1.78 -2.42
1995 +0.59 -0.92
1996 -4.73 -5.15
1997 -3.35 -4.30
1998 -4.05 -3.64
1999 -5.87 -5.94
2000 -1.46 -2.48
2001 -2.85 -3.64
2002 +1.72 +0.75
2003 -2.08 -2.29
2004 +0.72 +0.53
2005 -1.94 -1.45
2006 +0.48 -0.03
2007 +0.54 +0.82
2008 +4.38 +3.38
2009 +3.01 +2.65
2010 +3.30 +2.94
2011 +3.92 +3.27

So really the switch to having a offense better than the baseline and a defense worse than really only took off in 2008. I know that 02, 04, and 07 show this as well, but those are all pretty close to being baseline. But it is a big change from the majority of the 90s where teams had offensive below baseline and defenses better than baseline. I just thought the trend lines would look a bit different than what the data paints.

The 2011 numbers we can reasonable expect to both end up closer to baseline with what the last 3 weeks generally produce, I suspect that it will look more like 2010 by the end of the year and less like 2008 that it currently resembles.

by Perfundle :: Fri, 12/16/2011 - 4:49pm

"Yes I would think the league average should be taken into account to adjust for environment but it should be be split still, not just overall, you'd want over/under average for offense and for defense."

Thanks for doing the split, but the reason I didn't do it can be seen from your numbers. DVOA is basically a zero-sum stat, so the offensive and defensive DVOA average should be roughly equal. Thus the individual components should be both about half of the total imbalance.

by tuluse :: Fri, 12/16/2011 - 3:45pm

Really, you should be comparing number of standard deviations away from the mean. That would also account for more or less variation depending on year.

by DisplacedPackerFan :: Fri, 12/16/2011 - 4:56pm

No one will let me be lazy will they? :)

Assuming my math is right...

1998 Broncos 30.7 and 1.8 were 2.00 STD and 0.17 STD from the 1998 means
2006 Colts 29.0 and 8.8 were 2.08 STD and 0.88 STD from the 2006 means
2011 GB 35.1 and 9.1 are 2.27 STD and 0.88 STD from the 2011 means
2011 NO 33.7 and 16.1 are 2.18 STD and 1.55 STD from the 2011 means
2011 NE 35.3 and 15.0 are 2.29 STD and 1.49 STD from the 2011 means

I still want to see the DVOA and the absolute value from the mean I still like so I'll keep that and just extend the table.

2011 Patriots..31.4..11.7..42.8...2.29...1.49...3.78
2011 Saints....29.8..12.8..42.6...2.18...1.55...3.73
1998 Broncos...34.8...5.4..40.2...2.00...0.17...2.17
2006 Colts.....28.5...8.8..37.3...2.08...0.88...2.96
2011 Packers...31.2...5.8..37.0...2.27...0.88...3.15

So that makes the Broncos not really match any of the 2011 teams that closely as their D really wasn't that far from average and their offense was closer to average. The Packers now look to match the 06 Colts fairly well, oddly with the chance of a D improving (though not from an injured player returning, I talked about it in the Week 14 DVOA thread). The Colts are still the closest match to the 2011 Saints and Pats, but the defenses are significantly worse.

Hopefully I wasn't too sloppy on the STD from mean on these. I was really good at math 15-20 years ago then I stopped having to do anything real with it.

by tuluse :: Fri, 12/16/2011 - 5:11pm

Haha, very nice work. Yeah those Broncos don't really fit at all. The defense was basically average. On the other hand, all these other teams have better offenses then they did.

I still contend the Colts don't really fit either. Yes their regular season might match the other teams, but unless you have a defensive player who is HoF caliber joining your team for a playoff run, you are not going to play like they did.

As an aside, when Aaron works on changing the baseline for DVOA this year, I would really like 2 results to be reported. One for the old baseline or some "historical" baseline, so we can keep track of how the environment of the NFL is changing and one for more recent times so we can compare how teams are doing relative to each other better.

by Perfundle :: Fri, 12/16/2011 - 6:17pm

"1998 Broncos...34.8...5.4..40.2...2.00...0.17...2.17"

Wait, this can't be right. The league that year had negative offensive and defensive DVOAs. Those OSTD and DSTD numbers are definitely too low.

Actually, I don't think using standard deviations is going to affect your results from the previous post; it should still put them right in the pack as before.

by tuluse :: Fri, 12/16/2011 - 6:36pm

Running through really quickly with excel, I got 2.3 for Denver's offense and .5 for the defense. Which would give 2.8 total.

"Actually, I don't think using standard deviations is going to affect your results from the previous post; it should still put them right in the pack as before."

It could especially with say Denver's defense, if league wide defense had crazy variation, it would make them look closer to average than just subtracting the mean from their score, or vice versa.