You can determine what happened in each game by looking deeper. You choose not to do that because it keeps alive your narrative that Tannehill is the problem. For example, there was the INT in the Houston game this season that was totally on the TE. If he does his job, Tannehill ends up with a passer rating of 116. Tannehill played great in that game yet you want to imply he played poorly because of the passer rating. Passer rating is not a good measure of the QB's play in a single game. Hell, you spent the first 5 weeks of Tannehill starting saying that passer rating was not a good measure in so few games..... yet here you are..... BTW, even with the INT, Tannehill outplayed Watson (one of the guys you like to pimp) in that game.
No I didn't say that. What I said was that it wasn't yet different at that point from other similar stretches of play in his career, in terms of passer rating. That didn't mean his passer rating during those games was meaningless; it meant the length of that stretch of play didn't yet have distinctive meaning at that point, within the context of Tannehill's previous career.
That is the same thing weasel worded so that you can continue your narrative. At this point, you have exhausted all of your attempts to discredit Tannehill. You are standing alone with your opinion. An opinion that is unsupported by the statistical analysis you pretend to use. You've been proven wrong by the stats and the analysis of games.
I don't believe that's the case. But I'm happy to agree to disagree, and I don't mind standing alone with an opinion.
You are entitled to your opinion. You are not entitled to manipulate numbers to support your opinion. You have tried that multiple times in this thread alone and you have a long history of doing it.
No manipulation of others has been intended, and who supports or doesn't support my opinion isn't important. Like I said I don't mind standing alone with an opinion. I've simply stated my point and supported it with the information that I believe is relevant and accurate. Others are free to disagree, and as I said I'm happy to agree to disagree.
I will say...it was DAMN fun watching them beat the Patriots a few weeks ago. I certainly wouldn't complain about more of that!
Then why did you try it REPEATEDLY? You accidentally cherry picked the data in the same way multiple times? Why don't you compute the odds of that happening randomly?
He spent 7 years on Finheaven doing the same thing. He was caught using fabricated data to support his argument and got banned.
First, I don't believe I cherry-picked data. But like with anything else here, I'm happy to agree to disagree about that as well. That'll be my last post about this topic (i.e., my personal intentions and conduct). I'm not on the witness stand being cross-examined here, and I've made my position clear enough.
I was happy the team drafted him because at the time it needed to explore what a first-round pick at quarterback could do.
Oh, you certainly have. Pathologically so. You’ve steadfastly maintained your position despite being proven wrong on multiple occasions and despite numerous predictions being wrong.
If you are unwilling to discuss your points, and are unwilling to respond to specific questions about your positions, and you are unwilling to change your position when given facts that refute your position, then why are you on a message board? It would seem you’d be better off blogging about your positions, and turning comments on your posts off, so that no one argues with you.
You have to wonder how it is somebody could be so hung up on proving ONE guy sucks, for so long, directly in the face of mountains of evidence to the contrary...to the point a verified, expert stats guy showed him where his approach was wrong numerous times, yet he still goes back to it, and stands behind it as incontrovertible evidence. This level of obsession is mind-boggling to me. Either way, it's much nicer not seeing his posts...they're just the same unsubstantiated drivel over and over anyway. I recommend the block feature y'all...does wonders.
He was banned under two different accounts on Finheaven. His last attempt at discrediting Tannehill was using passer rating and more specifically passer rating differential. Funny how we haven't seen that argument this season..... Before that it was sacks, then deep ball passing. He has bounced around trying different things at different times but always attempting to show either the team's problems can be attributed primarily to Tannehill or the team's success is for reasons other than Tannehill. Sound familiar?
I'm going to address this in depth because cherry picking is serious business, and while I like the fact you've been willing to go to the lengths of adjusting ratings and using z-scores in some cases (not often enough though because those correlations you posted aren't correct once you remove the effect of passer rating inflation), it's important to understand how serious "cherry picking" is. Basically, as long as you want to claim you're using "valid statistical methodology" you can't hold the opinion that you didn't cherry pick data. First of all, let me explain mathematically (as in conceptually.. won't introduce actual math here unless you want to) what the problem is. Let's say you randomly choose a whole bunch of numbers from some distribution, like a standard normal distribution. You give that data to someone to analyze and that person sees what looks like an outlier in the data and wants to remove it. Maybe they try to justify it by saying it's more than X standard deviations away from the mean. OK, so they remove that data point. Now you're left with a new distribution with one less data point. Often, it turns out that if you apply the SAME rule again (e.g., more than X standard deviations away from the mean), you'll find a new outlier! So you remove that one. And the process repeats itself until you can't remove any more "outliers". There are examples of this occurring with simulated data where you can start off with 1000 data points and by removing "outliers" after each iteration you're left with just 3 data points in the end!! In other words, one huge problem with using ANY rule for removing outliers is that the systematic application of that rule leads to removal of tons of "non-outliers" that in fact constitute "real" data. That can be shown through simulations, which is one reason why you don't want to remove ANY data points if possible. When can you justify removing a data point? Basically you have to have a theory that is independent of your data for how the data should be distributed and is "known" to be valid (there's tons of evidence already showing that is the expected distribution), and show that with the number of samples you obtained it's too unlikely (choose some threshold like 0.1% likely or so – the threshold is arbitrary but it needs to be convincing) to have ever obtained that kind of data. For example, if you already know from some other source that the data should be normally distributed and you have 1000 data points and one data point is 4.7 standard deviations away from the mean, then I'd remove it because 4.7 standard deviations corresponds to 0.0001% probability of seeing that data point, meaning that you'd expect to see one such occurrence in 1 million samples. In other words, for 1000 data points it should occur only 1/1000 times, and that's probably convincing enough to whoever you're talking to. Of course, those probabilities are different if the data is supposed to be distributed differently, which should help you intuitively understand why there is no principled approach in statistics for determining what an outlier is because you don't a priori know what the distribution of that data should be like. And every "rule" for removing outliers you see used generally corresponds to assuming a specific distribution in some specific case. Back to the current situation. You don't a priori know what the correlation between passing attempts and passer ratings should be so there's no justification for removing data points. The only recourse is to use different methods of statistical analysis on the same data (there are usually multiple approaches that are "valid" given the data). A simple example would be using the median instead of the mean because the median is insensitive to outliers. Of course using the median with those correlations still gives you the same result. Finally, let me try to explain how serious this issue is in science. There's a LOT of bad science out there, most of it due to incompetence. Before a research paper gets published you have to convince a few reviewers (experts on the subject) that your paper is good enough for publication. Most papers get rejected, and many that get published still have flaws in them because the reviewers weren't up to the task. HOWEVER.. despite all the issues you might see in research papers, one you rarely see is cherry picked data because that's relatively easy to identify. You try cherry picking data without sufficient justification and the whole analysis is rejected right off the bat. And if you try defending cherry picking without sufficient justification your credibility takes a hit and people will think you're incompetent. Like I said.. this isn't an "agree or disagree" issue as long as you want to claim you're using valid statistical methodology. Having said all that, none of this precludes you from formulating a hypothesis of what is likely to occur in a Tannehill vs. Mahomes matchup or what a long-term trend (in your mind) is likely to look like for Tannehill in high-volume games. Maybe it turns out to be the case that in Tennessee over many years we get enough data to show that for Tannehill the distribution is different for high-volume than low-volume. But that's just a hypothesis for the future. The data today show he's performing (more or less) as expected for that stat.
Wow. I thought QBs couldn't be judged by wins or losses? Are you implying that other QBs would have fared much better with this minor league roster?
And that day I send you the PM asking if you thought that was him, I'd already made up my mind. The trend was transparent and staggeringly in line with his previous stuff. We could've written the script for how this thread has gone, at that point, and probably been within one standard deviation of how it went
cbrad first let me say I always value your contribution to the forum and the information you're generous enough to generate and share with us here. In my opinion it's the single best contribution made by anyone here (myself included), though I realize I value statistical information more than a good number of folks here, and so others may of course value other kinds of contributions (film study, for example). Second let me say that I agree with everything you stated above with regard to the manipulation of data. I've published several research papers and had to defend a quantitative dissertation, and so I'm fairly well-versed in the area you're talking about. However, in the post we're talking about here, I don't believe I cherry-picked data. What I believe I was doing is calling your attention to a curvilinear relationship between the variables involved, and how a correlation coefficient on non-transformed data won't reflect such a relationship. Here is the post of mine I'm talking about: I'd appreciate your feedback once again if you wouldn't mind taking a second look. Again I always value your contributions here even if your viewpoint and mine may not happen to jibe on occasion.
He’s a cutback runner. He’s very patient. He’ll find creases. And guys didn’t seem like they was too interested in tackling him. “I think our mindset is a little different." -Earl Thomas A week later this is what happend: El o el
My response to that post was that you could have cherry picked more efficiently. For example, why not separate the 3 lowest volume games from the rest instead of leaving the lowest volume in there. Separate the 3 and you actually get a stronger argument than what you were making because you then have two separate lines that fit the data, etc.. Even better.. you could separate it by lowest 7 volume and highest 6 volume games. For the highest 6 volume games you'd get a correlation of -0.9853!! That's explaining 97.1% of the variance in the top 6 volume games, etc... That's why I said the cherry picking could have been done better than it was. So why is that cherry picking? Because, like I said in my post, you don't a priori know what the relation should be. The best estimates would come from population stats which I've already shown Tannehill's isn't any outlier to. So if all you're doing is looking at data and trying to find some hypothesis that fits, well remember that mathematically there are an infinite number of hypotheses that will fit any finite amount of data, so you need evidence FOR your particular hypothesis. You can't just say here's one way to look at it. So absent knowing what kind of function to fit, you opt for the simplest approach that answers the question which is a correlation without removing any data points. Hopefully that explains why you can't take the approach you took and claim it's valid statistical methodology.
no. Right guy for the ask. But you can improve the roster and it won’t be long before you find out the qb play isnt good enough/sub par