I haven't said a single thing about the future. What I'm saying is that his performance so far this year, while better for him, hasn't elevated him beyond where he's been in relation to the rest of the league, because the rest of the league has improved as well. Whether that remains the same in the future is an unknown.
I know Tannephins (for good reason) like Y/A. So look at the average Y/A for the league (for starting QB's, specifically those that started at least 8 games in a season) from 2012. It's stayed very steady: 7.12 in 2012, 7.17 in 2013, and 7.26 in 2014. So far this year it's 7.28. Tannehill's Y/A for 2012-2014 are: 6.8, 6.7, and 6.9. So far in 2015 (after 2 games) he's 7.5! So from Y/A perspective, Tannehill has definitely gone way up this year after the small sample of 2 games. He's also NOT improved over his first three years in this important metric.
I have the league average YPA this year as 7.38, with a standard deviation of 1.27, and so Tannehill's YPA so far this year isn't above average. Unfortunately there you've chosen the statistic along which Tannehill has always been below average, and so his improvement there, while a good sign obviously, isn't distinguishing him from the average QB in the league right now.
Things is, you're not considering the upwards trend. At this point last season, Tannehill was below average in most statistical categories, now he's just average. We're both seeing the same things, you're just interpreting that it's bad while I'm considering it to be good. Yes, he is (slowly but surely) climbing the rankings. Who's to say that by the end of the season he won't be in the top 10?
I'm considering the upward trend for Tannehill, while also considering the upward trend for the league. Tannehill's current YPA, for example, would've been above average last year. This year, while improved, it's only average.
Yeah, it's hard to decide who to include and who not to include when calculating Y/A for an incomplete year. The 2012-2014 calculation I did was for all QB's that started at least 8 games, with data from pro-football-focus. What about 2015? I can't seem to reproduce your results, so how about we just decide on a criterion from here on out? We only count stats for QB's that started at least half their games, same as what I did for 2012-2014? I actually didn't do that in my 2015 estimate above. Recalculating using that criterion we have 34 QB's that satisfy that this year. The average Y/A I get is 7.35 with std 1.31. Either way Tannehill's 7.5 is above this year's average, whether we use 7.35 or your 7.38.
You simply do the team YPA across the league. Nobody else has thrown a pass for the Dolphins. Right, but not significantly so, that is if you define significance (as most do) as a standard deviation or more above the mean. The Dolphins could finish the season 9-7, for example, above the likely league average of 8-8, and nobody would be hailing that as some meaningfully positive result (assuming it wouldn't be associated with a playoff berth).
Team YPA is one option.. reason I didn't choose that is because there were many teams where you had another QB start a game or two. I guess we could use both approaches. Essentially no one defines significance as 1 std above or below the mean. 99.9% or more scientists will define it as either 2 or more std's above or below the mean (people use 90% CI or 95% CI.. basically no one uses 66% CI). Either way, Tannehill was within 1 std of the mean for all years 2012-2014. The std's for Y/A there (using the minimum 8 games starting criterion) were: 0.62 in 2012, 0.7 in 2013, and 0.61 in 2014.
What I mean is the definition of above the average range in just one distribution. Everybody within one SD on either side I would term as being in the average range.
Watching stats guys argue with each other, always reminds me the South Park "Cripple Fight" episode. No offense intended, I like seeing your statistical breakdowns.
Not sure exactly what you're referring to. The only things I can think of that might satisfy your "definition" are certain rating scales, such as those used in assessment tests or medical treatment outcomes questionnaires. In those cases, there can be a convention for saying something is "above average" and that could mean anything (e.g. 1 std above, but it could be anything). However, in none of those cases does one say something is "significantly" above or below average. That word carries a clear meaning of statistical significance, which means rejecting a (null) hypothesis in this case (or finding confidence intervals). So the moment you say "significant" you MUST use at a minimum 2 std or more away from the mean.
Given the correlation between YPA and win percentage, what I'm looking at here is the increase in win percentage a team would be expected to get with every standard deviation increase in YPA, hence the focus on getting to a SD or more above the mean in YPA. That's the "significance," or more informally, the "meaning," I'm talking about here. It's awfully tough to have a high win percentage in the NFL when you're mired in the average range of YPA. There are teams that buck that trend, but they're exceptions to the rule.
Hey I give you credit for pointing out the importance of Y/A to me, and std's are natural units to use. Just don't say something is "significantly" above average when you're making an argument with stats and it's 1 std above average because anyone who understand stats will point out you're wrong. Just say it was "well" above average or something else to emphasize the importance of it being 1 std away (actually just saying it was over 1 std away on its own carries the meaning you're looking for).
"Meaningfully" above average is what I'm getting at here. Once you're distinguished from the average range (at least a SD on either side of the mean), there's some meaning there.
Actually this is a good time to point out how something can be "statistically significant" but not "meaningfully" different, or vice versa. One thing about stats is that increasing sample size decreases variance. So if you have enough samples you can make ANYTHING statistically significant (smaller variance means less overlap between two distributions that are separated by a fixed distance between their means) even though there's nothing meaningful there. For example, if you took 10's of millions of samples, you could probably show there is a statistically significant difference between the heights of men east vs. west of ANY demarcation line through the US (e.g. the Mississippi river). Of course, something like that would be statistically significant but utterly meaningless because the only reason there was a difference statistically speaking is because you took too MANY samples! And then of course you have things that are "meaningfully" different but not statistically significant. What you're referring to here with an "average" Y/A range (defined by 1 std above and below the mean) might well fall into that category.
Well, there's stats and then there's real football. Do the pros study numbers or film. Yeah that's rhetorical...
Again, though, in this area, you have to consider the correlation between YPA and win percentage, and the expected increase in win percentage you're going to get with every SD increase in YPA. That's where it gets its meaning, or "significance."
That's what I meant. It's arguably a meaningful difference even though it's not statistically significant. All my previous post was trying to point out is how those two don't necessarily coincide.
No they sure don't. I suspect Dennis Lock is making more per year than any of us: http://www.miamidolphins.com/news/p...ditions-/e01d79f2-6daa-4289-a488-66840f94d7d5
Take a look at the research he's done: http://www.lockanalytics.com/football.html Basically, he's had experience estimating win probabilities, and trying to beat the spread. Note what he says about trying to beat the spread: "In the end I chose to concede and tip my hat to the folks in Las Vegas" Ha! Shows you how hard it is to beat Vegas. On another note, I doubt the experience he's had with estimating win probability will change any play-calling on the field. If that happens though, there's something very innovative the Dolphins will be doing, but I doubt you can get Philbin to go along with that even though he has the resources to make statistical arguments for what kinds of plays to call in which situations. Still, someday someone's going to have to have the guts to help coaches call plays based on statistical analysis. I bet (up to a point) it works better than some coaches expect.
If it wasn't, they would've shut down the sportsbooks years ago. Vegas doesn't do any losing propositions. The area where this has been pretty informative recently is in going for it on fourth down, where the stats show that coaches don't do it anywhere near as much as they should. But yeah, I doubt that's going to change.
Well, there's stats and then there's real football. Do the pros study numbers or film. Yeah that's rhetorical...
Over a season Y/A std across all QB's is something like 0.6 to 0.7. Why is it larger when computed across 2 games vs. 16? Let's suppose there is a "true" Y/A for each QB. The more attempts, the more data points, and the closer your sample Y/A is to the "true" Y/A. So compare two distributions: 1) distribution of "true Y/A + small error" for each QB (e.g. after 16 games), vs. 2) distribution of "true Y/A + big error" for each QB (e.g. after 2 games). Distribution 2 will have a larger spread, giving you a larger std.
I know my lane when it comes to those cats...I think its facinating stuff though...debating players strictly based on stats that is.
Makes sense that it would tighten up over the series. Just know my initial reaction was, "Wow!" And what are these std's you speak of? Get those off of toilet seats, right?
Not sure you can get them from toilet seats, but if you can.. as I said, having a larger "spread" means you get a larger std.
And what comes with that is the possibility that Tannehill's early-season YPA is an aberration, and may regress to where it was in the past as the season goes on. Certainly he's had two-game stretches before in his career where his YPA was around 7.5 on average, yet was less over the season as a whole. And again, that's only a possibility. Nobody can predict the future, obviously.
Once you have a pretty firm idea of what makes teams win in this league (and that knowledge is certainly readily available to anyone who seeks it), it makes sense to start looking at how key players are performing along those variables, in comparison to the rest of the league. Until this team has a passing game that's functioning at an above-average level, it makes little sense to expect great things from the team as a whole. They may surprise us and achieve great things despite having only an average passing game, but you'd be a fool to bet any money on that.
There are strengths and weaknesses of every approach. The weakness of film is that, unless you're superhuman, you can't possibly make standardized comparisons among players and teams on the basis of it. That diminishes considerably your ability to know where your team falls among the other teams in the league, and what to expect from it. Anybody for example who is managing to get one over on Vegas is doing so with a great deal of statistics and very little if any film.
The other weaknesses of film are: 1) observer bias gets in without knowing what the bias is, and 2) lack of ability to quantify (you can say A>B but not by how much). The main weakness of statistics is that it throws away tons of information that a human has access to (information used as input to a statistical analysis). However, what is done with that information is far superior to what humans would otherwise do.
Stats and game tape viewing go together, as a for instance. Tanne has a 7.5 avg, but his longest pass of the season was a pass that shouldn't have been completed, it was tipped into Matthews hands by a defender. Eliminate that 1 pass and Tanne's ypa is 6.9, so being able to take the knowledge you get from watching can help put stats into perspective, Tanne has been completing a lot of short passes, that's what you see in games, so the 7.5 isn't indicative of his play so far.
But even then you'd need stats to determine what would happen to the league average YPA if you did the same thing for every other QB. In other words, it's exactly the weakness of film I mentioned above, where you can't possibly make standardized comparisons among players and teams unless you're superhuman.
but this is not something you do, because that long pass happened, just like you don´t deduct freaky plays like tipped INT`s etc from a players resume
You can do it, but unless you do it for every QB, you're inherently acting in a biased manner against Tannehill, perhaps under the guise that the statistic you've chosen to manipulate is eliminating your bias. This is basically one version of cherry-picking.
Yeah, in general you do not remove data from analysis. Only justified exception is when you can independently show that a data point was collected under a different condition than other data points. Then by all means remove it if you want. So you'd have to remove all tipped passes that were completed to make it fair. That's OK if you want to do it.
Without all the whining... And you're not saying that's a bad thing are you? Now give Tannehill Bill Walsh and then you'd need to change your drawers more often old smeller!