Ryan Tannehill

I can tell you that right now I play basketball with the kids I work with. I won't be hitting any shots. Then I'll hit a bunch in a row. I'm telling you, it's a different feeling. When you're on, you KNOW it's going in. You can feel it. If you haven't played sports, then you can't understand what I'm saying.

Accepting some metrics, like with moneyball, is not what I'm talking about. Convincing managers to play averages and hire athletes based on specific metrics was extremely controversial...and frankly, I don't think it will work in football as well as it has in baseball. The season is too short.

"resnor wrote:

Another example...let's say a guy is a 65% comp rate for the season. Then one game he throws for a 90% comp rate. He then follows up with 4 more games of 85, 87, 75, 92. That's 5 games well above his average. Now, it may not be abnormal, in that he might follows that up with 5 games of 65, 55, 58, 63, 61, which could lower him back to his average (fake numbers, don't think they actually average out to 65, but you get the point I'm making). Would still make that 5 games streak significant. Of course you'd need to look at what was going on, try to figure out why he was better those 5 games.

IF you can identify some external condition that independently separates those first 5 games from the next 5 (independent of the ratings themselves!), then yes you can analyze different parts of the season separately. What you don't want to do is to invent conditions based on the data itself to support a desired hypothesis. That's called cherry picking.

As far as I'm concerned, the "conditions" here are performance by season and by team: easiest ones to justify looking at for every QB.

"cbrad wrote:

That's exactly what was being counted though.

The initial studies looked at the conditional probability of making or missing a basket after X consecutive successes or X consecutive misses. Actually, to be technical, they sometimes conditioned on "X or more" consecutive successes or misses, and later studies pointed out that there are some subtle changes in the probability that those Nobel Prize winners weren't taking into account lol. Either way, that was precisely the condition researchers were looking at to show that the probability of making a basket or missing it after X consecutive successes or misses was essentially identical.. with some random variation (binomial probabilities).

I see what you are saying, I just feel like it’s a case by case, person by person situation.

Everyone gets butterflies in their stomach, not everyone handles it the same.

One of those things that are always fun to discuss. Math/emotions/positive energy/negative energy etc etc etc

"cbrad wrote:

IF you can identify some external condition that independently separates those first 5 games from the next 5 (independent of the ratings themselves!), then yes you can analyze different parts of the season separately. What you don't want to do is to invent conditions based on the data itself to support a desired hypothesis. That's called cherry picking.

As far as I'm concerned, the "conditions" here are performance by season and by team: easiest ones to justify looking at for every QB.

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

"cbrad wrote:

You're not getting it. Take the ratings Tannehill has had so far in 6 games with the Titans: 78.1, 120.1, 109.8, 82.3, 133.9, 155.8.

Now.. that's the actual order of the ratings. To you that looks like the 5 game "streak" started on game #2. Now suppose you randomly reorder those SAME ratings. For example: 109.8, 82.3, 155.8, 78.1, 120.1, 133.9. What does it look like to you now? That it was a 6-game "streak"? I don't know.. doesn't matter. To the statistical test the ordering is irrelevant and there is NO streak. There is no "5 game streak", there is no "6 game streak", there is simply NO streak.

I used the word "streak" because others have used that and it usually helps with communication, but from the point of view of the statistical test there is NO streak. There are just sets of ratings (one for each season). So you can't start with the assumptions of the statistical test (the independence of observations assumption) and then say "suppose the streak started on game X". That's a meaningless assertion.

I think we are looking at 2 different points here now. I'm assuming you are looking at a accumulative point structure? while I'm looking at the games themselves as a stand alone data point.

Here is what I'm trying to say. If you go by the most basic model of tier for a QB game you end up with Good, average, bad.

If you have 5 starts you have a TON of possible outcomes: good, bad, average, average, average or good, good, good, good, good, and so on. Order/streak was never actually part of my argument, it was just a tool to show variances in a model. I'm good not talking about streaks going forward.

What I am saying and still standing by, is that Tannehill even using this basic equation that honestly robs him of his great games, there is no way that you can say that good, good, avg, good, good equals 25% probability. Especially if you believe he is an average QB. The most likely outcomes for an average QB would be the dead center of the probabilities, while this one is clearly near the top end of the likely outcomes.

I know you are great with stats man, but there is just no way that number is right for the above scenario.

Wait..... now the argument is that if a guys first game with a new team coming off the bench is not good, then he light's up the next 5 games, there's no statistical importance to that.....and that argument is somehow a positive for purely stat based view?

"The Guy wrote:

You could say the same thing however about Andy Dalton's 16 games in 2015, where he posted a passer rating about 18 points above his career number. What was the likelihood Andy Dalton would have a passer rating of 106.2 in 2015, when his highest season passer rating before that had been 88.8, and he'd had other seasons of only 80.4, 87.4, and 83.5?

Not the same thing. You are picking from the entirety of the NFL and all of history, while clearly my entire argument was only applicable IF Tannehill did it. You essentially just took the "there is Life somewhere in space" and tried to compare it apples to apples with my "there is life on this planet".

Even if we wanted to look at it as if it was comparable do you believe that Dalton's year was strictly a random series of events, and had nothing to do with any other influences in the NFL? My guess is if you dig deep enough you would find a couple things that were different that year. If not, then you found the guy that actually won the lottery that I was talking about. That still doesn't make a strong case that this is what happened to Tannehill.

"Phins_to_Win wrote:

I think we are looking at 2 different points here now. I'm assuming you are looking at a accumulative point structure? while I'm looking at the games themselves as a stand alone data point.

Here is what I'm trying to say. If you go by the most basic model of tier for a QB game you end up with Good, average, bad.

If you have 5 starts you have a TON of possible outcomes: good, bad, average, average, average or good, good, good, good, good, and so on. Order/streak was never actually part of my argument, it was just a tool to show variances in a model. I'm good not talking about streaks going forward.

What I am saying and still standing by, is that Tannehill even using this basic equation that honestly robs him of his great games, there is no way that you can say that good, good, avg, good, good equals 25% probability. Especially if you believe he is an average QB. The most likely outcomes for an average QB would be the dead center of the probabilities, while this one is clearly near the top end of the likely outcomes.

I know you are great with stats man, but there is just no way that number is right for the above scenario.

Well.. the first thing to note is that Tannehill's ratings aren't near the top end of likely outcomes when looking at HIS game-by-game ratings. When adjusted to 2019, Tannehill's average rating is 91.62 (league average is 91) while the standard deviation in game-by-game ratings is 26.76. And that standard deviation isn't abnormal either. For Tom Brady it's 27.94 and for Russell Wilson it's 28.01.

Tannehill's average rating in 2019 is currently 111.4 which is 0.739 standard deviations above his overall mean (z-score = 0.739). And a z-score of 0.739 corresponds to top 23rd percentile for HIS game-by-game ratings (not distribution of year-end passer ratings). That's a pretty rough way of getting an intuition for how "likely" performing at that level is (for smaller sample sizes). So intuitively 25.57% is about right.

Maybe second thing to note is that you have to consider all possible permutations of those "good", "average" and "bad" to calculate the probability. Let's just simplify this and test your intuition. Let's suppose we flip a fair coin 6 times. What do you think is the probability you get 4 heads and 2 tails? Here's the thing: you can't just imagine something like HHTHTH. You have to imagine all possible permutations of 4 heads and 2 tails, and people tend to not be able to do that correctly.

Answers (X heads out of 6 flips for a fair coin):
0 heads = 1.56%
1 head = 9.38%
2 heads = 23.44%
3 heads = 31.25%
4 heads = 23.44%
5 heads = 9.38%
6 heads = 1.56%

Is that similar to what you expected? In my experience people tend to think 5 heads is WAY lower than 9.38%, which would jibe with your intuition. Of course in your case we'd have to look at 3 possibilities, but there's no need to look at such artificial examples when we know the mean and standard deviation of Tannehill's ratings. The probabilities make sense.

"cbrad wrote:

Well.. the first thing to note is that Tannehill's ratings aren't near the top end of likely outcomes when looking at HIS game-by-game ratings. When adjusted to 2019, Tannehill's average rating is 91.62 (league average is 91) while the standard deviation in game-by-game ratings is 26.76. And that standard deviation isn't abnormal either. For Tom Brady it's 27.94 and for Russell Wilson it's 28.01.

Tannehill's average rating in 2019 is currently 111.4 which is 0.739 standard deviations above the mean (z-score = 0.739). And a z-score of 0.739 corresponds to top 23rd percentile for HIS game-by-game ratings (not distribution of year-end passer ratings). That's a pretty rough way of getting an intuition for how "likely" performing at that level is. So intuitively 25.57% is about right.

Maybe second thing to note is that you have to consider all possible permutations of those "good", "average" and "bad" to calculate the probability. Let's just simplify this and test your intuition. Let's suppose we flip a fair coin 6 times. What do you think is the probability you get 4 heads and 2 tails? Here's the thing: you can't just imagine something like HHTHTH. You have to image all possible permutations of 4 heads and 2 tails, and people tend to not be able to do that correctly.

Answers (X heads out of 6 flips for a fair coin):
0 heads = 1.56%
1 head = 9.38%
2 heads = 23.44%
3 heads = 31.25%
4 heads = 23.44%
5 heads = 9.38%
6 heads = 1.56%

Is that similar to what you expected? In my experience people tend to think 5 heads is WAY lower than 9.38%, which would jibe with your intuition. Of course in your case we'd have to look at 3 possibilities, but there's no need to look at such artificial examples when we know the mean and standard deviation of Tannehill's ratings. The probabilities make sense.

Actually I like to play craps, so yeah that's not far off from my expectation. 3 heads being the strongest and you get weaker as you make your way out from it. ITs the same with the number 7 in craps.

"Fin D wrote:

Wait..... now the argument is that if a guys first game with a new team coming off the bench is not good, then he light's up the next 5 games, there's no statistical importance to that.....and that argument is somehow a positive for purely stat based view?

To be technical, he didn't have a 5 game streak. He had one below average game, then two very good ones, then one below average game, then two very good ones. So there's no "5 game streak" here anyway, which in many ways supports the assumption of the statistical test.
https://www.pro-football-reference.com/players/T/TannRy00.htm

This may be the biggest thread ever

"cbrad wrote:

Well.. the first thing to note is that Tannehill's ratings aren't near the top end of likely outcomes when looking at HIS game-by-game ratings. When adjusted to 2019, Tannehill's average rating is 91.62 (league average is 91) while the standard deviation in game-by-game ratings is 26.76. And that standard deviation isn't abnormal either. For Tom Brady it's 27.94 and for Russell Wilson it's 28.01.

Tannehill's average rating in 2019 is currently 111.4 which is 0.739 standard deviations above his overall mean (z-score = 0.739). And a z-score of 0.739 corresponds to top 23rd percentile for HIS game-by-game ratings (not distribution of year-end passer ratings). That's a pretty rough way of getting an intuition for how "likely" performing at that level is. So intuitively 25.57% is about right.

Maybe second thing to note is that you have to consider all possible permutations of those "good", "average" and "bad" to calculate the probability. Let's just simplify this and test your intuition. Let's suppose we flip a fair coin 6 times. What do you think is the probability you get 4 heads and 2 tails? Here's the thing: you can't just imagine something like HHTHTH. You have to image all possible permutations of 4 heads and 2 tails, and people tend to not be able to do that correctly.

Answers (X heads out of 6 flips for a fair coin):
0 heads = 1.56%
1 head = 9.38%
2 heads = 23.44%
3 heads = 31.25%
4 heads = 23.44%
5 heads = 9.38%
6 heads = 1.56%

Is that similar to what you expected? In my experience people tend to think 5 heads is WAY lower than 9.38%, which would jibe with your intuition. Of course in your case we'd have to look at 3 possibilities, but there's no need to look at such artificial examples when we know the mean and standard deviation of Tannehill's ratings. The probabilities make sense.

Why would you use standard deviation as the mark of good or bad? That doesn't make any sense. That value has a purpose but determining what a good game is not it. 120.1, 109.8, 133.9, and 155.8 will be considered a GOOD game by any reasonable metric (3 of them are amazing).

So my original statement of 4 good games and 1 average game stands, so does the formula I provided.

average, average, average, average,average would be the dead middle of the probability triangle.

followed by average,average,average, average good and average, average, average, average bad
followed by average, average,average, good, good and average, average, average, bad, bad and average, average, average, good, bad

As you can see we still have a long way to go to get to good, good, good, good, average, and just like the graph above that you posted, the percentages start to plummet once you get away from the center. The only way to reduce that effect would be to place Tannehill at a higher starting level player, which changes the whole argument.

"Phins_to_Win wrote:

Why would you use standard deviation as the mark of good or bad?

I'm not using standard deviation as a mark of "good" or "bad". There is no "good" or "bad" per se here because this is a continuous scale. The mean and standard deviation specify (approximately) the distribution you're randomly picking individual game ratings from. So if the average of a small set of ratings you randomly select is top 23rd percentile, then that's a quick and dirty way of getting an intuition for what a true statistical test will say about the probability that small set of ratings comes from a QB with that distribution.

You have to get off this "good", "average", "bad" categorization because there's none in the ratings themselves. That kind of categorization is what I was talking about with ordinal ratings previously. We don't have ordinal ratings, we have a continuous scale. So you want to know the type of distribution those ratings come from and that's (approximately) specified by mean and standard deviation.

Anyway, the probability not only makes total sense it's actually correct.

"BrazForPhins wrote:

This may be the biggest thread ever

You haven't seen the ones in Club. I rarely post there anymore (and also rarely go there nowadays) but it's just a few gigantic threads lol. I stopped posting regularly after some unbelievably long threads about Kyler Murray being 1 inch too short or something like that. Went on forever!! Turned me off quite a bit, though I'll probably post there again someday.

"cbrad wrote:

I'm not using standard deviation as a mark of "good" or "bad". There is no "good" or "bad" per se here because this is a continuous scale. The mean and standard deviation specify (approximately) the distribution you're randomly picking individual game ratings from. So if the average of a small set of ratings you randomly select is top 23rd percentile, then that's a quick and dirty way of getting an intuition for what a true statistical test will say about the probability a small set of ratings come from a QB with that distribution.

You have to get off this "good", "average", "bad" categorization because there's none in the ratings themselves. That kind of categorization is what I was talking about with ordinal ratings previously. We don't have ordinal ratings, we have a continuous scale. So you want to know the type of distribution those ratings come from and that's (approximately) specified by mean and standard deviation.

Anyway, the probability not only makes total sense it's actually correct.

Okay so now there is no such thing as good, average, and bad. What your describing doesn't make sense, WE CLEARLY use good bad and average to describe QBs in the league, I have seen you use the term Average several times, but now all of a sudden, there is no such thing as a good or bad outing for the QB?

You are trying to eliminate good/bad because it ruins your 25% stance. If you admit to something that is clearly evident to ALL SPORTS fans your argument is ruined. There is such a thing as a good/bad outing by the QB, and there is literally no discussion on this board that an argument has been made that it doesn't. You can argue that the QB rating doesn't always capture the good/bad game, but as you like to point out it will average out.

So if I said that Ryan Tannehill was going to have 5 games that fall into the 80-100 range.
Then I say he is going to have 4 games that fall into that 80-100 range and 1 that falls into the above 100 range.
Then I say he is going to have 4 games that falls into the 80-100 range and 1 that falls into the sub 80 range.
I then say he is going to have 3 games that fall into the 80-100 range, and 2 games that fall into the greater then 100 range.
I then say that he is going to have 3 games that fall into the 80-100 range and 2 games fall into the sub 80 range.

I just gave you 5 scenarios that are more likely to happen (based on average) than what Ryan has actually done, but all could have easily happened in the time structure given. You are going to honestly tell me that given the above examples that Ryan's output falls at 25%?

The only way you can possibly say that is if you are using such a large generic umbrella of performance that it literally renders itself useless in any discussion.

"Phins_to_Win wrote:

Okay so now there is no such thing as good, average, and bad. What your describing doesn't make sense, WE CLEARLY use good bad and average to describe QBs in the league, I have seen you use the term Average several times, but now all of a sudden, there is no such thing as a good or bad outing for the QB?

You are trying to eliminate good/bad because it ruins your 25% stance. If you admit to something that is clearly evident to ALL SPORTS fans your argument is ruined. There is such a thing as a good/bad outing by the QB, and there is literally no discussion on this board that an argument has been made that it doesn't. You can argue that the QB rating doesn't always capture the good/bad game, but as you like to point out it will average out.

So if I said that Ryan Tannehill was going to have 5 games that fall into the 80-100 range.
Then I say he is going to have 4 games that fall into that 80-100 range and 1 that falls into the above 100 range.
Then I say he is going to have 4 games that falls into the 80-100 range and 1 that falls into the sub 80 range.
I then say he is going to have 3 games that fall into the 80-100 range, and 2 games that fall into the greater then 100 range.
I then say that he is going to have 3 games that fall into the 80-100 range and 2 games fall into the sub 80 range.

I just gave you 5 scenarios that are more likely to happen (based on average) than what Ryan has actually done, but all could have easily happened in the time structure given. You are going to honestly tell me that given the above examples that Ryan's output falls at 25%?

The only way you can possibly say that is if you are using such a large generic umbrella of performance that it literally renders itself useless in any discussion.

Unbelievable. Sorry dude you just don't understand the concept of randomly choosing from a continuous distribution. I mean.. I don't mind if people don't understand basic math, but how you don't even attempt to try and learn some elementary statistics based on what I wrote – just google this stuff!! – before you accuse me of doing something shady I won't understand. I'd NEVER do that myself.

But so be it. Believe what you want. Anyway.. the probabilities I listed are correct as anyone can verify themselves (data are all publicly available and I've explained precisely what I did). You go ahead and believe ANOVA is shady and wrong.

"cbrad wrote:

To be technical, he didn't have a 5 game streak. He had one below average game, then two very good ones, then one below average game, then two very good ones. So there's no "5 game streak" here anyway, which in many ways supports the assumption of the statistical test.
https://www.pro-football-reference.com/players/T/TannRy00.htm

Forget Thill. The premise in and of itself makes no sense, which is why I didn't;t mention Thill.

"The Guy wrote:

Sure, and that's true for all QBs, and then the questions become: 1) how likely are the surroundings to be optimized in that way for any QB, and 2) to what degree will that optimization enhance a QB's performance?

1) We would need to look at the specific strengths/weaknesses of the QB in question.

2) I think that calls back to #1

"cbrad wrote:

Unbelievable. Sorry dude you just don't understand the concept of randomly choosing from a continuous distribution. I mean.. I don't mind if people don't understand basic math, but how you don't even attempt to try and learn some elementary statistics based on what I wrote – just google this stuff!! – before you accuse me of doing something shady I won't understand. I'd NEVER do that myself.

But so be it. Believe what you want. Anyway.. the probabilities I listed are correct as anyone can verify themselves (data are all publicly available and I've explained precisely what I did). You go ahead and believe ANOVA is shady and wrong.

The problem with stats in general with sports is they are just a way to categorize results. They are not all that helpful in finding the reasons for the results. The best they can do is point us to what is and isn't likely reasons. But unlikely things happen quite frequently, especially in a situation like pro football and the virtually infinitesimal variables there are. Randomly choosing from a continuous distribution, in these kinds of cases, literally removes variables. Weather, newness to a team/system/staff, postseason chances, etc. all factor in to a given performance. By just randomly selecting numbers out of that, removes the context with which those numbers were created. I feel like you believe that's ok, because you have a long history of basing much of your approach on assuming all variables outside of the QB, come out in the wash. Your approach only works if all WRs, TEs, Olines, coaching staffs, opposing defenses are virtually identical. Even with this randomly choosing from a continuous distribution, you are literally saying, the ONLY variable that matters, is the only I'm accounting for, and that is individuals QB performance regardless of literally anything else, because every QB faces the exact same variables.

"cbrad wrote:

Unbelievable. Sorry dude you just don't understand the concept of randomly choosing from a continuous distribution. I mean.. I don't mind if people don't understand basic math, but how you don't even attempt to try and learn some elementary statistics based on what I wrote – just google this stuff!! – before you accuse me of doing something shady I won't understand. I'd NEVER do that myself.

But so be it. Believe what you want. Anyway.. the probabilities I listed are correct as anyone can verify themselves (data are all publicly available and I've explained precisely what I did). You go ahead and believe ANOVA is shady and wrong.

So what is the percentage that in a 5 game stretch Tannehill has 4 games above league average and 1 below. I think we can agree that this would be worlds easier to do then what he actually did.

"Fin D wrote:

The problem with stats in general with sports is they are just a way to categorize results. They are not all that helpful in finding the reasons for the results. The best they can do is point us to what is and isn't likely reasons.

True, and that's quite an accomplishment, given that what we're talking about is at its root only math.

But unlikely things happen quite frequently, especially in a situation like pro football and the virtually infinitesimal variables there are.

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.

"Fin D wrote:

The problem with stats in general with sports is they are just a way to categorize results. They are not all that helpful in finding the reasons for the results. The best they can do is point us to what is and isn't likely reasons.

Yes this is correct.

"Fin D wrote:

But unlikely things happen quite frequently, especially in a situation like pro football and the virtually infinitesimal variables there are.

No they happen infrequently.

"Fin D wrote:

Randomly choosing from a continuous distribution, in these kinds of cases, literally removes variables.

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.

"Phins_to_Win wrote:

Not the same thing. You are picking from the entirety of the NFL and all of history, while clearly my entire argument was only applicable IF Tannehill did it. You essentially just took the "there is Life somewhere in space" and tried to compare it apples to apples with my "there is life on this planet".

Even if we wanted to look at it as if it was comparable do you believe that Dalton's year was strictly a random series of events, and had nothing to do with any other influences in the NFL? My guess is if you dig deep enough you would find a couple things that were different that year. If not, then you found the guy that actually won the lottery that I was talking about. That still doesn't make a strong case that this is what happened to Tannehill.

Dalton's performance in 2015 could be entirely the result of non-random events (enhancements in his surrounding cast, better coaching, etc.) that are nonetheless unlikely to be replicated. His performance throughout the rest of his career is a testament to that.

Again, the point isn't that surroundings don't contribute to quarterbacks' performances. The point is whether the surroundings necessary to make the quarterback perform at the necessary level 1) are likely to be assembled, and 2) can be sustained.

I might be able to make a really good high school quarterback perform well enough in the NFL if I can assemble the best 10 other offensive players in the history of the league. But then of course the issue becomes that it's impossible to assemble those other players.

"Phins_to_Win wrote:

So what is the percentage that in a 5 game stretch Tannehill has 4 games above league average and 1 below. I think we can agree that this would be worlds easier to do then what he actually did.

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 (i.e., nowhere near the top end of likely outcomes) you'll get a pretty high probability.

"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.

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.