Very thorough and great statistical work, but it's all arguing against a weak premise.
His insight was that 2nd and 10 disproportionately followed an incomplete pass. This generated two hypotheses:
Coaches (like all humans) are bad at generating random sequences, and have a tendency to alternate too much when they’re trying to be genuinely random. Since 2nd and 10 is most likely the result of a 1st down pass, alternating would produce a high percent of 2nd down rushes.
Coaches are suffering from the ‘small sample fallacy’ and ‘recency bias’, overreacting to the result of the previous play. Since 2nd and 10 not only likely follows a pass, but a failed pass, coaches have an impulse to try the alternative without realizing they’re being predictable.
The argument boils down to coaches trying to catch the defense off-guard; they are either alternating to be random or "overreacting".
It may not be that complicated. Coaches hate 3rd and 10 or more specifically third and long. Analysts emphasize it every game and the reason why is because at that point there is no randomness. You pretty much have to pass on 3rd and 10 and the defense knows this, not to mention a play that requires ten yards takes longer to setup than a play which takes four and the defense is already rushing your QB...
If you have an incomplete pass on 1st and 10, you pass again on 2nd and 10 and have another incomplete, you're now looking at passing on 3rd and 10: 3 passes in a row, no time burned off the clock (time of possession is also highly valued), and no momentum - you've just fucked up on the same play two times in a row and you've got to run it (well pass it) again.
You run on second down, you'll usually at worst get a yard or two, but you very well could have a good run and you're then looking at 3rd and 6, third and 4, situations that open up your playbook a lot.
This might not be the reason coaches do it, but it's certainly a possibility. They seem to get into an argument of causality when they've got some interesting correlations. But I just armchair quarterback so don't mind me.
Yep.
From the beginning, the assumption that the coach or OC "randomly" calling plays is a particularly important goal is pretty flawed.
In high quality NFL offenses the QB will walk up to the line with 2-3 plays of varying types available and make a last-minute decision depending on how the defense sets up.
And surprise comes less from which play is called and more from showing run and then passing, or vice versa.
There is so much more that this doesn't account for, like matchups and how defenses/offenses handle runs or passes.
And unless any of this analysis is actionable, it doesn't really matter.
Does this mean an opposing defense can accurately predict when a run or pass will occur?
Will they actually be able to act on that knowledge?
Unclear.
If the fear of 3rd and 10 was the main factor, then you wouldn't see such an uneven distribution based on whether the 2nd and 10 came via incomplete pass, or run for no gain.
The difference is if a pass fails, you gain nothing, whereas even a mediocre run will probably give you at least a couple yards. So, if your goal is to avoid 3rd and 10, a run is more likely to accomplish this than a pass, even though a successful pass has a much higher ceiling in terms of yards gained.
Yes, I agree, but later in the article they show that if the second and 10 is due to a run on first down, they are more likely to run on second down than if they passed on first down. I.e., the coaches are looking backwards not forwards when they make the decision.
> they show that if the second and 10 is due to a run on first down, they are more likely to run on second down than if they passed on first down
I think you got this backwards. If the first down was a run, they were far less likely to run on second down compared to when the first down was a pass.
The other possibility is that the offense is reacting to what the defense is giving them. On a second-and-long, the defense might be anticipating a pass and responding with pass-defense personnel (lighter edge rushers, nickel/dime personnel substituting for a heavier lineman). The offense might then chose a run or draw play since it would be easier to run against smaller defensive personal.
Man, I am working on a project just like this, but with high school teams.
The problem is that these guys seem to lack how coaches make decisions when calling plays.
I have a few problems with their research:
1) There are A LOT more variables that go into coaching decisions. They picked QB completion percent as the one they correlated. This is just a way over simplification of the process.
2) They applied their findings on the whole league to an individual coach. They is too much variance between coaches that you just can't do this and be sound.
3) Even if they are right 90% of the time, this may not even be good enough. Football plays are sort of like no-limit poker. Any one play can determine a game. So, if you are wrong 10% of the time, this still may result in giving up a touchdown because you relied on the probability that they would pass/run the ball.
4) Coaches are humans and they can very readily make adjustments to their biases. I have seen it myself when coaching. A coach realizes he has called 55% of his plays to the right, so he ends up calling the next few plays to the left. All these tendencies are shown to coaches every week, as a result, they work very hard to counter act their own tendencies.
For me, more promising is using their anti-biases against them. In a truly random pattern, seeing a run of five plays to the Right isn't that unusual. However, in football play calling, you won't see this nearly as much as you think you would.
That's my idea. I am still working on the math, but it might take awhile to be usable.
Very thorough and great statistical work, but it's all arguing against a weak premise.
His insight was that 2nd and 10 disproportionately followed an incomplete pass. This generated two hypotheses:
Coaches (like all humans) are bad at generating random sequences, and have a tendency to alternate too much when they’re trying to be genuinely random. Since 2nd and 10 is most likely the result of a 1st down pass, alternating would produce a high percent of 2nd down rushes.
Coaches are suffering from the ‘small sample fallacy’ and ‘recency bias’, overreacting to the result of the previous play. Since 2nd and 10 not only likely follows a pass, but a failed pass, coaches have an impulse to try the alternative without realizing they’re being predictable.
The argument boils down to coaches trying to catch the defense off-guard; they are either alternating to be random or "overreacting".
It may not be that complicated. Coaches hate 3rd and 10 or more specifically third and long. Analysts emphasize it every game and the reason why is because at that point there is no randomness. You pretty much have to pass on 3rd and 10 and the defense knows this, not to mention a play that requires ten yards takes longer to setup than a play which takes four and the defense is already rushing your QB...
If you have an incomplete pass on 1st and 10, you pass again on 2nd and 10 and have another incomplete, you're now looking at passing on 3rd and 10: 3 passes in a row, no time burned off the clock (time of possession is also highly valued), and no momentum - you've just fucked up on the same play two times in a row and you've got to run it (well pass it) again.
You run on second down, you'll usually at worst get a yard or two, but you very well could have a good run and you're then looking at 3rd and 6, third and 4, situations that open up your playbook a lot.
This might not be the reason coaches do it, but it's certainly a possibility. They seem to get into an argument of causality when they've got some interesting correlations. But I just armchair quarterback so don't mind me.