Tannehill

[Dolfanalyst]

I'm discussing the fact that Tannehill has not had even a serviceable offensive line during his four year career. You cannot actually know how the quality of offensive line play affects Tannehill without seeing what Tannehill can do with an at least serviceable offensive line. Every season, there is at least one, usually two, glaring holes in the offensive line.

If what you're saying about the offensive line is true, then it shouldn't be difficult to produce a quantitative measure of offensive line play from some source that indicates the Dolphins' line is at least a standard deviation worse than the average line in the league.

And if what you're saying about the relationship between Tannehill's play and the offensive line is true, then it shouldn't be difficult to produce either a league-wide correlation between quarterback and offensive line play, or a Dolphins-specific correlation between its offensive line play and Tannehill's.

[Sunstroke]

No disrespect to the pursuit of statistical enlightenment, but I've always felt that I learn more about football players by watching them play football than I do from looking at their stat sheet.

[Dolfanalyst]

No disrespect to the pursuit of statistical enlightenment, but I've always felt that I learn more about football players by watching them play football than I do from looking at their stat sheet.

And no disrespect to learning things about football players by watching them play, but what we're seeing when we watch them generally translates to statistics.  Certainly you don't think the fact that Aaron Rodgers has the highest career QB rating in the history of the league, for example, is meaningless?  Certainly we see evidence of that when we watch him play?

In other words, the two approaches aren't mutually exclusive.  What we're seeing on the field is generally measurable statistically, and what we see statistically is generally reflected on the field observationally.

Where we get into trouble, however, is when we all very much want to see a certain thing on the field (Ryan Tannehill's success, for example), and then we're susceptible to confirmation bias (myself included, of course).  It's at those times that we should check ourselves with the objective data in my opinion.  Sometimes the data will confirm our perceptions, and sometimes it won't.  When it doesn't, and we're in an area where we're prone to bias, we ought to back away from feeling so certain about our perceptions in my opinion.

To each his own, however, of course.

In the end, having a polite and courteous discussion, regardless of our differences and disagreements, is what's most important, since none of what anybody thinks here really matters in the grand scheme of things.

[Rich]

No disrespect to the pursuit of statistical enlightenment, but I've always felt that I learn more about football players by watching them play football than I do from looking at their stat sheet.

Exactly. Data analytics is great for figuring out why widgets in an assembly line are defective and what can be done to reduce defects.

Watching film is the best way to decipher what is wrong with a team.

[Rich]

If what you're saying about the offensive line is true, then it shouldn't be difficult to produce a quantitative measure of offensive line play from some source that indicates the Dolphins' line is at least a standard deviation worse than the average line in the league.

Football Outsiders and PFF agree that this offensive line has sucked for a few years now. You have seen the numbers. I'm not sure why you are pretending they don't exist.

And if what you're saying about the relationship between Tannehill's play and the offensive line is true, then it shouldn't be difficult to produce either a league-wide correlation between quarterback and offensive line play, or a Dolphins-specific correlation between its offensive line play and Tannehill's.

All I said about Tannehill is that we cannot compare how he performs behind a poor offensive line to how he would perform behind even a decent offensive line because he has always played behind a poor offensive line.

When we get to see him play behind a decent offensive line (if that ever happens, I'm not holding out hope), then we can talk about whether it impacted his performance.

Serviceable blocking from all five spots may translate into a higher level of QB play or it may not.

[Dolfanalyst]

Football Outsiders and PFF agree that this offensive line has sucked for a few years now. You have seen the numbers. I'm not sure why you are pretending they don't exist.

I doubt I sound like someone who pretends that numbers don't exist.

What I'm saying, rather, is that the numbers don't indicate that the Dolphins' line is at least a standard deviation worse than the average one in the league.  The line is within the league norm in every league-wide statistical measurement of offensive lines that I'm aware of.

All I said about Tannehill is that we cannot compare how he performs behind a poor offensive line to how he would perform behind even a decent offensive line because he has always played behind a poor offensive line.

When we get to see him play behind a decent offensive line (if that ever happens, I'm not holding out hope), then we can talk about whether it impacted his performance.

Serviceable blocking from all five spots may translate into a higher level of QB play or it may not.

And all I'm saying is that the evidence we have at hand currently suggests he wouldn't.  That evidence is the following:

1) the weak correlation league-wide between offensive line and QB play.

2) the weak correlation between the Dolphins' offensive line play and Tannehill's.

3) the strong, positive correlation (0.97 -- correlations are very rarely that strong), ironically, between the frequency of pressure Tannehill has experienced year-to-year, and his performance.  He does better the more he's pressured!

What these data (and others) suggest in my opinion is that Tannehill's performance is far, far more a function of what's going on within him (his own personal development, for example) than it is a function of what's going on around him.

[Rich]

I doubt I sound like someone who pretends that numbers don't exist.

What I'm saying, rather, is that the numbers don't indicate that the Dolphins' line is at least a standard deviation worse than the average one in the league.  The line is within the league norm in every league-wide statistical measurement of offensive lines that I'm aware of.

The offensive line has been ranked in the bottom half of the league based on cumulative statistics every season that Tannehill has played. Both on FO and PFF.

By you adding the term "standard deviation", all you're doing is whitewashing their poor statistical ranking.

It is irrelevant. They rank consistently poor. That is all.

And all I'm saying is that the evidence we have at hand currently suggests he wouldn't.  That evidence is the following:

1) the weak correlation league-wide between offensive line and QB play.

2) the weak correlation between the Dolphins' offensive line play and Tannehill's.

3) the strong, positive correlation (0.97 -- correlations are very rarely that strong), ironically, between the frequency of pressure Tannehill has experienced year-to-year, and his performance.  He does better the more he's pressured!

What these data (and others) suggest in my opinion is that Tannehill's performance is far, far more a function of what's going on within him (his own personal development, for example) than it is a function of what's going on around him.

#1 I do not find relevant. Some quarterbacks do fine with poor protection because they are mobile enough to extend plays. Others do fine with poor protection because their offense requires them to get rid of the ball quickly, throw the ball away or take the sack rather than risk a bad throw. There are too many other variables to consider that this data point has any sort of credibility.

#2 Maybe it is a credit to Tannehill that he can perform well at times despite deplorable line play during his entire NFL career

#3 Pressure isn't always an indicator or poor offensive line play, pressure comes in many forms

You are relying on irrelevant or significantly incomplete data points to reach a conclusion. That is the problem with relying on statistics.

[Dolfanalyst]

Watching film is the best way to decipher what is wrong with a team.

I think it's another great way of getting to the bottom of things.

Certainly everybody who belongs to a Dolphins' fan forum watches all of the Dolphins' games intently.  If the forum is roughly split on the issue of whether Tannehill is a good player, then the truth, via all of their watching of the same film, would be that he's about average, no, since that would represent the most accurate combination of all of their viewpoints?

[Dolfanalyst]

The offensive line has been ranked in the bottom half of the league based on cumulative statistics every season that Tannehill has played. Both on FO and PFF.

By you adding the term "standard deviation", all you're doing is whitewashing their poor statistical ranking.

It is irrelevant. They rank consistently poor. That is all.

How much something deviates from the norm tells you whether it's "poor," not its ranking in a list.  If it doesn't deviate significantly from the norm, it's part of the norm, regardless of where it's ranked in the list.

[Rich]

How much something deviates from the norm tells you whether it's "poor," not its ranking in a list.  If it doesn't deviate significantly from the norm, it's part of the norm, regardless of where it's ranked in the list.

How consistently this offensive line has ranked in the bottom half of the league tells you that it is part of the norm for this offensive line to be below average. Because they have ranked in the bottom half of the league every season since 2012. Sometimes they ranked right near the bottom (like this season), sometimes they rank closer to the middle. But they consistently rank in the bottom half.

And therefore, we can effectively conclude that Tannehill has played behind a below average offensive line for his entire professional career.

It doesn't really matter of the standard deviation is wide or narrow.

The NFL is a league of parody where a loss or two can mean a top 10 pick or a playoff spot. And therefore having a narrow standard deviation is right in line with how the league operates so it doesn't really mean anything. It is what it is and it is the world in which NFL teams operate in, so it doesn't really add anymore weight to what the numbers mean or don't mean.

What means something, again, is that this offensive line is consisently below average.

And that is all I am going to say on the topic as we're going around in circles now.

[Sunstroke]

And no disrespect to learning things about football players by watching them play, but what we're seeing when we watch them generally translates to statistics.  Certainly you don't think the fact that Aaron Rodgers has the highest career QB rating in the history of the league, for example, is meaningless?  Certainly we see evidence of that when we watch him play?

True, though I had a good feeling about Rodgers chances of becoming a top-tier QB long before he put up elite NFL statistics.

In other words, the two approaches aren't mutually exclusive.  What we're seeing on the field is generally measurable statistically, and what we see statistically is generally reflected on the field observationally.

Wait...there's chickens AND there's eggs?

Look, I am not knocking statistical analysis. As someone who has played some form of fantasy football for close to 30 years now, I've devoured tons of statistical analyses in the pursuit of profitable knowledge. Where I scratch my bean is when I see people try to analyze football by looking solely at the stat sheet (and often only certain stats on that sheet) while ignoring the subtleties that can only be seen by watching the game itself.

[Run Ricky Run]

On one side we have Bill belichek and his statistical analysis on the other side we have Mike singletary who knows you just can't win with Vernon Davis.  This thread shws why smart coaches have to make dumb decisions if they want to stay employed i.e. Punting on 4th and short and ".taking the points"

[fyo]

What I'm saying, rather, is that the numbers don't indicate that the Dolphins' line is at least a standard deviation worse than the average one in the league.  The line is within the league norm in every league-wide statistical measurement of offensive lines that I'm aware of.

In almost all league statistics EVERYONE is within a standard deviation or so of the mean. So everyone is the same?

The problem in football is that there just aren't enough games, particularly games against the same opponent, to get a enough numbers.

The issue is further complicated by the number of variables, which is just staggering.

Stat padding is also a huge issue and pretty much kills any stat that doesn't adjust for garbage time (and, preferably, down and distance).

To get back evaluating line play vs quarterback play, I do think we could get some better numbers (how good is the question), but we'd need a lot more data. For example, defensive blitzing has a strong correlation with positive quarterback play. To some extent this makes sense, since using more players to go after the quarterback leaves fewer defenders to cover receivers. However, if a team can generate pressure using just 4 players, is this correlation still there? I very much doubt so, but I've never seen the numbers run.

To get even close to a meaningful statistical discussion of how Tannehill's play (or any quarterback's play) is affected by the offensive line, we would need more data than simply season totals. That's obvious. We would need a per-play metric to evaluate quarterback performance and a per-play metric to evaluate line performance.

One simple example of why this is crucial is the formation used. If a team brings in two tight ends to help block, along with the running back, and the quarterback still gets sacked by a defender in less than 3 seconds, that's a completely different situation than a 5+ second sack in a three-receiver set with a blitz from the D and one of the receivers going uncovered. In one case, the protection for the quarterback is horrendous and the line (which includes any "help" it receives) is probably completely fault. However, in the other case, the quarterback at the very least had time to throw the ball away and not take the sack (even if he doesn't spot the open receiver).

Any performance metric that scores those two plays identically is fatally flawed if your goal is to distinguish between line play and quarterback play.

Currently, most comparisons use metrics that not only score those two plays identically, they don't even score them except in aggregate.

[Dolfanalyst]

In almost all league statistics EVERYONE is within a standard deviation or so of the mean. So everyone is the same?

In many areas, yes, they're not significantly different, owing to the parity in the league.  Significant differences among teams are generated rather by multiple areas in which there are perhaps non-significant differences among them (i.e., pass offense, pass defense, running game, turnovers, etc.).

The problem in football is that there just aren't enough games, particularly games against the same opponent, to get a enough numbers.

The issue is further complicated by the number of variables, which is just staggering.

Stat padding is also a huge issue and pretty much kills any stat that doesn't adjust for garbage time (and, preferably, down and distance).

To get back evaluating line play vs quarterback play, I do think we could get some better numbers (how good is the question), but we'd need a lot more data. For example, defensive blitzing has a strong correlation with positive quarterback play. To some extent this makes sense, since using more players to go after the quarterback leaves fewer defenders to cover receivers. However, if a team can generate pressure using just 4 players, is this correlation still there? I very much doubt so, but I've never seen the numbers run.

To get even close to a meaningful statistical discussion of how Tannehill's play (or any quarterback's play) is affected by the offensive line, we would need more data than simply season totals. That's obvious. We would need a per-play metric to evaluate quarterback performance and a per-play metric to evaluate line performance.

One simple example of why this is crucial is the formation used. If a team brings in two tight ends to help block, along with the running back, and the quarterback still gets sacked by a defender in less than 3 seconds, that's a completely different situation than a 5+ second sack in a three-receiver set with a blitz from the D and one of the receivers going uncovered. In one case, the protection for the quarterback is horrendous and the line (which includes any "help" it receives) is probably completely fault. However, in the other case, the quarterback at the very least had time to throw the ball away and not take the sack (even if he doesn't spot the open receiver).

Any performance metric that scores those two plays identically is fatally flawed if your goal is to distinguish between line play and quarterback play.

Currently, most comparisons use metrics that not only score those two plays identically, they don't even score them except in aggregate.

What you said above is certainly reasonable.  The question would be whether that situational variation (blitzing, pressure with four-man rushes versus more rushers, more blockers, etc.) becomes roughly equivalent among teams when the data are based on season totals.  We're talking upwards of 600 pass dropbacks per team here, on average.  That's an awful lot of opportunity for that sort of variation to become roughly equivalent among teams.

[fyo]

In many areas, yes, they're not significantly different, owing to the parity in the league.  Significant differences among teams are generated rather by multiple areas in which there are perhaps non-significant differences among them (i.e., pass offense, pass defense, running game, turnovers, etc.).

Very, very few quarterbacks are more than 1 standard deviation away from the mean in passer rating. Are their differences insignificant? Tannehill, for instance, is solidly within 1 standard deviation. Brady just barely squeaks past the 1 standard deviation limit. Why are we even having this 9-page discussion if everyone's pretty much the same?

Clearly, they are not and either the metric used to measure the value of quarterbacks is rubbish or the insistence on a 1 sigma difference is rubbish. While we're at it, why not insist on the 5 sigma usually used in particle physics to be *really* sure there's a difference?

Okay, that was somewhat facetious, but the problem with sigma is that it doesn't reflect the magnitude of the effect. Take an extreme case of 100-meter sprinters. It really doesn't matter how narrow the distribution of performances are, Usain Bolt just keeps winning. (Actually running the numbers for some major 100m events, it turns out that the difference between the average and the top-end of the 1-sigma range is the difference between not qualifying for the finals at all and winning a medal, although usually not gold).

The question would be whether that situational variation (blitzing, pressure with four-man rushes versus more rushers, more blockers, etc.) becomes roughly equivalent among teams when the data are based on season totals.  We're talking upwards of 600 pass dropbacks per team here, on average.  That's an awful lot of opportunity for that sort of variation to become roughly equivalent among teams.

It probably would, if you looked at it naively. However, if a defense can generate pressure with just rushing 4 guys are they going to blitz as often as the same defense when it's having trouble getting to the quarterback?

Will the coach change the play-calling if the offensive line is atrocious? Maybe call a lot more runs, even when behind and passing is needed? Or calling for 5-yard hitch plays on 3rd and long?

Done properly, I have absolutely no reason to believe these effects are just going to magically balance themselves between teams over the course of 600 drop backs.