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

As boxers know too well, when your opponent is on the ropes, that is the time to finish them off. Texans head coach Romeo Crennel deployed this strategy on Sunday against the Titans by attempting a late two-point conversion that would have essentially sealed the win. However, the Titans managed to duck the knockout blow and come away with an overtime victory. The critical drive for the Texans began at 8:30 of the 4th quarter and found them holding a slim 30-29 lead.

Several key decisions ensued.

 

 

Crennel should be credited with two very good decisions on the 4th and 1s as he picked up a total of 11.7% in game-winning chance (GWC) by avoiding the more conservative field goal kicks. However, it was his decision to go for 2 that is being heavily scrutinized.

The GWC values listed above are generated by a team-customized simulation using the EdjSports model that incorporates important DVOA metrics from Football Outsiders. The underlying assumptions of team strengths and weaknesses on both offense and defense are taken into consideration. Additionally, the game state variables (clock, ball position, timeouts and yards to first) are all distinct inputs in the simulation. In the case of each of the two 4th down decisions, the choice to go for it is clear. This is due to the fact that even under alternative assumptions (weaker offense, stronger defense), the aggressive action still emerges as the optimal choice.

When Romeo Crennel faced his PAT decision, he may have felt empowered by his prior successes on 4th down. Of course, every decision is independent and has to be justified on the best information available at the time. Our simulation indicates his aggressive action cost the Texans -1.2% GWC, but this assessment is very dependent upon the assumption of the Texans’ 2 PAT success rate and warrants a deeper look.

To break down this interesting decision we must consider three key game states:

The Texans therefore risk (91.5 – 84.4) = 7.1% GWC to gain (98.1 – 91.5) = 6.6% GWC. Their required 2 PAT success rate is:

 

\frac{7.1}{(7.1+6.6)} = \textbf{51.8\%}(7.1+6.6)7.1​=51.8%

 

This success rate would appear to be too high for NFL average offenses, as the historical 2 PAT success rate for the past 20 years is only 46.4%. The Edj Power indexes show the Texans and Titans both to be about league average in pass and rush so far in 2020.

Many underlying assumptions go into this analysis, and each has its own uncertainty. However, some of the assumptions are identical regardless of the resulting game state after the PAT.

This table illustrates that regardless of the path for the remainder of the game, many of the underlying assumptions are relevant and consistent. Simulations indicate the Titans may be expected to score a TD about 30% of the time, but that will be required regardless of the outcome of the Texans’ PAT. Similarly, consistent assumptions exist for the kickoff (assume touchback) and overtime win rate (~50%). The Titans’ 2 PAT and onside kick success rates are only relevant to specific game states as indicated in the table. In both of those cases we would expect the Titans to be close to NFL averages of 46.4% for the 2 PAT and 15% for the onside kick recovery.

With all of this considered, we believe there is a compelling case that Romeo Crennel committed a small error with his 2 PAT attempt. However, we cannot fault him too harshly for attempting the knockout punch in what has been a disappointing season so far for the Texans.

For additional coverage of this analysis, read "Did a brilliant intentional penalty enable the Titans' comeback vs. Texans?" on Yahoo Sports.

 

Presented by:

 EdjSports

Comments

6 comments, Last at 19 Oct 2020, 7:31pm

1 "The GWC values listed above…

"The GWC values listed above are generated by a team-customized simulation using the EdjSports model that incorporates important DVOA metrics from Football Outsiders. The underlying assumptions of team strengths and weaknesses on both offense and defense are taken into consideration."

I didn't realize the GWC numbers were calculated this way.  I'm not sure what I think about this, particularly this early in the season when DVOA numbers have modest opponent adjustments included.

Although these numbers say Crennel made a small error in going for 2 points instead of 1, I still like his decision.  It almost certainly only comes into play if TEN gets their own TD.  So basically the decision is "do I want to try and let my O score 2 points, or rely on my D to stop them from scoring 2 points?"  If I was HOU's coach, I'd go with my O, too. 

 

2 Time Factor on Titans Drive

They didn't discuss this in the article but I'd be interested to see how they incorporated the time factor into the Titans ensuing touchdown drive. In the article they include a table showing their assumptions for the three possible game states after the PAT. The table makes it seem like the assumptions for a Titan's TD is the same no matter the game state, but that isn't quite the case. If the Titans only need one possession then they can use the full balance of remaining time and score with 4 seconds left like they did in the game. However if the Texans trailed by 9 it would have forced them to score with enough time left to attempt the onside kick, recover, etc... The Titans used the middle of the field to drive quickly and score on their final drive which wouldn't have been possible if they'd needed to preserve any clock. It wouldn't result in a huge swing but when our total difference is 1.2% GWC it could be significant.

 

I think that raises a larger point that 1.2% GWC is such a small margin that it's almost definitely within the error/variance of the model. It's the nature of modeling that you can't capture the complete game state and even with the benefit of hindsight I'd stand by Crennel's decision to go for 2.

5 "I think that raises a…

"I think that raises a larger point that 1.2% GWC is such a small margin that it's almost definitely within the error/variance of the model."

The other point is that the risk/reward is basically the same: they were risking about 7% to gain 7%, hence you needed ~50% success rate. It wasn't a pointless risk for minimal reward.

That's a totally acceptable gamble to me - it's just preference for what you want to do. Do you have a play for a 2-pt conversion you feel confident in? Cool, go for it. Do you feel more comfortable on your team's performance on more "normal" (non-goal line) plays? Cool, play for overtime.

6 Very fair point on the…

Very fair point on the difference between trailing by 7 or 8 vs trailing by 9 . The clock management would be more rushed at -9 but probably not likely to be a material difference on the Texans 2 PAT decision, which is pretty close. The Titans’ GWC is already very low ( <2%) in the scenarios when they go down by 9 after a successful Texans 2PAT.