Where does Tannehill rank among quarterbacks today?

I was responding to DolphinGreg, so the probabilities are ONLY with respect to post-season. I already pointed out that Tannehill and Flacco are comparable stats-wise in many ways in the regular season (though they are very different QB's), so the question was what to do with post-season data.

The probabilities are thus the probability of getting any desired measure of performance (e.g. QB rating) in the playoffs, for the two QB's.

 

No, in probability theory you use all the data given to you. So for Flacco you use his data from regular and post-season (+ everything from all other QB's if you know how to use that data) to predict post-season performance, while for Tannehill you can only use regular season and what other QB's (the population stats) do in regular vs. post-season.

The specific form of probability theory in question is here is what's called Bayesian inference. You start with prior probabilities and update those into posterior probabilities with every new piece of info.

 

If they are comparable in the regular season, then you'd have to predict they'd be the similar in the post season.....unless we had hard numbers from tannehill to say differently, which we don't.

That is why, until tannehill plays in the postseason some, the post season is irrelevant in this specific conversation.

 

You are using the theories incorrectly for the discussion. We are comparing one to the other as they stand, there's no need to predict anything. We are not saying who will achieve more team accolades in the future.

 

Like I said, you use population stats there. What happens to the average QB's regular season stats in the post-season?

To ONLY use Flacco to predict Tannehill I think would never fly because you can't show Tannehill is somehow more likely to be like Flacco (regular season vs. post-season wise) than other QB's.

 

I'm pointing out we always extrapolate. We often compare expected values to actual ones, and there actually exists a formal means of making that inference. So yeah it's absolutely standard to make comparisons between two events/processes/people etc.. where they were not necessarily in the same conditions. That doesn't mean it wouldn't be better if they were in the same conditions, just means people naturally make such comparisons and there are formal methods of doing so.

 

And yet, if you came into this blind and you were only given Flacco's regular season numbers and all the numbers for every other QB, using your method (formal or otherwise) you never would have predicted Flacco's post season success.

 

That is correct.

But that's just the way things are.. Bayesian inference guarantees that you are using the information available in the optimal way. Optimal may still be pretty bad.

 

Personally, I include all the data in my analysis, but I don't give playoff performance greater value. I don't believe that players are or aren't clutch. I believe that for almost all professional athletes they tend to perform to their averages. You'll have anomalous results but those are due to small sample sizes. Whenever you have a player lucky enough to have a ton of playoff appearances their numbers normalize given the larger set. So if a player had a rep for being incredibly clutch early, but had a ton of subsequent playoff appearances, his numbers come back down to earth. And conversely a player who has a reputation as a choker during early playoff appearances but is fortunate enough to have a ton of subsequent playoff appearances will see his numbers improve. I've seen studies argue both for and against "clutch", but the ones that look at larger samples seem to unanimously conclude that players play to their averages given a large enough sample size. For that reason, I don't weigh playoff performance differently.

Additionally, I don't tend to just focus on stats in my evaluations. That's where it becomes as much art as science. That's where the subjective part factors in and why for closely grouped players a definitive answer is impossible. That's why I stated in my first reply in this thread that an argument could be made for any order amongst the players ranked 9-14. It's a matter of opinion as to how much these player's stats and success or lack their of was the result of their supporting cast or situation. I'm of the opinion that Tannehill was helped the least and hindered the most by his supporting cast and situation amongst that group ranked 9-14. I don't think he's perfect, far from it. But I like that he's been progressing each year. Most players tend to plateau after year two, Tannehill didn't. He showed significant improvement from year 2 to year 3. It may be b/c he came in so raw compared to most QBs these days. I'm not sure he'll ever be elite, but I think it's obvious that he's already good enough to win with. I see him as a top 10 QB this year. I think he could potentially become elite, but reality is that few reach that level so you're never smart to bet that any QB will get there.

When I watched Flacco during his playoff runs my impression was that much of his success was the result of his team mates rather than him playing better than he had previously. Reality is that completing a few long passes for TDs will dramatically improve your stats. But long ball success is dependent on several factors that are outside of the QB's control. Assuming the QB has a strong enough arm then the skill requirement is small. QBs aren't throwing precise long passes. They're generally throwing to a large, general area and relying on they're WR to adjust and make them look good. My impression during those playoff runs was that Flacco just threw it up often and his receivers made a ton of plays for him. And looking at Flacco's long ball stats from last season I see that he's near the bottom of the league, well below Tannehill. I don't believe that Flacco forgot how to throw it deep. I think he did exactly what he did before but his receivers made a few less plays. Now I do credit Flacco with trusting his receivers and with moving around in the pocket to buy time when necessary. Those are areas where I believe that Tannehill can improve. And I also think he may have a few receivers worthy of being trusted enough to make a play when you throw it up deep.

My assessment of Eli is similar.

 

So, you're defending a system that has to make up numbers and doesn't work, just because you want to use every piece of data available? Seems...silly to me.

 

Almost completely agree with the rest of your post so I'll just respond to this part.

The reason I value playoff performances more is because it's win or go home, which is different and more important than in the regular season (you are penalized more for mistakes and the reward is greater). In other words, it's like whether you value success on a huge bet more or less than on a small bet. It doesn't matter if "clutch" exists or not, the value should be greater the more important the situation is.

 

The point is the system (when you can use it) works better than any other over time. Its absolute performance may still be crap but you still want to use all data when possible.

I should actually point out one exception to this "use all data" business but it's a bit technical. When you have too many parameters and have too many measurements, it's often the case that the noise (the uncertainties) in the measurements trumps the added increase in prediction accuracy. This can happen in modeling fluids like weather models etc.. You have to be careful that the added mechanisms don't add more noise than accuracy.

 

Then what research have you done that proves the use of the model is justified for this discussion? What data proves post season stats helps decide who is or isn't a better QB between two players?

 

Oh, just pointing out you can compare two things even when they aren't always occurring in the same conditions. Nothing more. That's a general statement for which something like Bayesian inference is a good reference.

 

But, I wasn't arguing that you can't compare things in general.

 

You were arguing you can only compare things that occur in similar conditions. And I pointed out we naturally compare things even when they don't occur in the same/similar conditions, and there are formal methods for doing so too. That's the only point of contention as far as I can tell. Applied to Flacco vs. Tannehill, it nudges Flacco above Tannehill in terms of overall ranking based on what they've done so far, but as I said in a post much earlier, you ask me which I'd take now and they're much more even.

 

But in an individual evaluation my concern is the individual's performance not the larger outcome. Is Brady a lesser QB if Vinateri misses a couple of key FGs? With the Flacco example I rated his performance to be generally equivalent to his other performance. The difference is that other players performed better and elevated his stats. Why would I bump Flacco up for the performance of others? And if given a large enough sample size there is no difference between "win or go home" performance and regular season performance, I see no reason to rate one as more important than the other.

 

No, I was arguing that you can't compare post season stats for one player to another player without post season stats in trying to decide who is the better player.

 

I think you're right if you can truly independently measure how well an individual component performs. But there are too many moving parts here to really know what contributed to the outcome by how much, so I think you have to condition on the environment (reg. season game vs. playoff).

 

Yeah that's what I just said.. "similar conditions" = went to and played in playoffs, in this case.

 

For me, there is such a thing as un-clutch. Look, people crack under pressure. to argue there is no such thing as clutch, you'd have to argue nobody cracks under pressure. Sometimes clutch is just performing to your average, in a high pressure situation.

Look at Nate Kaeding. The 2nd most accurate kicker of all time. Yet in every single pressure situation (not just regular kicks, but kicks that matter towards the end of the game) Charger fans were shaking in their boots, because he missed them. Comically. 2004, missed a game winning FG. 2006 missed game tying. 2007 missed a couple more. In 2009 against the Jets, he missed all 3 attempts. They lost, 17-14.

Nate Kaeding by himself proves there is such a thing as clutch.

Peyton Manning: Perennial underperformer in games that really matter. One and done games. Stats won't get you there. You have to look at each individual situation.

Playoff numbers usually take a hit. You're quite literally playing the best teams, and often the best defenses. Tom Brady's career TD% goes from 5.5% down to 4.9% in the playoffs. Sure it's a hit. Manning goes from 5.9% down to 4.0% and nobody on earth is going to say Brady has had better receiving options through their career. Not even close. Sample sizes are not small with around 1000 attempts from both.

Joe Montana has a 95 rating in the playoffs, which, pre-2004, is fantastic. Career 5% TD, in the playoffs 6%.

 

That's why I think it's subjective. But there's also the part that given a large enough sample size it's all equal. So giving greater weight to a playoff environment would seem to just be placing greater emphasis on the small sample size for no good reason.

 

Do you happen to have a link to some of those studies? I'll look for some myself later on (lol.. I'm trying to code something while commenting on a message board), but I'd actually like to see whether there are any methodological problems there or what the uncertainties are.

 

I see clutch and unclutch as very rare amongst professional athletes. Anybody who trains a lot at something tends to default to that training. People who haven't trained a ton worry about what to do or if they're good enough. People who have done something a million times generally know they can do something so the pressure isn't a factor.

And sure you're playing better teams in the playoffs so numbers will go down a bit. But if your numbers are shooting up, it's an anomaly. And in the case of a guy like Flacco where a few big plays substantially increase your stats then it's a tiny sample size.

 

Flacco has thrown almost a season of passes.

Andy Dalton has yet to perform even average. Most playoff numbers take a hit, yes. Most players will sit in the middle, yes. But there are some who suck when it comes to pressure situations, and some who excel.

 

I'm sorry. It's nothing I've looked at recently. It's probably been at least a decade since I last researched it.

I did the research b/c the arguments for "clutch" were always based on a few players, basically anecdotal evidence. I always found that to be a particularly unconvincing argument. The longer term study IIRC was looking at baseball stats. I don't follow baseball much so I may be wrong on the guy they pointed out but it may have been Arod. Did he have a reputation as a choker early on? I think they had looked at 1000s of players over decades. Anyway the upshot was that as sample sizes increased the playoff averages got closer to the player's regular season averages. I've seen other studies using baseball and golf that argued for a clutch gene, but they always used smaller samples.

 

And IMO outside of just luck, much of that difference is the result of things outside of the QB's skill level. With Flacco in particular his stats are greatly inflated by long passes. And long passes rely more than most on the performance of the WR.

 

To think this thread went 4 pages before sh*t hit the fan...wonder why.

 

The **** didn't hit the fan in this thread.

Though I suppose it will when you make a response to this.

 

Well I just googled through a bunch of studies (most of it fluffy to be honest) on "clutch" performance or "choking" under pressure. I don't think anyone really knows what's real and what's not. As you pointed out, sample size is a serious problem, so you get lots of seemingly anomalous performances, but it's hard to tell to what degree that's expected just due to "randomness".

It doesn't look like playoffs provide sufficient sample size anyway. You can get sufficient samples when you say "clutch" = everything from the 7th inning onwards in regular season baseball, but playoffs alone probably don't give you sufficient samples.

So.. back to the question of whether to weight playoffs more if you are only interested in measuring individual performance. Suppose we assume there is no psychological difference on performance between "win-or-go-home" vs. "regular season". Even then, there still is one measurable difference between playoffs and regular season that should lead you to weight playoffs more: the average strength of opposition is greater. In other words, getting the same stats vs. stronger opposition should count more than vs. weaker opposition.

 

Yeah we had a good debate.

 

lol..yes..if he's played very well in the playoffs and the Super Bowl, then yes I can trust him in the regular season..how are you gonna prove to me that its easier to perform in playoff conditions than it is then regular season.? you cant because we all know its a much more stressful tougher game..so that evidence should tell you that in the regular season his team needs to step up..not him..he's proven.

 

Nope. I like comedy...


Thanks

 

Obviously.

 

Andy Dalton is a great example of why regular season "stats" are not ideal measuring sticks.

 

Playoffs are generally tougher.

 

The only reason they are tougher is because of injuries and fatigue, but that could actually work in a team's favor if the other team is more injured/fatigued.

 

Yep. Def more that than just the quality of opponents.

You nailed it.

 

That's just stupid, as usual.

When was the last team that didn't play a playoff bound team during the season. I'll wait....

 

I can't wait for video of the magic that teams use to all of a sudden become better teams once the post season starts.