As we all know (or should know, anyway), the NFL passer rating is strictly a measure of passing efficiency; it does not take into consideration a quarterback's rushing statistics. The four statistical categories that comprise the NFL passer rating are completion percentage, yards per attempt, touchdowns per attempt, and interceptions per attempt. The formula was designed in the 1970s to rate an average performance for a quarterback of that era as 66.7 and an excellent performance as 100.0.

The nice thing about this formula is it is an "absolute" measure that allows us to compare quarterback passing performances across eras. The disadvantage is that it the scale is artificially flattened at both ends, such that it's impossible to get higher than a 158.3 or lower than a 0.0, and it doesn't take into consideration a quarterback's rushing attempts or his fumbles. It's a passer rating, not a quarterback rating. The other problem is that it fails to account for season-to-season variation in leaguewide performance, so it makes it more difficult to get a feel for how quarterbacks are performing relative to each other on a year-to-date basis. For example, a passer rating of 110.0 looks great on paper, and if the league average is 85.0, it really is impressive; but if the league average is 105.0 this year, a 110.0 rating doesn't stand out nearly so much.

For these reasons, I thought I would try my hand at designing a more open-ended, relative quarterback rating system that would take passing, rushing, and turnovers into consideration while minimizing the effects of changes in the landscape of the league from season to season.

Below is a table comparing three different preliminary versions of the rating system, so you can get a feel for how they vary in the results they return. I want to get your feedback as to which one you think returns the most useful results and/or your suggestions for ways to improve the formulas further.

Rating a weights the following statistical categories equally: yards per passing attempt, yards per rushing attempt, touchdown:turnover ratio, and completion percentage. The weighting of the rushing category hurts quarterbacks who are statues in the pocket, while the inclusion of completion percentage favors dink-and-dunk quarterbacks.

Rating b weights the following statistical categories equally: yards per play, touchdown:turnover ratio, and completion percentage. Lumping rushing and passing together in a single category like this inherently weights rushing and passing yardage by the number of attempts; this favors immobile quarterbacks at the expense of rushing quarterbacks, since not even the most run-happy quarterback rushes nearly as often as he passes.

Rating c weights the following statistical categories equally: yards per play and touchdown:turnover ratio. The advantage of this method is its simplicity; it measures a quarterback's efficiency at moving the ball down the field and avoiding turnovers. The disadvantage is that it could arguably be considered more of an offensive rating than a quarterback rating, although it really isn't, since it ignores rushes by anyone not the quarterback.

It's worth emphasizing that this is a relative quarterback rating. It's not based on a body of statistics from prior seasons -- it rates quarterbacks against each other with respect to their relative performances this year. It gives you the ability to more consistently and objectively define terms like "elite" and "mediocre" from season to season. If you want to define an elite quarterback as one whose performance is in the 95th percentile or above (my personal definition), then it's obvious that there is only one quarterback in the league this year who is playing at an elite level across statistical categories.

Player
Rating a
Rating b
Rating c

Drew Brees
81.05 99.43 104.49
Philip Rivers 51.86 65.77 69.45
Patrick Mahomes 62.02 64.85 68.53
Jared Goff
59.16 65.98 68.06
Ryan Fitzpatrick 61.65 63.52 66.38
Matt Ryan
61.16 65.37 62.97
Mitchell Trubisky 61.32 54.96 56.01
Aaron Rodgers 52.54 50.39 55.95
Russell Wilson 58.92 55.25 55.67
Deshaun Watson 56.45 53.73 55.09
Ben Roethlisberger 51.60 53.65 53.12
Jimmy Garoppolo 47.61 46.52 52.46
Carson Wentz 56.39 58.17 52.33
Jameis Winston 54.08 50.37 50.05
Derek Carr
50.03 55.31 47.27
Tom Brady
43.06 48.72 47.12
Cam Newton
54.44 51.94 46.84
Eli Manning
44.76 50.94 46.74
Andy Dalton
46.58 46.07 46.38
Kirk Cousins 50.89 54.15 45.84
Case Keenum
45.94 46.04 45.57
Ryan Tannehill 50.96 48.28 45.37
Marcus Mariota 53.22 48.98 44.72
Matthew Stafford 46.90 48.63 44.51
Blake Bortles 49.71 42.94 44.47
C.J. Beathard 43.25 41.42 43.58
Baker Mayfield 46.90 42.44 42.94
Dak Prescott 49.39 44.72 42.81
Alex Smith
45.39 44.02 41.76
Andrew Luck
44.55 46.15 41.56
Brock Osweiler 41.37 43.14 41.38
Joe Flacco
37.65 38.57 38.08
Sam Darnold
31.91 30.52 35.57
Tyrod Taylor 37.20 23.52 33.72
Josh Rosen
36.32 29.93 33.45
Josh Allen
33.77 25.63 29.77


Legend:

Standard deviations
Scale
from mean
Percentile
Evaluation
100
+3.0
99.9
Stellar

50
0.0
50.0
Average

0
-3.0
0.1
Shitty


If anyone wants an in-depth description of the mathematics behind the three rating systems, I'll be happy to provide it.

What do you people think? What did I get right? What did I get wrong? What did I miss that I should have included, or what did I include that I should have left out?

Somehow, plays per turn over at the fault of QB.
Passes that hit receivers hands and should have been caught, count as completion.
Passes that were tipped and picked off, do not count as even a pass attempt.
Passes thrown away due to coverage.
Sack ownership.

A few things that stand out to me:

1) When it comes to efficiency, Drew Brees is absolutely the class of the NFL this year. There is no one remotely on his level. These ratings look favorably on quarterbacks who protect the football, and no one is protecting the ball as well as Brees this season. His 24.0 TD:TO ratio is probably unprecedented in league history. (By way of comparison, Philip Rivers has the next best ratio at 5.3.) Not to mention Brees' 77.3 completion percentage is 6 points higher than Derek Carr's (71.5) and Kirk Cousins' (71.3). That is simply unearthly. Brees is probably going to run away with the MVP vote his year, if Mahomes doesn't steal it from under his nose.

2) Aaron Rodgers' pedestrian yardage averages are dragging down his ratings; in both passing and rushing, he has been thoroughly middle-of-the-road this year. More importantly, his three lost fumbles loom large. If it weren't for the lost fumbles, he'd be a top-three quarterback this season in all three systems. It's just hard to compete with a quarterback who's only lost the football once this season despite passing over 33 times a game. Remember, these are relative systems. If Brees suddenly starts losing the football , Rodgers' rating will shoot back up.

3) Rookies will be rookies. Three of the bottom five and four of the bottom ten quarterbacks so far this season are rookies. On the other hand, given the right coaching, quarterbacks can develop quickly: two of the top five quarterbacks have been in the league less than three full seasons.

Zero2Cool;406968Somehow, plays per turn over at the fault of QB.

What?

Zero2Cool;406968Passes that hit receivers hands and should have been caught, count as completion.
Passes that were tipped and picked off, do not count as even a pass attempt.

Ha! You have the money to hire an army of nerds to watch tape?

Zero2Cool;406968Passes thrown away due to coverage.

This would be feasible if I had a parser to collect that kind of data from the play-by-plays posted on the NFL website. I worked on one with a buddy a few years ago and even got it working reasonably well, but I eventually had to abandon the project.

Zero2Cool;406968Sack ownership.

Sacks aren't included in this rating system yet. I was thinking of putting them in there, but again, without someone to watch tape, it would be hard to decide which sacks were the fault of the line and which were the fault of the quarterback. In the end, I suppose, all quarterback rating systems end up being offensive rating systems.

Plays the QB is responsible for the turnover. Fumble because QB is dumb. Interception because he threw it into coverage. Exclude tip at line passes caught and picked. Let's get serious with this! Let's go!


Parsing the data isn't the hard part. It's storing in a manner that's retrievable that caused me headaches.

It's not hard designing a parser that will get the data 90% or 95% right. The hard part is that last 5% or 10%. There are so many subtle outliers. A few NFL statisticians seem to have a fondness for writing up plays in their own idiosyncratic ways; some of them use slightly nonstandard phrasing, others include details most other statisticians don't, and a few include less information than others. Those are the cases on which the parser tends to gag, and although it's only a few plays a week, it adds up over the course of the season. Some of the errors are subtle, so you have to be on the lookout and do a lot manual crosschecking while you're designing the parser rules. But it's definitely something I'd love to get operational one of these days. NFLGSIS makes it very easy to download the data.

Here's a fun example:

(5:04) (Shotgun) D.Brees pass short right to M.Thomas for 5 yards, TOUCHDOWN.
Brees throws 501st career touchdown pass.

Here are some examples (two consecutive plays, actually) where the statistician judges the receiver to be at fault for an incompletion:

(11:36) (Shotgun) D.Anderson pass incomplete short middle to C.Ivory.
Dropped at NE 41, crossing from right.
(11:31) (Shotgun) D.Anderson pass incomplete short right to K.Benjamin.
Through receiver's hands, along sideline at NE 30. Penalty on BUF-K.Benjamin,
Offensive Pass Interference, declined.

So it's certainly doable.

You can do it! I got faith in you.