I'll try to pour some cold water on the digression. So basically for every 10% increase in 8+ game winning percentage you can expect, roughly, a 3% in win% in 0-7 point games Since 8+ wins are generally considered to be more team wins than due to the vagaries if one player I think its reasonable to say that the 3% shift in 0-7 games for every 10% shift in 8+ win% is a signal of the overall strength or weakness of the team. So for fun lets compare some QBs. Peyton Manning 8+ win%: 73.6% expected 0-7 win%: 57% actual 0-7 win%: 65.8% difference +9% Tom Brady 8+ win%: 82.6% expected 0-7 win%: 60% actual 0-7 win%: 69.1% difference +9% So when you control for outside factors such as coaching and teammates Brady and Manning are close to identical in DPAP. John Elway 8+ win%: 72.8 expected 0-7 win%: 57% actual 0-7 win% 56.4% difference -0.5% Dan Marino 8+ win% 63.1% expected 0-7 win% 54% actual 0-7 win% 59.1% difference 5% So, Dan was more DPAP than Elway
Two issues if you want to do a DPUP calculation: 1) Correlation (how much one quantity varies with another) isn't linear regression (best-fitting line through data). You're I'm guessing using the correlation of 0.3 between 0-7 games and 8+ games to say a 10% increase in 8+ games win% gives you a 3% increase in 0-7 game win% (I see no other 0.3 there so that's my assumption). Can't do that. You need to actually do a linear regression and use the slope of the best-fitting line. Easy example to see the two are different: if all points lie on a line with positive slope, that will always give a correlation of 1, but the slope can be ANY positive number. 2) For DPUP one should also calculate how likely something was by chance, or more specifically how likely anything better than that result was by chance (if higher than the mean) or worse if lower, given the sample size. That way we can see if it's "statistically significant" on its own, or if not, we'll know what probability to use when combining multiple stats to determine overall statistical significance. If that's too hard a calculation to do, just report how many standard deviations above or below the mean the result is relative to the drop-off distribution and we can figure out those probabilities later (remember it's the distribution of drop-offs and its standard error that needs to be used).
Again, sugar to water is different than temperature. Dye to water is different than temperature. Your ignorance is astounding.
What you're talking about and the science version of cold are two different things. You are basically arguing about what a tackle is in football (US) with someone who was talking about a tackle in football (soccer to us). The best part is the belligerence with which you're doing it.
I have a new one to add to the list. Things that Tannehill homers believe: 1. if the stat makes the QB look bad then its a team stat. If it makes the QB look good then its a QB stat (wins/losses, passer rating (depending on the year), yac, etc, etc) 2. intangibles are make believe. 3. it is wrong to say ice makes drinks colder.
I'll just again voice my disagreement with using team win% as a measure of QB DPUP for all the reasons I have previously expressed. As far as statistical significance, I think there are two parts to that. First is whether any apparent difference between expected and actual 0-7 pt win for a given QB is statistically significant. I'm not sure even the 9% differences would satisfy that condition, and I'm pretty sure the 5% difference would not. Second is whether the difference between one player's differential and another player's differential is statistically significant. On this, I'm not sure its right to compare, for example, Marino's 5% with Brady's 9%, stated as percentages. I think one would get different results if one were to compare them as decimals -- .05 vs. .09. And I'm not sure even that is right. It's been 25 yrs since my last statistics class, so I could be wrong about that, but . . .
There should be an agreement on this issue too. Suppose you have 2 or more independent (i.e non-overlapping) team stats that are indirectly measuring one individual's DPUP for identical pressure differences (making this assumption so that all stats from a differential pressure point of view are considered equally important for now). The degree to which you'd weight each measure is precisely the degree to which that individual contributes to the team measure. So if for measure A you'd say 7% was due to the QB and for measure B it's 19%, the weights are 0.07 and 0.19. If the measures are not independent and partially measure the same thing, then you'd have to remove that, and of course if the pressure differentials (which we can't directly measure) are not identical you'd have to weight accordingly. Point is.. any team stat that the QB influences IS a valid measure of DPUP, but how strong a measure it is depends on its relative weight, which may be small. For win% maybe it's in that 15-20% range, who knows. So in the future, instead of saying one shouldn't use a team stat for DPUP, you just need to state the relative weights you'd put on it. Maybe in your mind the weight is very small and that's fine, but that's where the disagreement should be focused on. Yeah.. let's first make sure the actual numbers in Pauly's post #1004 are dismissed because the 10% to 3% coming from the correlation of 0.3 is a faulty deduction. Now.. to talk about whether an individual's stat is different from expected in a statistically significant way you can use hypothesis testing. The null hypothesis here is that the individual's drop-off is league-average. You want as small a probability it's not league-average, and we might as well use the standard 5% figure (called a p-value) as a threshold: if the probability it's due to chance is greater than 5% you can't reject the null hypothesis. The way to do that is graphically described here: https://en.wikipedia.org/wiki/P-value#/media/File:P-value_in_statistical_significance_testing.svg Just calculate the area under the curve past the point where the individual's stat lies (green area.. total area under the entire curve is assumed to be 1), and that's your probability the individual is league-average in drop-off (that distribution has to be the drop-off distribution). Sample size is naturally taken into account btw because with smaller sample size that distribution will be "fatter", meaning it's harder and harder to reject the null hypothesis. Note that if we're using multiple stats that measure different things, those probabilities are multiplied with each other, so you quickly get the needed level of "statistical significance" if you use multiple stats. If those stats partially overlap in what they measure, you'd have to remove the overlap. As to whether one individual's performance is statistically speaking different than another, that requires a different test. You'd have to know the standard deviations for both individual across all conditions for that stat (or all non-pressure situations.. whichever.. in this case both are defensible) then use something like a 2 sample t-test that essentially looks at the overlap in those distributions given the new means in the pressure situation. That directly takes sample size into account (can be unequal sample size) and the t-test will give you the p-value directly. A graphical representation from a different field (signal detection theory) using similar concepts: http://www.nature.com/nrneurol/journal/v4/n6/fig_tab/ncpneuro0794_F1.html Basically, the t-test is looking for that overlap between the distributions, which in the context of signal detection theory tells you things like "hits" vs. "false alarms", etc.. So the tests are different, but in both cases you can arrive at a DPUP (and you're right Pauly didn't do the second test for the two QB's).
But there is no way to know what % of a team stat is due to the QB and there is no standard figure that can be applied across the board. It could vary wildly from QB to QB and game to game. At one extreme, you might have a 52-45 shootout in which the QB throws for 500 yards, 7 TDs and no INTs and the team has only 11 rushing yards. That would presumably be a win in which the QB played a big role. On the other extreme, you might have a 10-0 win in which the QB was 5-25 for 38 yards, O TDs and 3 INTs with the only scoring by special teams or the defense. And of course a dominant defensive performance. That would presumably be a win for which the QB deserves no credit. And of course there are infinite gradations between those extremes. But there's no solid or reliable way to figure out the % QB impact in any such game.
I don't know why you think I don't know that, I never said it or even implied it. But what I also know is its correct to say ice can change the temperature of water.
Yeah, just pointing out there's nothing special about win% in that regard. The same critique can be applied to ANY team stat, including YPA, passer rating, etc.. The difference is the subjective weight you put on each, so that should be the point of disagreement instead of saying one team stat can't be used as a measure of an individual's ability. It can.. it's just a weak measure. Now, as far as a standardized way of estimating that, there's nothing I've seen in research papers in sports science (because they're tbh not that sophisticated on the stats side), but there believe it or not is elsewhere. It's called Rasch analysis, and what it does is take subjective ratings and estimates an objective measure based on them. This is how it works: you assume each person has their own subjective thresholds that define boundaries of rating categories on some axis. You don't know where those thresholds are, but you assume each threshold is normally distributed across the population (no way to directly test this but given how almost every trait is distributed normally in a population it's a decent assumption). Those distributions will overlap to some degree, meaning different people will rate the same item differently. The probability they'll do so depends on the overlap. Now.. you don't a priori know those overlaps but you can have a computer try out a massive number of possible locations of the mean thresholds used in the population to find out which set of possible thresholds predicts the ratings people assigned best. That set of thresholds - the estimated population thresholds - now help rate the difficulty of different items (e.g. on a test) and the abilities of different people rating those items (e.g. taking the test). The resulting measure is objective GIVEN the observed ratings, no different than a stat is objective given the observed data. I've never seen that applied in sports, but it should be. Right now you got guys at pff assigning subjective ratings and you have no idea how much difference there is between a "1" vs. "2" or "2" vs. "3", etc.. Rasch analysis teases that apart, assuming the people are measuring some common underlying property. If they're not (subjectively) measuring some common underlying property (in our case the % of the team stat due to the individual), then Rasch analysis actually tells you that too. So if I were serious about publishing in sports science, one thing I'd do is bring Rasch analysis to a lot of things they talk about. It would give us a standard for estimating the percent of a team stat an individual is responsible for, using what seems to be a counter-intuitive approach of subjective ratings across the population.
So you're just into full on troll mode? 1. Consistently we've argued that win/ loss is a team stat. Consistently we've argued that passer rating is not solely a QB stat. You, however, can't admit to either of those. 2. Intangibles, by definition, can't be measured. If it can't be measured, then how can you say some have it more than others? 3. From a scientific standpoint, which is what we've been saying, yes it is wrong to say that.
As long as people acknowledge the underlying mechanism of heat being transferred from water to ice, there's nothing wrong with saying "ice can change the temperature of water" or "ice makes the water colder". The former is technically correct anyway, though it doesn't specify the mechanism. The latter has to be understood with "colder" as a sensation. So as long as that's acknowledged, nothing wrong with those statements.
If you understand that the underlying mechanism is heat transferring to the ice cube, isn't it actually water changing the temperature of the ice?
It's really neat, the light switch in my house creates electricity that makes my light turn on. It's really neat, my phone charger creates electricity to charge my phone. It's really neat, my thermostat creates heat when I turn it up.
Each is changing the average kinetic energy of the other. So it's correct to say water changes the temperature of ice, and vice versa. btw.. this is ignoring the breaking of bonds during a state change, but that's another story (temperature remains the same).
http://indianapublicmedia.org/amomentofscience/ice-cool-drinks/ I know you love to troll, but acting like we're crazy is just ridiculous.
Thank you for that, it proves us all right. IT takes heat from the water. IT makes the water colder. Notice in your link, the quote you bolded and enlarged, that the actor, "IT" is the ice cube, and not the water. They didn't say ice sits there and receives hot from the water. Your link, I remind you again, says it takes the heat from the water. Ice makes your drink cold. please end this thread.
sort of ... . Ice is solid H2O. Liquid Water is liquid H2O. Vapor/Steam is gaseous H2O. Water, without any modifier, is the liquid form. EDIT: But I get what you're getting at. But in normal language, nobody calls ice, water. Like nobody calls vapor, water. Water vapor maybe.
What a fun thread to read through, lots of laughs, the fact that there are 2 posters in here that are saying ice doesn't make your drink cold has inspired me to create a new word, obtuseance, ob-two'-she-unts; when a being becomes one with obtuseness. A new high, or low, which...
On clutch; The fact that those who have argued there is no such thing as clutch, have admitted that there is such a thing as choke, are by that very fact admitting there is clutch. Definitions; Choke, to not perform well under pressure. Clutch, to perform well under pressure. Without clutch, there can be no choke, so admitting that choke exists, is admitting that clutch exists, because obviously, if no one ever performed well under pressure, there would be no need of a term for not performing well under pressure. Now we can all have a nice beverage with a sufficient amount of heat removed from it to make it cold
Silly boy, the ice doesn't make your drink cold. Your drink, made the ice warm! Now swallow them ice cubes!
Not really sort of. Ice is water. "without any modifier"? That is a weird way of saying it and not really correct.
OMG.. please let's not start debating the definition of water. jdang307 is right on this IMO. I mean yes, even in scientific circles you occasionally hear "water" = "H2O", but usually it means its liquid form. What's next? Asking whether "light" and "electromagnetic radiation" are the same thing? (that one's also like "water" = "H2O" in that sometimes people say "light" = "EM radiation", but generally light is only the visible part of the EM spectrum).
So you go around calling ice, "water?" THAT is weird. Again, ice is frozen water, there is liquid water, and there is water vapor. But nobody goes around saying water, when they mean ice. "Hey, can I get a coke, extra water please" said noone ever!
I was just eyeballing the trendline. Having played a bit more with the data the trendline is about a 2% increase in 0-7 win% for every 10% increase in 8+win% for all QBs. However by removing HoF QBs. The trendline is +2% in 0-7 for every 30% increase in 8+win%.
No what's funny is that there are 3 or 4 posters who don't understand the difference between science's cold and layman's cold, then proceed to make the STUPIDEST analogies to try and prove a point that science has already explained as wrong. Its.....Trumpian.
No. "Clutch" as commonly used refers to a player that steps it up or elevates his play in pressure situations. I have seen no evidence that such a thing exists on any sustained or consistent basis. "Choke" implies that it is the pressure that caused the poor performance. There's not much evidence for that either. Sure, players make bad plays and have bad performances under pressure, but they do it without pressure too. When a 75% free throw shooter misses a game tying free throw with 1 second left he didn't necessarily choke. He did what he does in 1 out of 4 free throws regardless of pressure. It just happened to happen at an inopportune time. Maybe you want to call that choking, but if so, then you have to admit that every player in every sport ever is a choker. Because a choker is presumably defined, consistent with your simplistic definitions, as one who chokes. And every player has bad plays or bad performances, both under pressure and without pressure. So while there are certainly bad performances under pressure, I don't think they are really chokes. And even if they were, the lack of evidence of players who consistently and systematically perform better under pressure means to be that there is no "clutch," at least as that term is used to mean some kind of inherent characteristic of a player.
Looking at Standard deviations above/below the trend line and win% in 0-7 games. Tom Brady +2.0 SDs Peyton Manning +1.7 SDs Jim Kelly +1.7 SDs Johnny Unitas +1.4 SDs Matt Ryan +1.0 SDs Dan Marino +0.9 SDs John Elway +0.3 SDs Joe Montana 0.0 SDs Tony Romo -0.2 SDs Aaron Rodgers -0.8 SDs Joe Flacco -1.0 SDs