Tannehill

[Rich]

The problem with that approach is that it wouldn't tell you how often the linemen on other teams are getting beaten similarly

I find this to be a completely irrelavent argument. It's like saying that its OK for Dallas Thomas to be one of the worst guards in the league because there are other teams that also have one of the worst guards in the league.

Besides, you can watch the games and then check the PFF rankings and see that what your eyes tell you align with how low Thomas and Fox are ranked at their respective positions.

There is no way to argue that Dallas Thomas and Jason Fox are not huge holes on this offensive line without grasping at straws.

[Dolfanalyst]

I find this to be a completely irrelavent argument. It's like saying that its OK for Dallas Thomas to be one of the worst guards in the league because there are other teams that also have one of the worst guards in the league.

Besides, you can watch the games and then check the PFF rankings and see that what your eyes tell you align with how low Thomas and Fox are ranked at their respective positions.

There is no way to argue that Dallas Thomas and Jason Fox are not huge holes on this offensive line without grasping at straws.

Again, I'm operating from the perspective that we're trying to ascertain the effect of the offensive line on Tannehill (i.e., the title of the thread).

If that's one's goal, then we can hardly say the offensive line has an adverse effect on Tannehill if in fact the offensive line is no worse than other offensive lines whose quarterbacks play significantly better than Tannehill.

And again, to get at that, you can't simply stop at the observation of Dallas Thomas and Jason Fox.  You have to know how offenisve lines function across the league as a whole, and whether there is any correlation between that and quarterback play.  For all we know, every team has a "Dallas Thomas," and there is a weak correlation between offensive line and quarterback play.

If you want the matter to rest at least partially on PFF rankings, as you've indicated above, then you'd have to obtain a strong correlation between PFF's offensive line rankings and quarterback play.  Again, if that correlation is weak or non-existent, it hardly makes sense to conclude that the offensive line is a significant contributor to Ryan Tannehill's individual play, based at least on PFF's ratings.

[Rich]

How can we conclude the impact to Tannehill's play caused by the offensive line if Tannehill has never had a good offensive line blocking for him?

If in 2014, he had a top 10 unit and had an 85 QB rating, and in 2015 he had a bottom 10 unit and still had an 85 QB rating, we could theorize that the offensive line has no impact and Tannehill is what he is.

But because he has had a terrible offensive line every single year he has been here, we have no differentiator to compare it to.

Tannehill has played behind the offensive lines he has played behind. It doesn't matter how other QBs are performing behind other offensive lines. It matters how Tannehill is performing behind his and how he would perform if he had better blocking. We've yet to see the better blocking, so there is no way to complete the analysis.

Also, looking at the offensive line as a whole can be misleading. You can have four linemen playing at a high level and one guy who can't block a corpse and it impacts the QB or you can have five linemen doing a serviceable job and no weak link and it has a totally different impact on the QB.

[Dolfanalyst]

How can we conclude the impact to Tannehill's play caused by the offensive line if Tannehill has never had a good offensive line blocking for him?

If in 2014, he had a top 10 unit and had an 85 QB rating, and in 2015 he had a bottom 10 unit and still had an 85 QB rating, we could theorize that the offensive line has no impact and Tannehill is what he is.

But because he has had a terrible offensive line every single year he has been here, we have no differentiator to compare it to.

Tannehill has played behind the offensive lines he has played behind. It doesn't matter how other QBs are performing behind other offensive lines. It matters how Tannehill is performing behind his and how he would perform if he had better blocking. We've yet to see the better blocking, so there is no way to complete the analysis.

Also, looking at the offensive line as a whole can be misleading. You can have four linemen playing at a high level and one guy who can't block a corpse and it impacts the QB or you can have five linemen doing a serviceable job and no weak link and it has a totally different impact on the QB.

If there is a weak correlation between offensive line and quarterback play league-wide, then yes, the argument could still be made that Tannehill is an outlier in that regard, and he (and perhaps he alone) is more dependent on offensive line play than the other QBs in the league.

In that event, what you'd do is take a look at Tannehill's game-by-game performance and see whether there's a correlation between the Dolphins' offensive line's play and Tannehill's.

In other words, the question becomes, does Tannehill play better when his offensive line does?  Does he play worse when it does?

If you have no strong league-wide correlation between offensive line play and quarterback play, and you have no strong game-to-game correlation between the Dolphins' offensive line play and Tannehill's play, then attributing Tannehill's play to the Dolphins' offensive line surely becomes the sort of "grasping at straws" you mentioned above.

[Rich]

We have no comparison between Tannehill's play behind a terrible offensive line and Tannehill's play behind a good or even moderately serviceable offensive line. So we do not know what kind of impact it would have if he played behind a somewhat compenent offensive line.

I'd like to see Tannehill play behind a somewhat remotely mediocre offensive line instead of a historically terrible one before determining if the offensive line impacts his game or not.

What we do know is that Tannehill has played behind a bottom of the league offensive line for 4 years and he has performed how he has performed.

As long as there continue to be holes in the offensive line, we won't know how he can perform behind an even serviceable offensive line.

And just to be clear, because this is a not a black and white argument. I'm not asking to see Tannehill behind an offensive line full of All Pros. I am just asking to see him behind an offensive line that doesn't have a gaping hole at 1 or 2 positions.

I've used several adjectives in this post to illustrate that I am not asking for worldbeaters at offensive line. Just guys that can block somebody for longer than 1 second.

[fyo]

The strongest argument that Tannehill's play is a function of the Dolphins' offensive line would come from both 1) a strong league-wide correlation between pass blocking and quarterback play, and 2) quantitative evidence that the Dolphins' offensive line functions significantly more poorly than that of the average team in the league.

Let's do some math then.

Actually, the math is trivial, the big question is, how do we evaluate the line?

This is far from trivial or obvious. Looking simply at sacks (or sack rate) is severely limited for a number of reasons. Sacks can be completely the fault of the line, completely the fault of the quarterback, or anywhere in between (and that's ignoring other sources of blame). There's just no way to know from just looking at the numbers. And you could argue that this "blame" is actually a large part of what we want to find out. A good quarterback will find a receiver, make adjustments, or just throw the ball away. Crediting the offensive line for that is counterproductive to what we wish to accomplish. As always "correlation is not causation".

One option would be to include only specific types of sacks (e.g. exclude "long" sacks, where the blame is rests more clearly on the quarterback), but there aren't a whole lot of sacks to begin with. Plus I don't have the numbers .

Another option would be to include hits and hurries. That would seem to at least lessen the impact of the line/QB blame issue. If I had the numbers, I'd be tempted to exclude "long" sacks.

Yet another option would be to attack the problem from a completely different angle and look at the overall quality of the line and how it contributes to other plays (that would be running plays). FootballOutsiders have a statistic called "adjusted line yards" which attempts to measure exactly this value. Basically, the offensive line is dinged for running plays that achieve negative yardage and is credited less and less the further the run (so lots of credit/blame to the line for loss or short yardage, not so much for a huge run). The entire thing is then adjusted for things like opponent, formation, down-and-distance, time left etc. using regression analysis.

Running the numbers returns the following correlation coefficients between adjusted line yards and XX:

DYAR: +0.48
DVOA: +0.49
QBR: +0.46
PASSER RATING: +0.32

Those are fairly decent correlations, which would seem to indicate that offensive line performance does indeed significantly impact quarterback play.

Looking instead at adjusted sack rate, the correlation just isn't there. In fact, it's only significant when looking at passer rating (+0.23) and actually turns negative for DYAR (-0.15).

Including hits (NFL.com lists this stat) didn't improve the correlation (DVOA and DYAR about -0.15, the rest close to zero).

[Dolfanalyst]

We have no comparison between Tannehill's play behind a terrible offensive line and Tannehill's play behind a good or even moderately serviceable offensive line. So we do not know what kind of impact it would have if he played behind a somewhat compenent offensive line.

I'd like to see Tannehill play behind a somewhat remotely mediocre offensive line instead of a historically terrible one before determining if the offensive line impacts his game or not.

What we do know is that Tannehill has played behind a bottom of the league offensive line for 4 years and he has performed how he has performed.

As long as there continue to be holes in the offensive line, we won't know how he can perform behind an even serviceable offensive line.

And just to be clear, because this is a not a black and white argument. I'm not asking to see Tannehill behind an offensive line full of All Pros. I am just asking to see him behind an offensive line that doesn't have a gaping hole at 1 or 2 positions.

I've used several adjectives in this post to illustrate that I am not asking for worldbeaters at offensive line. Just guys that can block somebody for longer than 1 second.

With regard to the part I bolded, we actually don't know that, unless, again, we objectively compare the Dolphins' line to those of other teams.

In the end this may come down to an epistemological issue, i.e., how people know what they believe they know.

Some people may be very comfortable feeling they know what they believe they know by simply observing the team on TV.

I for one am not comfortable with that.  I feel like we're very prone to confirmation bias in this area, and so I'd much rather rest on whatever objective data are available.

Errors can be made in either case (observation versus objective data), but in my opinion we'll make far fewer errors by resting on the objective data.

I think we have to start with a strong appreciation for the fact that there are certain things about the team we all want to see (like Ryan Tannehill's success, for example), and so, again, we'll be very prone to confirmation bias in those areas.

[fyo]

^ On  the offensive line... If you want statistics on how the Dolphins' o-line compares to that of other teams, you could look at e.g.:

http://www.footballoutsiders.com/stats/ol

28th in adjusted line yards, 27th in short yardage situations, 30th in tackles for a loss (that's the bad direction). Once the running back reaches the second level we're suddenly a top 10 team (same with "open field" runs).

[Dolfanalyst]

Let's do some math then.

Actually, the math is trivial, the big question is, how do we evaluate the line?

This is far from trivial or obvious. Looking simply at sacks (or sack rate) is severely limited for a number of reasons. Sacks can be completely the fault of the line, completely the fault of the quarterback, or anywhere in between (and that's ignoring other sources of blame). There's just no way to know from just looking at the numbers. And you could argue that this "blame" is actually a large part of what we want to find out. A good quarterback will find a receiver, make adjustments, or just throw the ball away. Crediting the offensive line for that is counterproductive to what we wish to accomplish. As always "correlation is not causation".

One option would be to include only specific types of sacks (e.g. exclude "long" sacks, where the blame is rests more clearly on the quarterback), but there aren't a whole lot of sacks to begin with. Plus I don't have the numbers .

Another option would be to include hits and hurries. That would seem to at least lessen the impact of the line/QB blame issue. If I had the numbers, I'd be tempted to exclude "long" sacks.

Yet another option would be to attack the problem from a completely different angle and look at the overall quality of the line and how it contributes to other plays (that would be running plays). FootballOutsiders have a statistic called "adjusted line yards" which attempts to measure exactly this value. Basically, the offensive line is dinged for running plays that achieve negative yardage and is credited less and less the further the run (so lots of credit/blame to the line for loss or short yardage, not so much for a huge run). The entire thing is then adjusted for things like opponent, formation, down-and-distance, time left etc. using regression analysis.

Running the numbers returns the following correlation coefficients between adjusted line yards and XX:

DYAR: +0.48
DVOA: +0.49
QBR: +0.46
PASSER RATING: +0.32

Those are fairly decent correlations, which would seem to indicate that offensive line performance does indeed significantly impact quarterback play.

Looking instead at adjusted sack rate, the correlation just isn't there. In fact, it's only significant when looking at passer rating (+0.23) and actually turns negative for DYAR (-0.15).

Including hits (NFL.com lists this stat) didn't improve the correlation (DVOA and DYAR about -0.15, the rest close to zero).

Excellent work there!  Thank you.

Pertaining to this part here:

DYAR: +0.48
DVOA: +0.49
QBR: +0.46
PASSER RATING: +0.32

I agree with you that those are decent correlations, but there is between 76% and 90% of the variation in the quarterback play variables you mentioned (DYAR, DVOA, QBR, and passer rating) that isn't accounted for by the offensive line variable you used.

So while the correlations are indeed decent, what we find is that there is a whole lot about QB play -- at least as measured by those variables -- that has nothing to do with offensive line play, as measured in the way you did it above.  What that suggests, of course, is that there are other variables that are driving quarterback play to a much greater degree.

As for the sacks and hits data, those correlations are very meager, as you pointed out.

[fyo]

^ On the topic of "other effects", one example would be play calling.

Conventional wisdom would have us believe (and this is indeed my own feeling) that the Dolphins have been horrible at calling plays. But how do you evaluate play calling? This is one of the eternal problems with analytics in football (as opposed to, for example, baseball); there are just so many interdependent issues that conclusively determining causal relations is exceptionally difficult.

One way to try and isolate performance from play calling would be to look at specific situations where the options in play calling are very limited. The most obvious example (non-special teams) would be very short yardage (power) situations. This is one case where success or failure is pretty much out of the hands of the play caller, resting solely on the field and probably (conventional wisdom again) in the hands of the respective lines.

Using this method to evaluate the Dolphins' offensive line doesn't paint a pretty picture, as illustrated in my previous post.

One might expect 3rd and long situations to also represent a good opportunity to reduce the role of play calling, but when you keep throwing 5 yard hitches, pitch outs, or draw plays in those situations (and never succeeding), I'm not convinced that's a viable approach.

[fyo]

So while the correlations are indeed decent, what we find is that there is a whole lot about QB play -- at least as measured by those variables -- that has nothing to do with offensive line play, as measured in the way you did it above.  What that suggests, of course, is that there are other variables that are driving quarterback play to a much greater degree.

Just a quibble here:

-> What that suggests, of course, is that there are other variables that combined are driving quarterback play to a much greater degree.

Offensive line play could well be the singles largest determining factor in quarterback performance. I'm not saying it is, just pointing out the logic.

[Dolfanalyst]

Just a quibble here:

-> What that suggests, of course, is that there are other variables that combined are driving quarterback play to a much greater degree.

Offensive line play could well be the singles largest determining factor in quarterback performance. I'm not saying it is, just pointing out the logic.

Right, and good point, since we don't know how large the variation accounted for by each of the other variable(s) is.

What we'd have to do is take a look at single QBs across variation in the play of their offensive lines, and determine whether their play varies accordingly.

I will say that one thing we know about Tannehill is that the percentage of pass dropbacks in which he was pressured increased from 2012 to 2014, yet so did his performance as measured by the sorts of variables you mentioned above.  His percentage of pressured dropbacks decreased somewhat in 2015, and so did his performance.

So based on that alone, we get the counterintuitive finding that as Tannehill is pressured more often, his performance improves!

[Dolfanalyst]

^ On the topic of "other effects", one example would be play calling.

Conventional wisdom would have us believe (and this is indeed my own feeling) that the Dolphins have been horrible at calling plays. But how do you evaluate play calling? This is one of the eternal problems with analytics in football (as opposed to, for example, baseball); there are just so many interdependent issues that conclusively determining causal relations is exceptionally difficult.

One way to try and isolate performance from play calling would be to look at specific situations where the options in play calling are very limited. The most obvious example (non-special teams) would be very short yardage (power) situations. This is one case where success or failure is pretty much out of the hands of the play caller, resting solely on the field and probably (conventional wisdom again) in the hands of the respective lines.

Using this method to evaluate the Dolphins' offensive line doesn't paint a pretty picture, as illustrated in my previous post.

One might expect 3rd and long situations to also represent a good opportunity to reduce the role of play calling, but when you keep throwing 5 yard hitches, pitch outs, or draw plays in those situations (and never succeeding), I'm not convinced that's a viable approach.

Good thought, and there is a related area where we get a relatively weak correlation regarding Tannehill on a game-by-game basis.  The correlation between the Dolphins' run-pass ratio and Tannehill's performance, game-to-game, is relatively meager.  Whether the Dolphins run the ball more or less of the time in a game, Tannehill's performance isn't strongly related to it.

There is also a relatively weak correlation between Tannehill's performance and how well the running game functions game-by-game, as measured by total yards and yards per carry.

[fyo]

So based on that alone, we get the counterintuitive finding that as Tannehill is pressured more often, his performance improves!

Using a pure "hit plus sack" rate as a measure of pressure, that observation holds generally. As noted above, the advanced metrics DYAR and DVOA have a negative correlation with "hit plus sack" rate, albeit a very weak one (-0.15).

This could be related to the often noted effect that almost all quarterbacks get more yards / play (and have a higher passer rating) when blitzed than when not blitzed.

[Dolfanalyst]

Using a pure "hit plus sack" rate as a measure of pressure, that observation holds generally. As noted above, the advanced metrics DYAR and DVOA have a negative correlation with "hit plus sack" rate, albeit a very weak one (-0.15).

This could be related to the often noted effect that almost all quarterbacks get more yards / play (and have a higher passer rating) when blitzed than when not blitzed.

And you'll never guess what the correlation is between the percentage of dropbacks in which Tannehill has been pressured, year-to-year, and his QB rating:

0.97!

On the other hand the correlation between the percentage of pressured dropbacks in which Tannehill has been sacked, year-to-year, and his QB rating is -0.61.

So, as he's pressured more often, his QB rating increases (and almost isomorphically), yet his QB rating decreases the more often his pressures result in sacks.

If we use a partial correlation to control for the percentage of pressures resulting in sacks, the correlation between his QB rating and his percentage of pressured dropbacks remains very strong -- 0.95.