Completed air yards per completion

[Spider-Dan]

I was browsing through some PFRef QB stats, and I found an interesting one on their Advanced Passing page:

"Completed air yards per completion - yards the ball traveled in the air past the line of scrimmage prior to a completion"

Now, as we all know, Tua only throws 2 yard passes to the left.  So surely this will show him at the bottom of the league, right?  Well, he's #23.  But then I took a look at the QBs surrounding him:



It looks like Tua's CAY/C is tied with MVP candidate Tom Brady, and 0.2 more than other MVP candidate Aaron Rodgers.  How can this be?

Well, maybe they are old men who have to save their arm for the playoffs.  But if you look down a little further, you see The Golden One Patrick Mahomes chilling at #30 with 5.0 CAY/C.  Again, how can he possibly be throwing shorter passes than noodle-armed waif Tua?

Or maybe this whole narrative is BS, and Tua plays for a team with 1) a terrible pass-blocking OL that affords him little time to throw, and 2) a defensive-minded coach who wants to play conservatively and not take shots down the field.

[fyo]

^ I was looking at those stats and wondering the same thing the other day, just at NFL's NextGenStats:

https://nextgenstats.nfl.com/stats/passing#yards

There are a few things to consider as to why the narrative is the way it is (and I agree that it's at the very least exaggerated):

1. Tua *doesn't* have elite arm strength. (Does he need it to be great? No, but he doesn't and that fact *enables* this narrative.)

2. Tua has the third lowest INTENDED air yards. So, in other words, while he completes at an average roughly equal to acknowledged great quarterbacks, he doesn't throw as many deep INCOMPLETE passes as they do.

3. YAC. Tua's receivers get the 7th lowest amount of yards after catch, 4.7 yards / completion. Two of the quarterbacks with similar average air yards as Tua get the 2nd and 3rd most YAC, respectively (Rodgers and Mahomes).

4. Touchdowns. Tua just doesn't throw many touchdowns. A large part of this is simply volume and his touchdowns per pass attempt is quite respectable. Not top-shelf (5.5% to 6%), but middle-of-the-road 4.6%. Certainly better than Mac Jones and his sub 4% or Derek Carr and his 3.6% or Matt Ryan and his 3.8%, but pretty much the same as a guy like Garoppolo.

5. Short-to-go high-probability conversions. This is tied to the touchdowns above. The simple fact that the Dolphins pull Tua in (some) short-to-go situations results in fewer touchdowns, fewer third down conversions, fewer chances to "score" in advanced metrics where these things count a lot. And when he does get put back in in goal-to-go, it's because our wild cat / Brisset lost a bunch of yards (meaning a real QB has to come in). A conversion/TD "takes away" that touchdown from Tua while a loss-of-yards results in a much lower probability conversion attempt for Tua. Neither looks good on the stat sheets or in the minds of fans and analysts alike.

6. Throwing short of the down marker. There isn't a huge difference between Tua and the middle-of-the pack where most of the big name quarterbacks are, but the fact is Tua is in the top 10 in yards thrown short of the sticks. This isn't a stat most normal people follow, but it is something a lot (if not most) notice during games. It's a natural product of the poor offensive line (and thus no running game and no time to throw), but that doesn't change the perception.

[Dolfanalyst]

Measures such as those are certainly "in the weeds" when it comes to quarterback evaluation.  These articles may help:

https://mfootballanalytics.com/2020/04/06/which-qb-stats-are-the-most-important/

https://mfootballanalytics.com/2020/08/29/creating-a-model-for-quarterback-rankings-in-2020/

The last five league MVPs have either led the league or been an extremely close second in the league in EPA per play.  I'd stick to that as a global measure in evaluating QBs.  Right now Rodgers is far and away the league leader, and Tua is 13th among QBs with sufficient pass attempts.

[Dave Gray]

I have confidence in Tua's arm.  I just wish we'd use it more to go downfield.  He makes mistakes, but it's part of the process and the trade-off is worth it.

[EDGECRUSHER]

I have confidence in Tua's arm.  I just wish we'd use it more to go downfield.  He makes mistakes, but it's part of the process and the trade-off is worth it.

He can't go downfield unless we sign Sonic and Usain Bolt in the offsason because our O-Line can't block long enough for our receivers to go further than 10 yards.

He has an accurate arm, seems like all his INTs this year were dumb mental mistakes where he forced the issue. I haven't seen them all, but the ones I have are of him throwing into double coverage. Not really missing the WR by an inch or two.

[fyo]

Measures such as those are certainly "in the weeds" when it comes to quarterback evaluation.  These articles may help:

https://mfootballanalytics.com/2020/04/06/which-qb-stats-are-the-most-important/

https://mfootballanalytics.com/2020/08/29/creating-a-model-for-quarterback-rankings-in-2020/

The last five league MVPs have either led the league or been an extremely close second in the league in EPA per play.  I'd stick to that as a global measure in evaluating QBs.  Right now Rodgers is far and away the league leader, and Tua is 13th among QBs with sufficient pass attempts.

EPA includes things like YAC and sacks, regardless of who is at fault. In general, it's a statistic that's geared towards a high correlation with scoring points (and thus winning), and no attempt is made to determine who is at fault on any given play, nor does it attempt to adjust for opponents. Over multiple teams and multiple seasons, these things average out, but for a single quarterback for a single season, it's not a great evaluator of the *quarterback*. It's a solid evaluator of the combination of quarterback-oline-receiver, at least within the confines of the play-calling and opponents.

Since what we are doing is specifically trying to evaluate Tua separately from the horrific offensive line, EPA isn't a good tool.

As for league MVPs have been tops in EPA, that's fairly obvious since EPA is geared towards maximum correlation with points scored (which correlates very well with winning). And who gets MVP? Usually the quarterback on the most winning or most scoring team.

[Dolfanalyst]

EPA includes things like YAC and sacks, regardless of who is at fault. In general, it's a statistic that's geared towards a high correlation with scoring points (and thus winning), and no attempt is made to determine who is at fault on any given play, nor does it attempt to adjust for opponents. Over multiple teams and multiple seasons, these things average out, but for a single quarterback for a single season, it's not a great evaluator of the *quarterback*. It's a solid evaluator of the combination of quarterback-oline-receiver, at least within the confines of the play-calling and opponents.

Since what we are doing is specifically trying to evaluate Tua separately from the horrific offensive line, EPA isn't a good tool.

As for league MVPs have been tops in EPA, that's fairly obvious since EPA is geared towards maximum correlation with points scored (which correlates very well with winning). And who gets MVP? Usually the quarterback on the most winning or most scoring team.

We're not going to find a valid measure of QB play over small samples for precisely the reasons you mentioned.  No one can sufficiently disentangle QB play from what's going on around the QB.  We simply have to wait until a larger sample is obtained, which is why you won't see me say much about how Tua has performed so far during his pro career.

What we do know however is that when you use EPA per pass dropback as the dependent variable and PFF's season QB grades and PFF's season team pass blocking grades -- which are probably the most valid measures we have of those parts of teams in isolation (purportedly) -- as independent variables in a multiple regression model, PFF's season QB grades are a very strong and significant predictor of EPA per pass dropback, whereas PFF's season team pass blocking grades are not.

So over the longer haul (a season, as opposed to a single play or a single game), the variation in team pass blocking throughout the league isn't sufficient to moderate the relationship between QB play and pass efficiency (EPA).  In terms of their season-long performance, there's a far bigger difference between the best and worst QBs in the league than there is between the best and worst offensive lines in the league, though this year's Dolphins team may in fact be an outlier in that regard, and Tua's performance may in fact be diminished by an offensive line this poor.

What we do know however is that if Tua has but an average offensive line, based on the finding noted above, no one should be complaining about how it interferes with his performance.  He should be perfectly capable of performing as well as the best QBs in the league with but an average offensive line, if indeed he has that ability.  If he has lesser ability, however -- i.e., Ryan Tannehill -- then he perhaps needs the stellar offensive line the league's best QBs do not.  At that point the complaints should be about him, not about the offensive line.

[masterfins]

I don't really care about stats like these because there can be so many other factors that the QB really can't control.  What interests me are TD's/Int's/and game wins.

[masterfins]

Then what interests you (implicitly) is EPA per play, since it'll do the best job of incorporating those variables as an overall measure of QB play (again valid only over larger samples).  Again the last five league MVPs have either led the league or been a very close second in the league in EPA per play.


And although we frown on ESPN's QBR because it's "black box," the last five league MVPs have finished either 1st (four of them) or 3rd (one of them) in QBR.  QBR is based very largely on EPA per play, however.  So QBR may be your preferred measure.

Okay, so I typed out this long response which I couldn't post because I got timed out, and I don't want to retype the whole thing.  So I'll respond with my last line of that post, which was: I don't believe in the "Moneyball" approach to judging NFL QB's.

[Dolfanalyst]

Okay, so I typed out this long response which I couldn't post because I got timed out, and I don't want to retype the whole thing.  So I'll respond with my last line of that post, which was: I don't believe in the "Moneyball" approach to judging NFL QB's.

Not a problem.  I deleted the post since it's of no use to you.

[fyo]

We're not going to find a valid measure of QB play over small samples for precisely the reasons you mentioned.  No one can sufficiently disentangle QB play from what's going on around the QB.  We simply have to wait until a larger sample is obtained, which is why you won't see me say much about how Tua has performed so far during his pro career.

What we do know however is that when you use EPA per pass dropback as the dependent variable and PFF's season QB grades and PFF's season team pass blocking grades -- which are probably the most valid measures we have of those parts of teams in isolation (purportedly) -- as independent variables in a multiple regression model, PFF's season QB grades are a very strong and significant predictor of EPA per pass dropback, whereas PFF's season team pass blocking grades are not.

So over the longer haul (a season, as opposed to a single play or a single game), the variation in team pass blocking throughout the league isn't sufficient to moderate the relationship between QB play and pass efficiency (EPA).  In terms of their season-long performance, there's a far bigger difference between the best and worst QBs in the league than there is between the best and worst offensive lines in the league, though this year's Dolphins team may in fact be an outlier in that regard, and Tua's performance may in fact be diminished by an offensive line this poor.

We do have better statistics available, though. I've been involved in football analytics for the better part of two centuries now, although I've pretty much dropped it completely over the past 4-5 years (bad timing in some ways), and the data has always been there to do better, yet we're still stuck using what essentially amounts to scraping the play-by-play. Part of the problem is the NFL's reticence to allow access to their (awesome) data, although you can get access to it in "dumps" if you smile just right and promise not to actually DO anything with it. But as analytics has moved past a few guys aggregating data as a side gig to a significant industry, it's perfectly possible to recreate much of the detailed NFL data from the all-22 video, so the lack of access cannot keep being used as an excuse.

Even that isn't really necessary to go significantly beyond what we have now (publicly available, anyway).

You mentioned EPA and Tannehill. Well, allow me to seize on that for a second. Back when Tannehill was pretty unanimously considered a horrible QB by advanced metrics, there were some pretty obvious indicators that Tannehill wasn't the main issue with the Dolphins offense. Tannehill lead the league (by a good margin) in what PFF at the time called "quick sacks", which basically amounted to "untouched to quarterback", yet he was dinged as fully at fault in all advanced metrics at the time.

Another indicator, one that hasn't even been properly explored to this day, was the issue of play-calling. The Dolphins brain trust insisted on calling receiver screen after receiver screen. This isn't a play where the quarterback can just go to his second or third read, it's pretty much set in stone. Yet WR screens are consistently among the worst plays in every success metric (aside from dumb ones like "completions"). One of the stat sites at the time, probably PFF again, but not sure, tabulated every pass of the season according to play and the Dolphins ran more receiver screens than anyone. And with below average results - but below average results was horrible compared to pretty much any other play call. Yet, again, metrics like EPA or DVOA ignore the play call completely, looking only at the average (i.e. expected) outcome from a similar pre-snap game state.

PFF actually upped the analysis earlier this season by calculating the EPA of the various different types of screens . Unsurprisingly, the type of screen the Dolphins ran with Tannehill (outside WR screen) scored the absolute worst EPA. (The article is quite interesting and certainly worth it if you haven't come across it already: https://www.pff.com/news/nfl-outside-wide-receiver-screens-and-ineffective-2021 )

Anyway, just a long-winded way of pointing out that we DO have better data than straight EPA and we CAN at least attempt to make adjustments for the stupidity of the play-calls. Perhaps we cannot always quantify these things (yet) to a sufficient degree, but - as with 4th down decisions - we CAN make a solid case for certain plays being called WAY too often at the wrong times.

Sadly, I don't have a team of people to go through game tape and while I do enjoy the once-a-year data dump of NFL data, it's frustrating not to have it week-to-week.

As for EPA as a measurement of quarterbacks, I would remind you that EPA is a *horrible* predictor when a quarterback switches teams. The EPA of the previous quarterback on the new team is actually a better predictor than the quarterback's own EPA with his old team. So maybe, just maybe, that offensive line and supporting cast has a larger influence than one might expect?

[Pappy13]

I don't really care about stats like these because there can be so many other factors that the QB really can't control.  What interests me are TD's/Int's/and game wins.

Because a QB is DIRECTLY responsible for all game wins? I don't understand what the other 40 or 50 players are doing out there then. Why are they even out there?

[masterfins]

Because a QB is DIRECTLY responsible for all game wins? I don't understand what the other 40 or 50 players are doing out there then. Why are they even out there?

I never said a QB was DIRECTLY responsible for all game wins.  Nor did I discount the contribution of the other 40 or 50 players.  What I did say was that game wins are ONE of the statistics that I prefer to use to judge a QB.

[pondwater]

I never said a QB was DIRECTLY responsible for all game wins.  Nor did I discount the contribution of the other 40 or 50 players.  What I did say was that game wins are ONE of the statistics that I prefer to use to judge a QB.

You have to be objective and use game wins and stats compared to the other top half QBs. Tua isn't bad when you take it all into account. Now the question is if Flores buys the shit he's selling. He tells the players to buy into the system. Does he buy into the system? Is him and Grier buying into that same system? Are they going to upgrade the positions that it takes for Tua to succeed?

That's on Ross. Their last Super Bowl win was 47 years ago, I was 2 years old. He's 81 years old, the party don't go on forever. If he wants to do something. I would say the next couple years is the time for a Miami Super Bowl win before before he dies. If he can do that, he will be remembered. If he can't he will be forgotten. That's basically what it amounts to. We have the 2nd highest cap space, it can be done.

[Pappy13]

I never said a QB was DIRECTLY responsible for all game wins.  Nor did I discount the contribution of the other 40 or 50 players.  What I did say was that game wins are ONE of the statistics that I prefer to use to judge a QB.

But you preceded that statement by saying that you don't like stats where there are so many factors that a QB doesn't control. He doesn't control the outcome of games either, all the players do.