The keys to winning with Taylor Heinicke

The Commanders have found themselves in quite a predicament. The starting quarterback they traded for in the offseason has turned out to be just what his last two teams thought he was, and now he’s injured. For the second season in a row, they have been forced to start their backup quarterback, who, as luck would have it, turns out to be no worse than the injured starter. In some important respects he is better.

While no one ever planned for Taylor Heinicke to be the starting quarterback, he has ended up starting 21 games since he was called up as an injury replacement for the 2020 Wild Card playoff game. The Commanders were already facing a difficult decision, with Carson Wentz set to return from injury in a few weeks, and the team playing significantly better with Heinicke at QB. The pressure associated with choosing the right starting QB has ratcheted up since Heinicke led the team to the upset of the season on Monday night against the previously undefeated Philadelphia Eagles.

Prior to the Eagles game, fans had expressed frustration that Scott Turner didn’t really seem to adapt his offense to make use of Heinicke’s strengths or to scheme around his weaknesses. That seemed to change on Monday night when Turner leaned more heavily on the run than in previous weeks, which took the pressure off Heinicke to win with his arm.

The resulting win over a heavily favored opponent, by the second-largest spread of Heinicke’s tenure, would seem to validate what a lot of us thought. That is, that the keys to winning with a QB with limited arm talent should be an emphasis on the running game, controlling time of possession and allowing the QB to function as a game manager, distributing the ball to his playmakers on short and intermediate routes. Of course, a good defense and limiting mistakes by the QB and everyone else on the field always help.

I think most fans are happy when the results on the field seem to reinforce their favorite narrative. However, a game of football is very complex. A lot of different variables on both sides of the ball can influence game outcomes, and many of those variables interact with one another, providing ample scope to develop misconceptions. For example, we might give credit to the QB for a victory, when it was really the defense shutting down the opponent’s passing attack that made the crucial difference.

With that in mind, I decided to have a look at some of the more obvious variables to see if I could identify key variables which correlate with winning and losing when Taylor Heinicke is playing QB. The results won’t really prove or disprove anyone’s pet theory about why the Commanders win more games with Heinicke under center than Wentz, because correlation does not prove causality. One the other hand, if two variables are not correlated, it is a good bet that they are unrelated.

Nevertheless, despite these limitations, I suspect that most fans will find plenty of food for thought regarding the ongoing QB debate, and the Commanders’ future at the position.

Photo by Dustin Bradford/Icon Sportswire via Getty Images

Determining Which Factors Lead to Success and Failure

My basic approach was to use correlation analysis to identify key variables which seem to be related to winning and losing in the 21 games that Taylor Heinicke has started in Washington.

I used Point Differential as the measure of team performance because it has two advantages over simple Wins vs. Losses. First, it has greater dynamic range and allows us to differentiate between a close game and a blowout. As a result, it is the more sensitive indicator for detecting factors which might influence team performance. Secondly, it is a continuous variable, which allows me to use simple correlation analysis which should be familiar to many readers. If I had decided to go with Wins vs. Losses, I would have had to use a mathematical technique called logistic regression. I doubt many readers would have the interest or patience to wade through an explanation of what that means.

I kept the search for success factors fairly basic and did not delve into intricacies like play selection or down and distance. In addition to the obvious variables, I also included one that comes up in discussions on Hogs Haven from time to time: scoring early. Unfortunately, I was not able to evaluate offensive line play, which has to be a critical variable, because I don’t have access to proprietary Blocking Win Rate or OL Rating data.

I evaluated the following variables:

Offense

• Offensive Production: Rushing & Passing Yards, Rushing & Passing TDs
• Run/Pass Ratio
• 3^rd Down Conversions
• Scoring Early: Time to First Score, 1^st Quarter Points
• Getting the ball to playmakers: Antonio Gibson & Terry McLaurin targets & carries

Defense

• Rushing yards allowed, Passing yards allowed
• 3^rd Down Stops

Whole Team

• Turnover Differential
• Time of Possession
• Penalty Yards

The emphasis was on finding factors that might help Heinicke succeed, so I went into more detail on offense than elsewhere. But I also wanted to see if other factors might be contributing to the team’s success when Heinicke is playing, so I also covered the basic non-QB-related factors.

To evaluate the strength of correlation between these variables and team performance, I calculated the Pearson’s correlation coefficient between the variable in question and point differential across all 21 of Heinicke’s starts. For those of you who slept through Math class in school, the correlation coefficient (r) is a standard statistical measure of closely related two sets of data are to one another. It varies from 0, indicating no relationship, to 1 when there is an exact one to one mapping of one set of data on to the other. As a really gross rule of thumb, a correlation of r > 0.5 indicates a strong relationship between the two variables.

I will also talk about another measure called Explained Variance (R²), which is also sometimes called Effect Size. This is the amount of variance in one variable, for example point differential, that is explained by another variable, such as passing attempts per game. Explained Variance is just the square of the correlation coefficient.

If anyone is wondering why I didn’t test Points Scored or Points Allowed, it is because those are the two components of Point Differential, and it does not make mathematical sense to test the correlation between a variable and one of its component elements.

For statistics nerds: The more sound approach to asking these questions would have been to perform a multivariate analysis. I decided that would be more trouble than it was worth. I am happy for you to discuss the potential limitations of my more simplistic approach in the comments, if you like.

Photo by Jonathan Newton/The Washington Post via Getty Images

What Are the Keys to Winning with Heinicke?

OFFENSE

The obvious place to start on offense is Rushing Yards and Passing Yards. The following graph plots those two variables on the horizontal axis against Point Differential on the vertical axis:

In the 21 games with Heinicke at QB, Passing Yards was only weakly correlated with Point Differential (r = 0.21). Rushing Yards was more strongly correlated (r = 0.44) than Passing Yards, but even so, it is not a particularly strong effect. Rushing Yards can explain 21.2% of the variance in Point Differential across Heinicke’s 21 starts, while Passing Yards only accounts for 3.7% of the variance.

What this means is that Heinicke’s passing appears to be having little influence on game outcomes. There is not much difference in game outcomes when Heinicke passes for 120 yards or 336 yards. On other hand, there is a stronger trend for the team to win when they rush for over 150 yards and to lose when they rush for less than that.

These results would seem to indicate that the team’s chance of winning with Heinicke at QB should increase when they lean on the run more than the pass. As you might expect from these findings, the Run/Pass Ratio was also positively correlated with Point Differential (r = 0.51).

There are two possible explanations for why Rushing Yards and high Run/Pass Ratios could be correlated with Point Differential. First, success in the running game could be leading to scores and limiting opponents’ scoring opportunities by burning up clock time. Alternatively, teams tend to run the ball more later in games after they have built up sizeable leads. Since Washington has seldom developed commanding leads during Heinicke’s time with the team (check out the Point Differential axis of the graph to confirm for yourself), it is probably more the former than the latter.

Since Passing Yardage was only weakly related to Point Differential, there was no real point in testing the correlation with Intended Air Yardage per Pass Attempt to get at the question of whether it matters if Heinicke keeps his passes shorter or longer. Whatever he does passing does not seem to have much influence on game outcomes.

3^rd Down Conversion Rate. I recently showed the Commanders have a particular problem with finding themselves in third and very long yardage situations. Of the two QBs, Heinicke is better than Wentz at converting on third and fourth downs. I also showed that Heinicke is better at avoiding sacks on third down that lead to punts. It was therefore of interest to see if 3^rd Down Conversion Rate had a particularly strong relationship game outcomes when Heinicke was playing. There was a modest level of correlation between 3^rd Down Conversion rate and Point Differential (r = 0.36). That is stronger than the correlation with passing yards, but a bit weaker than the correlation with rushing yards.

Getting the Ball to Playmakers. Another common suggestion is that one reason that Heinicke has been more successful than Wentz with the Commanders is that he is better at getting the ball to his playmakers. To see whether that might be an important factor, I calculated the correlation between Point Differential and Opportunities (rush attempts + pass targets) for the two main rushing and receiving playmakers throughout Heinicke’s starts, Antonio Gibson and Terry McLaurin. Gibson’s Opportunities had a moderate level of correlation with Point Differential (r = 0.34). The correlation with McLaurin’s Opportunities was even weaker (r=0.24).

It might strike some readers as surprising that the number of Gibson’s touches and targets is more strongly correlated with game outcomes than McLaurin’s, particularly after Monday night. However, this is consistent with the earlier finding that the passing game appears to be less influential on game outcomes than rushing when Heinicke is starting.

Scoring Early. It has been suggested a few times that, since neither of Washington’s QBs is capable of winning a shootout, it is essential for the team to score early. To test out whether that’s true, I evaluated the correlation between point differential and two measures of scoring early: Time to First Score and Points Scored in the 1^st Quarter. There was essentially no correlation with either variable (r = -0.05, r = 0.09, respectively).

In summary, the variable with the strongest correlation with Point Differential on offense when Heinicke is playing is Rushing Yardage, to which he makes a relatively small contribution. The passing game appears to have a much weaker influence on game outcomes. The team’s ability to convert on 3^rd downs is correlated with game outcomes but that effect, if real, is smaller than Rushing Yardage. Getting the ball to his two playmakers might have some influence on game outcomes, but any effect there is fairly weak. Scoring early does not appear to have any relationship with whether the team wins or loses.

DEFENSE

The variables that most strongly correlated with Point Differential when Heinicke was playing were all related to the defense. Total Yards Allowed by the defense showed a strong correlation with Point Differential (r = -0.75). To find out which aspect of the defense might be most important, I broke that down to Passing and Rushing Yards Allowed:

Interestingly, the pattern was opposite to what we saw on offense. Pass defense was strongly related to team performance; whereas, run defense was only weakly related. The correlation between Passing Yards Allowed and Point Differential was the strongest in the entire data set (r = -0.79), and accounted for 63.1% of the variance in Point Differential. The correlation coefficient is negative because larger values of yardage allowed are associated with more negative point differentials (i.e. the slope of the trend line is negative).

Rushing Yards Allowed was weakly correlated with Point Differential (r = -0.21), and only accounted for 4.3% of the variance in team performance.

3^rd Down Efficiency: The inability to get opposing offenses off the field on 3^rd downs was a major issue on defense in 2021, and has improved significantly as the 2022 season has progressed. To see if that might be a crucial factor to game outcomes, I calculated the correlation between 3rd Down Stop Rate and Point Differential. 3^rd down efficiency on defense had a reasonably strong correlation with Point Differential (r = 0.48).

In summary, defensive performance was more strongly correlated with overall team performance than offensive performance when Heinicke was playing QB. The strongest effects were associated with pass defense and third-down stops. Run defense was only weakly related to game outcomes.

WHOLE TEAM

Three other factors that have significant impacts on game outcomes are turnovers, time of possession and penalties. Quarterbacks can make some contributions to these variables, but they are also affected by other players on offense, defense and special teams. To check whether any of these factors might be contributing to Heinicke’s wins and losses, I calculated their correlations with Point Differential.

Turnover Differential was strongly correlated with Point Differential (r = 0.63), and accounted for 34.6% of the variance in game outcomes. The Commanders have won six out of nine of Heinicke’s starts when the Turnover Differential was negative and have won three of four games when it was positive.

Time of Possession and Penalties. Fans who watched the Commanders dominate the Eagles on Monday night will not be surprised that Time of Possession was also fairly strongly correlated with Point Differential (r = 0.54). It might come as more of a surprise that the correlation with Penalty Yards was weaker (r = 0.37), though still likely significant.

Summary and Conclusions

The correlations between all the variables that were examined and Point Differential are summarized in the table below, sorted by strength of correlation. Positive and negative values are irrelevant, as they indicate the direction of the trend, not the strength of the correlation.

This started as an exercise to discover what factors might be important to giving Heinicke the best chance to succeed as the Commanders’ starting QB. The results turned out somewhat differently than what I was expecting.

Most of the variables that were strongly associated with winning and losing when Heinicke was starting relate to things that are not related to QB play or to which the QB only makes indirect or small contributions.

The strongest correlations were with Passing Yards and Total Yards Allowed by the defense, which are not directly related to QB play. It can be argued, however, that the QB can contribute to these defensive stats indirectly by sustaining long offensive drives which keep the opposing offense off the field. Doing so can lead to an advantage in Time of Possession, which I showed is also fairly strongly correlated with winning games.

A great example of this effect was Monday night when the Commanders held the Eagles’ offense in check, partly by great defensive play, but also by dominating the Time of Possession 40:24 to 19:36.

I have previously shown that Heinicke is better than Wentz at keeping drives alive because he is less prone to making costly mistakes that lead directly to change of possession and is better at converting from third and very long yardage situations. As a result, it is fair for Heinicke to get some credit for the effects of Turnover Differential, Time of Possession and even apparent defensive production on wins and losses when he starts.

However, the facts that game outcomes are most strongly correlated with defensive production statistics suggests that Heinicke is being carried by the rest of the team more than the other way around. Furthermore, the fairly strong correlation between defensive stops on third downs and game outcomes has nothing to do with Heinicke. On offense as well, rushing stats (Rushing Yardage, Run/Pass Ratio) are more strongly correlated with game outcomes than passing production.

Getting back to the original intention of this analysis, the main things that the team can do to maximize the chance of winning while Taylor Heinicke is the starting QB are:

1. Commit to the run. Running the ball effectively keeps Heinicke out of situations where he is likely to commit turnovers by keeping third downs manageable, and helps to maintain a positive time of possession balance which, in turn reduces scoring opportunities for opposing offenses.
2. Don’t rely on Heinicke to win with his arm. Passing production is only weakly related to game outcomes when he plays.
3. Continue to field a strong defense.
4. Cut down on penalties.

It might not be rocket science, but the results on Monday night would seem to validate this approach. On balance, it appears that the team is winning despite Heinicke rather than because of him. He is a better option than the other options at QB at present, but there is little evidence that his contribution on offense is the major driver of the team’s success.

Acknowledgement: Thanks to James Dorsett for editing

---