WP: Better Play Calling with Twister

This week's post at the Washington Post's Redskins Insider site takes a look at tendencies of NFL signal callers and Kyle Shanahan in particular.

The laws of zero-sum game theory say that there are only two requirements for optimizing payoffs. First, plays should be called relative to their expected effectiveness. Offenses should proportionately favor their own strengths relative to its opposition’s weaknesses. And second, play calls should be unpredictably random. There could be no play caller in the league as unpredictable as a spinner from the game Twister. As silly as it sounds, whenever people are making decisions, they can’t help but create predictable patterns — patterns that can be exploited by the opposition. But even the most prescient defensive coordinator in the world can’t predict a Twister spinner.

  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

20 Responses to “WP: Better Play Calling with Twister”

  1. Anonymous says:

    What are your thoughts on creating a gameplan based on your strengths and opponent weaknesses, then in certain situations like 1st and 10, have a 60-40 pass ratio. But have some random calculator determine if you will pass or run, then call that play no matter what. Then adjust the situation like say 3rd and 10 with a 95% pass ratio. It's impossible to give away tendencies because your playcalling is completely random.

    This would work even better in baseball. Just imagine if a pitcher called his pitches based on strengths and batters weakspots, then randomizes his pitch selection in certain counts.

  2. Anonymous says:

    Definitely an interesting article. A few additional thoughts:

    -As maniacal as NFL coaches are in wanting to control every possible situation, I can't imagine anyone giving up such control.

    -A coach would have to be smart enough to know when to go away from the spinner (sucks if it keeps spinning Run during a 2 minute drive) or how to modify it during the game to reflect changing conditions (big changes in weather, injured players, etc)

    -Shouldn't these Tendencies be discovered when teams Self Scout themselves which should smooth out such trends over the course of a season (although not necessarily within a particular game)

  3. Adam Davis says:

    Years ago when I was frequently playing a football video game, I noticed right away that my ability to call "random" plays was nothing of the sort. I would start a game thinking that I should have a certain mix of plays, and when it was finished I had allowed my own sense of randomness to warp this outcome entirely.

    To fix this, I built a model that said on X down with Y distance to go, I wanted to pass Z% of the time. I then literally used a random number generator to determine run/pass mix. The process was extremely effective and took almost no time at all. The random number generator was the same every time - all it had to do was randomly generate a number from 1-100. So there's no reason why an offensive coordinator sitting in the booth couldn't do the same kind of thing. He would see that it's 2nd-and-7, he would know from a chart that he had created before the game that he wants to pass in 60% of these scenarios, and then he would hit the button on his generator (it would be an extremely simple app for a mobile phone). The number would come up, say, 32, and he would then proceed to pick a pass play.

    This would still leave *almost* all of the play calling in the "human element". Every team has dozens of different run/pass plays to choose from. And I would never propose that anyone uses a random number generator to randomly pick the actual play. I just think it would be waaaaay more effective if they used a random number generator to pick the basic type of play (run or pass).

  4. stevekirsch says:

    I've mentioned on here before that predictability is a death sentence in the NFL. The level of parity makes it difficult to execute against a defense that knows what's coming. The days of imposing your will on a team are nearing an end. Coaches need to do the NFL equivalent of pitching a batter backwards every once in a while. Part of the Tebow effect (and the wildcat before it) is that defenses just don't know what to expect and that give the offense an edge.

    As for the spinner, I think it needs to be a bit more complex than just run and pass. Deep passes are defended differently than short passes. Spread run are different than power runs. All runs and passes are not created equal. OCs should make sure that they are attacking all areas of the field and defense.

    Furthermore, formations could be varied as well. Plenty of teams run from the spread, but more could throw from 2 TE sets. Teams could also mix up what plays they call based on field position. Even individual players routes could be tweaked. Send Wes Welker deep once in a while. There are a ton of variables to play with here and NFL coaches should be looking to make themselves much less predictable.

  5. Boston Chris says:

    Just a thought, but if you did this, and advertised that this is how you were going to call plays...it might give you a further advantage on the defense. Now that the defense knows that you are playing some preset mix, they can no longer try to "guess" what is coming. Therefore, they start having tendencies that might become obvious in how they defend. Allowing the offensive playcaller at opportune moments to perhaps override the random generator, and call plays based on the defensive tendencies against a random offense. Maybe I'm over thinking it....

  6. Bob says:

    I suggested this for tennis a while back:


  7. Anonymous says:

    I enjoy playing table top football board games. These games usually overemphasize the effect of play-calling in determining play result, with some games, like Football Strategy, going as far as having the play result determined by cross-indexing the offensive play with the defensive play - with no random element.

    To play the games solitaire, I have created play-calling charts, where the play is selected by a 2d10 dice roll (I prefer dice to electronic random number generators). In these games I always find it easy to outperform the random playcaller, since I then don't have to worry about revealing my own trends (since I know the random dice cannot pick up on my trends - so as a defender I simply call the play the yields the lowest YPA - except on 3rd down when I call the play with the best chance of stopping a 1st down.

    For example, against a "random" offense that is going to go 60/40 pass/run, I'd always defend against the short pass on 1st down. Occasionally I'd get burned on a run from time to time, but I'd "know" that the offense is not going to run it down my throat forcing me to step up and stop the run. In other words, if one "knows" that the other side is random, one can allow oneself to become entirely predictable.

    Of course in real life, one could never trust the other side to stick with its random playcall strategy so (a) one would still try to mask one's own trends and (b) one would waste time trying to predict the other side's trends.

  8. Jeff Fogle says:

    Would a twister spinner work with chess? Could an opponent successfully defend against a computer using random chess moves? Could an opponent successfully launch attacks against an opponent who was using random moves.

    How much of pro football is chess on turf?

    If an offense tries to "take what the defense is giving," wouldn't randomness lessen the ability to do even that much?

  9. ff says:

    Jeff, I'm no chess expert but I doubt a random computer could succeed in Chess. You got supergenius computers making right moves that lose to top level players. If the random move is bad, it's kinda hard to recover from them because an expert can exploit that situation.

  10. Anonymous says:

    Chess is different from football. In football, both the offense and the defense must decide at the same time what play to make. In Chess, only one person moves at a time, so a random move has no "surprise" value.

  11. Jeff Fogle says:

    Agree with you ff. And, anon, I don't think football is as cut and dried as both teams deciding "at the same time what play to make." Defenses are moving around before snap trying to disquise their intentions...QB's can audible...QB's have check-down options based on how defenses are covering a play as it unfolds...some could argue each play is speed chess...or, at least many plays are speed chess when a defense is attacking and offenses are trying to defend against the attack. Each side is reacting to that the other is doing in real time both before and after the snap.

    Can see randomization possibly leading to optimal performance against a very passive defense that's just trying to tackle after they see where the ball is going. Tougher to be random against a defense that's launching missiles at your QB's head.

    Elements of both sides are "under attack" on a given play. It looks like evolutionary force has led to giving a QB options to quickly consider under duress rather than just hoping a random play call hits the mark.

    Being predictable leads to problems...but being random may just lead to different problems rather than providing a solution. A QB with options at least has the ability to read and react.

  12. Boston Chris says:

    Jeff, that sounds good, but defenses have already responded to this "read and react" from the QBs. As soon as the QB audibles they change defenses. The offense is the team fighting the play clock, so they end up losing this battle, b/c the defense can always respond to the offense changing its look (or its call). The offense would run out of time trying to make a call based on the "last move" of the defense.

    In fact, the response to this is already being found. Some of the top offenses are no running the super-hurry up. If you randomize the play-calls out of a super-hurry up, it can be quite effective as the Patriots have shown when they go with this look this year. In fact, they probably get the most yards of any team using QB sneaks on plays that aren't "...and 1" to go situations, using them as one of their possibly plays in the super-hurry up.

  13. Jeff Fogle says:

    Fun and interesting to think through the possibilities BC. Do you think the QB sneak would be most optmized if run randomly? Or, if Brady reads the opening in the middle of the field and scoots up the middle based on his read?

    Wouldn't running it randomly sometimes get into trouble? But, running it when Brady reads the opening will almost always work for a few yards?
    Is assigning random plays to Tom Brady better than trusting his very experienced reads?

  14. KJ says:

    Thinking about this is analagous to thinking about poker strategy (in which more plays than you'd think devolve into randomization (I usually use the seconds indicator on whatever clock I can see as the input). Randomization, as a mechanism to avoid predictability, is a very good ploy in poker, particularly when playing with few people or even 1 opponent (less information available).

    All that said, I think the better plan by far would be what I expect coaching staffs are doing. That is, spending as much time as possible determining the patterns of their opponents, and removing patterns of your own as much as possible. This would also be a capital plan for poker as well. The difference is that one has so very much more information available on past actions in football compared to poker.

    This is a rather long-winded way of saying that randomization in the sense we're talking about is a defense mechanism, not an offensive plan - rather the disguise of the offensive plan (not using those terms in the sense of football offense or defense mind you).

  15. Anonymous says:

    How could you write an article at Washington Post without mentioning the play-calling tendencies of the Steelers (my favorite team)? What column is complete without mentioning the best franchise ever? Steelers.


  16. JamesM says:

    A few comments firstly surely the game theory optimum mix is only optimum in the sense that it means the opponent cant exploit your strategy so if you run too much on 2nd and 10 the defense can cheat towards the run if you pass too much they can cheat towards the pass and so it is only when your mix is optimal that the defense cant get an edge (the same applies to the defense as well) however if your opponent is not at the right ratio of pass/run offense or defense then it makes sense that you abandon your optimum and exploit their error so if they keep running on 2nd an 10 then the defense should switch to more run defense even if game theory says this is not the right mix and could be exploited by better play calling.

    Another issue is that for offenses it is a qualitative decision between pass/run whereas a defense is more of a quantitative shift from pass to defense except for obvious run or passing downs. So whereas on 1st and 10 in midfield the offense should pass x% of the time and run y% the defense shouldn’t be in nickel X% and goalline package Y% of the time as this will lead to an audible rather they should have a mixed defense. Does this mess up the simple game theory ideas. As instead of pass/run vs. pass defense/ run defense it is pass/run vs stronf run D–medium run D–slight run D–balance D–slight passD etc etc

    Finally I read somewhere that computers are banned from sidelines and coach’s boxes during a game so they cannot use a computer to play call. Although imagine if they had a twister spinner (or better still the ultra-nerdy 10 sided Dungeons and dragons dice) in the coach’s box, spun it, looked up the chart and then called the play the outrage would be unbelievable even though it is perfectly correct. But surely the laminated sheets where they have the plays written for each situation could be altered so that you have the first play to be called on 1st and 10 and then the second etc etc listed in order after you randomise them, as surely they have so little time they cannot spend long agonising over play calling but have to pick one straight away. The coach would have the option of skipping a play and going to the next one but the default would be randomisation.

  17. Adam Davis says:

    One more thought on predictability: I have often been frustrated by teams' tendencies to use a wide array of formations. The problem is that they use so many different formations - and there are only so many different plays in the playbook - that with good scouting it can be determined sometimes that X formation equals Y play. I'm not making this up - I have heard some players say directly after a game that we knew when they were in a given formation that they were going to run a given play.

    I think there would be great value in having only a handful of formations, chosen because they are versatile enough that you can run a great number of plays from them.

  18. Tomm Caudillo says:

    I've wanted to do this for ages: pull the plays out of a hat. Sure, the realities of the game dictate that some considerations be made...for example, you'd have several hats for different situations: One for first and 10, one for 3rd and short, one for the two-minute drill, etc. Think it out for a while and you can decide how you want to go about it. Maybe your "halftime adjustment" would be putting in a few plays into the hat that you think are more likely to work, and taking out some that you have less faith in now than you had before the game started. In effect, you'd get "weighted randomness". You have most of the plays in the hat be sensible plays for the situation, but there would be a few wacky ones that you'd run every now and then to mess with the defense a little.

    The issue though, really, is the practical application. Let's say it's 3rd and short and you need that yard. So, you reach into the 3rd and shrot hat and you pull out a long bomb. Wait...what?! this is a big game and you can't afford to get this wrong...except, you had to put a few oddballs in the hat to make it random (that's the whole point). As the OC, do you have the guts to call in the oddball play at a moment like this? Hey, it might work. In fact, I'd argue that the odds of a wacky play succeeding go up in high-pressure situations, like the Patriots used to by throwing to defensive players who lined up on offense just now. (That'd be a fun study...how do the odds of oddball plays like onside kicks and flea flickers rise and fall with the game situation?)

    Then again, it might not, and if it doesn't, you'd have quite a storm on your hands wondering why you did what you did, like the cowboys "icing" fiasco" or the Patriots going for it on 4th against the Colts in 2009.

    In poker, sometimes you need to bluff with all your chips with nothing. At least you have no one to answer to but yourself. But in the NFL? Yeah, you've got to have quite a bit of courage to pursue this strategy. It's just the Secret of Football: "It's not about football."

  19. Anonymous says:

    With everyone getting so caught up in tendencies, here is something to consider: On occasion, it makes sense to intentionally set up tendencies and seem predictable.
    Let's say you are Team A with the best record in the NFL. In weeks 6 & 7, you are facing relatively mediocre teams, Teams B & C, while in week 8, you are facing Team D, which is the 2nd place team in the league. Why not establish some tendencies against B & C, and then go against those tendencies against Team D? It will likely throw Team D for a loop, especially if they have over-gameplanned for them.

  20. wangtangkiki says:

    Have you done any analysis on the most predictable offenses? That seems like it could be interesting.. and I'm sure NFL teams would pay a lot for that information lol.

Leave a Reply

Note: Only a member of this blog may post a comment.