≡ Menu

You remember 1976, don’t you? Two teams — the Colts with Bert Jones and Roger Carr, and the Raiders with Ken Stabler and Cliff Branch — stood out from the pack when it came to pass efficiency that season. The Colts led the NFL in passing yards, ranked 2nd in passing touchdowns, and threw just 10 interceptions, tied for the fewest in the NFL. Oakland threw 33 touchdown passes — nine more than the Colts and 12 more than any other team in football — while ranking 3rd in passing yards. Both teams averaged 7.5 Net Yards per Pass Attempt, while every other team was below seven in that metric. Those two teams went a combined 24-4.

The next four best passing teams were St. Louis, Dallas, Minnesota and Los Angeles. Each of those teams went 10-4 or better. In fact, the linear relationship between pass efficiency and team record was quite strong that year. Take a look at the chart below, which plots Relative ANY/A — i.e., Adjusted Net Yards per Attempt relative to league average — on the X-Axis, and Winning Percentage on the Y-Axis:

RANYA 1976

This shows an incredibly tight relationship between just one metric — Adjusted Net Yards per Attempt — and team success. Now, let’s run those same numbers for the 2014 season. And, to compare apples to apples, let’s use the same dimensions, with the X-Axis running from a RANY/A of -4.0 to 4.0, and the Y-Axis from 0 to 1:

RANYA 2014

Pictures can tell a 1,000 words, and one look here is a great representation of how the spread of both passing efficiency (and, to a less extent, winning) is more compressed now than it was in 1976. But what’s interesting is that the correlation coefficient between pass efficiency and winning was the same in both years: 0.75. In other words, pass efficiency was about as responsible for team success in 1976 as it was in 2014.

For 2014, the best-fit formula to predict winning percentage was 0.50 + 0.147*RANY/A. For 1976, it was 0.50 + 0.113 *RANY/A. This means it “costs” less RANY/A to gain wins in 2014, but since it’s harder to get RANY/A — because pass efficiency is more compressed — it winds up canceling out. In other words, while it takes less RANY/A or ANY/A to gain wins, it also costs more to “buy” pass efficiency now.

Now, how about 1953?

1953

Here, the CC was 0.82, which means pass efficiency was even more importan to team success in 1953 than it was in ’76 or ’14. The best-fit formula for ’53 to predict winning percentage was 0.49+ 0.094 * RANY/A. Again, we see that in this era, you had to buy more RANY/A to get wins, but it was also relatively cheap to buy RANY/A — you would never see a team with a RANY/A of +5.0 today the way we did with the ’53 Browns.

1953 was somewhat of an outlier year: the decade of the ’50s was not like that. The graph below shows the correlation coefficient between RANY/A and team winning percentage for each year from 1950 to 2014. The blue line represents the NFL, the red line displays the AFL.

cc 1950 anya

This is just the first step in trying to answer the question of how much does passing matter now versus at various points in NFL history. That’s enough for one day, but before we move to Part II, I need some help. What sort of queries would you be interested in seeing? How would you go about answering the question of how much passing “matters” — whatever that means — now versus in 1976?

  • Passing efficiency has always been important to winning, but to what extent do you think that the relationship is reciprocal? (i.e., Efficient passing teams win games, but teams with a lead throw safer passes and produce more efficient numbers.) I’m almost positive this is why passer rating correlates so strongly with winning, but I am less certain about ANY/A. Since you’re the King of ANY/A, your opinion matters, Chase.

    • Kibbles

      The beauty of YPA is that it’s supposed to be scheme-neutral. A team can get a good YPA by throwing a lot of high-percentage, low-reward passes… or it can get a good YPA by throwing a lot of low-percentage, high-reward passes. In theory, YPA should measure team passing quality regardless of *how* the team is passing.

      In practice, YPA actually does show a very mild bias towards deep-passing teams. A team with a deep passing identity will average just a hair better YPA than a comparably good team with a short passing identity. Of course, this would actually work against teams with the lead, since playing with a lead tends to mitigate deep passing in favor of lower-risk, higher-percentage throws.

      I also don’t know how NY/A works, but I’d imagine including sacks would help wash out some of that deep-passing bias, if not overcome it entirely. No hard data on this, but conceptually it would make sense that more deep throws would correlate with a higher sack percentage, which would help eat away at the YPA advantage.

      I also don’t have a clue how the “adjusted” part impacts the comparison, and I couldn’t begin to speculate. I would guess that deep-passing teams have more TDs, but I’d also guess that they have more interceptions, and I wouldn’t have the slightest idea which force would be stronger.

      • Richie

        I did a quick query of 2014 teams. In the first half of all games, the league average was 7.18 yards per attempt. When teams are trailing (by any margin) in the second half, the league average was 7.06 yards per attempt. So there was a negative impact, but not a huge one. But that impact varied wildly by team.

        Of course, some teams had way more pass attempts while trailing in the second half (Oakland led with 295) than others (Seattle had 84 and Green Bay 87). The 49ers struggled the most, dropping from 8.29 to 6.99, while Carolina turned it up the most, improving from 6.30 to 8.05.

        My guess for the long run would be that the team-to-team variation is probably random, with a general trend of slight decrease in the second half.

        Although, I took a quick look at the Packers and Patriots over the past 6 years. The Packers Y/A has been better in the second half while trailing over that period, by about half a yard per attempt, but the Patriots have been worse by about .7 Y/A in the second half. But then, Rodgers may be one of the most accurate deep passers ever(?).

        Here’s the full list:
        http://i.imgur.com/Zg3K0zp.png

        • Adam

          That surprises me. I figured Y/A would slightly increase for trailing teams, given their more aggressive strategies and facing prevent defense. But I I like the truth better – sounds like garbage time doesn’t help the losing team very much.

          • Richie

            My threshold was any point margin. Maybe it would be a different story if the team was down by 14+ points, and the defense didn’t care as much about 20 yard completions.

            I wish the PFR play finder would let me search by win probabilities.

            • Adam

              Down 14+ points in the fourth quarter, the average Y/A was 6.9. Of course that’s biased by the sample being heavily weighted with bad teams, but for individual QB’s, there doesn’t seem to be any noticeable garbage time inflation. The only glaring exception is Blake Bortles, who averaged 9.4 Y/A on 75 attempts.

        • Perfundle

          I don’t think that’s the best comparison. These four numbers should be more useful:

          YPA, trailing in first half: 6.94
          YPA, leading in first half: 7.62
          YPA, trailing in second half: 7.06
          YPA, leading in second half: 7.43

          The first half numbers are what you expect: teams with bad YPA numbers fall behind. In the second half the trailing team does slightly better compared to the first half, possibly because of prevent defense, and the leading team does slightly worse compared to the first half, because they’re trying to run out the clock.

      • Actually, Y/A isn’t quite as scheme-neutral as you might think: http://www.footballperspective.com/yards-per-attempt-where-does-it-go-wrong/

        Howeer, as you note, I think including sacks and INTs does help make ANY/A more scheme-neutral than Y/A. I’ll also note that being scheme-neutral is a different (but equally interesting) question than whether ANY/A is biased by teams playing with or without the lead.

        • Kibbles

          I don’t think that’s incompatible with what I wrote. It is “in theory” scheme neutral, but in practice biased towards deep-passing teams. YPA is much higher on deep passes than it is on shallow passes, but since I believe the difference in percentage of deep passes by a team with a “deep passing identity” and one with a “short passing identity” is actually fairly small, I think in practice this bias is fairly mild.

          According to PFF, last year 23 of 25 ranked quarterbacks threw between 9.7% and 15.3% of their passes “deep”. In 2013, 19 of 21 quarterbacks threw between 10.1% and 14.7% of their passes “deep”. It seems that we’re seeing maybe 90% of quarterbacks falling in a very narrow range from between 10% and 15% “deep” passes. Despite its substantial bias towards deep passes, I think YPA’s bias towards deep passers is probably pretty mild.

          I could very well be wrong on that, though.

          • Richie

            I figured the data in the old post that Chase linked to needed to be graphed.

            http://i.imgur.com/UT56s4E.png

            Throwing deep will (at least in 2012) get you more average yards per attempt than throwing short.

            But I agree with you, the distribution of pass distances probably doesn’t vary a whole lot. Also, this could be a diminishing returns problem. For one thing, if you keep sending your receivers on deep routes, they will probably get tired. For another, once you start getting in that 30-40% completion range, you risk having to punt before you can pick up the first down. If you keep throwing 5-yard passes, the completion percentage means you will hardly punt. One more problem is the “game theory” aspect of possibly becoming easier to defend if you keep bombing away.

    • Good question, Bryan. I think there is some reciprocal element here, but not nearly as much as for stats like completion percentage. It would be a fun experiment to really dig in to the stats to try to answer that question, but for now, I’ll just put that on the ever-growing to-do list.

  • It’s interesting that efficiency correlates about equally well with winning each year, even as passing undoubtedly is a larger part of offenses today. Roughly (with scrambles and kneels unaccounted for) 49% of plays were passes in 1953, vs. 44% of plays in 1976, and 58% in 2014.* With each extra decision the QB is given, you’d think that’s more chances for differences in team passing ability to affect each game.

    Anyway, hopefully those thoughts give you something.

    * (Pass Att+Sacks)/(Pass Att+Sacks+Rush Att)

  • John

    CHFF has been pushing this idea for a decade

    • They’ve also been confusing correlation and causation for a decade.

  • bobrulz

    This is my theory about the 1970s. The 1970s was obviously a defensively dominant decade, which led to the major rule changes late in the ’70s to open up the passing game. Most passing games were very poor statistically, but this just made the good passing games stand out more. If you had a quarterback and a scheme that could do better against the dominant defenses of the era, you REALLY stood out. In the early ’80s, the correlation decreased – I theorize this is because passing became easier, so it became easier for quarterbacks that weren’t quite as great to gain these huge chunks of yardage against defenses that weren’t as dominant. This brings the average up, but makes the best passing offenses stand out less. I also believe it’s possible that it took a little bit of time for defenses to adjust to the new rules (although I wasn’t around at the time, so this is just a theory and I admit I could be wrong).

    Now, the rules are very passing-oriented, which has led the average to tick up. But because of how systematized passing has become, combined with the increasing complexity of defenses and other factors, the gap between the best and worst QBs is arguably bigger than at any point since the early days of the passing game. There’s more good QBs than ever, but that makes the bad ones stand out, so it kind of creates an opposite effect from the ’70s.

    A lot of this is conjecture, so hopefully I’m not way off base.

    I would be curious to see an article that shows the gap between the best and worst passing games in a year and see if there’s any truth to what I’m saying.

    • Adam

      I’ve noticed the same trend with today’s passing game. The best QB’s are hugely advantaged by the friendly rules, but the worst QB’s don’t seem to benefit much at all. I’m thinking of Ryan Lindley and how spectacularly awful he played even with all the rules slanted in his favor.

  • AgronomyBrad

    Would be interesting to see how rushing efficiency correlates to winning percentage, and then determine which was more influential for a given year. I would expect the years where the correlation with passing efficiency dips to have a stronger correlation with rushing efficiency.

  • Adam

    I would love to see the correlations broken down by each component of ANY/A (comp %, Y/C, Y/A, TD %, INT %, sack %). Even if the overall effectiveness of passing has remained steady, I wonder if different aspects of the passing game have become more or less important at various points in NFL history?

  • Tim Truemper

    Some may “pooh-pooh” the correlation model used above as not showing causation. Ok, sure. We know Football is a dynamic, multi-factorial system. Still, I like Adam’s explanation below about quality passing game of the 70’s overcoming (to some extent) dominant defenses and thus, leading to improving probability for win outcome. One take away from this analysis by Chase is getting an idea of the amount of variance to win outcome could be accounted for by passing efficiency. Looking at run game efficiency too would be interesting. Kind of trying to see what confounding variable is consistently involved that invalidates or minimizes the effect of passing efficiency on winning, and I don’t see one.

  • James

    I’d like to see a comparison against Value – how much can you make up for a pedestrian ANY/A with volume?

  • sacramento gold miners

    Sad news today, Ken Stabler has died at the age of 69. RIP