## A Passing League

In some ways, the premise of this post is geeky even for this site. And that’s saying something. There is a debate over the proper way to measure league average. For example, when we say the average completion percentage in the NFL is 61.2%, this is generally assumed to reflect the fact that in 2013, there were 18,136 passes thrown in the NFL, and 11,102 of them were completed.

An alternative method of measuring completion percentage in the NFL is take the average completion percentage of each of the 32 teams. That number won’t be very different, but it won’t be identical, either. The difference, of course, is that this method places the same weight on each team’s passing attack when determining the league average. The former, more common method, means that the Cleveland Browns make up 3.755% of all NFL pass attempts and the San Francisco 49ers are responsible for only 2.299% of the league-average passing numbers. The latter method puts all teams at 3.125% of NFL average.

Wow, Chase, is this really a football blog? Two paragraphs on calculating the average in a data set? Believe it or not, that background presents an interesting way to look at how the NFL has become more of a passing league.

For example, let’s look at the 1972 season. Miami led the NFL in points scored and in rushing attempts, while ranking 24th out of 26 teams in pass attempts. Does this mean the Dolphins weren’t a good passing team? Of course not; in fact, Miami had the highest Adjusted Net Yards per Attempt average of any team that season! That year,only two teams threw over 400 passes: New England and New Orleans. And both teams were below-average in ANY/A, with the Patriots ranking in the bottom three.

In 1972, the average pass in the NFL gained 4.28 Adjusted Net Yards.  But an average of each team’s ANY/A average was 4.34, because good passing teams like Miami and Washington passed less frequently than bad passing teams like New England and New Orleans.  The league-wide average was only 98.5% of the “average of the averages” average; whenever that number is less than 100%, we can conclude that the better passing teams are passing less frequently.

Fast forward 39 years. In 2011, three teams topped the 600-attempt mark: Detroit, New Orleans, and New England. Tom Brady’s Patriots and Drew Brees’ Saints ranked in the top three in ANY/A (and the Lions in the top 7), while Aaron Rodgers’ top-ranked Packers in ANY/A still finished above average in pass attempts. The Tim Tebow Broncos were last in pass attempts, and in the bottom ten in ANY/A. The Jaguars, who finished last in ANY/A by a large margin, were in the bottom five in pass attempts, too, as Maurice Jones-Drew led the league in rushing. In 2011, the league-wide average ANY/A was 5.90, while the “average of the 32 teams” ANY/A was 5.85; that’s because the best passing teams were throwing more frequently than the worst passing teams (the ratio here was 100.8%).

The graph below shows the ratio of “average of each team’s ANY/A” relative to “league average ANY/A” for each year from 1970 to 2013. Remember, a percentage below 100 indicates more passes from inferior teams, while a percentage above 100 indicates the better passing teams are passing the most:

Note that the scale is from 97.5% to 102.5%; as you can quickly see, most of the action takes place in the bottom half of the graph. That is as good a picture as any to show why for quarterbacks, rate stats matter more than gross stats. Poor passing is part of a positive feedback loop for gross pass attempts: bad pass efficiency early in a game leads to a large number of pass attempts later in the game, and vice versa. And for nearly all of the ’70s, bad passing teams tended to throw more passes than good passing teams.

As you can see, though, the ratio has been on the rise over the past 44 years (the black line is a best-fit linear trend line). But the effect is not enormous, and we see this in 2013. While Peyton Manning had an incredible season, the Broncos finished six pass attempts behind the Browns, who finished 22nd in ANY/A. Russell Wilson didn’t throw very often, but Seattle ranked 5th in ANY/A while ranking 31st in pass attempts. The 49ers and Colin Kaepernick were 7th in ANY/A and last in pass attempts. Nick Foles led the NFL in ANY/A, but the Eagles ranked 27th in pass attempts. In that sense, 2013 wasn’t that out of whack with historical numbers.

Still, the general trend is clear. Two years ago, I wrote that the correlation between NY/A and pass attempts was on the rise; of course, at the time, the 2011 season looked like a new peak that was part of an ever rising standard. But over the past two seasons, that trend has curtailed a bit. Teams like Seattle and San Francisco are efficient at passing but aren’t pass-happy, which is why they seem like “throwbacks.” And Houston, despite finishing 5th from the bottom in ANY/A, was fifth in dropbacks. That’s because the Texans were the worst team in football, of course, but it also means many better passing teams chose to allow the game script to shift towards the run.

• “That is as good a picture as any to show why for quarterbacks, rate stats matter more than gross stats. ”

Couldn’t agree more, but this whole NFL.com bracket madness has got me thinking about QBs an awful lot lately. When you look at a passer’s stats, at what point do volume stats trump rate stats? For instance, Aaron Rodgers blows Brett Favre out of the water on rate stats, but I don’t know a lot of people who would argue his place in history is higher than Favre’s. Would you use something like career RANY, or would just be happy just saying it’s apples and oranges?

And for the Kerry Byrne types (as well as your post regarding Manning winning the Super Bowl to make this the greatest passing season ever), how can you possibly quantify something like championships? Do you think reasonable people could agree on some value to assign to playoff wins and titles, MVPs, All Pro selections, etc.? I saw some backlash when you decided to make playoff passing stats worth more than regular season ones, so my initial thought is no. Maybe I should just go to bed.

• Richie

Favre has 18 seasons of 9+ AV, while Rodgers only has 6. Favre gets an extra legendary boost because he led a team that had been mostly terrible for the previous ~25 years and helped turn them into perennial contenders. Rodgers just carried on the tradition.

It’s a little bit like Montana and Young. Young generally had the better rate stats, but much shorter career and a bit of a reputation for just continuing what Montana built.

I think if Rodgers can put together another 4-5 years of ~15 AV seasons that his historical reputation will improve quite a bit.

• I like AV as a quick stat, but a stat that gives more credit to Jeff Garcia’s 2000 than Peyton Manning’s 2013 is lacking a bit. For QBs at least, I think I prefer RANY.

In regards to Favre-Rodgers specifically, I am not sure if Rodgers will ever match the reputation of Favre, since there will be no Madden to evangelize for him.

• James

“When you look at a passer’s stats, at what point do volume stats trump rate stats?”

Let’s look at an example.

QB A passed 400 times and gained 3200 yards at 8.0 YPA but then got hurt and didn’t finish the season. Of the following QB seasons, which would you say is the closest to QB A’s?

QB B: 600 attempts, 4800 yards, 8.0 YPA
QB C: 600 attempts, 4200 yards, 7.0 YPA
QB D: 600 attempts, 3600 yards, 6.0 YPA
QB E: 600 attempts, 3000 yards, 5.0 YPA

It can’t be QB B since he produced at the same rate on more attempts; clearly that’s more valuable. It also can’t be QB E since he produced fewer yards despite more attempts. So it has to be close to either C or D, but which one?

QB A had 200 fewer attempts than them so one way to look at it is what he have needed to do on those extra 200 attempts to match their production. To match QB D he needed 400 more yards on 200 attempts. That’s only 2.0 YPA which is extremely low. It’s so low that no QB in NFL history with at least 30 attempts has averaged only 2.0 YPA, so it’s fair to say QB A had a better season since literally anyone could take over for those final 200 attempts and combined they’d easily outperform QB D. To match QB C he needed 1000 yards on 200 attempts, which is 5.0 YPA. 5.0 YPA is interesting – it’s close to what Akili Smith, Jimmy Clausen, and Charlie Whitehurst averaged in their careers. It’s what Kerry Collins averaged when the Colts pulled him out of retirement. 5.0 YPA seems like a reasonable approximation for how the best unsigned free agent QB would perform, a minimum baseline of QB performance.

That minimum baseline is generally called replacement level, and it’s the perfect tool for determining when volume stats match rate stats. Of course, this particular case is a bit harder because the league has changed so much from when Favre began playing to today, but the concept of replacement level is still the same.

To use a quick and dirty method, Favre averaged 108 ANY/A+ over 19 years, while Rodgers has averaged 124 ANY/A+ over the past 6. For Rodgers to match Favre, he’d have to play at a league average level (100 ANY/A+) for the next 13 years to match Favre. This is very straightfowardly correct – Favre’s best 6 seasons averaged 123 ANY/A+, matching Rodgers, and without those 6 seasons he was essentially league average for the other 13 years (101). Of course, that bypasses the replacement level consideration because playing time is equal. How many more seasons would Rodgers have to play at his current level to match Favre?

Well, based on Whitehurst, Clausen, and Collins it looks like replacement ANY/A+ is around 60 (maybe a bit higher). So from there it’s solving a simple equation: 108*19 = 124*X+60*(19-X), where X is the number of seasons Rodgers plays, and the second term indicating how much extra ANY/A+ Rodgers’ replacement provides. Or you could take ANY/A+ over Replacement, in which case the equation is: 48*19 = 64*X +0*(19-X). In both cases X is 14.25, so Rodgers would need to maintain his current level of play for another 8 seasons to match Favre’s career output.

Of course, different people might have different replacement levels if they happen to want to give extra reward to excellence or longevity – for instance, if Rodgers played between a Pro Bowl and All-Pro level for 14 straight seasons I think many people would consider that better than Favre’s 19 years at solidly above average play.

• James

Sorry, I didn’t mean to write an essay, but I just kept thinking of things and adding to it as the morning went on.

• I’d much rather read a thoughtful essay than a pithy statement that ultimately says nothing.

One of the issues I have, which I may not have touched on enough in my original comment, is the propensity for people to overrate (I think) championships. If Jackie Smith holds on to a pass, does it somehow make Staubach any better or Bradshaw any worse? If Vinatieri was a terrible kicker, would that make Brady a bad player? Since you read Chase’s articles, I assume you already agree with me that the answer is no. But I don’t think you can read an ESPN or NFL message board without seeing the RINGZ argument.

Two actual statements that were said to me on NFL.com (my fault for even going there):
“Montana, 4-0, end of discussion.”
“Unitas played before Montana, so he was bigger and tougher.”

And let’s not forget the epic battle between Tom T and Brock on Chase’s Elway story.

This is the sort of mentality that guys like Chase probably have to deal with on a constant basis (although it may bother me a lot more than it bothers him). My thought is this: if we want to use stats to measure a player’s career, would there be any agreeable way to say something like “1 playoff win = X ANY+” and “1 Super Bowl win = Z ANY+?” Something that would appease the masses, if you will.

I don’t even necessarily agree with doing that, but I can see some merits in it. By somehow measuring both achievement and performance, you get sort of a legacy score (like Madden, but not broken).

• James

“if we want to use stats to measure a player’s career, would there be any agreeable way to say something like “1 playoff win = X ANY+” and “1 Super Bowl win = Z ANY+?” Something that would appease the masses, if you will.”

It’s certainly doable. I bet the hardest part would be coming up with or agreeing to a set of groundrules to base the process on, since there are tons of different ways to do it and each indicates something slightly different. Of course, someone is always going to be upset that “playoff adjusted ANY/A” isn’t the same thing as “best”, much less those that don’t understand why “Montana: 4-0” isn’t the end-all, be-all to the debate.

I have three basic ideas:

1) Championship Probability Added. Your typical win probability added (clutch and all), but each regular season game is also weighted by impact on making the playoffs and each playoff game weighted by impact on winning the SB. My guess is this is the closest approximation to how the typical fan looks at players (a couple Eli drives at the end of the two Super Bowls are more important than a 14-2, one-and-done season by Peyton, even though the latter requires more talent) but it could make some very counter intuitive results. I bet most people don’t have a good appreciation for how little regular season and even early playoff rounds would impact the final results, and a extreme result in the SB could make or break a career.

2) Weighted Passing Stats. Lots of different ways for the specifics, but essentially weight a player’s playoff stats by the importance of a game. Very similar to my first idea, but not dependent on time and score. Chase did this recently, although for playoffs only: http://www.footballperspective.com/the-best-playoff-quarterbacks-in-the-super-bowl-era/ It would be easy enough to add regular season value to get an overall career value.

3) Legacy Tracker. Based off Bill James’ original HOF Tracker: QBs get points for various benchmarks, such as 3 points for each 3,000 yard season, 5 for a Pro Bowl, 10 for an MVP, 15 for winning a Super Bowl, etc and just rank by most points. This allows you to give out points for ‘big picture’ things beyond the box scores, like awards or being Top 5 in TD passes that add to a players reputation. Maybe even give out bonus points for great seasons that people will remember, like the Greatest Show on Turf, or single season TD record.

• I’ve decided to work on number 3 for a future article on my site. If you have any opinions on proper values for awards/milestones/championships, I’m all ears.

• James

Cool! Let me know when you publish.

This was originally designed for baseball, (and you can find more at the bottom of this page http://www.baseball-reference.com/about/leader_glossary.shtml) but here are some of James’s values for baseball pitchers:

– 8 for MVP, 5 for Cy Young (1st team All-Pro?), 3 for All Star (Pro Bowl)
– 2 points for best ERA (TD passes?), 1 point for other pitching stats (passing yards?)
– 2 points for each WS start (so maybe 5 points for a SB start?), 2 for a win (so maybe another 3-5 for a SB win?)
– Then give out other benchmarks for stat compiling – X number of starts, 3,000 yard seasons, 4,000 yard seasons, 30+ TDs, 40+ TDs, etc.

It’s worth noting that James’s goal wasn’t quality, but to accurately reflect if a player would make the HOF or not.

• I read the BBR link a bunch of times trying to find proper analogs in the NFL. Like PFR counts AP1, but AP2 should count for something, right? But then how many AP2s equal an AP1, and how many AP1s equal an MVP. So much room for subjectivity.

I went with some black and grey ink tests, like 5 points for leading the league in certain things and 2 points for being in the top 3.

1.5 points for a Pro Bowl
3 for AP2
5 for AP1
7.5 for OPOY
10 for MVP
1 for each playoff win
7.5 for a SB loss
20 for a SB win (Aikman had to get in the HOF somehow)
15 for a “legendary” season, like Marino 84, Warner 99, Young 94…whatever. This is super subjective. for instance, do you count Brady’s 2010 or just his 2007?

I also have compiling benchmarks, but they are a little different. Half a point for each:
25 completions over average
500 yards over average
10 TDs over average
10 INTs below average
1000 ANY over average
Where numbers are calculated similar to the way Chase does RANY.
There is also something to be said for holding records, but then you get into which records are important and which aren’t. Plus, Bill James is probably smarter than I am.

• Posted the initial model for the HOF Monitor, if you’re still interested.

http://nflsgreatest.co.nf/2014/05/hall-of-fame-monitor-defense/

• Richie

Being born in 1972, I wish that 2011 was only fast-forwarding 29 years. 🙂

• Chase Stuart

Nobody comes here for the math, Richie.

• Michael Terry

Clever post. Who would have thought you could tease out interesting meaning from the difference between calculating averages?

• Chase Stuart

Thanks, Michael.