Writers of Pro Football Prospectus 2008

06 Aug 2009

Indianapolis: Aaron Radio Interview, Injury Analysis

18to88.com links us to a downloadable version of Aaron's interview with Derek Schultz of XL 950 in Indianapolis.

They also dare question our infallible injury research, noting that Colts' starters missed 64 games and that it would have been extremely difficult for them to only accrue 6.4 AGL over the rest of the season.

Of course, we make plenty of mistakes and are constantly adding and adjusting injury information -- I spent nine hours on Sunday, for one, heading through the Wayback Machine for 1998 IR data. (You, too, can be a Football Outsiders writer! Just write to...)

In this case, it's a simple misinterpretation of how AGL works. 18to88.com assumed that a player that missed a game was credited with one point of AGL, and that's not the case.

A player that is listed as Out or on IR/PUP is credited with 1 AGL, regardless of whether he misses the game or not. (Something like 99.8% of all players listed as Out actually do miss games.) A starter listed as Questionable, though, is assigned .38 AGL, regardless of whether he actually does play or not, because starters listed as Questionable end up playing 62% of the time.

And, to provide a resource, we have the Colts' starters credited for 69 missed games. They include:

- Mike Pollak (missed three games where he would've started had he been healthy)
- Joseph Addai (4)
- Gary Brackett (4)
- Dallas Clark (1)
- Tyjuan Hagler (7)
- Marvin Harrison (1)
- Kelvin Hayden (6)
- Marlin Jackson (9)
- Freddy Keiaho (2)
- Ryan Lilja (16)
- Gijon Robinson (1)
- Bob Sanders (10)
- Jeff Saturday (4)
- Tony Ugoh (1)

Posted by: Bill Barnwell on 06 Aug 2009

12 comments, Last at 08 Aug 2009, 11:34pm by Bobman


by starzero :: Thu, 08/06/2009 - 3:48pm

bob sanders, iron man

by DZ (not verified) :: Thu, 08/06/2009 - 4:12pm

Wait, so if a player is questionable, but does not play, he only gets credit for .38 AGL?

I understand the idea of AGL approximating reduced performance by indicating that a player is playing hurt and giving the team AGL points for that.

But what possible value is it to say that a questionable player who did not play in a game gets LESS than a full point? Is it merely to reflect that there is some practice time gained during the week and that has value? The team didn't have the player that day. I'd say that was worth a point.

I also question if the radical change in the Colts position from 2007 to 2008 marks some change in the way teams reported injuries.

How does a team go from 2nd most injured to 9th most injured in just one year, when they had the same amount of injuries both years?

In the interview, Schatz claimed they weren't decimated by injuries in 2008, but they were in 2007 even though they lost the same number of AGL.

It seems inconsistent to me.

by Pat (filler) (not verified) :: Thu, 08/06/2009 - 11:28pm

"But what possible value is it to say that a questionable player who did not play in a game gets LESS than a full point?"

Because AGL is not intended to be descriptive. It's intended to be predictive. If you assume that coaches don't fudge injury reports (or at least, fudge them consistently) then a player listed as questionable plays 62% of the time. Missing the game means the team was unlucky - a questionable player *should* play, but for whatever reason, the guy wasn't able to do it.

If you want to compare a team's injury rate to the league average, the additional probable/questionable/doubtful/out information allows you to filter out some of the luck (well, unpredictability) in injury recovery.

by buzz :: Thu, 08/06/2009 - 7:03pm

I am also confused by this "A starter listed as Questionable, though, is assigned .38 AGL, regardless of whether he actually does play or not, because starters listed as Questionable end up playing 62% of the time." My understanding of how the AGL worked was that any player who didn't play a game would be given credit of 1 and then the probable, questionable etc rating would give them an additional point because they were probably playing at something less than 100%.

So if I understand correctly when Matt Schaub ended up not playing one game last year because he was sick at the last minute (and rosenfels lost the game for them) he was not given any credit for AGL since he wasn't on the injured list but in years past when brady was consistently listed as probable was given some points? If this understanding is correct then I am definetely moving the AGL down a tick on my reliability index.

by Theo :: Thu, 08/06/2009 - 7:49pm

Info > sloppy stuff

by buzz :: Thu, 08/06/2009 - 8:00pm

Just to note in saying that I am moving the reliability index down a bit i am not necessarily bashing the measure as i think it is very critical and is great information. It just goes from great information to good if it is really calculated how it now sounds like it is. I would love to hear the reason that someone being inactive isn't considered a value of 1.

by Joseph :: Fri, 08/07/2009 - 3:06pm

I'm going to say it like this: Every week, 8 (or is it 7?--I forget) players on the ACTIVE roster don't get to play--they're INACTIVE. Sometimes those players are a healthy scratch (think DNP--coach's decision in the NBA), other times it is a bench player with a sprained ankle who can't go that week, other times it may be a starter who has been listed as "doubtful" for 3 weeks. So if starter X was "questionable"--and then didn't play on Sun. because he was inactive, it is VERY PROBABLE that another player who would have been inactive is active for that week--probably the 2nd or 3rd stringer at the position of the starter. But, as this backup is not as good as the starter, probably 80% of the starter's value, the AGL is a way to compensate for the loss of that player. If I understand correctly, that's what the inactive list is for--so that you aren't short a player for that particular game/games that someone is slightly injured (think mild sprain) but in no way needing to be sent to IR. Sure, that changes in weeks 15-17, as a player with a minor injury might go on IR so that the coach can get a 3rd stringer some reps as his team plays out the string (think 6-10 team)--but by and large, a team with quality depth will not lose as much--because a healthy, normally INACTIVE player will be activated to compensate for the injured starter. INDY would be a good example of that. (Sure, if PM goes down, it's not quite the same--but for other guys, they get plugged in and produce almost as good as the starter.)

by DZ (not verified) :: Fri, 08/07/2009 - 5:16pm

Apparently my entire understanding of AGL is wrong. I was operating under the assumption generated by this line in FOA 09:

Unlike a count of “starter games missed,” this
accounts for the fact that a player listed as questionable
who does in fact play is not playing at 100 percent

So I figured that if a starter missed a game, his total was 1. If he was on the injury report, but played, he received some percentage of a point. That would account for players like Tony Ugoh who missed 1 game, but was forced to limited duty in 3 or 4 others. That would reflect his diminished capacity.

Obviously, that assumption was not correct. I still don't understand why this method is less accurate than the actual AGL formula, but I guess I have to take your word for it.

Is there any place to get a complete explanation of this without buying last years PFP?

by Bill Barnwell :: Thu, 08/06/2009 - 11:43pm

The measure's based on historical trends of players participating or not participating. That's the whole point.

Right now, a player who's marked as questionable is assigned an AGL of .38, because that's the historical rate of player participation. If we simply give that player 1 AGL when he doesn't play, then we can't assign him a value of .38 when he does play, because that value is irrelevant. Since every player left to compare him to has a value of 0, he would be assigned 0 AGL. A 1/0 played/not played metric is still useful, but not as useful.

I've already reported its reliability as a metric, a reliability which has only improved since that last discussion. If you think less of it because it doesn't fit your specifications of what it should look like, well, you're fooling yourself.

by DZ (not verified) :: Fri, 08/07/2009 - 6:50am

Part of the problem is that the problem is that I don't have access to the original AGL article.

Every time I try to track down the reasoning, it refers back FO 2008. That's fine, but people that don't have FO 2008 are left to accept your word that it's a valid metric.

Surely you can admit that it is at the very least a counter intuitive system. That does not make it wrong, but you have to forgive readers for needing more explanation than "I explained it in an old book. If you want to understand it, then buy it"(which is the explanation given both in the 2008 AGL article and the glossary). If there is a more clear explanation on line, please point me to it. This site does incredible research, but unusual ways of generating statistical data will struggle to gain acceptance if readers have to purchase out of date copies of the Prospectus to understand the methodology.

You say the reliability of the metric has improved. I would love to know why. It might answer my second question. Did team reporting of injuries change from 2007 to 2008. How did the Colts drop 7 spots while suffering the same number of injuries.

Was there a league-wide rash of injuries in 2008? Did the way teams report change?

by DeltaWhiskey :: Fri, 08/07/2009 - 7:19am

Apparently the "I Believe" button is broken again.

by Bobman :: Sat, 08/08/2009 - 11:34pm

I believe you're right.

I side with DZ on the surface logic aspect--conceptually, if you are trying to measure something, isn't it better to actually measure it rather than measure what is said about it? (Think Plato's Cave from philosophy 101). If a coach hands you a dozen pieces of 2x4 and say's "they're all 1 foot long" but when you line them up, you KNOW they are all not the same length. AGL accepts what he says while logic suggests you should actually measure them to get your answer. No? Clearly FO had an accurate count of actual games missed for the Colts and presumably the other 31 teams--isn't that more useful?

If it IS trying to be predictive rather than historically descriptive, then maybe two meaures are called for--an actual historical account of games missed (which I am okay with including partial counts for "questionable" guys not being 100%) as well as AGL as it stands to be predictive.

As it stands, it is really relying on consistency in reporting from game to game and season to season. With the revolving door for coaching staffs (what's average tenure for a HC in the NFL? 3.5 years or something?) that seems a little uncertain, no? Will Caldwell report injuries the same as Dungy? Maybe. Maybe not.