One: Each First down give you 4 opportunities to score points. The more first downs you have the more opportunities you have to score points. Two: Each time you give the ball back to the opponent you are giving them an opportunity to score points. Three. Each first down moves you closer to your opponent's end zone. It is easier to score points the closer you are to your opponent's end zone. They call it the red zone for a reason, even though not every trip to the red zone results in points. If yards per play is the only important consideration why not eliminate the rushing game entirely? No running backs on your roster. Put an extra OL on the field for every play. That would be the logical conclusion to make if YPA is your only criteria for assessing success.
Good QBs don't need a good rush attack to effectively run the play action. Does it help? Sure. Colts in 2014 were 22nd in total rushing, 22nd in ypc. Luck was one of the best playaction QBs in the league. It's not how effective the runner is, as Trent Richardson is probably the least effective runner after Daniel Thomas http://www.si.com/nfl/2015/06/04/be...acks-andrew-luck-russell-wilson-philip-rivers Philip Rivers, didn't use it a lot but when he did, he killed it. Do I need to tell you that they were 31st in ypc? Naw. NY Giants, 28th in ypc
That certainly is the way the game is trending. Passing ratio climbs year over year and is at an all time high. RBs are used in pass pro more than ever, and the value of RBs is very low. The NFL is not very innovative, so you will always see sub optimal strategies, and changes happen slowly. Teams punting way too often being a primary example. Either way, you will never see the running game disappear completely, because teams will always have the necessity to keep the clock running in various situations. At the farthest end of the spectrum, you will always have the QB kneel. Sent from my SAMSUNG-SM-G920A using Tapatalk
While I was using football examples, I was referring to the psychology of misdirection. This applies to any sphere of competition from tiddly winks to total war. It doesn't address how well each team uses play action, or whether play action is the optimal call in a given situation. You can't be surprised if you have no expectations.
The threat of the run may not even be of benefit. The QB turning his head away from the defense may be what makes PA successful. It prevents the QB from staring down receivers. Sent from my SAMSUNG-SM-G920A using Tapatalk
It would be interesting to know what the yards per play would be in that scenario. One thing you would have going for you is because every defense you face would have resources allocated to stopping the run for the rest of their schedule you would have a distinct advantage over them in being able to allocate 100% of your resources to the passing game, so it could work. It could be revolutionary. Or it could be a total disaster. Would love to be able to simulate it though.
That quote bugged the hell out of me when I saw it at the time and still bugs me. There is no paradox. Misdirection is more successful when it is unanticipated. Either you have a QB with special skills selling the fake (Luck, Troy Aikman etc.) or it is a low use weapon (Rivers, Marino). Dan Marino had success with play action even though he probably had the worst fake of any QB I've seen simply because people knew he didn't like doing it and didn't expect it.
Play to play success may not be optimal. If you only use it one time on the season and gain 90 yards you would have more play to play success than any other QB. But if another QB uses it 300 times a season to the tune of 2 extra yards per attempt he might have had the worst play to play success but might have also gained the most extra yards from the play. So whats more successful, the team that unleashed it at the exact right moment for a one time huge gain, or the team that used it on almost every passing down churning out a low but consistent extra 1 or 2 yards per attempt?
So, I calculated the probability of making a 1st down for all possible 3-play sequences of run or pass, using the actual distributions of run/pass from 2005: http://postimg.org/image/3tuuehrsx/ The percent of times you'd make a first down with the following play sequences (over 10,000 simulations): Run-Run-Run = 53.6% Run-Run-Pass = 59.8% Run-Pass-Run = 58.5% Run-Pass-Pass = 64.5% Pass-Run-Run = 59.1% Pass-Run-Pass = 64.2% Pass-Pass-Run = 63.7% Pass-Pass-Pass = 68.2% So even sampling from the actual run/pass distributions, it looks like pass-pass-pass gives you (marginally) the highest probability of making a 1st down. This all assumes the defense is exactly the way it was in actual games with no adjustment. Not saying that's a realistic assumption, but at least now you have some numbers to the probability of making a 1st down given different play sequences and actual run/pass distributions (all this obviously includes cases where you made the first down with only 1 or 2 plays and didn't need more plays).
What counts is the Defenses' expectations. It's what made Zach Thomas great,the ability to diagnose and recognise both patterns and misdirection. Sometimes you use a tactic sparingly because you are bad at that thing. Other times you want to save it for critical situations. For example Marino's fake spike. If that had been used every other game it wouldn't have worked when we needed it. So it depends on the purpose that you want to use play action for. If it's to churn out someextra yards, then you might do it for as high as a percentage as it gives you benefit,but if you want it for big play deep strike plays you use it more sparingly.
May be off topic but misdirection is an equalizer to a defenders athleticism, it is why I want this new coach and QB to run read option, and to evolve around the brilliant formula..everything can be designed from it as long as you have a QB who can execute it and threat the defense..it is why I hated Lazor from the get..
I've played board games most of my life and looking at that raw distribution I would say your optimal strategy to maximise yardage would be around 2 passes/1 run. Once you start going higher than that ratio defenses will adjust, and lower than that ratio is suboptimal. Fortunately for us, courtesy of Bill Lazor, the Dolphins have conducted the experiment to show how Defenses adjust when you start going more that 2/1 pass/run ratio. Firstly let uslook at Tannehill's splits between when he was tied/ahead and when he was behind *(we all know Bill Lazoe went pass wacky when behind) In 2014/15 Tannehill had 438 pass attempts when tied/ahead for 3322 yards. 7.6 ypa In 2014/15 He had 738 attempts when behind for 4931 yards. 6.7 ypa. As a control to check if Tannheill is adversely affected by pressure to a significant degree let's use his 2012/13 stats as a control. In 2012/13 Tannehill had 538 pass attempts when tied/ahead for 3493 yards. 6.5 ypa In 2012/13 He had 534 attempts when behind for 3714 yards. 7.0 ypa. The higher ypa when trailing in 2012/13 is consistent with throwing longer higher risk passes when trailing. 21 Interceptions when trailing compared to 15 when tied/ahead is consistent with this theory. My conclusion is that the drop in ypa in 2014/15 when trailing is more likely to be the result defenses adjusting to Lazor's playcalling than to Tannheill performing badly under pressure. So to check this let's compare Tannehills ypa to his run/pass ratio. Short version: in games where Miami was ahead most of the time Lazor called a relatively high% of runs and games where Miami was behind he called a higher % of passes. Top 8 games of run/pass ratio. 64.8% runs to 47.2% runs. Average 53.4% run Tannehill: 234 passes for 1960 yards ypa: 8.4 Miami rushing: 268 rushes for 1278 yards ypa: 4.8 Middle upper 8 games of run/pass ratio. 45.3% runs to 36.8% runs. Average 40.8% run Tannehill: 286 passes for 2157yards ypa: 7.5 Miami rushing: 189 rushes for 978 yards ypa: 4.5 Middle lower 8 games of run/pass ratio. 36.8% runs to 32.9% runs. Average 34.4% run Tannehill: 292 passes for 1910 yards ypa: 6.5 Miami rushing: 160 rushes for 722 yards ypa: 4.5 Bottom 8 games of run/pass ratio. 32.7% runs to 11.3% runs. Average 25.7% run Tannehill: 364 passes for 2178 yards ypa: 6.0 Miami rushing: 126 rushes for 514 yards ypa: 4.1 In games where Miami rushes 36%+of the time Tannehill's ypa is 7.9. Interceptions 1.9% Rushing ypa is 4.67 In games where Miami rushes less than 36% of the time Tannehill's ypa is 6.0. Interceptions 2.1% Rushing ypa is 4.32 Essentially rushing efficiency doesn't change meaningfully with a change in rush/pass ratio. However a change to pass heavy that is predicable by defenses reduces passing efficiency drastically. The difference in interception% is minimal, so the biggest effect is that opposing defenses can reduce yardage allowed. PS Cbrad, I know this might be difficult but would it be possible to re-run your simulations, assuming same distribution, but with Miami's above 36% and below 36% averages? PPS My view is that the difference rushing ypa is due the effect of several exceptional rushing games (good and bad) than any direct effect from the passing game had on the rushing game.
Another way to look at this however, is that QB17 plays better the more you run the ball, because it's only in the top half where he is performing well, and he gets progressively worse. Middle lower is that 2-1 pass-run ratio, and QB17 is still not performing well.
I think my last post shows a clear relationship. Rushing more doesn't necessarily make the passing game better. Failing to rush will harm your passing game. As for what%, I don't think that there is a magic % number that applies to all teams. I believe that the minimum run/pass ratio to allow your passing game to operate the most efficiently will change depending on your personnel and coaching. In Miami's case over the last 2 years the cutoff was around 35% rushing, and below that number the passing game was harmed.
Not difficult.. just have to divide all values by current mean and multiply by a new one. However, there is one disclaimer here: the average YPA of 6.57 for all passes thrown in 2005 is different than the NFL stated average YPA of 6.8 because they are averaging all teams' YPA while this just combines all data. Anyway, if you adjust just the passing distribution to a mean of 7.9, you get: Run-Run-Run = 53.5% Run-Run-Pass = 63.0% Run-Pass-Run = 63.7% Run-Pass-Pass = 71.3% Pass-Run-Run = 62.6% Pass-Run-Pass = 70.7% Pass-Pass-Run = 70.3% Pass-Pass-Pass = 76.8% If you adjust the mean of the passing distribution to 6.0 you get: Run-Run-Run = 53.1% Run-Run-Pass = 56.9% Run-Pass-Run = 55.6% Run-Pass-Pass = 59.9% Pass-Run-Run = 54.7% Pass-Run-Pass = 59.3% Pass-Pass-Run = 57.8% Pass-Pass-Pass = 63.4% As far as your theory goes, it's certainly possible, but I think you'll need a lot more evidence to make the case because there are a lot of possible explanations there.
Yes and no. There definitely is an effect from Bill Lazor's playcalling becoming predictable. It kind of becomes a chicken-egg argument as to how much of the reduction is due to the QB performing poorly or the OC becoming predictable. Maybe Cbrad has some ideas on whether QB performance and OC playcalling can be separated by the stats more definitively, but I'm at the end of my rope.
Really tough. I mean, you're not just asking for conditioning on play-calling, which one can't do with the stats I've seen. You're now asking for conditioning on the degree to which the defense wasn't surprised by it!! No, I have no idea on how to analyze the stats I've seen at pro-football-focus to get at that answer statistically (though play-calling analysis might help by actually analyzing how random it was.. that definitely can be done).
I think I deserve a plausible not a possible The next place to look for more data would be ypa in obvious passing downs, although that may be skewed by OCs calling plays with deeper pass routes. Maybe second and long would be better than 3rd and long.
I couldn't even imagine a possible solution or partial solution, so that's a lot better than I could do.
Awesome answer! OK.. first thing that would concern me is the "control". You need to get the comparable data for 2012-2013 at the very least. Second thing is the claim that the 7.0 YPA in 2012/2013 while trailing is consistent with more risky passes. Yes, but how do the distributions for passing yards from 2012-2013 compare to 2014-2015 when trailing? Depending on that answer you have evidence or not. There are more things, but those are my initial questions.
Yeah, I mean theoretically you can do it if someone gets the necessary data. Suppose you have a finite number of possible plays that could be called. There are tests for the randomness of a sequence of numbers (each play gets a number) that can serve as a quantitative measure of how unpredictable you are (I actually think sports teams would benefit from using random number generators to select from a set of possible plays tbh). Once you have quantified the degree to which a set of plays was predictable or not, you can track that over games or seasons and quantify the degree to which Lazor's play-calling was random at different times. Well.. then you just look at your favorite stat conditioned on the level of randomness and you have your answer. Problem is of course lack of necessary data, but yes it can be done.
That sounds very similar to the type of things my wife's professor told her when she was getting her Ph.D.
Easy thiings first. Looking at 2012/2013 in a bit more detail. Passing when ahead. 328 completions of 538 passes for 3493 yards. 13 Ints. 6.5 ypa 60.9% completios 2.4% Interceptions Passing when behind 309 completions of 534 attempts for 3717 yards. 21 Interceptions 7.0 ypa 57.9% completions 3.2% Interceptions More yards/attempt with a lower completion% and higher interception%. I would submit this as a textbook example of higher risk passes being called.
It's no-holds barred once you try to publish something claiming you have a better method than someone else or their "discovery" really wasn't. The level of evidence required can be huge, especially when the previous method/claim has already been accepted. It's worth it though, even if progress is so slow nowadays in science (problems on average are just becoming too difficult).
Yeah I'd agree, I just think that leaves open a bit too many possibilities (but yes your theory is plausible). Personally, I'd like to see the distributions of passes, but I know of no way to get that without play-by-play parsing.
The inherent problem with this is, who is responsible for it being predictable? The D may just be reading QB17, he may be the one mainly responsible for predictability, and that may not only be on field, that may be due to his limitations at the position, so certain types of plays, in certain situations may get "pidgeonholed" because of that. Then there are the rest of the players, is there a certain "body english" that Cameron gives off when he has to block, or when the play goes to him, and for any other player, like the RT lining up a little more turned to the right. These are things that are studied by teams, just like poker they are looking for tells, so it becomes difficult trying to identify just what may be causing it.
No, I meant the sequences of play-calls themselves could be highly random or not so much. So you don't even need to look at the game to determine that, just the sequences of play-calls. It's like saying one sequence of 1's and 0's can be summarized more easily (and thus the next number in the sequence is more easily predictable to a computer) than other sequences. For example, you can summarize 10101010... more easily than 110100001 (you need fewer "bits" to represent the first string). So that's an objective measure and it's clearly due to Lazor. Then, you can see how other stats depend on that. Maybe for the same level of randomness in play calls you see one QB perform better than another. That's a starting point for saying maybe one of the various things you mentioned might be going on.
The AB test for Tannehill has been done, with his performance under Sherman. If he was consistently giving more clues that tipped off opposing Defenses when behind then you would expect an across the board decrease in his stats, especially YPA and int%. Instead you get a variance, under Sherman, more in line with higher risk/higher reward passes being called when behind. Under Lazor the int% between being behind and being tied/ahead is for practical purposes identical. Which is inconsistent with Tannehill tipping off defenses more than usual when he is behind. I'm not suggesting that RT17 never tips off opposing Ds with his tells. What the data suggest is that the extent of whether or not he tips off the opposing D doesn't change significantly according to game situation. This might change under the very specific case of being behind with 2 or 4 minutes left since we know RT17's performance in that specific case is sub par, but in the general case of being behind the data doesn't show signs of RT17 tipping off defenses more than he usually does.
Ah, but sequence is dictated by down and distance, field position and weather, and also by personnel, as in who's available, and what "packages" are on the field, O and D. So we are talking about a massive amount of data that would have to be taken, every single play, then the down and distance, personnel, weather, field position, then figure out how many times in similar circumstances, the same play was called. It's tough enough just trying to predict run vs pass on many plays, never mind what particular plays would be called when and where. Obviously it's easy enough to predict a run on 3rd and short, or a pass on 3rd and long, but the real trick is that you're dealing with people making these calls on the fly, so the unpredictability of the human mind is an X factor that can't be quantified. You may be able to come up with avgs, but in the long run they wouldn't be reliable, as every game itself is different.
I understand what you're saying, but you aren't accounting for QB17s experience, comparing his rookie and 2nd year to his 3rd and 4th year isn't really a fair comparison in regards to that.
Understand and valid point. That's why I was more concerned about differentials (how much things change according to situation) rather than amounts (how much there is). If it's something that's an inherent trait, maybe he stares down receivers more than the average QB when under pressure, then I would expect that you could find a pattern in variance according to situation even if its a part of a game that he improves over time. You also have experienced gained by RT17 and experienced gained by opponents facing him. The well known sophomore slump for example.
I'll tell you what Pauly, between you and Brad, I seriously doubt that you could find another fan site with statistical theory and analysis to rival you guys. It's not just food for thought, it's a smorgasbord, pretty amazing actually.
I would really be surprised if he goes to a or runs a read option offense. I think he has a clear idea of what he wants to do in his offense, and since he has not done the read option in the past, I do not see him using it here. He wants a QB that can pass.
Oh I disagree completely. The average level of randomness is probably highly dependent on who the OC is. All those factors you mention are being processed by the same brain, with the goal of choosing a play that helps win the game the most. You're going to get very idiosyncratic behaviors here (and very stable). No, in fact this would be one of the few tests that actually isolates the contribution of one person on the team. Even if the HC designs the overall game plan, the randomness of the sequence of specific plays will tell you something about the OC. btw.. one thing that's important to know here is how predictable human behavior is, probabilistically that is (so not precisely what they do next, but what they tend to do over some period of time). Whether it's humans choosing "random" numbers, deciding which paths they take, strategies used in competitive games, or countless other stuff that has been studied by all kinds of researchers, you can often get a computer to predict those patterns with high probability (often better than other humans). So based on all kinds of research already done on how patterned individual behavior is (remember those algorithms are trying to predict each person's behavior), yeah this test would not only work but give you stable results in most cases.
This is a valid counter-argument. It's not valid if Lazor gives a set of options and the QB chooses from them. However, what that does is introduce a correlation between the contribution of the OC and the contribution of the QB to the final play called. Not much different than how QB and WR combine for YPA, so it's still a useful stat. But yeah.. good job in identifying a case where the method would not capture what it's intended to. You'd have to then look at how relevant that really is in practice. I mean if it happens only 10% of the time, then the stat is 90% as valid as it would have been otherwise, which is still pretty damn good.
Well I'm not sure random is what we're looking for here, I don't think there is any random in it at all, we're talking about predictability, no one randomly chooses plays, there is a definite purpose for every play called, it whether or not it's predictable. Predictability of play calls would still have to entail all the things I mentioned, for instance; a 3rd and 2, on the 50yd line, in the 4th quarter with 5:00 mins left in a tied game, vs the Pats on a hot day in sept in Miami with a full squad. a 3rd and 2, on the 50yd line, in the 4th quarter with 5:00 mins left in a tied game, vs the Jets on a cold, wet and windy day in NJ with injuries to certain players. Even though down, distance, field position, time in game and score are all identical, these are different situations and therefor could not have similar predictability expectations. The complexity of the game, combined with weather, opponent and player availability, make trying to find an answer difficult. Some coaches will be very predictable, and some coaches have had a lot of success even being predictable, especially when you have a great team, that makes it much easier to be predictable, because you're simply better than the other team. There is a certain amount of predictability that even a casual fan picks up on, but overall, trying to pin that down with a comp wouldn't be very useful imo, because what would it actually show you that you don't already know. Who was guessing a pass at the end of the Seattle, Pats SB? lol.
Yeah, gotta be careful we're not talking past each other here. First of all, if something is not perfectly predictable (deterministic, given the information we have), then there is by definition a level of randomness in it. So talking about randomness is almost by definition the correct thing to do. And there are ways of measuring that. The basic theory (most commonly accepted) is based on something called Kolmogorov complexity that essentially asks how many "bits" does it take to represent a sequence (a sequence of numbers that has a smaller summary representation is less "complex" and less "random", i.e. takes fewer resources to predict). Now you mention all these things about play based on down, distance, weather, etc.. but that's true for practically every summary stat like YPA, passer rating, etc.. There's nothing fundamentally different here. Just as those summary stats are useful and don't require astronomical amounts of data, you won't need it here. Also, note that we're talking about the randomness of the sequence, which can be very stable across different conditions (unlike YPA or passer rating). For example, suppose there are three types of plays: A, B and C. Now, suppose in one situation you call AAB, in another you call CCA, and in yet another you call BAA. The level of "randomness" is identical (in this example) even though the plays and situations are different. YPA or passer rating will often change in those situations even when "randomness" will not because randomness depends more on how a person processes information and there's a lot of research showing people are much more predictable than they think. Oh, and the utility of measuring this is you have a way of objectively quantifying how predictable for example Lazor was relative to other OC's. You can't do that if you just use a subjective thought process.
