Ryan Tannehill

"cbrad wrote:

Since you asked.. lol

Regarding the 1st graph:

1) What kind of function are you fitting to that first plot? You should be fitting a logistic regression model, not some kind of polynomial-type function, because logistic regression actually has a meaning: it's the (log of the) "odds of winning". In other words the mathematics has a meaning for dichotomous data (0's or 1's).

2) Also, if you have lots more 0's than 1's, or vice versa, then that alone will bias the fit because the fitting procedure is trying to minimize cumulative error, which is dominated by whichever is more numerous (in this case 0's). So it's better to do this for regular season wins (1's) and losses (0's), not SB winners vs. everyone else.

3) The result is obvious. If you have 0 passer rating differential in the regular season, then that probably means you didn't even make the playoffs (most likely you're an 8-8 team)!! I think it's better to plot win% against PRD. That's more informative.

Regarding the 2nd graph:

1) Technicality: you say you're plotting a correlation, but a correlation equals the slope of the best-fitting line ONLY when both axes are equal, which they're not here. So what is being plotted? The best-fitting line, or did you just take the correlation and assume it's the slope of the best-fitting line? If it's the best-fitting line, then you should tell us the slope.

2) What does a hypothesis test on that correlation say. What's the p-value? For only 14 data points, even with a correlation of 0.52 I'd bet the p-value is greater than 0.05 and you can't reject random variation around a true correlation of zero being responsible for the result.

3) Point #2 brings up the key point: we need larger sample size because correlations can vary tremendously over seasons. I know it requires some work, but since you seem to be interested in this, I'd suggest looking at the regular season because win% is more fine grained than looking at just the SB winner. Should be minimum 5 seasons. Maybe you can plot that data for 2014-2018.

4) You say you're interested in whether Tannehill will replicate his 2020 passer rating. That's offensive passer rating, and if your theory about salary cap% is correct, then the result should also be seen for offensive passer rating. So I'd extend #3 to offensive passer rating.

Anyway, good job getting the data, but I'm not yet convinced salary cap% for QB's is that important for a winning team. It sounds good in theory, but: 1) the resources saved through a rookie QB contract aren't THAT huge and I doubt most GM's would increase win% by a huge amount, and 2) too many SB's are won by QB's that weren't on their rookie contract. But yes, it's good you're gathering data here.

Appreciate your feedback!

The first graph was a logistic regression model. Here is the plot for season win percentage and passer rating differential for the 480 team seasons since 2004 (r = 0.81; p < .001):

The p-value for the correlation between the percentage of the salary cap absorbed by the starting QB on Super Bowl-winning teams since 2004 and those teams' defensive passer rating surrendered is 0.055 (again r = 0.52). And there are only 14 data points because one of the seasons since 2004 was uncapped.

If I'm reading it correctly, the function is y = 1.16x + 72.83

The correlation between percentage of the cap absorbed by the starting QB and offensive passer rating among Super Bowl-winners since 2004 is 0.28.

"The Guy wrote:

The first graph was a logistic regression model.

No, a logistic regression is a single sigmoidal curve asymptoting at 0 on the left and 1 on the right. Look at the graphs here:
https://en.wikipedia.org/wiki/Logistic_regression

Your graph has a kink in it so it can't be logistic regression. Either way, you want similar numbers of 0's and 1's so regular season is best.

"The Guy wrote:

The p-value for the correlation between the percentage of the salary cap absorbed by the starting QB on Super Bowl-winning teams since 2004 and those teams' defensive passer rating surrendered is 0.055 (again r = 0.52). And there are only 14 data points because one of the seasons since 2004 was uncapped.

Right.. not statistically significant. So the most important thing here is to get a large enough sample size using regular season and win%.

"The Guy wrote:

The correlation between percentage of the cap absorbed by the starting QB and offensive passer rating among Super Bowl-winners since 2004 is 0.28.

Which I'm sure is also not statistically significant, but that points in the opposite direction of your theory. That would be suggesting that the more you pay QB's the better the passer rating, which in some sense makes sense because so many SB winning QB's weren't on rookie contracts and the ones that get paid more tend to be the better ones.

@9201

Mahomes hasn't faced a defense lately that's anywhere near the level of SF. And they will have no answer for the Niner run game, I don't think thats really debatable.

"The Guy wrote:

I invite @7089 to critique the following if he has the time and is willing.

Here is the plot for Super Bowl wins as a function of adjusted (to 2019) regular season passer rating differential in the NFL since 2004 ("PRD" on the X-axis indicates passer rating differential):

The noteworthy thing there in my opinion is that teams have almost no chance of winning a Super Bowl if their regular season offensive passer ratings and defensive passer ratings surrendered are equivalent or nearly equivalent.

For example, let's say Patrick Mahomes pumps out a regular season passer rating of 115 and leads the league in that regard. Well if his pass defense happens to be exceptionally poor and it surrenders a passer rating around 115 (which is highly unlikely but used here just to make the point), the Chiefs' have almost no chance of winning a Super Bowl. Likewise, if Andy Dalton pumps out a 2019 league average passer rating of 90.4, the Bengals stand almost no chance of winning a Super Bowl if their defense surrenders a passer rating around 90.4.

In other words, passer rating differential (offense minus defense) is what wins.

Now, let's add the context of Super Bowl-winning quarterbacks' salary cap hits since 2004.

Since 2004 the correlation between the percentage of the salary cap absorbed by the regular season starting quarterback on Super Bowl-winning teams and those teams' defensive passer ratings surrendered is 0.52. Here is that plot:

So what that means is that the more of his team's salary cap the regular season starting QB on a Super Bowl-winning team absorbs, the higher his team's passer rating surrendered will likely be.

This is exactly what I'm getting at with regard to Tannehill, his need for a strong pass defense to help him keep pace with the league's best QBs in the effort to win a Super Bowl, and the likelihood that his future salary cap hit will interfere with his team's ability to surround him with what he needs to win (a very strong pass defense).

Tannehill's passer rating in 2019 was 117.5, and Tennessee's passer rating surrendered was 90.4, which was precisely the league average. Notice Tennessee didn't win the Super Bowl despite having a very good passer rating differential of 27.1.

Now, the important questions are the following in my opinion: 1) what is the likelihood that Tannehill in 2020 will come close to replicating his 2019 passer rating? and 2) what is the jump in his salary cap hit likely to do to Tennessee's ability to add defensive talent and improve the passer rating it surrendered in 2019?

I think this is an economic question and not a statistical question at this point. If you are paying Tannehill for the value he brings to the team and not more, then you can build a solid team around him.

I remember making the same argument you are making now about Suh. I thought it was insane to pay what we did for a DT, no matter how good he was.

After those graphs and charts. it is all so much more clear now. I was lost until I saw those. Thanks so much!

https://boston.cbslocal.com/2020/01/22/espn-story-deflategate-scandal-nfl-patriots-national-narrative/

It’s again worthwhile pointing out how useless and non-credible “ESPN” is as a sports news source.

"cbrad wrote:

Right.. not statistically significant. So the most important thing here is to get a large enough sample size using regular season and win%.

Man, you are a scientific stickler! p = 0.055 "not significant." :smile:

Which I'm sure is also not statistically significant, but that points in the opposite direction of your theory. That would be suggesting that the more you pay QB's the better the passer rating, which in some sense makes sense because so many SB winning QB's weren't on rookie contracts and the ones that get paid more tend to be the better ones.

Yeah when you extend the correlation between salary cap percentage and defensive passer rating beyond the Super Bowl winners, it collapses.

"Phins_to_Win wrote:

I think this is an economic question and not a statistical question at this point. If you are paying Tannehill for the value he brings to the team and not more, then you can build a solid team around him.

I remember making the same argument you are making now about Suh. I thought it was insane to pay what we did for a DT, no matter how good he was.

That is indeed the theory, and I felt the same way about Suh. In fact in 2017 Tannehill accounted for 12.2% of the team's cap, and Suh accounted for 11.4% of it. Two players with nearly a quarter of the team's cap. And of course the team was 6-10 that year, with Jay Cutler.

"Cashvillesent wrote:

@9201

Mahomes hasn't faced a defense lately that's anywhere near the level of SF. And they will have no answer for the Niner run game, I don't think thats really debatable.

Be careful if you're betting on the game, because the Chiefs just dealt successfully with Derrick Henry, and in terms of passer rating surrendered on the season the 49ers' pass defense isn't even a standard deviation better than the Titans', who Mahomes just posted a 120+ passer rating against.

"The Guy wrote:

Be careful if you're betting on the game, because the Chiefs just dealt successfully with Derrick Henry, and in terms of passer rating surrendered on the season the 49ers' pass defense isn't even a standard deviation better than the Titans', who Mahomes just posted a 120+ passer rating against.

Their pass rush is much better than what Titans had to give Mahomes.

"Cashvillesent wrote:

Their pass rush is much better than what Titans had to give Mahomes.

You're right about that. They had a decent bit higher QB pressure and hurry percentages than the Titans.

RT at pro bowl mic'd up.

"PhinFan1968 wrote:

RT at pro bowl mic'd up.

He has a good personality, I can see why Titans offense changed so much when he took over.

Mariota is probably the classiest guy in the league... but his personality was super dry and never commanded the huddle like I seen real QB do.

Why would such an average to below average QB be at the Pro Bowl?

Did the voters not see his lack of leadership? His lack of a deep ball? His lack of accuracy? His lack of touch? His lack of scoring? His lack of red zone efficiency? His lack of clutch? His inability to call an offense? Did they not know he was a former WR? That he was carried by the entire team?

LOVE this thread!!

:jt0323:

Here's an interesting and important finding in my opinion.

As mentioned above, the correlation between win percentage and adjusted passer rating differential for the 480 team seasons from 2004 to 2018 is 0.81.

As we know, sacks aren't included in the calculation of offensive or defensive passer rating.

If we add sacks differential (offensive minus defensive sacks) to the model, both adjusted passer rating differential and sacks differential are statistically significant predictors of win percentage (p < 0.001 for both), and the R for the model increases from 0.806 to 0.852. You get about an 8% increase in the ability to predict win percentage on the basis of adjusted passer rating differential when adding sacks differential to the model.

"Fin D wrote:

Why would such an average to below average QB be at the Pro Bowl?

Did the voters not see his lack of leadership? His lack of a deep ball? His lack of accuracy? His lack of touch? His lack of scoring? His lack of red zone efficiency? His lack of clutch? His inability to call an offense? Did they not know he was a former WR? That he was carried by the entire team?

Gee Tony, you’re just so...so...negative! :sidelol:

"Cashvillesent wrote:

He has a good personality, I can see why Titans offense changed so much when he took over.

Mariota is probably the classiest guy in the league... but his personality was super dry and never commanded the huddle like I seen real QB do.

Funny you say that.

People here knocked Tannehill for not being a leader or showing emotion.

"Mcduffie81 wrote:

Funny you say that.

People here knocked Tannehill for not being a leader or showing emotion.

I think that is because the coaches here didn’t allow him to lead.

Under Philbin/Sherman Tannehill was the rookie who was getting chewed out for making off-script plays, even when they were successful. The guy hey told everybody was a work in progress, the guy they brought in his college HC to hold his hand in the transition to the pros.

Under Philbin/Lazor Tannehill was the dud who wasn’t allowed to audible. The guy they wanted to replace with Derek Carr. The guy Lazor was reported as saying “would get them all fired.”, which if we heard that as fans then the players on the team all knew it.

Under Gase he was the golden child who was immune from criticism. Gase was the one who jumped in and responded to queries about Tannehill’s leadership, and didn’t allow Tannehill to set his own agenda.

None of those situations allow for Tannehill to be himself and to establish his own leadership style. One thing I’ve learned in my corporate life is that if you want leaders to lead, you have to give them the freedom to do so. When the big boss is constantly looking over the shoulder and micromanaging that cuts off all lower leadership at the knees.

"Pauly wrote:

I think that is because the coaches here didn’t allow him to lead.

Under Philbin/Sherman Tannehill was the rookie who was getting chewed out for making off-script plays, even when they were successful. The guy hey told everybody was a work in progress, the guy they brought in his college HC to hold his hand in the transition to the pros.

Under Philbin/Lazor Tannehill was the dud whi wasn’t allowed to audible. The guy they wanted to replace with Derek Carr. The guy Lazor was reported as saying “woukd get them all fired.”, which if we heard tgat as fans then the players on the team all knew it.

Under Gase he was the golden child who was immune from criticism. Gase was the one who jumped in and responded to queries about Tannehill’s leadership, and didn’t allow Tannehill to set his own agenda.

None of those situations allow for Tannehill to be himself and to establish his own leadership style. One thing I’ve learned in my corporate life is that if you want leaders to lead, you have to give them the freedom to do so. When the big boss is constantly looking over the shoulder and micromanaging that cuts off all lower leadership at the knees.

I don't mean to be argumentative, but I think you get what you got out of Tannehill this year by "decentralizing" him as the hub of the wheel of the offense, both in terms of physical play and leadership. In my opinion he needs to experience himself within his team as a complementary piece and not the driving force. Physically speaking that manifests itself in low-volume passing games, as I've mentioned extensively here, and leadership-wise I suspect that manifests itself in having other players around him who are understood as heavy-duty load-carriers and therefore leaders (like Derrick Henry).

I don't think Tannehill is either a leader or a great player who was "lying in wait" so to speak. I think he's neither, and the 2019 Titans just so happened to have just the right mix of ingredients to get out of him what was never gotten out of him before, and likely will never be gotten out of him again.

"Pauly wrote:

One thing I’ve learned in my corporate life is that if you want leaders to lead, you have to give them the freedom to do so. When the big boss is constantly looking over the shoulder and micromanaging that cuts off all lower leadership at the knees.

You're gonna have to back that up with stats sir.

"PhinFan1968 wrote:

You're gonna have to back that up with stats sir.

:sidelol:

"cbrad wrote:

No, a logistic regression is a single sigmoidal curve asymptoting at 0 on the left and 1 on the right. Look at the graphs here:
https://en.wikipedia.org/wiki/Logistic_regression

Your graph has a kink in it so it can't be logistic regression. Either way, you want similar numbers of 0's and 1's so regular season is best.

I took another look at this and used teams' making the playoffs since 2004 as the dichotomous dependent variable and passer rating differential as the predictor. This is the graph for that, similar to the one you mentioned above:

However when I replicate that process for the Super Bowl winners as the dichotomous dependent variable, I continue to get this graph:

"The Guy wrote:

I took another look at this and used teams' making the playoffs since 2004 as the dichotomous dependent variable and passer rating differential as the predictor. This is the graph for that, similar to the one you mentioned above:

However when I replicate that process for the Super Bowl winners as the dichotomous dependent variable, I continue to get this graph:

Right, just like I said in the post you quoted, for SB winners you have too few 1's (SB winners) and far too many zeros (non-SB winners) to have any confidence in the curve that's being fit. Look at the gray area (which I'm assuming is the 95% confidence region) in the SB winner curve.. it's massive! That means that any logistic function in that region could be the "true" curve => no reliability in the fit.

The playoff graph is fine because you have enough 1's (12/32 playoff teams). And notice that it "looks" right too because 12/32 = 37.5% and you'd expect somewhat less than 37.5% of teams with PRD = 0 to make the playoffs, and that's what you see.

Anyway, take home message here is that you have to report stats where the reliabilities are high enough for the argument you're making, and that's not the case with a logistic function fit to SB winners vs. all teams that didn't win the SB.

"The Guy wrote:

I took another look at this and used teams' making the playoffs since 2004 as the dichotomous dependent variable and passer rating differential as the predictor. This is the graph for that, similar to the one you mentioned above:

However when I replicate that process for the Super Bowl winners as the dichotomous dependent variable, I continue to get this graph:

Why start in 2004?
The only rationalization appears to be that it is a round number. The problem with this is that you are truncating the data, and whenever you truncate the data you run the risk of an anomaly changing the outcome.
Dates that make sense for a cutoff dat, depending on what you are looking at.
2011 - introduction of the rookie salary pool.
2002 - the realignment into 4 divisions of 4 teams.
1994 - introduction of the salary cap
1978 - introduction of new passing rules that dramatically affected passing stats.
1970 - AFL - NFL merger