Aggressiveness Index 2020
Football analytics people have been beating the drum for fourth-down aggressiveness for two decades now. The good news from the last three years is that, slowly but surely, we're winning the argument.
Football Outsiders introduced the concept of Aggressiveness Index way back in Pro Football Prospectus 2006. The goal was to find a way to rank coaches based on their tendencies on fourth downs in a manner that was easy to understand but accounted for the different rates at which the average coach will choose to "go for it" in different situations. Although no NFL coach is as aggressive as the data suggests he should be, we discovered there is quite a wide range of fourth-down tendencies among coaches.
Aggressiveness Index numbers center around 1.0 and generally describe how much more (or less) likely each coach is to go for it on fourth down compared to his peers; for example, a coach with 1.20 AI is roughly 20 percent more likely to go for it than an average coach historically would in equivalent situations. There are other methodologies now for measuring fourth-down aggressiveness, mostly based on win probability analysis: for example, the EdjSports head coach rankings or results spit out by Ben Baldwin's fourth-down simulator. Each methodology will have small differences in how it ranks the coaches, but Aggressiveness Index differs from the others by measuring coaches not against what they should do but against the actual decisions made by coaches themselves.
That's based on historical averages, of course; the numbers on aggressiveness have been much higher over the last three years. Take a look at the leaguewide numbers for the last seven seasons. This year, the overall attempts on fourth down were 60% higher than expectation based on historical norms.
|NFL Aggressiveness Index, 2014-2020|
The Aggressiveness Index excludes obvious catch-up situations: third quarter, trailing by 15 or more points; fourth quarter, trailing by nine or more points; and in the last five minutes of the game, trailing by any amount. It also excludes the last 10 seconds of the first half, and it adjusts for when a play doesn't actually record as fourth-and-short because of one of those bogus delay of game penalties that moves the punter back five yards. Only regular season is included.
One thing you might notice is that both actual fourth-down attempts and expected fourth-down attempts were up in 2020 despite overall fourth-down opportunities being down. This is what happens when offenses are at their highest level in league history. Offenses were more likely to get a new set of downs before ever reaching fourth down, which meant fewer overall opportunities. But when offenses did reach fourth down, they were more likely to get eight or nine yards before they got there, which meant more expected attempts. Total qualifying fourth-and-1 opportunities went from 302 in 2019 to 366 in 2020. Total qualifying fourth-and-2 opportunities also went up, from 202 to 229.
The league averages for aggressiveness have changed so much over the last three seasons that we may need to re-figure the baselines at some point in the near future. Eventually, we're going to end up with two different versions of Aggressiveness Index: one that's centered around 1.0 for the years since 2018 and one that's centered around 1.0 for history so that we can still compare current coaches to coaches of the past.
Aggressiveness Index comes out with Kliff Kingsbury of Arizona as the most aggressive head coach of 2020, at least when compared to historical averages. Kingsbury's AI of 3.12 is the second highest ever measured, trailing only John Harbaugh's 3.95 in 2019. Doug Pederson's AI of 2.64 is the third highest ever measured. 11 of the 20 historical seasons over 2.0 (minimum 60 opportunities) have taken place since 2018.
Matt LaFleur of Green Bay went for it the most often on qualifying fourth downs, 25% of the time. However, Kingsbury finishes No. 1 in AI because of some very unexpected go for it decisions in fourth-and-mediums. Four times, Kingsbury had Arizona go for it on fourth down with 3 or more to go before the fourth quarter while losing by less than 10 points. He also went for a fourth-and-12 in Week 13 from the Rams 40 while losing by 10, and a fourth-and-4 from the Washington 37 in Week 2 in the third quarter of a game Arizona was winning.
Kingsbury is a surprising No. 1 in AI because he is not judged kindly by WPA-based measures of fourth-down aggressiveness. He went for it a lot, with a league-leading 21 fourth-down gos that qualified for AI. But those were balanced out by some egregious errors, in particular the decision to have Zane Gonzalez try a 49-yard field goal with 1:58 left, losing to Miami by 3 in Week 9. Both the EdjSports model and Ben Baldwin's model had this as the worst fourth-down mistake of the season, and that counts a lot more in a WPA-based measure than it does in Aggressiveness Index, where every decision has the same weight.
An even more surprising rank in 2020 AI may belong Frank Reich of Indianapolis, who has a reputation for being very aggressive but somehow comes out only 23rd in AI. It turns out that Reich went for it a lot in 2020 because he had a lot of opportunity to go for it. Mike Tomlin, who is not known for being an aggressive head coach, ends up in roughly the same boat as Reich. Reich had 13.2 expected gos based on historical baselines. Tomlin had 12.4, while no other coach was above 11. Reich went for it on 14-of-21 qualifying fourth-and-1s. Tomlin went for 12-of-22 qualifying fourth-and-1s. No other coach was above 17 qualifying fourth-and-1s.
Where Indianapolis didn't go for it was in the unexpected situations, such as fourth down with more than 2 to go. In the first three quarters, the Colts only did this once for the entire season: Week 13 against Houston, when the Colts went for fourth-and-4 on the Houston 39 in the second quarter and Philip Rivers threw a 39-yard touchdown pass to running back Jonathan Taylor on a wide-open wheel route.
With the dramatic change in frequency of going for it on fourth down over the last couple years, 33 of 35 head coaches come out with an AI above 1.0. The exceptions are Vic Fangio and Brian Flores. Flores is a surprise because he was ranked ninth in his first season, 2019, but he had a lot of fourth-and-1 opportunities last year and didn't use them wisely, going for it in just 6 out of 17 qualifying opportunities. Fangio is easily last in Aggressiveness Index for 2020, and his AI of 0.54 would have been very low even for one of the seasons before 2018 when the whole league was going for it on fourth down less often. In a couple of days, I'll be running another article with historical Aggressiveness Index numbers going all the way back to 1983. There have been 192 men since 1983 with at least one year as an NFL head coach, and out of that group Fangio's AI over two seasons ranks 185th. Denver only went for it four times last year in the first three quarters of a game. One of those was a fourth-and-5 against Kansas City, losing by 18 in the third quarter, so it doesn't count for AI. The other three times were fourth-and-1, and unfortunately the Broncos didn't pick up any of them.
Here are the full Aggressiveness Index results for 2020.
|Aggressiveness Index, 2020|
Thanks to Jim Armstrong, who came up with the idea of Aggressiveness Index and computes it for every season. Later this week, we'll return with a look at historical Aggressiveness Index going all the way back to 1983.
17 comments, Last at 21 May 2021, 11:47am
#1 by BigRichie // May 11, 2021 - 11:32am
I'm more curious about the data base than the plays. In computing how often teams convert 4th-and-1/and-2/etc., are the "third quarter, trailing by 15 or more points; fourth quarter, trailing by nine or more points; and in the last five minutes of the game, trailing by any amount" results included in that calculation or excluded from that calculation?
#2 by Aaron Schatz // May 11, 2021 - 11:43am
Those trailing plays are excluded from the baseline of how often teams go for it in these situations. Aggressiveness Index is about how often teams go for it, though, not how often they actually convert.
#3 by Joseph // May 11, 2021 - 12:23pm
This might be a big ask, but could you add three additional columns--actual attempts (that are part of the sample, not those excluded); actual conversions; and rate?
It may not be, but it seems logical that coaches whose personnel give them a slightly better chance to convert 4th and short (QB's Taysom Hill, Kyler Murray, Lamar Jackson, and Cam Newton come to mind) might be inclined to go for it more often. It also seems logical that teams who are more successful at those 4th down attempts would tend to go for it more, because of the "hot hand" theory.
#4 by BigRichie // May 11, 2021 - 12:26pm
The analytical raison 'd etre underlying Aggressiveness Index is that teams Should! go for it, leastwise more often than they are/historically have. That's certainly predicated on how often you do convert a (for example) fourth-and-2.
So when you guys state that 'teams have converted 4th-and-2 62% of the time', the outcomes from 'those trailing plays' are being included rather than excluded in that?
#6 by jheidelberg // May 11, 2021 - 5:12pm
"The league averages for aggressiveness have changed so much over the last three seasons that we may need to re-figure the baselines at some point in the near future."
Yes, and as soon as you do the recalculation, the bar will keep moving so much that this will need a constant recalculation of baselines. Once something analytical catches on in sports the trend continues on and on. I expect the AI to accelerate greatly again during the next 3 years.
In a number of years your grandchildren will be saying, "Grandpa you have to be kidding me, they actually used to punt in this situation back in the old days?"
#7 by Bryan Knowles // May 11, 2021 - 5:53pm
Heck, we already do. I love this strategy chart from a 1940s-era guide, imploring young coaches to punt on first down when pinned inside their own 20.
#15 by Noahrk // May 14, 2021 - 1:03pm
That's pretty great, thanks for sharing.Funny that the trick-play zone is inside the 5 and yet they advice to keep using successful plays. I wonder if gainin three yards qualifies as a successful play for them.
#13 by Bryan Knowles // May 12, 2021 - 4:51pm
I should clarify that they're not saying bring on the punting unit, as they would today -- there were no punting units; these are all suggestions for your back to kick the ball away rather than throw or run as an element of surprise.
Teams wouldn't just punt on early downs, but when you go to VERY old rules (the 1910s), you actually had teams opting to kick off after their opponents scored, going right back out there on defense. Possession was considered far less important than field position. Offenses were far less efficient, too -- the odds of a turnover or some other disaster were relatively higher than the odds of moving the ball forwards then they are today. This page is from 1947, which was the first year scoring in the NFL went over 20 points per game -- and it's based on wisdom from the 1930s, where the average was in single digits. In that environment, giving up a field goal or, heaven forbid, a touchdown could be catastrophic.
So you had far more cases of quick kicks -- punt the ball out there when your opponents are in standard defense, and win that field position battle. With the right bounce and roll, you could gain 70 yards of field position. Yeah, you lose the ball, but considering their offense isn't likely to be much better than yours, that's not a bad trade. If your kicker was better than their kicker, you could gradually move the ball down field just by exchanging punts until you were far enough from the shadow of your own end zone to switch the risk-reward ratio in your favor.
I believe Vic Fangio operates under the same basic principles to this day.
#10 by Joseph // May 12, 2021 - 10:18am
I wonder if it shouldn't be calibrated around 1.0 every year, or a rolling average every 2-3 years. Only downside would be looking at Fangio and realizing his lack of aggressiveness is mind-boggling. Although, given the QB situation in DEN last year, somewhat understandable.
#17 by jgibson_hmc95 // May 21, 2021 - 11:47am
I took the rate percentage in the last column for the years displayed here (2014-2020) and compared it to the mean average points/drive for the same years (just summed up the individual drive stats and divided by 32, so it's not exactly the true league average but should be close), and I got the following: PTS/Drive = 0.0521*(4th Down Percentage) + 1.55; R^2 = 0.896. Pts/Drive clearly going up as aggressiveness is going up.