Tannehill

[Dolfanalyst]

Those don't sound like ringing endorsements.

Yet when you look at those same folks' (PFF) data about the pass blocking of the Dolphins offensive line as a whole, those non-ringing endorsements don't distinguish it significantly from the other lines in the league.  Obviously there are "duds" elsewhere in the league, as well.

[Dolfanalyst]

Here's where you go wrong - assuming that PFF's rankings are normally distributed and that your assertion of "not more than one standard deviation away" has any significance.  That just speaks to a large standard deviation in a sample set that is too small to be statistically significant.  

You can not apply statistical analysis to PFF's ranking and rating and draw any reasonable conclusion.  To do so would essentially say that every offensive line in the league is equivalent in terms of performance.  You're a smart enough guy to know that this statement is false, right?

Why should that be false?  Is it not possible that there is virtually meaningless variation among the offensive lines in the league?

First, the league is built on a principle of parity overall.  Second, you have five guys out there, the biggest single unit on either side of the ball, on the field.  When you consider the functioning of a line as a whole, there is obviously vast potential for one guy's strengths to compensate for another guy's weaknesses.

It isn't inconceivable for that to result in very little meaningful variation among the lines in the league.

[Rich]

Is it not possible that there is virtually meaningless variation among the offensive lines in the league?

No. It is not.

[Rich]

Very little, when we're watching the team we care deeply about with much greater interest and intent than we are the rest of the league's games, and doing no systematic comparisons among those teams based on our observations.  In fact, I think the "eye test" can have negative value, in that it invites confirmation bias under those circumstances.

Told ya. He knows better than the rest of us how we evaluate our own team. 

[Dolfanalyst]

No. It is not.

Why not?

[Tenshot13]

WHAT A BUNCH OF NERDS!!!!
actually, I find this very entertaining
I had to get that out of the way.  Continue...

[Dolfanalyst]

WHAT A BUNCH OF NERDS!!!!
actually, I find this very entertaining
I had to get that out of the way.  Continue...

Nice!

[Dolfanalyst]

Told ya. He knows better than the rest of us how we evaluate our own team. 

I certainly hope that isn't the way you perceived it, Dave Gray.  If so, I apologize.

[Dolfanalyst]

Here's where you go wrong - assuming that PFF's rankings are normally distributed and that your assertion of "not more than one standard deviation away" has any significance.  That just speaks to a large standard deviation in a sample set that is too small to be statistically significant.  

You can not apply statistical analysis to PFF's ranking and rating and draw any reasonable conclusion.  To do so would essentially say that every offensive line in the league is equivalent in terms of performance.  You're a smart enough guy to know that this statement is false, right?

I wanted to address this once more with greater clarity because in my opinion this is where the crux of this matter lies, and I appreciate the post above for inspiring it.

When you're relating two variables to each other -- in this case offensive lines and quarterbacks throughout the league -- a change in one variable creates a change in the other to the degree that the two variables are related, or correlated.

If two variables are related very strongly, at let's say a correlation of 0.90, even a miniscule change in one variable would create a relatively large change in the other.

On the other hand, if two variables are related much more weakly -- let's say a correlation of only 0.30 -- it takes a much greater change in one variable to cause a relatively large change in the other.

What we find in this area of offensive lines and quarterbacks -- at least quantitatively, based on the available numbers -- is that the relationship is relatively weak.  The correlation between them is relatively weak.

Therefore even a relatively large improvement in the Dolphins' offensive line -- again, at least as measured quantitatively -- is very likely to produce a relatively small change in its quarterback play.

The difficulty we're having in this area, in my opinion at least, is that the relationship between these two variables -- offensive lines and quarterback play -- is perceived or assumed to be strong, when in reality it is not.  It's actually relatively weak.

Therefore some people are operating in this area from the assumption that a change in the Dolphins' offensive line play would cause a relatively large change in its quarterback play.  The correlation needed for that effect to occur between those two variables just isn't found league-wide, however.  The relationship (i.e., correlation) between the two variables just isn't strong enough for that sort of effect to be likely.

I hope that provided some clarity about my personal position on the issue, for what it's worth.  Of course people are free to believe whatever they'd like -- my intent isn't to cause any unwanted change in anyone else's beliefs.

[EKnight]

I'm gonna save you some headache here, Dolfanalyst. These guys who are in opposition to your data believe strongly (or want to believe strongly) that Tannehill was not only worth the #8 overall pick (In a year they could have drafted both Russell Wilson AND Luke Kuechly, if I recall), but also worth the absurd contract he was given in the offseason. They will find excuse after excuse to explain away his mediocrity because they don't want to believe their team made not only one, but two mistakes on RT. So, no matter how strongly your evidence demostrates that Tannehill's struggles are mostly on him, not his line, they will either debate this in the face of your direct evidence, or they will conjure up other reasons (he hasn't had quality receivers; he ate too much meatloaf and it slowed him down) to excuse the fact that he has not demonstrated that he is the QB they hoped and wanted him to be. He isn't an "absolute rockstar," as a former member thought he would be and that's not what they want to hear. Save yourself the headache of trying to convince them.

[Run Ricky Run]

100% agree with eknight however I appreciate your good posts. All you have to do is go back to the Henne threads were Brian and others were saying how he is a great qb who can make every throw and his only problem is he has the worst receiving corps in the nfl. 

[Dolfanalyst]

I wanted to address this once more with greater clarity because in my opinion this is where the crux of this matter lies, and I appreciate the post above for inspiring it.

When you're relating two variables to each other -- in this case offensive lines and quarterbacks throughout the league -- a change in one variable creates a change in the other to the degree that the two variables are related, or correlated.

If two variables are related very strongly, at let's say a correlation of 0.90, even a miniscule change in one variable would create a relatively large change in the other.

On the other hand, if two variables are related much more weakly -- let's say a correlation of only 0.30 -- it takes a much greater change in one variable to cause a relatively large change in the other.

What we find in this area of offensive lines and quarterbacks -- at least quantitatively, based on the available numbers -- is that the relationship is relatively weak.  The correlation between them is relatively weak.

Therefore even a relatively large improvement in the Dolphins' offensive line -- again, at least as measured quantitatively -- is very likely to produce a relatively small change in its quarterback play.

The difficulty we're having in this area, in my opinion at least, is that the relationship between these two variables -- offensive lines and quarterback play -- is perceived or assumed to be strong, when in reality it is not.  It's actually relatively weak.

Therefore some people are operating in this area from the assumption that a change in the Dolphins' offensive line play would cause a relatively large change in its quarterback play.  The correlation needed for that effect to occur between those two variables just isn't found league-wide, however.  The relationship (i.e., correlation) between the two variables just isn't strong enough for that sort of effect to be likely.

I hope that provided some clarity about my personal position on the issue, for what it's worth.  Of course people are free to believe whatever they'd like -- my intent isn't to cause any unwanted change in anyone else's beliefs.

I did some calculations to illustrate this further.  Bear with me while I drag this out fully.

In 2015, Ryan Tannehill was pressured on 38.6% of his pass dropbacks.

The quarterback who was pressured the least frequently in the league was Ben Roethlisberger -- 27.2%.

The correlation between frequency of pressured pass dropbacks and QBR (ESPN's QB rating) in 2015 was a mere 0.10.

Based on that information, we can calculate the predicted QBR Ryan Tannehill would've had, had he been pressured with the same frequency as Roethlisberger, and that QBR is 50.3.

Ryan Tannehill's actual QBR in 2015 was 43.1.  The average QBR in the league in 2015 was 58.5.

So if one is to consider that the least frequency of pressure on the QB reflects the very best offensive line functioning in the league, if we were to give Ryan Tannehill the very best offensive line functioning in the league (27.2% of pressured dropbacks), then by way of the calculated predicted change in his QBR, we would end up with an increase in his QBR from 43 to 50.

So, as I said above, that's the kind of relationship we have been offensive line functioning and quarterback play -- a very weak one.

Now, consider what would happen if instead the Dolphins' line improved from where it was to something less than the very best in the league, which of course is the more likely proposition.

Obviously Ryan Tannehill's predicted improvement would then be even less, from 43 to something even less than 50.

[Alwaysdullfan]

...They should let him run more! He's got good speed and throws well on the run. Esp considering how much we sucked on 3rd downs I dunno why they didn't let him run more. O well, hopefully Adam Gase can get more out of him!

[fyo]

When you're relating two variables to each other -- in this case offensive lines and quarterbacks throughout the league -- a change in one variable creates a change in the other to the degree that the two variables are related, or correlated.

If two variables are related very strongly, at let's say a correlation of 0.90, even a miniscule change in one variable would create a relatively large change in the other.

On the other hand, if two variables are related much more weakly -- let's say a correlation of only 0.30 -- it takes a much greater change in one variable to cause a relatively large change in the other.

Two words:

confounding variables

Also: QB pressure just isn't a very good way to estimate offensive line effectiveness. It doesn't reflect the number of defenders needed to create the pressure, it doesn't reflect how long the quarterback holds onto the ball (and why), it doesn't reflect offensive formation (extra men kept in to help block, which can reduce pressure, but also cripple the offensive production), it doesn't reflect play calling.

These are not minor issues. These are huge issues and every bit of football I've watched tells me it isn't something that just magically averages out with enough plays. As they say, extraordinary claims require extraordinary proof.

Personally, I think the best, easily available estimate of offensive line ability is Adjusted Line Yards. This was my initial suggestion and of all the things I've subsequently looked at, it also happens to have the strongest correlation with all measures of quarterback value, particularly the advanced metrics: between 0.46 and 0.48 for QBR, DYAR, and DVOA.

A correlation coefficient of nearly 0.5 is huge in football. Because of the sheer number interconnected and interdependent variables, it's very difficult to find anything that high. As always, correlation doesn't equal causation.

Adjusted Line Yards also just happens to be a metric by which the Dolphins' offensive line this year was more than 1 standard deviation below league average (if only just).

All that said, I don't think ALY is a GOOD way to estimate an offensive line's effect on the quarterback. Just looking at ANY game and you can see the problems. A good start would be to look at formations (O and D), play call (grouped), "quarterback time", and play result (including dropped receptions, dropped interceptions etc) -- along with the usual suspects (score, down and distance, time left, opponent etc). This is perfectly possible now that the all-22 tapes are public and some efforts (including footballoutsiders) are getting close.

[Dolfanalyst]

Two words:

confounding variables

Also: QB pressure just isn't a very good way to estimate offensive line effectiveness. It doesn't reflect the number of defenders needed to create the pressure, it doesn't reflect how long the quarterback holds onto the ball (and why), it doesn't reflect offensive formation (extra men kept in to help block, which can reduce pressure, but also cripple the offensive production), it doesn't reflect play calling.

These are not minor issues. These are huge issues and every bit of football I've watched tells me it isn't something that just magically averages out with enough plays. As they say, extraordinary claims require extraordinary proof.

Personally, I think the best, easily available estimate of offensive line ability is Adjusted Line Yards. This was my initial suggestion and of all the things I've subsequently looked at, it also happens to have the strongest correlation with all measures of quarterback value, particularly the advanced metrics: between 0.46 and 0.48 for QBR, DYAR, and DVOA.

A correlation coefficient of nearly 0.5 is huge in football. Because of the sheer number interconnected and interdependent variables, it's very difficult to find anything that high. As always, correlation doesn't equal causation.

Adjusted Line Yards also just happens to be a metric by which the Dolphins' offensive line this year was more than 1 standard deviation below league average (if only just).

All that said, I don't think ALY is a GOOD way to estimate an offensive line's effect on the quarterback. Just looking at ANY game and you can see the problems. A good start would be to look at formations (O and D), play call (grouped), "quarterback time", and play result (including dropped receptions, dropped interceptions etc) -- along with the usual suspects (score, down and distance, time left, opponent etc). This is perfectly possible now that the all-22 tapes are public and some efforts (including footballoutsiders) are getting close.

Great work.

You know me though, I'm going to rail at anything observational that doesn't allow for the systematic comparison of all the teams in the league, so I'll stick with your Adjusted Line Yards and contribute the fact that if Tannehill would've been afforded the league's best ALY by his offensive line in 2015, statistically speaking, his QBR would've most likely been 60.8 instead of 43.1.

That would've brought him from 30th in the league to 18th, just a tad above the league average of 58.5.

So even the much stronger correlation in the 0.50 range produces much less than the effect some people seem to believe exists between offensive lines and quarterback play.  Even that strong a correlation doesn't vault Tannehill to among the league leaders in quarterback play, even when you replace his offensive line's ALY with that of the best team in the league.

Now, one should ask themselves whether the Dolphins are likely to have the best offensive line in the league in that area.  Obviously that isn't likely.

So again, if you were to replace the Dolphins' offensive line's functioning with something better but much more likely to occur, you'd get an increase in Ryan Tannehill's QBR from 43 to something even less than 60.  And again, 60 is but a tad above the league average.

The problem of course is that Tannehill is starting so low in the league in this area that it would take an uncommonly strong correlation between QB play and offensive line play, as well as a highly unlikely degree of improvement in the Dolphins' offensive line, to get him where he needs to be for the team to be highly competitive.

Again, some people seem to believe that correlation indeed exists, and that improvement in the offensive line is likely, but neither seems to be the case, unfortunately.