Each week, we’ll highlight a few of the most important coaching decisions that contributed heavily to the final result. While many of these high leverage situations occur on fourth down, we’ll introduce some other situations where a coaching decision swung his team’s chances of winning in a major way as the season goes on.
There are a few main tools for doing this analysis: The most obvious is win probability, which takes into account the situational components of the game along with the strength of each team in each facet to predict the chance of winning. Another data point that is useful and in most cases aligned with win probability is expected points, which takes into account the field position, down and distance to predict the number of points that we expect the offense to score on the next offensive play. We can also compare the expected points for a team before and after a play to measure how successful a play was. We’ll also make use of conversion probability, which leverages the strength of each team in each facet to predict the chance a team picks up a first down on a given play.
During the season, more analysis may work its way into this column, but this gives us plenty to start with. While we leverage machine learning to make sound predictions for each of the above, we will do our best to make this article accessible to all, and we’ll do our best to answer any questions.
The decision of the week
John Harbaugh going for 2 down 8
Sunday was not the best of days for John Harbaugh and the Baltimore Ravens. Big favorites to win the game at home and come out nearly a two-thirds favorite to win the AFC North a short four weeks in the season, Baltimore fell 40-25 to the previously struggling Cleveland Browns.
However, the Ravens’ second loss of the season comes with their second spot on our AWS Decisions of the Week column, as John Harbaugh went with the math — and against intuition — after a Lamar Jackson touchdown pass to Mark Andrews cut the Browns' lead to 24-16 with just under 10 minutes remaining in the fourth quarter. Instead of kicking the PAT and going down seven (as almost everyone else does), the Ravens head coach called for the team to go for two, and after (another) penalty on Cleveland’s Myles Garrett, Mark Ingram II plunged into the end zone for the conversion, cutting Cleveland’s lead to six.
As our own Kevin Cole has written about in the past, this is clearly the right move for the team down two touchdowns. While this decision was not as asymmetric as some, the Ravens would have only needed to have a baseline conversion rate of over 52.8% for this decision to “break even” (we’ve estimated their base rate to be about 57%). A conversion by Baltimore gave them a win probability of 38.6% at the time, needing just a stop, a touchdown and a PAT to go ahead. If the Ravens would have kicked the PAT, their win probability would have been 36.1%, needing a stop, a touchdown and the choice of a PAT or a 2-point conversion to win or lose the game in regulation or send it to overtime. If the Ravens had missed the conversion, their win probability would have fallen to 33.3%, but they would have known a priori that the next touchdown required a two-point conversion to send the game into overtime. In addition to giving his team a better chance to win, Harbaugh compressed his team’s future uncertainty, opting to determine how many points they would need on their next score well before it ever (if it ever) happened.
While, for the second-straight week, the Ravens were not paid off for their shrewd decision making, the process was good and bodes well for them as they try to dig their way out of what is now underdog status in the AFC North.
The Decision that wasn’t
Kellen Moore and Cowboys Stop Running Play-Action in Primetime
Going into Week 4, all anyone could talk about was Kellen Moore and the Cowboys' new-look offense. There was a good reason to do so, as the Cowboys entered Week 4 with the third-best expected points added per play average and second-best rate of successful plays (plays that increase the number of points we expect them to score on the next scoring play). A particular area of improvement was the play-action game, which the Cowboys were using on 39% of their passing plays (second-highest, and 16% above last season’s rate) entering the Sunday night tilt against the Saints and Teddy Bridgewater.
Running a play-fake on nearly half their passes coming into the week, the Cowboys were remarkably successful; they averaged 10.1 yards per play-action play (4th), and Dak Prescott ranked second in completion rate and third in passer rating when using the play-action fake. Despite all of this, the Cowboys used play-action just seven times — accounting for 20% of pass plays — nearly half as frequently as they have through the first three games. It was not as if the play-fake wasn’t successful, either, as Prescott went 6/7 averaging 9.3 yards per play taking no sacks, three more yards per pass than without play-action. On the lone touchdown drive of the game, the Cowboys' first drive out of halftime, Dak dropped back to pass six times, two of which were play-action dropbacks that resulted in two completions for 44 yards. The Cowboys ran play-action only five times outside of that drive and ultimately lost to a backup quarterback who was unable to lead his offense into the end zone.
One might say that play-action opened Dak up to pressure. This is just incorrect; play-action leads to more time spent in a clean pocket when accounting for the designed dropback depth (between 0.2 and 0.4 seconds more per dropback). In this particular game, Prescott was under pressure on 41% of his dropbacks without play-action and 43% of his dropbacks with play-action (though he didn’t take a single sack and went two for three on pressured play-action plays). Further, there is no mathematical evidence that rushing success leads to improved play-action success (https://www.footballoutsiders.com/stat-analysis/2018/further-research-play-action-passing), and this game was a great example. The only failure was simply not running a key component of your offense enough, and no, I’m not talking about the Cowboys' most efficient runner, Tony Pollard.