≡ Menu

Adjusted Completion Percentage

In 1991, Dave Krieg led the NFL in completion percentage. He completed a career-high 65.6% of his passes, and while that mark was very good for that era, it doesn’t mean Krieg was great that season. In fact, he arguably wasn’t even good: Krieg actually finished just 24th in ANY/A that year.

One reason, I think, that Krieg was able to lead the NFL in completion percentage is because Krieg “ate” a lot of his incomplete passes. What do I mean by that? Krieg took a ton of sacks — he was sacked every ten times he dropped back to pass. When under duress, some quarterbacks eat the ball, to avoid an interception; that’s bad (well, it’s better than n interception) but it doesn’t get graded that way when calculating completion percentage. Other quarterbacks will throw the ball away; that’s good (assuming it isn’t intercepted) because no yards are lost, but it does hurt the quarterback’s completion percentage.

Even ignoring the yards lost due to sacks, fundamentally, a sack is no better than an incomplete pass. So why are quarterbacks who take sacks rather than throw the ball out of bounds given an artificial boost when it comes to completion percentage? Well, that’s largely just an artifact of how the NFL always graded things. The NFL was not always good at recording metrics, and somewhere along the way, sacks were either included as running plays, ignored, or included as pass plays. I don’t think a lot of thought went into it, but in my view, it makes the most sense to include sacks in the denominator when calculating completion percentage. Otherwise, we give undue credit to quarterbacks that take a lot of sacks, and penalize quarterbacks who throw the ball away when under pressure.

Take at the top 7 leaders from 1991 in completion percentage.    Aikman, Young, and Kelly all finished with 2.5 percentage points of Krieg but with noticeably better sack rates.  And Moon managed to have a sack rate that was a third of Krieg’s, too.

Rk Tm Age G QBrec Cmp Att Cmp%
Sk Sk%
1 Dave Krieg SEA 33 10 4-5-0 187 285 65.6 32 10.1
2 Troy Aikman* DAL 25 12 7-5-0 237 363 65.3 32 8.1
3 Steve Young SFO 30 11 5-5-0 180 279 64.5 13 4.5
4 Jim Kelly*+ BUF 31 15 13-2-0 304 474 64.1 31 6.1
5 Jeff Hostetler NYG 30 12 7-5-0 179 285 62.8 20 6.6
6 Bernie Kosar CLE 28 16 6-10-0 307 494 62.1 41 7.7
7 Warren Moon* HOU 35 16 11-5-0 404 655 61.7 23 3.4

Completion percentage is one of the simplest stats in all of football. It’s a binary stat that doesn’t tell us anything in the way of magnitude — a 90-yard completion is treated the same as one that loses 11 yards — but that doesn’t mean completion percentage is without its advantages. On the plus side, it is not very sensitive to outliers, is relatively consistent from year to year, and is easily understood.

Given how basic a metric it is, there are compelling reasons not to mess with it too much. For one, nobody gives completion percentage that much weight, and if we are trying to make the stat “better”, why not just use a better stat? But I do think including sacks1 in the denominator makes sense and is worth the tradeoff. If we do that, it’s now Young who leads the 1991 NFL season in adjusted completion percentage, at 61.6%. Kelly is second at 60.2%, Aikman is at 60.0%, and Moon is at 59.6%. Krieg falls to fifth, as adding 32 sacks to his 98 incomplete passes drops him to 59.0%.

The most egregious example of what happens when you don’t include sacks in the denominator occurred in 1988. That year, Wade Wilson led the NFL with a 61.4% completion percentage, and the top three leaders in that metric all had below-average sack rates.  But if you scroll down to #8 on the completion percentage rankings, you find the leader in adjusted completion percentage:

 

Rk Tm Age G QBrec Cmp Att Cmp%
Sk Sk%
1 Wade Wilson* MIN 29 14 7-3-0 204 332 61.4 33 9.0
2 Bernie Kosar CLE 25 9 6-3-0 156 259 60.2 25 8.8
3 Joe Montana SFO 32 14 8-5-0 238 397 59.9 34 7.9
4 Jim Everett RAM 25 16 10-6-0 308 517 59.6 28 5.1
5 Jim Kelly* BUF 28 16 12-4-0 269 452 59.5 30 6.2
6 Dave Krieg* SEA 30 9 6-3-0 134 228 58.8 12 5.0
7 Bobby Hebert NOR 28 16 10-6-0 280 478 58.6 24 4.8
8 Dan Marino MIA 27 16 6-10-0 354 606 58.4 6 1.0

Marino was sacked just 6 times that year, for a sack rate of better than one percent.  His adjusted completion percentage was 57.8%, over on percent better than any other passer in the league.

Overall, the leader in completion percentage would change in 13 seasons since the merger, including in 3 of the last 4 years, if we included sacks in the denominator.  So what do you think — is it worth including sacks in the denominator? And is this something that we should devote future posts to?

  1. And, arguably, scrambles. We don’t have that data going back historically, but we probably should include that now. And, I suppose, should remove spikes from the numerator. []
  • Clint

    Maybe I missed it somewhere, but what is the formula for the Adjusted Completion %?

    • Adam

      completions / (pass attempts + sacks)

      • Correct.

        • Clint

          I like this whole concept. I’ve always felt that completion % isn’t very important. You can be successful with what’s considered to be a low completion percentage, that’s just not the NFL’s style currently.
          You can complete 50% of your passes and be fine if they’re for bigger gains and touchdowns. If there are a lot of throwaways that otherwise could’ve been sacks/interceptions in there too, even better.

          • Adam

            Yes. People hammer Joe Namath for his low completion %, but fail to account for him attempting a high number of downfield (more difficult) passes. Any statistic that can be “gamed” loses much of its value. David Carr completed 68% of his passes in 2006, but most of them were meaningless short gains, and Houston’s offense was terrible.

            • Ryan

              Namath “only” completed 53.57% of his passes in game 2 of the 1972 season:

  • Adam

    Yes, it absolutely makes sense to include sacks in the denominator. Sacks are a failed pass play, so of course they should count against completion percentage. Sacks should also count against TD%, as again, they’re a pass play that doesn’t result in a touchdown. I would not include scrambles, though, because the data only goes back a few years. I don’t see much purpose in a stat if we can’t compare it historically.

  • Joe Wright

    This is nitpicking a really good piece, but I have two very minor questions.

    1. Does this account for–and is it even possible to account for intentional grounding penalties?
    2. What’s the logic, in general, of excluding spikes? If it’s that the pass isn’t intended to be completed, then the same is true of throwaways. But if it’s something else, then it could make sense to exclude.

    • You mean you didn’t read the footnotes? 😉

    • Richie

      I think the difference between a spike and a throwaway, is that a spike was never intended to be a pass play. The ball was only “thrown” in the sense that a rule exists that allows the team to stop the clock by spiking the ball to the ground. That spike play tells us nothing about a passer’s ability to complete a pass.

  • I’m all for it (which is why I use C/DB in my seasonal stats http://www.thegridfe.com/2015-stats/). If we measure net yards per attempt, it follows to measure completions per dropback.

  • eag97a

    Don’t mean to be pendantic but including sacks to the completion percentage means its now a measure of qb play NOT passer play and my view is if we are measuring qb play then we try to include all the variables we can include. Technically getting sacked means a qb has not thrown or carried the ball yet and this means its a qb play but necessarily a passing or rushing play IMO.

    • Adam

      Not sure I agree with that. As soon as the QB begins his dropback, it becomes a pass play. If he gets sacked because he can’t read the defense or his release is too slow, that doesn’t change the intention of the play.

      • eag97a

        If he drops back then pump fakes then scrambles then that is a running play. If he gets sacked because his o-line did not give him 2 seconds of protection then that changes the equation. In the same manner that not all interceptions are created equal, neither are sacks. Unless we have data stretching back to at least the merger about different kinds of interceptions and different kinds of sacks it doesn’t make sense to add all sacks to a measure of passing unless we are measuring qb play taken as a whole.

        • I could see this argument, but to be consistent, you would have to also remove all passes that the quarterback throws away, correct? If he is under pressure and just fires it out of bounds, would including that be a measure of passer play?

          • eag97a

            Agree to be consistent you have to do it that way. That is why I think if you include sacks in the completion percentage then you have to say its a measure of qb play NOT a measure of passing. I have always consistently pointed out QB eval is not always equivalent to passer eval though one is the subset of the other. And even under pressure doesn’t necessarily mean you can allocate credit/blame as well. We all know if a qb experiences pressure after 3 seconds of protection then its more his fault compared to being pressured in 2 seconds and throwing the ball away. We might want to subdivide the passing and running game to further refine our qb evals but its gonna be hard without extensive charting and film room data and even that will be incomplete without the playcalls.

        • Adam

          You’re right. We can’t really determine the play’s intent until it’s clear the QB isn’t trying to scramble. However my larger point is to question the utility of completion % when it’s confined to literal throws of the ball. Since comp % is a poor proxy for accuracy, I don’t see any advantage in ignoring sacks.

  • Inspired by that Eli Manning play, I had to look up what the worst “completions” were in the Play Index era (1994-2015).

    http://www.pro-football-reference.com/tiny/0FvcM

    The big loser is a -22 yard play by the Titans in their 59-0 thrashing against the Patriots in 2009. The bottom 3 all involve fumbles, which makes me wonder if some of the negative yardage comes from the ball bouncing backwards. Then you have a QB-to-himself “reception” (Christian Ponder, natch) for -15.

    Vick’s involved two plays, a -14 completion to himself and a -12 yarder that is lateraled back to him and he gains 16 yards. The lowest “natural” pass plays are a pair of -12s: Brooks Bollinger to Marion Barber and Alex Smith to Jamaal Charles.

    • Adam

      The NFL has a bizarre method of tallying yards on pass play fumbles behind the LOS. Instead of marking the yardage where the player lost the ball, they credit/debit the player from where the fumble was recovered. For example, Kurt Warner’s sack on the final play of SB43 credits him with positive yardage even though he fumbled behind the LOS, because the Steelers recovered his fumble a couple yards past the LOS.

  • sacramento gold miners

    Not sure how I feel about this, sometimes sacks are far better than attempting to intentionally throw a pass away. There are also times when sacks are better than incompletions, especially when one team is nursing a lead late. We should also factor in the value by the top QBs in extending plays, which sometimes result in game-changing plays. The annoying eight yard sack means little when the QB extends another play and finds the WR for a 30 yard gain.

  • Richie

    6 sacks in over 600 pass attempts is one of the crazier stats out there.

    It was also one of the higher interception % seasons of his career. I think his attempts to throw the ball away that season (and for much of his early career) may have actually been a negative. He started taking more sacks in the early 90s, and throwing fewer interceptions AND winning a few more games.