Last night’s Colts-Dolphins game was a statistical anomaly. Miami had the ball for over three quarters of the game and yet lost. Time of Possession (TOP) is viewed by some as one of the keys to victory in football, but I’ve got a different take.
Coincidentally, just yesterday I received an email from a reader asking me if I had done much research on TOP stats. I haven’t done much, and I’ll explain why.
I believe TOP is an “intermediate outcome” in a football game. In other words, it is a natural byproduct of being good at something else. You can’t be good at “time of possession.”
The way I conceptualize the sport, there is basically running and passing, and teams can have varying levels of ability based on their players’ talent plus coaching and other factors. It’s these core abilities that lead to things like 3rd down percentage, red zone percentage or TOP, and then ultimately lead to winning.
If we want to measure the strength of a team and how likely they are to win games—future games—we don’t want to look at the intermediate outcomes. Intermediate outcomes inject noise from sample error into the process and obscure the root causes of success or failure. To me, TOP is a side-effect of being good at other, more fundamental things.
Of course, late in a game with a small lead, the ability to run out the clock can be critical. But there is no such thing as “running out the clock.” The clock just runs. To run out the clock, a team needs to simply run, or sometimes pass, successfully. Even in this situation, when TOP appears critically important, it’s a byproduct of basic abilities.
There is also an issue of causation. Just like total running yards, lopsided advantages in TOP tend to appear late in games when offenses that are already ahead let the clock run down while offenses that are behind try to move the ball quickly. So it’s winning that often leads to TOP and not necessarily TOP that leads to winning.
I think efficiency stats are the best way to measure core football ability. There are some other more complex stats out there with various advantages, but they add tremendous complexity without an equivalent increase in value. Yards Per Carry (YPC) and Net Yards Per Attempt (YPA), on both offense and defense, can tell us volumes about a team. Efficiency stats also avoid the causation problem. While teams with leads rack up rushing yards and teams that are behind rack up passing yards (without necessarily being very good at either), their efficiency stats are closer to representing their true ability.
In last night’s anomaly of a game, let’s look at which team led in the “intermediate stats” compared to the core efficiency stats. We already know Miami dominated TOP 45:07 to 14:15. Miami had 27 first downs, 14 by rushing and 13 by passing. Indianapolis totaled 14 first downs, 3 by rushing and 11 by passing. Miami’s 3rd down percentage was a phenomenal 73% compared to 43% for Indy.
The Dolphins outgained the Colts 403 yards to 356. The Dolphins ran an amazing 84 offensive plays compared the Colts’ paltry 35.
Stat | Colts | Dolphins |
Time of Possession | 14:53 | 45:07 |
1st Downs | 14 | 27 |
Rush 1D | 3 | 14 |
Pass 1D | 11 | 13 |
3rd Down % | 43% | 71% |
Red Zone % | 100% | 100% |
Yotal Yds | 356 | 403 |
No. Plays | 35 | 84 |
Passing Eff (Net YPA) | 12.3 | 4.7 |
Rushing Eff (YPC) | 5.5 | 4.9 |
Admittedly, this game was very fluky, and I don't intend to draw any conclusions from any single game. Indianapolis had a couple of very quick scores, including one on the first play from scrimmage. But that's part of the problem when looking at TOP. In an example taken to the extreme, consider a team that scores a TD on every play. They would never lose, and their TOP might not total five minutes in an entire game.
So next time you hear a coach or analyst talking about needing to improve a team's 3rd down percentage or focus on "ball control," you can tune him out. Sure, there are some differences in particular situations, but to get better on 3rd down or to control the ball at the end of games, a team simply needs to be better at passing and running in general.
I think you are wrong about your 3rd down point. The Colts would benefit enormously with a better 3rd down defense. Over the last 4 seasons, the Colts have been last in the league in possessions. If the defense stopped a few of the 3rd downs, that would lead to more possessions, and the Colts are the team that would benefit the most from this because of your analysis on possessions. They are usually the favorite so having more possessions in the game benefits them because it gives the offense more chances to score and keep the game out of reach. The Colts play a style similar to underdogs (ball control O and D), yet are the favorites 75% of the time.
Right, but again, "3rd down defense" is another example of an intermediate stat. How do you get better at 3rd down defense? You get better at other things: pass rushing, run stuffing, coverage, etc. It's so annoying when analysts say "the key to this game will be Team X's third down defense, they've GOT to get off the field on third down." Well yeah, that's nice, how do they do that? They get better at the other stuff. "Get better at 3rd downs" is not a strategy, it's a goal.
Anon2, I disagree.
One of the key features of Dick Lebeau and Jim Johnson's defense has always been to use a different strategy on 3rd down than first or second down. Namely "knock the passer down" (assuming the offense is expected to pass). They both have historically been much more agressive on third down, one by using a new blitz, the other by sending like 14 pass rushers.
I agree with Anon2.
While I haven't looked at the numbers at all, assuming that Indy actually has been really bad at third down defense over the past few years, I'd wager the explanation is that:
1) Indy has had a terrible run D most of those years, which leads to lots of 3rd and short situations
2) Simpson's Paradox -- Indy might be league average on 3rd and longs, but because they face more 3rd and shorts (which all teams convert at much higher rates), they appear to have a bad 3rd down defense; and
3) Because of Indy's bad run-D, they've probably been really bad on 3rd and shorts.
Now if they've been one of the worst teams in the league in 3rd and 8s over the past few years, then maybe Dr Obvious is right.
Last year they were better on defense on 3rd down than on 1st or 2nd down, according to Football Outsiders Almanac. Defensive DVOA was -9.4% on 3rd down, 3.3% on 1st, and 2.5% on 2nd (remember, negative is better).
I've always been skeptical of TOP as an important indicator of a team's skill. The only time TOP matters for TOP's sake is holding onto leads late in the game. I've often wondered why points per drive isn't a more publicized summary stat.
I believe Anon1's notion of more possessions is a straw man. If possessions are the issue, in a game there is a certain possession equity. If I stop you on 3rd down, I may have an additional possession but after that possession you get the ball back and the number of possessions equalizes. At game's end the number of possessions will be at most 1 more than your opponent.
It comes down to what you do with the possession. Cutting your opponent's possession short is likely good for you. If the Colts are a team with the fewest possessions, their collective opponents (let's count the possessions their opponents had during those games) will also be low. That the Colts had a winning season indicates that they used those possessions more efficiently than their opponents.
Ed, I think what Anon1 wants to say is the Colts defense is bad, allowing longer drives by the opposition. These longer drives limit the number of drives by the Colts' offense. Since the Colts' offense is superior to most, they would benefit by having more overall possessions. Thus, reducing the Colts offensive possessions allows for greater variability, allowing the opposing team a higher shot at success.
Ultimately, you are correct: the Colts are more efficient with their possessions. However, the team would be even better (more likely to score/win) if the defense was average, giving them even more possessions to score. Still though, I think Brian's correct about TOP and other "intermediate" stats.
I'm glad someone finally commented on this.
After the Bills/Patriots MNF opener, I found it odd that the consensus among commentators was that the Patriots "dominated" that game(~38:22 in TOP being used as justification). They won on what was essentially a fluky special teams play. Just as with MIA/IND, the Bills had better YPA and YPC.
I agree that time of possession is probably bunk as a predictor. You might say that it's conditionally independent of winning given efficiency stats.
However, I'm not sure that efficiency stats are everything you need to know to make a prediction. First, obviously they leave out special teams, which I think deserve mildly closer examination, although that examination might be difficult.
Second, and more interestingly, there is this concept raised by the CHFF people and also possibly KC Joyner, which is that of scoring efficiency. There are teams, most spectacularly the 2007 or 2008 Cutler Broncos, which have if I recall correctly quite high passing efficiency, but score very little per game; and then there are the 2007 Patriots, who apparently have stats fairly similar to the Cutler Broncos, but then score hugely more points per game. I can't personally verify the efficiency stats myself, since I haven't found the play-by-play information on the NFL site yet, but it seems like if this were true, there would be some other factor needed to predict win probability, beyond just the efficiency numbers.
It could also be that the people who made these claims were looking at nonpredictive offensive stats, like passing yards, instead of passing efficiency, but again, I need to find some data to look at.
Elliot,
Your point about special teams (and, therefore, field position) is definitely correct; efficiency stats are not 100% of what you'd need, and I think Brian was being a bit hyperbolic.
Red zone efficiency is important, but sites like Football Outsiders have generally found that it's not very predictive year-to-year, and that teams with red zone offensive efficiency significantly lower than their overall offensive efficiency are good "bounce-back" candidates the following year.
I'd also refrain from citing anything CHFF has published; it seems to me that they don't quite understand that correlation is not causation. They also seem to make claims just for the sake of getting page views.
If you look at high time-of-possession teams over the course of a season, I predict you'll discover that their defenses get better (relative to the league) as the season wears on but their offense gets worse, both as a result of wear and tear.
But as a predictor of wear and tear, time of possession is still not as accurate as number of offensive plays. An all-passing team will stop the clock much more often than an all-running team, so if two teams with disparate offenses played the same number of plays, I think you'd see a big difference in TOP by the end of the game.
The way I think of red zone stats is that they are mostly just a sub-sample of overall performance. I think about 10 or 15% of all scrimmage plays occur in the red zone. If you took any random 15% subset of a teams plays, the performance level in that subset would be somewhat different than the overall set. This is simply and naturally due to sample error.
That said, I do believe that there are unique aspects of the red zone that matter. The field is compressed. Down-field speed is not as important, and height becomes more important. Risk tolerance usually decreases in play calling because turnovers are more costly.
But I think these aspects are often over-stated, and most of the differences between overall performance and RZ performance are just sample error. That's partly why teams more often than not see a "bounce-back" regression-to-the-mean in RZ performance from year to year.
Elliot-
Special teams does matter...with your own kickoff distance and coverage being the most important.
The one thing Brian also tracks is offensive turnover percentage since its has been shown that turnovers are more likely the fault of the offense. That's another part of efficiency although not quite as predictable.
so the difference between the Pats and Broncos is rather simple
1. The Patriots were first in turnovers per drive. The Broncos were last
2. THe pats average starting field position was the 31 (7th). The Broncos was the 25(last)
To sum up further the Patriots had a historically amazingly efficient low turnover offense coupled with an average defense. The Broncos had an efficient but turnover riddled offense coupled with a historically bad defense.
It's hard to say the Broncos were efficient since they turned the ball over so much but the plays that weren't turnovers were likely very efficient. Thus it's rather obvious to see why the Broncos didn't score as many points.
James has a great point about number of overall possessions. Say the Colts have an 80% chance of scoring on any given drive, and the Dolphins have a 50% chance... the Dolphins would be wise to cut down BOTH teams' overall number of possessions, since the more chances the Colts get, the more it benefits them (just like the house always wins in roulette - if your odds are lower than the house's, you want to put all your money on one spin). That's to say nothing of "wearing out the defense" or anything like that... the Dolphins certainly did exactly what they wanted to do, they just gave up some big plays and had bad luck in the end.
As an aside, I'm a bit surprised to find out that the Patriots were 2nd to only the Steelers in 2007 in overall time of possession, since they scored so quickly and so easily. The Saints, this year, are 7th... which makes sense to me now, thinking of it as an "intermediate outcome," since scoring early and often will not only allow you to run more later in the game, but will also cause your opponent to throw lots of incompletions as they try to play catch-up.
I would agree on most of the points above. Here are a few related posts on TOP and variance.
Keeping an Offense Off the Field
Underdogs and Reducing Possessions
Risk and Variance
Time of possession serves one very good purpose. When you see the number in the box score it helps paint a better picture of what happened in the game
When used as a measure of ability, a predictor of future success, or as a key point a team needs to focus on, it is just hot air
Another thing about efficiency and people who don't see it as important as others see it, is that yards per attempt is not the only measure. Both mean and median are measures of efficiency, just that one is easily available, while the other isn't
A team gives up 60 yards once every 10 plays and zero on the other nine. That is an extremely efficient defense that will be near perfect on third down (and the redzone) even though the yards per play is near the bottom
the reality is that a team has a wide variety of plays and they are better at some than at others. And although one situation might look the same as another to us, the coaches have a different batch of mostly unique plays to consider for every situation that comes up. Even if a team is generally good at passing, they might have some passing plays in their book that they just aren't that sharp on. All it takes is one player who was a little brain-dead when the play got practiced. When a coach says "we need to work on third down conversions" it probably isn't the statistic that he's worried about. He probably saw a receiver make the wrong cut on one 3rd down play, the QB make the wrong look on another, and the center miss his assignment on a third, and maybe the coach decided that a 4th play they ran just wasn't that well conceived. From all of that he has concluded that the team needs to work on its third down conversions. Because those were the plays he saw poorly executed.
Remember, whenever a coach talks to the media, they have to massively dumb down whatever they are trying to say. Often, what's left isn't much worth learning.
3rd down does have a different strategy because the defense knows how many yards the offence needs to get. If it's third and 10, you're not calling a play for 0-9 yards so the defense doesn't have to defend against that. As well, teams are less likely to go deep as that would lower the percentage of the 3rd down conversion. So in theory it cuts down the play book and should make it easier for the defense to defend. Red zone defense is the same sort of thing, as the defense no longer needs to defend against the deep pass.
That being said I don’t think that, adjusted for the advantages above, neither stat should differ from 1st or 2nd down or non-red zone stats.
I think that when analysts speak of 3rd down or red zone efficiency, they're trying to remark on the added pressure that these plays bring. Better performance in pressure situations could be said to show greater confidence of the players, and confidence, if it were measurable, would probably be the best predictor of a team's success.
CURSE YOU IE!!! I wrote a big long response, and it crashed before I could post. Brilliant.
ANYWAY, I 100% agree with the author. TOP is just an observed statistic. It's the result of good play.
While on a GAME by GAME basis, there is likely a high % of wins for the team that has most possession (anyone know the figure?? 75%???), the long run correlation (wins per season vs. avg TOP/GAME) is not very strong.
From 2006-2008, here are R^2 correlations:
(Cumulative): .2820
2008: .3187
2007: .2269
2006: .3067
The media is gonna say it over and over -- TIME OF POSSESSION wins games. Why? Because, if you have the ball more, you'll most likely win the game. However, this is not causation, and it's not something you get without doing something else. SCORING (net) has a 100% correlation with winning games. NET POINTS from 2006-2008 has an R^2 correlation with WINS PER YEAR of .8393. So, would a team rather preach SCORING and STOPPING? Or having the ball a lot?
Here is forum post that addressed this a few years back:
http://www.thefootballscientist.com/discussion/index.php?showtopic=602
As the OP said, "efficiency stats are the best way to measure core football ability." You don't have a coach going into practice saying, "We need to have more time of possession to win the game. Let's go practice time of possession!" No, instead, they profess efficiency -- getting first downs, executing plays, establish yards per play, and scoring.
So far through week 7, team that "wins" time of possession (TOP) has won: 77/103 (74.8%)
34 games a team has led TOP by more than 10 minutes. 30/33 were won (88.2%). (PIT beat MIN, CHI beat DET, OAK beat KC, IND beat MIA)
Winning teams have held the ball an average of 5:09 longer than losing teams.
Other quick stats through week 7:
Team with most YARDS wins game: 65/103 (63.1%)
Team that wins TO battle wins game: 67 (65.0%)
NOTE: 22 games have had even TOs. In games with UNEVEN TO, the team that won the TO battle: 67/81 (82.7%)
Of the three stats above (TOP, YARDS, TO Battle), team that has dominated:
3 of 3 won: 35
2 of 3 won: 39
1 of 3 won: 24
0 or 3 won: 5
But, most importantly:
Team that has most points wins game: 103/103 (100%) :) You can teach players to get aggressive, try to knock the ball loose, execute plays, improve accuracy, etc. But the stats are just observations (measures of effectiveness).
As for as "running out the clock". Do I have it figured right that you could use up almost 20 minutes in a single drive? Say you start at your 1 yard line, ran the ball the min. Distance to get 1st downs. Stopped at opponents 1 yard line with no time on the clock.
Scoring generally takes longer than three and outs. I think that's why winners of time of possession win games. I think that winning causes you to have longer time of possession- not the opposite way around. I definitely don't think that purposefully holding the ball longer or taking longer to do a play would help you win games. (Which is the way some people talk about time of possesion.)
Desirable field position can also be viewed as an intermediate outcome, based on a defense's ability to force three-and-outs and return punts effectively. Should we downplay the importance of starting field position based on the fact that these other variables play into it?