Don’t Show Your Stupidity By Wasting Timeouts
Peyton Manning is the smartest quarterback to ever play and he’s made this mistake over and over again. Tom Brady has six super bowl rings despite committing this error countless times. Russel Wilson is regarded as an outstanding decision maker (except for that one goal line pass in the super bowl) yet constantly engages in this miscalculation.
There is one play in football that drives me absolutely crazy. It’s not the short screen pass on 3rd down. It’s not a punt on 4th down and one. It’s not taking a field goal for a point after attempt instead of going for two. It is so much worse than all of those combined.
I can’t stand it when the play clock is counting down to the final seconds and a quarterback decides to call time out.
It’s bothered me for as long as I can remember. Like, really? You would rather waste a timeout in the 3rd quarter of the game on your own 40 yard line? You don’t think that timeout would be helpful to you or your defense later in the game?
Everyone knows how important a timeout is at the end of the game. If a team is trailing by a few points in the last minute of a game, there is a huge advantage to having 2 timeouts instead of 1. It gives you the ability to throw the ball in the middle of the field instead of only having to throw to players near the sidelines. But god forbid that you don’t have that extra timeout because you didn’t want to take the 5 yard delay of game penalty!
I decided that I needed to know what a team was giving up in return of that 5 yard penalty. I haven’t played any level of competitive football, so maybe I was missing something? Maybe it’s a smarter idea to keep the drive alive instead of preserving a timeout? How big of an effect does keeping the drive alive compare to needing an extra time out at the end of the game?
Since I can get a little deep with the data, I wrote a short analysis and a longer analysis for those who are the true data nerds like me.
For those of you who would like to see my code, here is a link: https://github.com/anchorP34/NFL-Analytics
Short Analysis
I decided to look at offensive possessions and break them into “drive sets” and “drives”. Drive sets are each set of downs 1–4, while a drive is every play from when an offensive team receives the ball to the final outcome (punt, turnover, touchdown, etc). Regardless of whether it is a drive set or a drive, the minimum objective is to get a first down. A touchdown would be nice, but acquiring a first down will guarantee the drive to continue. As you can see from the box plot below, successful drive sets tend to be closer to the first down marker than further away, which is what we would expect.
For some overview statistics on drives and drive sets, here are the average success rates for both drive and drive sets:
Drive Sets
- End in a 1st down 64% of the time
- End in a touchdown 8% of the time
- End in points scored 15% of the time
Full Drives
- Have at least 1 1st down 66% of the time
- End in a touchdown 21% of the time
- End in points scored 37% of the time
Since each drive set in the NFL has roughly a 64% success rate of gaining a 1st down, does the location of where the drive set starts make a difference in the outcome? For 1st downs, not really, but for points scored and touchdowns, absolutely. Drive sets that start more than 50 yards away rarely end in a touchdown or field goal, but as you can see, each 10 yards closer to the end zone start to have increasing exponential trend of likelihood. Which makes sense: the closer you are to the end zone, the more likely a team will score.
What about a full drive? Since a full drive can have multiple first downs and points being scored aren’t limited to the first 4 downs, the success percentages have a different trajectory than the drive sets. From 2014–2018, for a team to have greater than a 50% chance of scoring points on a drive, the drive needed to start somewhere closer than or between the 40 and 50 yard line.
So for a quick recap on first downs, scoring points, and touchdowns:
- First down’s are consistently around 65% regardless of where you are on the field
- For the odds to be in favor of scoring points on a drive, your drive needs to start in enemy territory
- For drive sets, there is an exponential increase in likelihood to score a touchdown the closer you get to the end zone
That’s great…but what does this have to do with timeouts?
After running further analysis, a team’s odds of winning a game increase by about 23% if they have an extra timeout in the last minute of a game. Not a 23% increase in odds of scoring a touchdown or getting a first down, but winning the game!
If your odds of winning increase by 23% by having that extra time out, then how much does a 5 yard penalty hurt your chances of gaining a first down and keeping your drive alive?
This heat map shows the likelihood of achieving a 1st down depending on the down and how many yards until a successful 1st down or touchdown. For instance, you can see that 2nd down and short has a much higher success rate than 3rd down and very long. Using this heat map, lets use this example:
It’s 2nd down and 4 yards to go, and the time is running out on the play clock. On 2nd down and 4, the likelihood of that down set ending up in a successful 1st down is 73%. If you take the 5 yard penalty for delay of game instead of using a timeout, the probability of success drops to 54%. If you spike the ball into the ground, your chances decrease to 51%. In my opinion, I would rather take my chances with a better than 50% chance of converting a first down rather than using a timeout.
Here’s the same values as the heat map, but a different visualization:
Here it’s a little easier to visualize the difference of percentages between downs. Now let’s think about a goal line stand. If you’re on the 2 or 3 yard line, there is almost no difference between 1st and 2nd down in terms of the success of converting a first down (would be a touchdown in this case). If you were running out of time on the play clock, it would almost make more sense to spike the ball than to call a timeout or take a delay of game penalty. There’s almost no difference. However, 3rd down is a different story. Taking a penalty on 3rd down could significantly decrease your chances of achieving a first down. For example, just taking a 5 yard penalty from 3rd and 2 to 3rd and 7 is a drop from 60% success to 40% success. Yet under normal circumstances, I still don’t believe that warrants the use of a timeout.
Conclusion
I was excited to confirm my initial hypothesis that timeouts are much more valuable than keeping a drive alive. With the odds increasing by 23% of winning the game, it is crucial to any team to hold onto their second half timeouts.
When it comes to taking the 5 yard penalty, it is up to the team to where they believe their threshold of acceptance lies. The difference between 1st and 10 and 1st and 15 is a drop from 65% success to 51% success. If you still have better than a 50/50 shot at making a first down, I would think that you should accept that penalty in regards of maintaining as many timeouts as you can. If it is an important drive later in the game, closer to the end zone, and you have to score points, then I can understand using a timeout. However, using one in the third quarter because the play clock is running out or the defensive alignment isn’t the best for the play you’re trying to run, it’s not worth giving up that leverage of timeouts at the end of the game.
Sports fans can almost all agree that winning a game in the NFL is harder than in any other sport. The rules are complicated, the margin of error is razor thin, and it truly takes all 11 players on the field for the team to succeed.
Don’t make it harder on yourself. Take the 5 yard penalty for delay of game, and give yourself a fighting chance at the end of the game.
Long Analysis — For all the true nerds out there
So how did I calculate that a team has 23% better odds if they have an extra timeout at the end of the game? Using the same dataset, I ran a logistic regression algorithm that used the time left remaining in the game (in seconds), if the team with possession of the ball was the home team, the net points the team was up or down, the down, yards away from a first down, yards away from a touchdown, timeouts left of the possession team and timeouts left of the defensive team. My response variable was whether they won the game or not.
Since my focus of this article is “should you call timeout to keep the drive alive or save it for when you need it”, I looked at when there was 600–100 seconds left in the game broken into 100 second increments. That would be the most influential time to use a time out when a team is making their final drives and they have to use it to stop the clock from running if a player gets tackled in bounds.
From the graph above, each point on the X axis relates to another logistic regression model that had records with that amount of time or less left in its game. The dashed lines are a 95% confidence interval of statistically significant parameters, so being outside of those intervals means that parameter is highly significant.
Starting with the eyeball test to make sure that things make sense, the negative parameters are yardline_100 (yards away from the end zone), down, and ydstogo (yards away from a first down). All three of those make sense to have your chances of winning decrease the higher those values are (further away from the end zone, further away from a first down, deeper into a drive set rather than 1st or 2nd down).
We also see that by far, the score difference is the most influential factor in the model, which makes sense given that blowouts will give an obvious answer to who will win. What I found interesting was until the last model, time left in the game was not considered a significant variable. But do you know what was?
Number of timeouts of the team in possession of the football was statistically significant for every model!
As for how accurate the model fit the data, it wasn’t half bad! Having R-Squared models in the .65-.80 range for each model was higher than I anticipated.
This makes sense since the closer you are to the end of the game, the more realistic the outcome should become with most games. Some games might be 1 or two points apart and making a determination can be difficult, but most games have a clear winner regardless of the last remaining minutes.
From here, the posteam_timeouts_remaining parameter is a significant factor and has a coefficient of .2061. The .2061 can be interpreted as “A one unit increase in posteam_timeouts_remaining, keeping the other parameters constant, would yield an increase in log odds of .2061.” If we take the exponent of .2016, we get 1.2289, meaning an increase of approximately 23% in the odds ratio.
Fun Visualization
One more visualization that I thought was interesting regarding teams first down success rates year over year:
First down rates seem to have an up and down trend for most teams in the NFL. One year they’re up, the next they’re down. You see a few of them (check out Kansas City and the LA Rams) where there is a huge increase over a two or three year span, but I found it interesting that regardless of how awful a team is, over the year they are more likely to get a first down during a drive set than not.
