Ryan Tannehill is Typically a Loser in NFL "Quarterback Duels"

[Dolfanalyst]

How do you make sense of the very strong correlation between QB rating differential and winning, over a very large sample?

Also, the point here for me originally was how the Dolphins would do in the playoffs, when facing teams that are likely to have QBs who can play as well as what we're talking about here.  So while the Dolphins regular season record may not improve a great deal if Tannehill's QB rating were elevated in such games, of course in the playoffs it's merely one and done, and the team with the QB that's likely to put it on the far shorter end of the QB rating differential is at a huge disadvantage.

I'm not saying anything nobody knows here.  If the Dolphins were to face the Patriots in the playoffs, for example, everyone's immediate thought would (or should) be that Tannehill is going up against Brady, and would have to put up an unusually good effort on his part to compete with what Brady is likely to do.

[fyo]

Well hello straw man.
I've never said quarterback performance doesn't matter.
I've never said passer rating doesn't matter.
I've never said difference in quarterback rating doesn't correlate highly with wins. Of course it does, only an idiot would suggest otherwise.

The issue here is your completely fictitious concept of quarterback duels, as if Tanehill could somehow influence the opposing quarterback to a significant degree. The simple truth is, no quarterback appears to significantly influence the outcome of games where the opposing quarterback eclipses 110 passer rating. The only way to win is to not let the opposing quarterback get there in the first place...

And the way to prevent quarterbacks from getting high passer ratings is defense, not offense.

That also happens to nicely explain your correlation... When your defense isn't laying an egg, a good performance by your quarterback wins the game.

I wish you'd admit that your quarterback duel concept is bogus. Despite trying to criticize the Dolphins and Tanehill with some nice cherry picking, the numbers show they've managed to win MORE of these games than average despite facing one of the highest average passer ratings. If course, I'm not saying that means anything... There's just not nearly enough data.

[Dolfanalyst]

Well hello straw man.
I've never said quarterback performance doesn't matter.
I've never said passer rating doesn't matter.
I've never said difference in quarterback rating doesn't correlate highly with wins. Of course it does, only an idiot would suggest otherwise.

The issue here is your completely fictitious concept of quarterback duels, as if Tanehill could somehow influence the opposing quarterback to a significant degree. The simple truth is, no quarterback appears to significantly influence the outcome of games where the opposing quarterback eclipses 110 passer rating. The only way to win is to not let the opposing quarterback get there in the first place...

And the way to prevent quarterbacks from getting high passer ratings is defense, not offense.

That also happens to nicely explain your correlation... When your defense isn't laying an egg, a good performance by your quarterback wins the game.

I wish you'd admit that your quarterback duel concept is bogus. Despite trying to criticize the Dolphins and Tanehill with some nice cherry picking, the numbers show they've managed to win MORE of these games than average despite facing one of the highest average passer ratings. If course, I'm not saying that means anything... There's just not nearly enough data.

And those (bolded) things are true, and in fact my conclusion in the OP is that the Dolphins, with Tannehill, need to amass a tremendous pass defense.  But it's also true that a good performance by your QB makes you far more competitive against the good performances of other QBs in the league.

The quarterback duel concept is simply the correlation between QB rating differential and winning.  It isn't a statement that one QB causes the play of the other.

[Dolfanalyst]

Here's a startling statistic for you:

In the 12+ years since 2004 (the year the league changed the rules to favor the passing game), the correlation between QB rating differential and win percentage across all 32 teams in the NFL is 0.92.

That's just incredible.

New England leads the way with a 15.9-point QB rating differential and a 0.775 win percentage during that period.

Green Bay is second with a 14.2-point QB rating differential and a 0.61 win percentage.

The Dolphins' QB rating differential is -5.7, and their win percentage 0.426.

Obviously if you have a QB who can post high QB ratings, a la Tom Brady and Aaron Rodgers, you stand a much better chance of being atop the league in win percentage.

[Dolfanalyst]

KC just overcame a -3 turnover margin to beat Oakland tonight.

How?

Alex Smith:  95.7 QB rating, 10.2 YPA

Derek Carr:  49.1 QB rating, 2.9 YPA

[Dolfanalyst]

Here's a startling statistic for you:

In the 12+ years since 2004 (the year the league changed the rules to favor the passing game), the correlation between QB rating differential and win percentage across all 32 teams in the NFL is 0.92.

That's just incredible.

New England leads the way with a 15.9-point QB rating differential and a 0.775 win percentage during that period.

Green Bay is second with a 14.2-point QB rating differential and a 0.61 win percentage.

The Dolphins' QB rating differential is -5.7, and their win percentage 0.426.

Obviously if you have a QB who can post high QB ratings, a la Tom Brady and Aaron Rodgers, you stand a much better chance of being atop the league in win percentage.

More on this.  X is win percentage, Y is QB rating differential, Z is turnover differential, covering the sample above.

Correlation r(xy):  0.92
Partial Correlation r(xy.z):  0.83
Correlation r(xz):  0.76
Partial Correlation r(xz.y):  0.39
Correlation r(yz):  0.71
Partial Correlation r(yz.x):  0.02

Obviously QB rating differential is far more important than turnover differential.  In fact when you enter both QB rating differential and turnover differential into a regression equation with win percentage as the dependent variable, the multiple R is 0.93, whereas with QB rating differential alone it's 0.92.

Amazing the impact of that one variable.  Even when controlling for turnover margin, it correlates with win percentage at 0.83.

Ironically enough, do a Google search on Joe Philbin and QB rating differential.  You'll find that when he was hired to coach the Dolphins, he mentioned the importance of that variable with regard to winning in one of his first press conferences with the team.

Remember also that he was in favor of ditching Ryan Tannehill in favor of Derek Carr.

[fyo]

There's some amount of self-fulfillment at work with the quarterback ratings. If your team is way behind, the optimal course of action is to take bigger chances to try and win. If it works, great, and the quarterback rating increases, if it doesn't you lose anyway and quarterback rating tanks. Optimizing for quarterback rating is not the same as optimizing your team's win probability, particularly in the situation where you are already behind.

[Dolfanalyst]

There's some amount of self-fulfillment at work with the quarterback ratings. If your team is way behind, the optimal course of action is to take bigger chances to try and win. If it works, great, and the quarterback rating increases, if it doesn't you lose anyway and quarterback rating tanks. Optimizing for quarterback rating is not the same as optimizing your team's win probability, particularly in the situation where you are already behind.

Right, but if that had a significant impact, would you expect the correlation between QB rating differential and win percentage to be a startling 0.92?

If teams are able to increase QB ratings significantly during the garbage time-type play you're mentioning above, that correlation would diminish.

Remember also that when teams need to use the passing game to mount a comeback, the opposing team knows it, and defenses can be called that mitigate the probability that garbage time passing game stats are accumulated.

In fact, that may be when INTs are much more likely to happen for the team down big on the scoreboard, thus bolstering even further the correlation above, perhaps.

[fyo]

Right, but if that had a significant impact, would you expect the correlation between QB rating differential and win percentage to be a startling 0.92?

Yes, that is in fact exactly what one would expect if the game were purely quarterback (passing) driven.

To me, it's a completely worthless statistic, on par with the old correlation between 100+ yard rushing and winning.

You know what? If you take the point differential instead of the passer rating differential, you get a 1.00 correlation. That's pretty amazing

[Dolfanalyst]

Yes, that is in fact exactly what one would expect if the game were purely quarterback (passing) driven.

The game is very largely passing driven!

To me, it's a completely worthless statistic, on par with the old correlation between 100+ yard rushing and winning.

You know what? If you take the point differential instead of the passer rating differential, you get a 1.00 correlation. That's pretty amazing

Right, but now we're back to the notion that QB rating either is or is not a valid measure of QBs' ability.  Do you believe it isn't?

Because if it is, we've discovered an extremely strong correlation between QBs' ability and winning, over a very large sample of games.

What that would say, of course, is that the ability of just one player on the field has a great deal to do with whether teams win in the league.

And of course that would also be highly consistent with how well they're paid by the people most strongly invested in winning (owners, general managers, and head coaches), and how highly sought after they are by those individuals.

[Dolfanalyst]

Right, but now we're back to the notion that QB rating either is or is not a valid measure of QBs' ability.  Do you believe it isn't?

Because if it is, we've discovered an extremely strong correlation between QBs' ability and winning, over a very large sample of games.

Do you have any interest in answering the question above and continuing on, or are we done?

[fyo]

^ That doesn't make sense Baba, unless you're saying Dolfananalyst is my alter ego... the Jekyl to my Hyde. No, I can guarantee that isn't the case .

Right, but now we're back to the notion that QB rating either is or is not a valid measure of QBs' ability.  Do you believe it isn't?

It is certainly one measure. Is it perfect? No. In fact, it is in many was quite flawed, which is what has lead to the proliferation of "advanced" metrics.

What ALL of those metrics TRY to measure, however, isn't actual quarterback performance. It's the "passing unit" performance, i.e. with the inclusion of receivers, offensive line, and even play calling. Unfortunately, the two main ones (at least in our discussions) are proprietary and include large win-biases due to asymmetric weights. This is logical in that both justify their existence by pointing to how much better they correlate with wins than "old-fashioned passer rating" and any adjustments you can make that depress the score of quarterbacks in games that they lose does exactly that. I would very much like to see some advanced metrics, open preferably, that used either win-probability, scoring-potential, or some "success measure", and applied symmetric weights (if any, the raw score might actually be more interesting).

Even this wishful thinking of mine doesn't begin to attempt to separate the quarterback from his offensive line and receivers. It doesn't account for play calling or coaching, it doesn't account for drops, for receivers slipping, bad snaps, and so on. And these effects are tremendous, let there be no question about that. It's obvious to anyone watching the game. It's obvious by looking at the success Matt Cassel had in Brady's absence (Garappolo only had 60 attempts, but he was certainly putting up some great numbers). Cassel, btw, had a better passer rating than Brady had in 4 of his previous 7 seasons.

If the NFL ever allows the public access to their advanced data from each game, we might be able to get a lot closer to a measurement that attempts to separate quarterback performance from that of offensive line and receivers while still maintaining complete objectivity. At the moment, however, even NFL teams aren't granted access until after the season (and, I believe, only for their own team or possibly games).

As for your correlation, I don't think it says a whole heck of a lot and I think I've argued why.

[Dolfanalyst]

^ That doesn't make sense Baba, unless you're saying Dolfananalyst is my alter ego... the Jekyl to my Hyde. No, I can guarantee that isn't the case .

It is certainly one measure. Is it perfect? No. In fact, it is in many was quite flawed, which is what has lead to the proliferation of "advanced" metrics.

What ALL of those metrics TRY to measure, however, isn't actual quarterback performance. It's the "passing unit" performance, i.e. with the inclusion of receivers, offensive line, and even play calling. Unfortunately, the two main ones (at least in our discussions) are proprietary and include large win-biases due to asymmetric weights. This is logical in that both justify their existence by pointing to how much better they correlate with wins than "old-fashioned passer rating" and any adjustments you can make that depress the score of quarterbacks in games that they lose does exactly that. I would very much like to see some advanced metrics, open preferably, that used either win-probability, scoring-potential, or some "success measure", and applied symmetric weights (if any, the raw score might actually be more interesting).

Even this wishful thinking of mine doesn't begin to attempt to separate the quarterback from his offensive line and receivers. It doesn't account for play calling or coaching, it doesn't account for drops, for receivers slipping, bad snaps, and so on. And these effects are tremendous, let there be no question about that. It's obvious to anyone watching the game. It's obvious by looking at the success Matt Cassel had in Brady's absence (Garappolo only had 60 attempts, but he was certainly putting up some great numbers). Cassel, btw, had a better passer rating than Brady had in 4 of his previous 7 seasons.

If the NFL ever allows the public access to their advanced data from each game, we might be able to get a lot closer to a measurement that attempts to separate quarterback performance from that of offensive line and receivers while still maintaining complete objectivity. At the moment, however, even NFL teams aren't granted access until after the season (and, I believe, only for their own team or possibly games).

As for your correlation, I don't think it says a whole heck of a lot and I think I've argued why.

Certainly it's flawed, but I don't see an argument there for why we shouldn't consider traditional QB rating a valid measure of QBs' ability.

Let's say there is a 10% error rate in the ability of QB rating to represent QBs' ability (hypothetically).  Then given the already very strong correlation between QB rating differential and winning (0.92), we're still talking about something that indicates a very strong relationship between winning in the NFL and the individual ability of quarterbacks, even minus the 10% error.

[fyo]

That's not how error rates work. This discussion is circling, I'm sure we'll pick it up again some time later.

[Dolfanalyst]

That's not how error rates work. This discussion is circling, I'm sure we'll pick it up again some time later.

I'm not sure I see it circling.  The issue is whether QB ratings are a valid measurement of QBs' ability, and I don't see you taking a position on that.