Ryan Tannehill

"cbrad wrote:

That depends on which range of ratings "above average" and "average" refer to. Post #658 is the key. When "above average" is ONLY at 0.739 standard deviations above the mean you'll get a pretty high probability.

The issue for that fellow is that he isn't sufficiently appreciating the history of variation involved, which makes the z-score of the recent performance less than a standard deviation from the mean. He appears to be assuming that Tannehill's recent performance is well beyond what would be expected given the history of (what he thinks is smaller) variation in his performance.

"Phins_to_Win wrote:

No I'm talking exact average. which is 91 or 92 I think? With Tannehill being average my guess is this is practically the flip of the coin scenario that you referenced before.

OK we can define "average" as exact average which right now in 2019 is 91. So how do you want to define "above average"? The probability you seek totally hinges on that.

"The Guy wrote:

The issue for that fellow is that he isn't sufficiently appreciating the history of variation involved, which makes the z-score of the recent performance less than a standard deviation from the mean. He appears to be assuming that Tannehill's recent performance is well beyond what would be expected given the history of (what he thinks is smaller) variation in his performance.

So are you saying that an Average QB isn't equally likely to have a below avg passer rating then above?

"cbrad wrote:

OK we can define "average" as exact average which right now in 2019 is 91. So how do you want to define "above average"? The probability you seek totally hinges on that.

91.0001 and up

Here's one of the issues here with regard to quarterbacks and surrounding casts.

Take the following two quarterbacks and their season passer ratings:

QB 1
80.4
87.4
88.8
83.5
106.2
91.8
86.6
89.6
79.2

QB 2
100
101.2
95
110.1
92.6
95.4
110.9
112.1

Those are for Andy Dalton and Russell Wilson.

So what we see here is that Dalton can perform at Wilson's level for a season, perhaps if he has the necessary surroundings. The fact that he's performed at that level for only one of his nine seasons in the league begs the question of how likely those surroundings are to be assembled and sustained.

Likewise, Ryan Tannehill may be able to play at Russell Wilson's level if he has the requisite surroundings. But if we look up in two or three years and find that his 2019 season was merely what Andy Dalton's 2015 season was (a single season of elevated play that wasn't sustained), what will that tell us?

In my opinion it'll tell us that Tannehill can play at a high level, but only if he has the kinds of surroundings that 1) aren't likely to be assembled in the first place, and 2) can't be sustained.

And if so, then what good will it be to be a quarterback who is dependent on surroundings that aren't likely to be assembled or sustained? Of course you'd rather have the Russell Wilson, the guy whose performance varies at a level much higher, who isn't so dependent on his surroundings.

"Phins_to_Win wrote:

91.0001 and up

Good, so here's a nice lesson in how probability is calculated with continuous distributions. The probability of a rating being "average" the way you defined it is zero. You wanted to defined "average" as precisely a rating of 91. In ANY continuous distribution where every value has some non-negative probability of occurring, the probability of randomly picking any specific value (e.g., precisely 91) is zero.

However.. the probability of 91.0001 and up refers to a range of possibilities, and in a symmetric distribution (which for simplicity we're assuming here – if you look at a large enough sample it is slightly skewed but symmetric is a good approximation) the probability of the entire range of 91.0001 and up is 50% probability.

So.. based on how you want to define "average" and "above average", the probability of randomly choosing 4 out of 5 passer ratings that are "above average" and one precisely "average" is zero percent because you NEVER (probabilistically) can flip "average" = precisely 91.

So it never happens basically, given how you want to define "average". In general, you have to choose ranges of values to get non-negative probabilities. But why do that in the first place? The thresholds defining any range as "average" or "above average" will be your subjective ones. It's much better to just take the ratings as is and calculate probabilities based on those. AND.. in ANOVA what you do is you look at properties of the distributions of those ratings, like their variance. In other words, for continuous values you don't care about the probability of every single value because it's always zero. Instead you look at the set of values as a unit.

"resnor wrote:

And maybe you can't find anything. Because it's simply an athlete being in the zone.

Quite possible, and that would be an internal attribution, as opposed to one centered on the athlete's surroundings. Again we have no idea the effect of the Titans' surroundings on Tannehill when we aren't measuring them.

The questions then would become, how likely is the athlete to experience "the zone," and for how long can he sustain it?

Unlikely things happen all the time. Let’s say there are 50 unlikely things that can happen. Anyone of them is occurring in a game. Say there are 1000 unlikely things. Let’s say 10 happen in a game. That’s still unlikely. You guys love to play freaking semantics and avoid the actual discussion.

"resnor wrote:

Unlikely things happen all the time. Let’s say there are 50 unlikely things that can happen. Anyone of them is occurring in a game. Say there are 1000 unlikely things. Let’s say 10 happen in a game. That’s still unlikely. You guys love to play freaking semantics and avoid the actual discussion.

Read post #680. Lots of unlikely things probably happened for Andy Dalton in 2015. But they've happened for only one season because they're unlikely.

"resnor wrote:

Unlikely things happen all the time. Let’s say there are 50 unlikely things that can happen. Anyone of them is occurring in a game. Say there are 1000 unlikely things. Let’s say 10 happen in a game. That’s still unlikely. You guys love to play freaking semantics and avoid the actual discussion.

Semantics matters. I mean.. if you can't precisely define the question you can't answer it either because you can always keep saying "no that's not what was meant". Fin D was talking about random sampling from ONE continuous distribution. In that context unlikely things happen infrequently by definition. You're talking about the probability of one of X unlikely events occurring. That can be highly likely to occur of course. So.. either state things precisely or accept that people will infer what you meant based on context.

And none of us are avoiding actual discussion. Just read the pages worth of actual discussion to see that.

"The Guy wrote:

Unlikely things don't happen frequently. They happen infrequently. That's why they're unlikely.

What happens frequently is a bias people experience that makes them believe their preferred result will happen, despite its low likelihood. The quarterback of my team will be the Drew Brees/Steve Young, the guy who started his career not so great and then changed teams and did very well.

"cbrad wrote:

No they happen infrequently.

No. Unlikely things happen all the time, every second of every day. It may not be the same unlikely things (hence those things being unlikely), but things we don't expect to happen, happen all the time. Every game, something unlikely happens at some point.

The problem is people are using stats as definitive explanations. Statistically, there should be no Tom Brady. Statistically, at the time, Drew Brees career should have been over after that surgery. Stats only give an idea, not a reason. You have to look at more. That's why I keep asking you guys to explain the ways a team can counter the pass rush. If you don't understand or account for those things, all you're doing is making assumptions based on LESS data.

"cbrad wrote:

OK.. this is a misunderstanding. The passer rating is the effect, not the cause. So NO causal variables are being removed as long as you are randomly choosing from the exact same distribution as the actual distribution. Thus, the only source of uncertainty is the degree to which the set of observed ratings is representative of the actual distribution. But at this point at least no variables are being removed.

The passer rating can be affected by numerous things from surrounding talent, opposing talent, to playoff chances to weather and all of that is based on when the game is played. Removing the "when" by randomly taking those numbers from anywhere, removes the effect of all those variables. You are literally removing data points.

"Fin D wrote:

The problem is people are using stats as definitive explanations. Statistically, there should be no Tom Brady. Statistically, at the time, Drew Brees career should have been over after that surgery. Stats only give an idea, not a reason. You have to look at more. That's why I keep asking you guys to explain the ways a team can counter the pass rush. If you don't understand or account for those things, all you're doing is making assumptions based on LESS data.

When do I make these "definitive explanations"? I always make it clear we're looking at the likelihood of a hypothesis being true based on historical data. It's actually the guys that don't rely on statistics that tend to make definitive explanations.. you know like asserting that Tannehill had it worse than any other QB in the league here in Miami. Statistics actually keeps you more honest about what level of certainty you can apply to different hypotheses (because it allows you to quantify them).

"Fin D wrote:

The passer rating can be affected by numerous things from surrounding talent, opposing talent, to playoff chances to weather and all of that is based on when the game is played. Removing the "when" by randomly taking those numbers from anywhere, removes the effect of all those variables. You are literally removing data points.

You're removing information when you use statistics, yes. But there are no causal variables being removed. Removing a causal variable (e.g., weather, crowd noise, etc...) would mean you adjusted the stat for that causal variable. That's not happening here. The stat is simply the effect of ALL those causal variables. None are being removed, even if you are removing information.

"Fin D wrote:

No. Unlikely things happen all the time, every second of every day. It may not be the same unlikely things (hence those things being unlikely), but things we don't expect to happen, happen all the time. Every game, something unlikely happens at some point.

The problem is people are using stats as definitive explanations. Statistically, there should be no Tom Brady. Statistically, at the time, Drew Brees career should have been over after that surgery. Stats only give an idea, not a reason. You have to look at more. That's why I keep asking you guys to explain the ways a team can counter the pass rush. If you don't understand or account for those things, all you're doing is making assumptions based on LESS data.

Whether unlikely things happen doesn't make them frequent. Unlikely things can happen and nonetheless happen infrequently.

Statistics indicate that a Tom Brady-caliber quarterback is highly unlikely to be drafted in the sixth round. Teams would be very foolish to believe they can wait until the sixth round to draft a great quarterback, and it's statistics that determine that foolishness. Statistics do account for a Tom Brady, and they allocate the appropriate probability to such an event.

The passer rating can be affected by numerous things from surrounding talent, opposing talent, to playoff chances to weather and all of that is based on when the game is played. Removing the "when" by randomly taking those numbers from anywhere, removes the effect of all those variables. You are literally removing data points.

Again take a look at post #680. Andy Dalton apparently got a whole lot of unlikely events like the kind you mentioned to happen in 2015. Nobody is discounting the effect of those things when they occur.

However, if you're using a quarterback whose performance at the necessary level hinges on the occurrence of such unlikely events (Andy Dalton), obviously you're doing a whole lot worse than you would be if you had a quarterback whose performance at the necessary level doesn't hinge on the occurrence of such events (Russell Wilson).

Coming back to the topic at hand, this is why in my opinion it'll be necessary to wait two or three years to determine exactly what's going on with Ryan Tannehill right now. If he's essentially replicating an Andy Dalton 2015 season -- whether that's due to internal and/or external variables -- it won't mean much in terms of his overall career.

"cbrad wrote:

When do I make these "definitive explanations"? I always make it clear we're looking at the likelihood of a hypothesis being true based on historical data. It's actually the guys that don't rely on statistics that tend to make definitive explanations.. you know like asserting that Tannehill had it worse than any other QB in the league here in Miami. Statistics actually keeps you more honest about what level of certainty you can apply to different hypotheses (because it allows you to quantify them).

We tell you X is the cause, based on actual facts, and then you tell us nope that's likely not it based on statistical data. Data, which doesn't account in anyway for X. So we argue. You then go down rabbit holes and tell us definitively X is either not an issue or cannot be accounted for or is a constant across the league.

"cbrad wrote:

You're removing information when you use statistics, yes. But there are no causal variables being removed. Removing a causal variable (e.g., weather, crowd noise, etc...) would mean you adjusted the stat for that causal variable. That's not happening here. The stat is simply the effect of ALL those causal variables. None are being removed, even if you are removing information.

Again this only works if the causal variables are consistent across the league or games. They aren't.

Its entirely possible to pull three games from a QB and those three games are Game A= first game with team so everything is off, Game B= playing injured and Game C= playing the best defense in the league and the playcalling was off.

"The Guy wrote:

Whether unlikely things happen doesn't make them frequent. Unlikely things can happen and nonetheless happen infrequently.

Statistics indicate that a Tom Brady-caliber quarterback is highly unlikely to be drafted in the sixth round. Teams would be very foolish to believe they can wait until the sixth round to draft a great quarterback, and it's statistics that determine that foolishness. Statistics do account for a Tom Brady, and they allocate the appropriate probability to such an event.

The way stats are being used in here, is that while Tom Brady's success is happening, we're being told it likely isn't.

No one in here is really projecting anything, like your example that we shouldn't draft a 6th round QB expecting Tom Brady. All we are saying is that Thill had issues beyond his control that caused his time here to be less than what it could be. We are being told it likely isn't the case.

"The Guy wrote:

Again take a look at post #680. Andy Dalton apparently got a whole lot of unlikely events like the kind you mentioned to happen in 2015. Nobody is discounting the effect of those things when they occur.

However, if you're using a quarterback whose performance at the necessary level hinges on the occurrence of such unlikely events (Andy Dalton), obviously you're doing a whole lot worse than you would be if you had a quarterback whose performance at the necessary level doesn't hinge on the occurrence of such events (Russell Wilson).

Coming back to the topic at hand, this is why in my opinion it'll be necessary to wait two or three years to determine exactly what's going on with Ryan Tannehill right now. If he's essentially replicating an Andy Dalton 2015 season -- whether that's due to internal and/or external variables -- it won't mean much in terms of his overall career.

No one is hinging anything on Thill putting up good numbers dude to unlikely events. We have explained that he had obstacles most QBs don't have beyond his control. Then we are told that's likely not true due to stats that don't in anyway account for the reasons we gave.

I've asked 2 questions numerous times now and in all the years we've been arguing Thill no one actually answers them, including you....

1. What are the ways a QB/team can counter a pass rush?
2. What modern era QB had a terrible line, but was also not allowed to audible and had no commitment to the run game, but was successful?

"Fin D wrote:

We tell you X is the cause, based on actual facts, and then you tell us nope that's likely not it based on statistical data. Data, which doesn't account in anyway for X. So we argue. You then go down rabbit holes and tell us definitively X is either not an issue or cannot be accounted for or is a constant across the league.

Yeah.. replace that word "definitely" with "likely" in the last sentence and that's more or less correct. Point is, I'm not providing definitive explanations of "why" we observed what we observe. However, I do make definitive claims about the math and sources of uncertainty.

"Fin D wrote:

Again this only works if the causal variables are consistent across the league or games. They aren't.

Its entirely possible to pull three games from a QB and those three games are Game A= first game with team so everything is off, Game B= playing injured and Game C= playing the best defense in the league and the playcalling was off.

My response is very specific to your claim that the stat is NOT taking into account all the possible variables that could affect it. The stat literally is the effect of ALL possible variables that could affect it. All you're trying to say (though it's not what you actually wrote) is that we don't know how the myriad of variables affect that stat. And that's correct. But the stat does NOT remove the effect of those variables.

"Fin D wrote:

We tell you X is the cause, based on actual facts, and then you tell us nope that's likely not it based on statistical data. Data, which doesn't account in anyway for X. So we argue. You then go down rabbit holes and tell us definitively X is either not an issue or cannot be accounted for or is a constant across the league.

Again this only works if the causal variables are consistent across the league or games. They aren't.

Its entirely possible to pull three games from a QB and those three games are Game A= first game with team so everything is off, Game B= playing injured and Game C= playing the best defense in the league and the playcalling was off.

The point you're making is essentially that statistics are less powerful with small sample sizes, which nobody would argue with.

The point cbrad is making however is that Tannehill's recent performance can't yet be meaningfully distinguished statistically from his previous performance, which involves a large sample size.

So whatever the causes of Tannehill's previous performance -- including his ability level, his surroundings, his opponents, injuries, weather, etc. -- those causes can't be reliably ruled out as explanations for his current performance because his current performance can't yet be meaningfully distinguished statistically from his past performance.

"cbrad wrote:

Yeah.. replace that word "definitely" with "likely" in the last sentence and that's more or less correct. Point is, I'm not providing definitive explanations of the "why" we observed what we observe. However, I do make definitive claims about the math and sources of uncertainty.

I'm fairly certain you have told me that things like oline play is definitely a constant.

"cbrad wrote:

My response is very specific to your claim that the stat is NOT taking into account all the possible variables that could affect it. The stat literally is the effect of ALL possible variables that could affect it. All you're trying to say (though it's not what you actually wrote) is that we don't know how the myriad of variables affect that stat. And that's correct. But the stat does NOT remove the effect of those variables.

It does remove it, effectively.

It is yet another way of saying all those things are constant and therefore don't matter, which removes them from the equation.

"The Guy wrote:

The point you're making is essentially that statistics are less powerful with small sample sizes, which nobody would argue with.

The point cbrad is making however is that Tannehill's recent performance can't yet be meaningfully distinguished statistically from his previous performance, which involves a large sample size.

So whatever the causes of Tannehill's previous performance -- including his ability level, his surroundings, his opponents, injuries, weather, etc. -- those causes can't be reliably ruled out as explanations for his current performance because his current performance can't yet be meaningfully distinguished statistically from his past performance.

Except this isn't the only run he's had. In this one and all the others, there are things present, things a few off us said he needed before he got them. Then got them and had these runs.

"Fin D wrote:

I'm fairly certain you have told me that things like oline play is definitely a constant.

Nope, never. In fact I've repeatedly said OL is one of the hardest to measure aspects of football, probably second only to coaching.

"Fin D wrote:

It does remove it, effectively.

It is yet another way of saying all those things are constant and therefore don't matter, which removes them from the equation.

Nope.. it literally is the effect of all those variables. This distinction is important. A stat like passer rating fully incorporates the effect of everything that led to that stat, which includes all the variables you're referring to. However, passer rating does not allow you to infer how those variables affected it. It's the latter you're talking about, not the former. But the stat does NOT remove the effect of those variables.

"cbrad wrote:

Nope, never. In fact I've repeatedly said OL is one of the hardest to measure aspects of football, probably second only to coaching.

Ok.

"cbrad wrote:

Nope.. it literally is the effect of all those variables. This distinction is important. A stat like passer rating fully incorporates the effect of everything that led to that stat, which includes all the variables you're referring to. However, passer rating does not allow you to infer how those variables affected it. It's the latter you're talking about, not the former. But the stat does NOT remove the effect of those variables.

You are effectively removing the specific variables that went into that specific passer rating.

If you randomly compare Joe Whathistits against Bobby Whateverthehell and the game you randomly grabbed for Joe was his worst ever because he was injured playing in the terrible weather against the best defense in history and the random game you grabbed for Bobby was his best game ever in great weather during perfect health against the worst defense ever.....then what you've done by comparing them is completely ignore (or remove) opponent, health and weather as factors.

"Fin D wrote:

You are effectively removing the specific variables that went into that specific passer rating.

If you randomly compare Joe Whathistits against Bobby Whateverthehell and the game you randomly grabbed for Joe was his worst ever because he was injured playing in the terrible weather against the best defense in history and the random game you grabbed for Bobby was his best game ever in great weather during perfect health against the worst defense ever.....then what you've done by comparing them is completely ignore (or remove) opponent, health and weather as factors.

Yeah.. read the portion of my post you quoted again. Your example is showing that we cannot infer what the effects of those factors were just by looking at the passer ratings. But clearly those passer ratings incorporated those effects. The effects aren't being removed.

This is a crucial distinction because it IS possible to remove effects from passer rating. For example, when I remove the effect of the defense based on average effect of points allowed on passer rating, that's literally removing the (average) effect of the defense from passer rating.

"cbrad wrote:

Yeah.. read the portion of my post you quoted again. Your example is showing that we cannot infer what the effects of those factors were just by looking at the passer ratings. But clearly those passer ratings incorporated those effects. The effects aren't being removed.

This is a crucial distinction because it IS possible to remove effects from passer rating. For example, when I remove the effect of the defense based on average effect of points allowed on passer rating, that's literally removing the (average) effect of the defense from passer rating.

You are removing their importance. Again, look at the extreme hypothetical I used. By comparing those two passer ratings you are effectively removing the opponents, health and weather from the equation because are not making allowances for them. You are essentially saying they don;t matter so we will compare as if they aren't a factor because they aren't.

"Fin D wrote:

You are removing their importance. Again, look at the extreme hypothetical I used. By comparing those two passer ratings you are effectively removing the opponents, health and weather from the equation because are not making allowances for them. You are essentially saying they don;t matter so we will compare as if they aren't a factor because they aren't.

Yes that phrasing is acceptable. So.. what has to happen then is to measure these other variables and see what the effect of removing them actually is (e.g., removing the effect of the defense). For some factors it's negligible, for others it's not.

Again there are 1,000s of unlikely events that can happen at any given point in a football game. When one happens, it doesn't make it any less unlikely.