ATVSQBPI Gets a Facelift

One of the only good things about the offseason is that we all get a chance to take a step back and a deep breath. Maybe we get to start that project we've been thinking about all season, but never have the time. I have all of these untended Excel spreadsheets that need my attention and, finally, I have the time.

Richard Pittman created the ATVSQBPI formula back when this site was in its infancy. It was a useful tool, and the two of us have been trying to refine the formula ever since. When he turned over editorial control to PodKatt and moved into the role of beloved Blogfather, he left the formula in my hands, as well as the responsibility to keep refining it if need be.

I've been spending the last few weeks looking over historical data, trying to back calculate ATVSQBPI over the years, which has also allowed me the opportunity to see how the formula works. Peaking under the hood means I've had the chance to make a major tweak to the formula, which has radically changed the results. Let's look at the whole formula, and why I've made the change.

WARNING: there will be math. If you don't want to deal with math, skip ahead until after the pretty charts, when we'll get into results. The math part is totally optional. I promise there will be fun conclusions as we apply the new formula, and I'll sum up everything in a simple sentence or two for those who skip this part.

OK, ATVSQBPI is built on the principle of yards/attempt. Due to wildly different pace across offenses, advanced stats have tended to look at football on a yards/play basis. The S&P+ formula is yards/play, and if you'll note, when I rank offenses and defenses, I tend to do so on a yard/play basis rather than simple net yardage. This prevents us from downgrading a quarterback simply for not throwing as often or rewarding someone for throwing a ton. Efficiency matters.

Now, I know the general complaint here is that yards/attempt overrates guys who throw it deep. Throw It Deep guy will have a low completion percentage but a huge yards/completion while Dink and Dunk guy will have a huge completion percentage but relatively low yards/completion. And while that's true, yards/attempt deals with this because over enough attempts, the formula punishes a quarterback for lots of incompletions because that is a lot of zeros on the spreadsheet. Meanwhile, a high completion percentage doesn't mean anything if you don't gain any yards. It should punish guys who complete a bunch of passes for no gain. The point is to gain yards. Yards/attempt does not reward a QB for a bunch of completions, but it does punish a guy for a bunch of incompletions. I'll show some examples when we put the full formula together, because ATVSQBPI will further punish Throw It Deep guy.

Pro football does not really need to consider the rushing ability of its quarterbacks. The guys who do run are usually encouraged not to, and eventually that part of their game is filed away, except for a few notable exceptions. Passer rating does not account for rushing, nor should it, as a running quarterback is still considered the outlier in pro football. Not so much in college. There are pocket passers to be sure, but the average college starter had 81.8 rushing attempts last year. That's too big a part of a college quarterback's production to ignore.

Also, college stats count sacks as rushing attempts, so tracking a player's rushing yards also allows us to further punish the immobile quarterback who cannot avoid the rush. Yes, this is partly dependent on offensive line play, but a receiver's catches are dependent on someone throwing him the ball and a running back needs the line to open up holes. Everyone's numbers in football are interdependent. So ATVSQBPI is not just yards/pass attempt but (rush yards + pass yards)/(rush attempts + pass attempts). This is how many yards a quarterback gains every time he calls his own number, whether it be by his arm or his legs.

ATVSQBPI takes the further step of yardage bonuses for touchdowns and interceptions. A touchdown is worth 20 yards, based upon the research of Chase Stuart over at sports-reference. Actually, it's worth between 19.3 and 20.3 yards, but I'm going with the integer value. I'm comfortable with the 20 yard TD bonus in the formula.

However, the value of an interception in the formula has been fixed at 30 yards. This has been based upon the value of a net punt being about 35 yards in college football, plus it maintains a TD/INT ratio for a quarterback of 3:2, making a perfectly average quarterback having his bonuses cancel each other out. The problem here is that, after looking at the data more closely, the average quarterback does NOT have a 3:2 TD/INT ratio. The average college starter last threw for 17.8 TD and 8.6 INT, much closer to a 2:1 ratio than a 3:2 ratio.

Because the ATVSQBPI formula also includes rushing stats, the ratio gets even bigger when we consider rushing touchdowns. The average QB scored 3.6 rushing touchdowns, giving us a total of 21.4 TD. Last year's TD/INT ratio was not quite as steep, though. The average QB threw 9.6 picks to 22.4 total touchdowns. Our TD/INT ratio for the average QB has now grown to a little under 5:2.

Re-reading The Hidden Game of Football, I find that Pete Palmer and crew calculated the value of the average interception not to be 30 yards, but 45 yards.  I can't link to the article, but go get a copy from Amazon.

Anyway, this 45-yard value for an interception makes more sense, and it fits the actual TD/INT ratio of college quarterbacks much better. 45:20 is not quite 5:2, but it is much closer and has basis in research on the value of the play, albeit in the pro game rather than the college one. Also, it occurs to me that we should not be looking at net punt yardage but the average punt without a return. This accounts for the eventuality of interception returns. The average punt last year was a shade over 41 yards. The 45-yard value sits right between the 50-yard bonus to perfect the ratio and the 41-yard value of the average punt.

This convinced me to change the value of interceptions in the ATVSQBPI in the name of accuracy. No longer is a pick worth -30 yards, it is now a 45 yard penalty. The rest of the formula is the same. This gives us the final formula of:

(Pass Yards + Rush Yards + 20*Touchdowns - 45*Interceptions)/(Pass Attempts + Rush Attempts)

It's important to me that the formula stay simple and doesn't stray too far from what it is attempting to measure: how many yards is a quarterback worth per each attempt. Changing the penalty for interceptions accomplishes this by making the formula more accurate without adding any complexity.

An added benefit is this: know who tends to throw lots of interceptions? Guys who throw it deep. An increased penalty for throwing an interception hurts quarterbacks who inaccurately throw deep a lot.

And look at the results. Of the top 20 players in completion percentage, ten of them rank in the top 10 in ATVSQBPI. ATVSQBPI doesn't simply reward throwing a bunch of completions, but the most efficient quarterbacks rank highly. However, ATVSQBPI destroys guys who can't complete a pass. Of the bottom 20 in completion percentage, 15 rank in the bottom 20 of AVSQBPI.

Of the five guys who don't rank in the bottom 20, one missed by one ordinal rank. The others missed the bottom not because of their ability to throw deep, but because they could run the football. They all had over 120 rushing attempts at over 3.0 yards/carry. That sounds like it would drag your numbers down, but here's the thing, that's 120 attempts in which you did NOT throw an interception.

This is illustrated by two guys with near identical ATVSQBPI's despite polar opposite completion percentages. Blake Kemp (ECU) completed 69.3% of his passes and Thomas Woodson (Akron) was at 52.8%. How did they drift to 6.172 and 6.086 ATVSQBPI's, respectively?

Both averaged about the same yards/attempt (7.1), and had near identical TD/INT rations (18/10 and 19/11). But the difference was running the ball. Kemp had 39 carries for -8 yards while Woodson had 603 yards on 135 carries. And those carries count the same as pass attempts, so Woodson actually had more yards per attempt despite a terrible completion percentage. However, Kemp's slightly better ratios resulted in a higher rating.

Let's look at the national numbers:

Welcome back. Here's what you missed: ATVSQBPI is modified yards/attempt now with a larger penalty for throwing an interception. Furthermore, it is not biased in favor of inefficient quarterbacks who throw it deep a lot. It is, however, biased slightly in favor of guys who run the ball a lot because those are attempts in which a guy can't throw an interception. End of summary.

The first that jumps out is that Seth Russell is all out by his lonesome. His ATVSQBPI is a good 2 yards better than the second ranked guy. He's even doing that with a fairly pedestrian completion percentage. 59.5% is the lowest completion percentage of any QB in the top 20, and he's not just sitting at the top, he's doing it by a huge margin. Part of that is an absurd 10.5 unadjusted yds/att. The only other QB above 10.0 is Vernon Adams. He also boasts a 35/6 TD/INT rate, and he can run a little bit. He checks almost every box.

Also hanging out in his own area code is Brandon Doughty. His 9.714 ATVSQBPI is 0.70 yards ahead of the field, still a massive gap. The cluster starts at Vernon Adams, right behind him. Even more impressive, Doughty completed over 70% of his passes at 71.9. He's nearly four percentage points ahead of anyone. If the guys at Football Outsiders are right, this makes Doughty at least an intriguing draft prospect, as the three most important stats for evaluation of a prospect are completion percentage, adjusted yards per attempt (ATVSQBPI without the rushing), and team S&P passing efficiency. Doughty ranks 1^st, 2^nd, and 4^th in those stats. I'm just saying. He's worth a flier, NFL guys.

The surprising name near the top of the ATVSQBPI ranking is Brandon Allen. I previously have described him as the platonic ideal for an LSU quarterback, and he lived up to that billing this season. Allen was top 20 in both completion percentage and ATVSQBPI, flashing his efficiency. He didn't run that well, but he didn't hurt the team there, but most importantly, he did not turn the ball over. Sure, the 9.3 yards/attempt unadjusted was nice, but he added to that by throwing for 30 touchdowns and only 8 interceptions. Efficiency, efficiency, efficiency.

Looking at just the SEC, only 13 quarterbacks had the minimum number of attempts to qualify thanks to the musical chairs at Auburn. Of those qualifying quarterbacks, they fall into a few distinct groups.

Our first group is the elite quarterbacks. Allen, Chad Kelly, and Dak Prescott were all top 20 in completion percentage and all but Dak were top 20 in ATVSQBPI. Prescott slips a bit due to his lower yards/attempt, but picks up ground by being able to run the ball and a phenomenal 39/5 TD/INT ratio. There was just too much ground to make up on Kelly and Allen, who both also had very good ratios.

The next group is our cluster of above average guys. They all posted ATVSQBPI ratings about the national average, but all have different shapes to their production. This is the bulk of the SEC: Brandon Harris, Grayson Lambert, Jake Coker, Kyle Allen, and Josh Dobbs. Now, I will point out that an above average ATVSQBPI got Allen run out of Dodge, so take that for what it's worth.

Coker had the highest completion percentage of the group, even ranking in the top 20 among qualifiers, but he didn't do much else that well. His 21/10 TD/INT ratio didn't help him compared to the guys ranking ahead of him (Lambert and Harris), and he was a statue in the pocket. Of course, he ended the season covered in confetti, so I doubt he's complaining. He's the perfect Alabama quarterback: he didn't hurt the ballclub. Coker did, however, have a ton of attempts, though some of that is a function of playing two extra games.

Lambert was even more of a liability in the pocket, posting the only negative rushing yards among SEC quarterbacks. However, he took taking care of the football to an art form. He threw two interceptions in his 256 attempts. Two. Dobbs, on the other hand, was the least efficient of the group, but managed to climb above the average by scoring a whole bunch of touchdowns, primarily on the ground. He threw for 15, but ran for another 11. If anyone is a mirage of the rankings, it's Dobbs, who racked up the rushing TD bonus like a gamer exploiting a glitch in a video game.

Perry Orth and Treon Harris are the next tier of below average but not terrible. They were bad in the most anonymous and uninteresting ways. Let's move on.

Finally, we have the true dregs of the conference. Three SEC QB's ranked in the bottom 20 of qualifying players in the nation. Patrick Towles showed some semblance of life at least, but Johnny McCrary and Drew Lock both found themselves below the 4.0 threshold, which is unspeakably awful. These three guys all have one thing in common: they threw more interceptions than touchdowns. You simply can't do that and succeed.

Which brings us to Brandon Harris, who was in that second group. First off, Harris' 17/6 TD/INT ratio helps him a lot. He's got a good unadjusted yards/attempt of 7.8, above the national average of 7.3. That's certainly something to build on. He can run a little bit, he takes care of the football, and he has decent averages. But his 53.8% completion rate is a glaring red flag on an otherwise quality resume. An average starter in college now completes 59% of his passes, and it's nearly impossible to post good numbers elsewhere on the stat sheet with a completion percentage so far below the average.

Harris is one of only eight quarterbacks to post a completion percentage under the national average of 59% and still have an ATVSQBPI above the national average of 6.60. The highest ranked QB with a completion percentage under 59% is Quinton Flowers of South Florida, who ranked 27^th in ATVSQBPI with a completion percentage of... 58.9%. The highest ranked guy from a Power 5 conference is 36^th ranked Connor Cook with a percentage of 56.1%. That would be our closest comp, but Cook lacked the running ability of Harris. Harris isn't going to be confused with Cam Newton, but he's a wide margin ahead of Cook's 56 yards on 52 carries.

Harris' closest comp might be Lamar Jackson of Louisville. Jackson posted an ATVSQBPI of 7.07 with a percentage of 54.7%. Jackson is a much better runner than Harris, but he is more careless with the football. Still, they are within the same ballpark of one another.

But the point here is that there really aren't a lot of players who can complete so few passes and still be as effective as Brandon Harris was. He cannot expect a repeat performance of an above average yards/attempt, adjust or not, without improving his accuracy. If he can improve that accuracy, well, Brandon Allen had an ATVSQBPI of 6.56 with a 56.0% completion rate in 2014.

There is platonic ideal. Brandon Harris should put a picture of Brandon Allen up in his locker. That's the exact developmental curve LSU needs his statistical profile to take. Harris can be the platonic ideal, he just needs to channel his inner Brandon Allen.