Dallas Cowboys Film Study: Second Down Play-Calling

In our first statistical analysis of Jason Garrett’s play-calling, we noticed that he was tipping plays via the formation .

As we studied the Cowboys’ game film this offseason, we also noticed a play-calling trend on second down.  On the majority of second down plays, it appeared as though Jason Garrett called a run if the team passed on first down, and vice versa.

We looked into our stat database and the results are shown below.

 

 

As you can see, our hunches were correct.  Other than on 2nd-and-1 or 2nd-and-2, the Cowboys ran significantly more after calling a pass on first down.  There were actually only 12 plays called on 2nd-and-1 or 2nd-and- 2 after a first-down pass, so that sample size is probably too small to make significant conclusions.

In the other scenarios, though, the sample size is plenty big enough to conclude that Garrett was tipping plays via the down and distance.  On 2nd-and-3 through to 2nd-and-7, for example, Garrett dialed up a run on only 23 of the 78 (29.5 percent) plays that followed a first-down run.  After first-down passes, though, the Cowboys ran on second down on 26 of 34 plays (76.5 percent).  Thus, Dallas was 2.95 times more likely to run on 2nd-and-3 to 2nd-and-7 after a first-down pass than after a first-down run.

On 3rd-and-8 to 3rd-and-10, that trend, surprisingly, did not get much better.  The team ran on only 10 of 50 plays (20 percent) in these scenarios following a first-down run.  After passes, Garrett called a run on 32 of 58 second down plays (55.2 percent), meaning the team was 2.76 times more likely to run on 2nd-and-8 to 2nd-and-10 after a pass than a run.

On 2nd-and-11 or more, the team was still 2.33 times more likely to run after a first-down pass than after a run.  Obviously Garrett did some things right in the past few years, but this sort of predictability is unacceptable.  Perhaps the offense’s success in recent years is not because of Garrett, but in spite of him.

If we are obtaining these numbers and noticing these trends, you can bet opposing defensive coordinators are aware of them.  Imagine if the Cowboys passed for five yards on first down and lined up in Double Tight Right Strong Right on 2nd-and-5.  The defensive coordinator could be all but sure that the Cowboys would run up the middle.

So why is Garrett so predictable in these situations?  Our hypothesis is that, ironically, he is trying to be unpredictable.  In an effort to seem “random” in his play-calling, Garrett is “mixing it up.”  True randomness, though, means each event is independent of the previous one .  What is the chance that a flipped coin will come up heads after it came up heads six straight times prior?  Still 50 percent, because the previous events have no bearing on future ones.

While previous football plays do have an impact on future ones (a team is more likely to pass on second down after a sack than a nine yard rush), if Garrett wants to become unpredictable, he must not allow the previous play call affect the current one.  In situations such as 2nd-and-3 through 2nd-and-7 when a team is probably about as likely to pass as it is to run, Garrett is particularly predictable. 

He thinks he is “mixing it up” by calling a play that is the opposite of the previous one, but in doing so, he is actually dissolving any chance of randomness for which he is shooting.

In fairness to Garrett, he is not the only offensive coordinator that suffers from this delusion.  Studies have proven that it is impossible for human beings to produce a truly random sequence .  We naturally assume that rather long strings of the same occurrence are unlikely in a random sequence, when in fact they are to be expected.

For example, which series of five coin flips is more likely:  “heads, heads, heads, heads, heads” or “tails, heads, heads, tails, tails”?  The answer is that they are both equally likely, yet this doesn’t seem to mesh well with common sense because of our notions of causation.

It is nearly impossible for humans to separate previous events from current ones, a task that is imperative to create randomness.  Thus, almost paradoxically, for play-calling to be as random as is humanly possible, the play-caller would have to try to forget about being random. 

He would have to implement an almost “non-aware awareness,” meaning he has to be aware of natural human tendencies concerning randomness without letting this awareness adversely affect his own ability to randomly call plays.  This is evidenced by the fact that it is Garrett’s attempt to call plays randomly that is hindering his ability to do so.

We doubt many offensive coordinators take this approach when calling plays.  Still, the failures of other coordinators do not justify those of Jason Garrett. 

If the Cowboys want to maximize the productivity of their potentially explosive offense, Garrett is the first person that needs to change.  Unfortunately, if his play-calling does not become less predictable, neither will the team’s fate in the playoffs.