I generally agree that W/L records tell you something important about a QB's ability (otherwise there wouldn't be such a large correlation between what we'd say are good QB's and W/L records), but there is at least one HoF QB with an overall losing record: Warren Moon http://www.pro-football-reference.com/players/M/MoonWa00.htm 102-101 in regular season and 3-7 in postseason. Just a technicality
I don't consider that a difference in difficulty. The task is the same and the difficulty is the same. The ramifications of success and failure may be different, but it doesn't sound like Fitts' law takes that into account, at least the way you describe it. So Fitts' law may apply to football generally in the way you describe in your first paragraph, but it is because of that that professional athletes are already at the elite end of the Fitts' law curve with respect to their sport. Now maybe if one somehow modified Fitts' law to account for different ramifications for success/failure it would measure what we call clutch. But none of the descriptions I have seen of it include such a component.
Never saw him play so I can't speak to him being an outlier. But I will say the league has changed a lot and the QB is much more important to a teams success than it was even just 15 years ago. Average passer rating has been rising dramatically almost year to year.
The correlation is just that, a correlation, and not causation. QBs with good W/L records are perceived as good because of those records. It's hard to be truly bad at QB but have a very good record, but we all know that the record doesn't mean the QB is actually good. We see it more clearly in college football where you often have QBs with outlandishly good records but who weren't themselves very good -- Kellen Moore, Eric Crouch, Tommie Frazier, Tony Rice, Ken Dorsey, AJ McCarron, etc.
I know, it's so crazy not to believe in things that can't be seen, felt, or otherwise identified. But, yeah, it's even crazier to say that win/loss is a QB stat.
Difficulty is operationally defined by the researcher. There's no independent measure of it (it's not like you're measuring distance or luminosity, etc..). Generally you'll see people operationally define A to be more difficult than B if you can do A only if you can do B (so A is a subset of B), or they'll say A is more difficult than B if a smaller percentage of people can do it (so both A and B are mapped to some other measure where that subset definition applies). So in this case, as long as you can say that in a pressure situation the set of outcomes you'd define as successful is a subset of those you'd define as successful in a non-pressure situation, you're doing the same thing as saying completing a task within a more restrictive set of parameters is more difficult.
Well, anyone who watches the game knows this isn't JUST a correlation. There is no question there exists a causal relationship between QB ability and W/L record. Furthermore, I'd bet you'd get the highest correlation for your own subjective rankings of QB ability (which now implies causal) with W/L compared to that of any other position.
There is some partial causation between winning and QB performance/efficiency, but QB performance/efficiency is not the same as QB ability. My own subjective rankings of QB ability aren't that highly correlated with W/L.
But if you are operationally defining a task, say completing a pass, as more difficult in pressure situations, you are artificially defining "clutch" in a circular way that will make you prone to a false conclusion.
I seriously doubt this to be true. You wanna give it a try? Gotta include both good and bad QB's so it's not just one end of the scale (and maybe rank order at least 20, preferably more). btw.. you'll probably get even better correlations if you just rate QB's instead of rank ordering them. So assign a number from 1 to 5 (or whatever range) with 5 being the best to every QB. That correlation I'd bet is even higher (rating is often better than ranking if we're talking about ability).
Right, but are you going to argue the assumption isn't accurate? Are you saying completing the same pass with defenders in your face is not more difficult than with little defensive pressure? Or that there is a smaller subset of outcomes that one would define as "successful" when the game is on the line than when it isn't? Those are accurate assumptions the great majority of the time.
By pressure situations, I am referring to "clutch" situations, not pass rush pressure. I think throwing that same pass is just as difficult in the 1st quarter of Game 1 as it is in the last 2 minutes of the Super Bowl. By "same pass" I mean same level of pass rush pressure, same receivers, same level of coverage, etc.
The physical side of it might be the same, but mentally it's not. To make sure the mental side isn't just ignored here, try throwing the same pass where you also have to do a math problem that you'll have to answer right after the throw vs. not doing it. I guarantee you accuracy for the throw will tend to go down because of cognitive load. Stress beyond some point often does the same thing to people (there's a ton of research on effects of stress). So to say you can just look at the physical side without the mental side isn't going to work. Either way, the operational definition of defining more difficult to mean do the same thing under a smaller range of parameter values in general works, and works here too because not every outcome that would be considered a success in non-clutch situations would be considered a success in clutch situations. The example I gave of having to get a TD on the final drive vs. just getting a FG or just settling for good field position should show that (can't just focus on one play here btw).
Again, clutch doesn't always mean perform even better (sometimes it does) it means not choking. And for those who argue there are no chokers (if chokers exist, then clutch is also real, yin/yang) I have two words for you. Nate Kaeding.
red herring. His clutch shot is just one shot. Missed or made it barely registers on his %. It's just one shot. And he needs to sink it. and he did, several times.
http://miamiherald.typepad.com/dolp...the-game-for-passing-yards-or-touchdowns.html As fans of the Dolphins how can we not obliged him? If you don't judge Tannehill on his wins, you're just a hater.
Unconvincing to me, because I don't believe Moon belongs in the HOF. Is Moon's career better than Ken Anderson? Why is Anderson not in the HOF while Moon is? Moon's accumulated a lot of stats but his career was not all that impressive.
I love how this gets trotted out all the time, as if it means something. WTF is Tannehill supposed to say? If he said, "Screw wins, judge me on stats," people would lose their minds and argue that Tannehill is a selfish prick. This discussion has devolved, as it always does, into QB being responsible for wins and losses, as if the rest of the team doesn't matter. All you need is a great QB, and you'll win games. Except Philip Rivers disagrees with that. And Peyton Manning last season shows you it isn't true. Or Mark Sanchez with the Jets shows you it isn't true. Or any other number of examples.
IMPORTANT NOTE Data has been updated to include Tony Romo, Ryan Fitzpatrick and Philip Rivers.\ This causes a significant upwards change in the correlation between overall 4th quarter passing and win% There is no significant change to the correlation between passing when behind in the last 4 minutes.
My problem with that approach is that mostly it used as post-hoc justification to fit a pre-determined narrative. People are looking for something that justifies their opinion. Often when people do it they only look at one QB in one situation. I much prefer to look at what the whole league does in similar situations, otherwise the temptation to cherrypick can become overwhelming (example: The Denver Broncos cherrypicking 4th Quarter comebacks to make John Elway's seem a better QB than how he really performed under Dan Reeves)
As per the original post: The correlation between team passer rating and team win% over the last 6 tears is 0.676, which is a ludicrously high correlation for the impact of one individual on the outcome of a game between two teams. Of course many things contribute to passer rating such as coaching, WRs and OL. But the biggest factor in passer rating is QB ability. Basically you have the QB's "real rating" to which you can add or subtract effects of above or below average contributions. You can surround Mark Sanchez with HoF OL, receivers and coaches and he'll struggle to get to an 85 rating. You can put Drew Brees in a situation where he has a poor OL, poor receivers and a dumbcluck coach and he'll still post an 85 rating.
Just how important is "clutch", really? It's funny that the Tannehill haters want to judge him on wins and losses, yet were bashing him after we beat Baltimore last season. Even though "he" won the game.
For those who remain unconvinced about HoF QBs. http://bleacherreport.com/articles/...terbacks-not-in-the-pro-football-hall-of-fame Taking that list we have Kosar, Hart, Gabriel, Hadl, Simms, Esiason, Cunnungham, Stabler, Brodie and Anderson nominated as being worthy of HoF induction. I make no comment on whether they are worthy but I will accept that opinion. There are 20 QBs who played post 1960 in the HoF, with 2 of them playing the majority of their careers prior to 1960 (Layne and Tittle), plus Manning and Favre who I included as guaranteed HoFers in my initial analysis. Of the proposed HoFers: Career win% in 0-7 point games: 51.6% Standard deviation. 5.45 Looking at the individuals, The ones I have heard proposed as having the strongest cases for getting into the HoF Phil Simms: 56.0% in 0-7 games Ken Stabler: 61.1% in 0-7 games Ken Anderson: 48.8% in 0-7 games Jim Hart and John Brodie (well before my time) have career win% in 0-7 games of 54.4% and 58.0% respectively. Randall Cunnimgham is at 50% exactly and the rest below. 4 above 50% 1 50% 5 below 50% The guys in the HoF + Manning and Favre 3 below 50% 19 above 50%
After adding some more points to the data it becomes clear to me that for QBs not named Andrew Luck that overall performance in the 4th quarter is the strongest and best indicator of an ability to win 0-7 point games. So let's look at what Tannehill has done in his career in the 4th quarter: 2012. 83.0 rating 2013. 64.3 rating 2014. 79.9 rating 2015. 84.9 rating
Just double checking as usual, nothing personal... but using the values you have listed for 4th quarter ratings/win%, I'm getting slightly different numbers. I guess you only updated the bold, but just for completeness I'm getting: Average 4th Q rating 91.63 Std Dev: 8.04 Correlation to win %: 0.385 Removing Luck.. Average: 91.88 Correlation to win %: 0.512 The big difference is between 0.512 to your 0.55 (what you have in bold). Not sure if your posted numbers are the same as the ones you're using but after double-checking to see if I typed them in right, that's what I get. Also.. Dalton is still missing on the 2nd list. If you can add that, I can double-check that too I know you're using a slightly different set of QB's, but the 0.676 is strikingly high from what is usually the case. Generally, I'm used to this: http://www.footballperspective.com/is-espns-qbr-the-best-measure-of-quarterback-play/ From 2006-2013 QB's with 14+ games had a 0.56 correlation to wins. Not saying your figure is wrong, but that's a huge difference if true so maybe you can double-check that one also just to be sure?
At this point Ryan doesn't understand what it means when he says judge me on wins..He hasn't shown that he knows how to translate the phrase to the field.. when qbs say that, I watch for one thing, and that is, are you willing to do whatever it takes to convert, are you conscious of how to convert, if your just thinking you can sit back in a pocket and do it with just your arm, your mistaken..you dont' get it. converting third and 4th downs by maneuvering the pocket when protection breaks down, scrambling and finding a different platform to make the throw, running for a damn first down, this is how you become a winner..do whatever it takes on those downs..so far he doesn't understand how to do it..until that changes we will continue to come up short.
I didn't want to use that number because 1) it didn't adjust for rising passer ratings. 2) I saw no reason to exclude QBs. Poor QBs have poor ratings and get benched, so removing them from the data will weaken the correlations. 3) Using team data removes any biases that may occur from partial games. Andy Dalton, dammit forgot to do him in my second spreadsheet. I might try posting the data in a fresh spreadsheet in case there's some problem from formatting when I inserted the extra data
Just to comment a little more on that football perspectives article. The league average passer rating in 2006 was 78.5 The league average passer rating in 2013 was 84.1 He's comparing ESPN's QBR, which is adjusted to league average to non-adjusted passer rating. If I wanted to make ESPN's QBR look better that's exactly what I would do. I'm not saying he has made a choice to present dodgy data, but like most people he's used to passer rating being a fairly stable measure so he has overlooked this important point. Also as far as tossing out samples involving less than 20 passes and restricting it to 14 games played. I can understand the logic of doing that if you are comparing QB1 to QB2. However he's comparing rating system A to rating system B against win%, so there's no need to cull QBs who have played fewer than 14 games from the sample, in fact doing so removes a big chunk of low performing QBs who count to establishing the efficacy of the rating system.
Yes and no. Yes to the second point, no to the first point. Yes, you're absolutely right that it's not justified to cull QBs who played fewer than 14 games because we're looking at passer rating over some period instead of comparing two QBs. Same point I was making in previous debates about not removing QBs from data when comparing based on some condition (e.g. trailing vs. tied). However.. I could see culling at some level (maybe 1 or 2) to remove probably tons of 0% and 100% win percentages that would destroy any correlation. But.. no to your first point. There's no benefit or cost to adjusting passer rating to some league average (in a given year) if your goal is to correlate it with wins. Correlation is insensitive to adding a constant or multiplying by a scalar. Try it yourself. Take two vectors that you're calculating the correlation between. Now add a constant to one vector, or multiply one vector by a scalar.. no difference in correlation. Why? Correlation is covariance divided by the product of the two standard deviations. Every component of that formula removes the effect of changing the mean so it doesn't matter if you add a constant to all values or multiply them all by some value. Thus, adjusting to a new mean actually has no relevance in this debate. Where it does have relevance is if you're comparing QB's over time (e.g. Marino to Brady). Then you want to see how many standard deviations above or below their respective means they were.
If he's comparing win% to passer rating in each given season, then it shouldn't be a problem. However I'm not sure he's done that, but if you're comparing it to win% from multiple seasons where the average is rising then there is a problem. 1n 2006 78.5 is the average rating, meaning that a 78.5 rated QB should have an 8-8 record (assuming everything else about the team is average). By 2013 78.5 is well below average and a QB posting that record probably posts a 6-10 record Conversely an 84.5 rating would get you a 10-6 record in 2006 but only 8-8 in 2006. I'll do some correlations based on each year in my spreadsheet, as well as unadjusted passer ratings.
That's a very political answer from a guy who has been dogged for not making the playoffs. Optics are important to keep fans in the seats.
Yeah, whether correlation increases or decreases if you adjust each year separately to the same mean depends on the data. Let's just do this for 2 years only because the same principle holds for multiple years. Let the x-axis be passer rating and the y-axis be win percentage. For year 1, say X1 goes from 50 to 100 in increments of 5 (so 11 data points), and for year 2, X2 goes from 60 to 110 in increments of 5. It's easy to concoct examples where your intuition is correct (without aid of a computer in generating the points). If for both years, Y goes from 0% to 100% in increments of 10, then with no adjustment correlation is 0.9535 while with adjustment (for simplicity, just subtract 10 from all values in X2) correlation is 1. So that's a case where your intuition holds: adjust each year separately and the overall correlation increases. Your intuition however doesn't hold in many cases, but I need a computer to generate the points (though I know what I want to generate). Here is some simulated data: Y1 = 21, 15, 19, 5, 64, 29, 54, 70, 50, 54, 45 Y2 = 13, 50, 86, 88, 28, 21, 57, 65, 42, 21, 95 Correlation between X1 and Y1 is 0.6766, with X2 and Y2 is 0.149. Now, if you combine the data with no adjustment of X2 to the mean of X1, the correlation is 0.4167 while adjusting gives you 0.358, so adjusting HURTS! Now, I generated that data deliberately so that I got the result I wanted (high correlation for 1st year, low correlation for 2nd year), but if you just do this randomly so that both years are not too dissimilar in correlation, you'll get many examples like that. For example: Y1 = 43, 89, 40, 77, 40, 81, 76, 38, 22, 80, 95 Y2 = 33, 68, 44, 84, 77, 17, 87, 99, 52, 89, 59 Individual correlations are relatively low at 0.15 and 0.32, while combined with no adjustment you have 0.243, and with adjustment you have 0.238. So what you get depends on the nature of the data. Conceptually, the bigger question is whether one should or should not adjust for different years separately. I think it's fine if you can truly argue the conditions are different, but if you can't you shouldn't. Anyway, I never criticized you for adjusting so from my perspective it's fine if you do, but over a short period of time (several years) I personally would just leave the data as is.
Again, I'm talking about ability or talent, as opposed to production or efficiency. And I think the translation of that ability/talent to production/efficiency is heavily dependent on surrounding talent, coaching, scheme, etc. So using last season's starters and based on the ability/talent I saw as of last year: Rodgers -- 5 Wilson -- 5 Luck -- 5 Newton -- 5 Brady -- 4 Palmer -- 4 Roethlisberger -- 4 Rivers -- 4 Brees -- 4 Mariota - 4 Stafford -- 4 Romo -- 4 TTaylor -- 3 Winston -- 3 Bortles -- 3 Tannehill -- 3 Ryan -- 3 Carr - 3 EManning -- 3 Cutler -- 3 Flacco -- 3 Kaepernick -- 3 Dalton -- 3 ASmith -- 3 Cousins -- 2 Bridgewater -- 2 Bradford -- 2 PManning -- 2 (obviously a big dropoff from previous years) Fitzpatrick -- 2 Foles -- 1 Hoyer -- 1 JMcCown -- 1 I could go back and forth on these quite a bit, but that's an initial cut. I value mobility and guys who can make things happen with their legs, so I probably have those guys a point or even two higher than many would. Similarly, that puts my scores on some of the non-mobile older guys (Brady, Brees, Palmer) lower than most might have them. There are a few guys who I think benefit more from coaching, system and/or surrounding talent and their scores may reflect that. And again, this is ability/talent, not production/efficiency and not necessarily how I'd rank them if I had to choose a guy to win a particular game today. I tried to keep it a reasonably equal distribution between/among the grades, so a certain number of guys had to get 1s and 2s. That doesn't mean that I consider them bad or terrible QBs at all -- it just reflects how I view their pure talent relative to other starting QBs.
Division games are the most important games on the schedule, win your division and you get in the playoffs, the only games that are more important are any late season "must win" games, and unless you go back through every season and look at what were the "must win" games, you can't get that info.
By the time they reach the NFL the stage gets bigger, so therefor the pressure gets higher, it keeps going up, thats why OTA superstars sometimes fade as preseason starts, and then preseason superstars fade when the regular season starts, then the playoffs, SB, pressure keeps getting higher.
Understood, and sound reasoning, however, sometimes you do have to look at certain situations in particular, like high pressure games, and how individuals handle those situations that are pertinent to the discussion.
I love the two word, one player answer that you think tells it all. There's a difference between "chokers" and guys who may occasionally fail to perform well in pressure situations. Everyone falls into the latter category. Every single athlete in every single sport has had pressure games or moments when they didn't come through. But I don't think that makes them "chokers" as if it did them every single athlete would be a choker. And no, even if chokers exist, it does not mean that clutch does too. If 99% of players perform exactly the same under pressure and 1% perform worse, the 99% are not all "clutch." Nate Kaeding had a bad game. It was in the playoffs so you remember it and seem to believe that game makes him the poster child for choking. But prior to that game, he had made his previous 6 playoff field goals and all of his playoff extra points. Prior to that game, he had made 9-11 of his career 4th quarter FGs with a chance to take the lead or tie the game. One of those two misses was a 54 yarder, so hardly a gimme. Of course, it's quite easy to find a bad game in the playoffs for virtually every supposedly clutch player who played a decent number of playoff games. Joe Montana put up a 98 yard, 2 INT, no TD stinker in a 49-3 blowout loss to the Giants in 86-87. Tom Brady posted passer ratings of 49.1, 57.5 and 62.3 in 3 playoff games against the Ravens. Marino, of course, had that 17/43 for 141 yards, 0 TD and 2 INT game against the Pats in the 1997 playoffs and that 11/25 for 95 yards, 1 TD, 2 INT horror show in the 62-7 loss to the Jags. Michael Jordan had a 5-19 12 pt game in the '95 playoffs, a 5-19 5 TO game in the '96 playoffs, a 6/11 from the FT line in the 96 playoffs and a 9/35 game against the Heat in the '97 playoffs. Etc.
I don't agree with that. It's not about pressure. OTa superstars aren't performing under real football conditions. Preseason superstars are typically playing against backups. And most players play at the same level in the regular season and playoffs.
He also missed many times. People forget them because they are unremarkable, but they happen. With any given 3 pt shot (which is typically what we are talking about with Horry), he was going to make 34-37% in line with his career average. So making any one is entirely unremarkable. Making two in a row is similarly unremarkable. Even having a 6-10 streak is pretty unremarkable because mere chance says that will happen from time to time for any event with a 34-37% likelihood. Not only is there no real correlation between Horry's 3 pt shooting in clutch situations and regular situations, there are many much stronger correlations that illustrate how random correlations can be and how completely different from causation they are. For example, Horry was a .388 3 pt shooter on Thursday's, but only a .297 3 pt shooter on Saturdays. Both samples have 900+ attempts, so they are much, much, much bigger than his "clutch" sample. So do we conclude that Horry is a very good Thursday shooter and a bad Saturday shooter, as if that has any meaning at all? He shot .485 from 3 against Golden State over a long career, but only .260 against Seattle/OKC. That's a huge difference with 300+ attempts in both cases. Do we conclude that he shoots better in certain states, or against certain color uniforms, or against teams whose names begin with certain letters? Of course not. And yet the sample sizes in that case are 20 times the size of his "clutch" shot sample size and the correlation is many times stronger.
