Risky Business
EdjSports examines critical decisions and their impact on GWC (Game-Winning Chance).
Amari Cooper
Photo: USA Today Sports Images

As the 2020 NFL season gets underway this weekend, analytics are a hot topic. Many teams are putting increased effort into understanding how to optimize GWC (Game-Winning Chance or win probability) – particularly at highly leveraged decision points. However, being equipped with the math only matters if you have the courage to deploy the proper tactics. Often these tactics can appear controversial, and when they don’t result in a win, it can be a very bitter pill to swallow.

Such was the case last night when the Cowboys faced the Rams in LA. Two pivotal decisions defined the 4th quarter and illustrated a stark difference in coaching philosophies.

  • The Cowboys’ Dak Prescott came up just short of a first down on a 4th and 3 pass attempt from the Rams 11-yard line when trailing 20-17 with 11:46 remaining in the game.
  • Facing a 4th and 1 from their own 49-yard line with 2:23 remaining in the game and a 20-17 lead, Rams punter John Hekker successfully punted the ball to the Dallas 9-yard line.

On the surface, we can look at these two key decisions and construct a line of reasoning that suggests:

  1. The Cowboys were too aggressive and failed,
  2. The Rams smartly executed a great punt,
  3. The Rams deservingly won the game.

However, in the cruel world of analytics, good decisions do not always get rewarded and poor ones sometimes do. Let’s see why this was precisely the case last night.

Extensive, customized simulations by the EdjFootball model indicate Mike McCarthy’s choice not to kick the field goal improved the Cowboys winning prospects by 3.5% GWC.Additionally, knowing in advance that the 4th down pass attempt was justifiable adds credence to what may have appeared as a questionable rushing attempt on 3rd and 6.

On the other hand, Sean McVay’s decision to punt on 4th and 1 was a very large blunder. On average, this choice will cost the Rams 18.5% GWC! An average NFL offense would be expected to convert this first down more than 70% of the time. The Rams expectations would likely exceed those levels. And with Dallas down to one timeout, a successful rushing play would have essentially put the game away.


This decision should have been clear for McVay; the Rams were risking 17% GWC to gain 30% GWC. This means the required success rate of the first down attempt is only 36%:

\frac{17}{(17+30)} = \textbf{36\%}(17+30)17​=36%

If they knew the punt would be perfectly executed, they could still have justified going for the first down, even as a significant underdog to convert it.

These two decisions resulted in a net advantage to the Cowboys of 22% GWC. It may not have made the difference in week 1, but over the course of a long season this type of decision can accumulate to make or break playoff prospects.


Presented by:



5 comments, Last at 15 Sep 2020, 11:59am

1 Intuition.

Analytics is only a tool.  Those of us who work in the Mental Health field know this.  Empirical evidence in and of itself doesn't do anything but inform you of how to practice, whether you're a Psychologist or Bettor.

McCarthy is too conservative generally, which is why he was fired in Green Bay.  In those handful of plays he needed to be aggressive, he was conservative.  That is what he is.  

The Rams won the game as much as McCarthy lost it by staying true to their edict.

If the Rams had lost that game, it would have told me more about them than the Cowboys.  

2 Yes, analytics is only a…

In reply to by DIVISION

Yes, analytics is only a tool. Empirical evidence is better... but analytics is the source of the only actual evidence we have.

McCarthy ranked highly in Aggressiveness Index on 4th, in tendency to pass more than run both overall and while ahead, and aggressiveness in playcalling in recent seasons. He was "conservative" in some ways not adding very many changes to the scheme, but I think that is overblown. McCarthy was booted because he lost the confidence of the QB and management, while being handed 0 weapons with which to work.

4 McCarthy...

....for me is just another version of Jason Garrett.  I don't see any difference.  I also predict he will suffer the same fate in a few years.  JJ doesn't want someone who will garner more attention than himself.

 Cowboys fans are trapped in a dead-end marriage going nowhere...

3 but probability

problem is that evaluating analytics against a binary set of outcomes is in itself flawed.  you can hit on a hard 18 and get a 3, does that make the decision good?  of course not, because you were very likely to lose.  how is this so hard to understand? 

edit: my bad, that seems to be precisely the premise of the article. serves me right for headline reading.

5 "over the course of a long…

"over the course of a long season this type of decision can accumulate to make or break playoff prospects"

"can", yes, "will", not so certain.  NFL season's are extremely short.  The cumulative effect of good or bad in game management decisions might never show up in a statistically meaningful way, even over multiple years.

Compared to other coaching duties - pre-game preparation and in game playcalling - this aspect of coaching is, I suspect, of little consequence to a coach's overall effectiveness.  It's easy to get caught up in "too conservative", "too aggressive" - hey, this is a stats site, it's what we love! - but I often think that's just a stick to beat the coach with for failures in their other, more impactful coaching duties.  Maybe poor decision-making here is indicative of poor decision-making in other, less easily observed areas, but that'd be speculative and difficult to confirm.