In our unending quest to make this preview as dorky as possible, we're kicking off the NFL review not with an overview of the leagues, but an overview of stats. In the offseason, BartCop Sports scientists (er, me) have worked a round the clock (um, when I have bouts of insomnia from eating too much cookie dough) to develop two new stats.
First off, raise your hand if you understand the QB rating. Anyone? Bueller? Bueller? So, I've come up with a new QB rating system. Also, there is no way to rate offensive linemen, so I came up with a system there as well. I couldn't rate the individuals because I couldn't think of a way, but I'm open to suggestions. Check it out.
>BartCop NCAA page">NFC preview
QB RATINGThe biggest problem is that the QBR means nothing. I have no idea what a 90 means vs. a 100, other than the fact a 100 is better. But why? Look, Pennington and Bulger were the two highest rated passers and that cannot be right.
As Gioia knows, I love the stats yds/attempt. It's a blunt tool, but it rates QB's the same way we rate RB's or WR's. It's also fair to a guy in a high-completion but low-yardage/reception West Coast offense vs. a guy in a low-completion but big gain Vertical Passing attack. They'll rate the same if they are both of equal quality (really). However, it doesn't include three factors which are vital to a QB's success (or failure): INT's, TD's, and sacks. The problem has always been coming up with a value for each of these events. Well, I've crossed that bridge and come up with a modified yds/attempt.
GIOIA: For QBs, one stat isn't enough to provide a semi-comprehensive evaluation because of the nature of the job. QB rating is an attempt to express it as such, but it fails because we can't express yards, attempts, interceptions, etc. in the same terms mathematically. Any attempt to do so is going to fail no matter how well-thought.
This is a good point, and one I concede. The One True Stat is sort of the Holy Grail of baseball sabermetricians, and I'm sort of putting the cart before the horse. Baseball has a terrific statistical record, and because of that, they can work on combining these huge and varied statistical records to come up with a singular stat: whether that is TPA, Linear Weights, RC/27, or Win Shares.
But I'm not trying to come up with the One True Stat. what I am trying to come up with is the Best stat to evaluate QB's AS PASSERS. The NFL came up with the passing rating system because it could never make up its mind how to rank the QB's, unlike the running backs (by yards) or the receivers (by catches).
GIOIA: Yards per attempt is a great measure of a QB and it is very easy to derive this since both fundamental stats are well-documented. I propose you augment this to include yards lost. Report the yards/attempt and also the yards/attempt - (yards lost to sacks, interceptions and fumbles). An even better idea would be to make an account of all yards he generates: (yards passing + yards rushing) - (yards lost to sack, fumbles & INTs).
Perhaps it's too hard to get those numbers.
Yards rushing is a pain in the ass. I can't find a source which lists rushing yards by QB. They are just lumped in with all of the RB's and it's too time-consuming to do it myself. But the data does exist.
First off, the current formula for passer rating is:
Rate= 100/24*(Comp*20+Yds+TD*80-INT*100)/Att+50/24
Ignore the constants, they just make the final number more readable. Broken down, the formula is YDS/ATT. Except with bonuses:
- 1 yard for each completion
- 80 yards for each TD
- 100 yard penalty for each INT
So, my system is different in four ways.
- I penalize the QB for sacks
- there is no bonus for completions
- I've changed the bonuses for INT's and TD's
- My presentation of the result is expressed still as yds/attempt
Why I did each of these things. I'll modify the formula as we go, starting with simply YDS/ATT.
SACKS
Believe it or not, the same QB's get sacked the most often, regardless of their o-line. Rob Johnson is a statue with a sslllllllloooooowwwwwww release. So I subtract their yardage lost due to sacks from their yards gained. I also count each sack as an attempt. Just because he didn't get the ball off didn't mean he wasn't trying to throw. A sack is a result of a pass play.
(YDS-SCK YDS)/(ATT-SACKS)
COMPLETIONS
Essentially, a completion is meaningless. According to the NFL system, a QB who throws two completions in two attempts for a total of 20 yards is more valuable than a guy who went 1 for 2 for 30 yards. Accuracy is often a function of the system, particularly the West coast offense. A vertical passing game sacrifices accuracy for big gains, while the West Coast sacrifices big gains for accuracy. It evens out in yds/att, but if you start giving completion bonuses, it will favor the west coast QB.
INT's and TD's
This was the big hurdle. What is the bonus and penalty for a TD and an INT. 100 yards is ridiculous. Unless the QB throws a pick on the goal line and the other team returns it 100 yards, an INT is not worth 100 yards. But obviously, it does cost a team yardage. Also, a good TD/INT ratio is 2 to 1, so if I could determine a value for INT's, a TD would be worth half as much in order to balance the traditional ratio.
Here, I simply stole someone else's work. Pete Palmer has studied the yardage cost of a turnover, and every turnover costs a team between 50 and 40 yards. Actually, fumbles cost 50 and INT's 40, since the INT tends to happen further downfield from the yard of scrimmage. So, if an INT is worth 40 yards, a TD is worth 20, and I can plug those figures into my formula.
(YDS-SCKYDS-INT*40+TD*20)/(ATT+SACKS)
PRESENTATION
I kept it as a modified yards per attempt because it is easier to
conceptualize the data that way. Example:
- Kerry Collins has a passer rating of 85.4
- Kerry Collins will average 6.58 yards for his team every time he
drops back to pass.
Both say the same thing, but one is much easier to conceptualize.
GIOIA: I also suggest an adaptation of the assist-turnover ratio from basketball. Essentially, it would be a points-turnover ratio.
We already do that. It's called TD/INT ratio. And look, I'm not suggesting yds/att should replace every stat, it's just a way to interpret all of this disparate data and put every QB on the same scale. But the raw data is still the most important. I'm just trying to make sense of it.
O-LINE
How do you rate lines? Yds/rush is an easy way to rate their run
blocking, and it's a pretty effective way to rate them. That leaves
passing, which has always been a huge problem. The line can't
control yds gained or even completion percentage. Once the QB gets
rid of the ball, it's out of the line's hands. So the only way to
rate them is based on the sack. At what rate do they allow sacks?
So I ranked the lines by att/sack. Which is a good method of evaluating pass blocking.
The problem here is that I couldn't put those two numbers together to come up with an O-Line rating. Why not? They are on different scales. And while there is wide variance in sacks allowed, most teams are tightly clustered around the mean in rushing. What does this mean? It's a chance for me to brush off my statistics textbook and come up with the standard deviation!
I can come up with an SD score for rushing and passing which are inherently on the same scale. By adding together these two figures I come up with an O-line rating. Then I multiply it by 100 for ease of reading. Any team above 0 is, by definition, above league average on the o-line while every team in the negative is worse. My real problem here is that the Texans were so friggin bad it's skewing my data set.
GIOIA: I think you are on the right track. The dream stat for the pass protection would be either the time to throw, and it's impossible to know that. The problem with the rating is that it is only relative to the other members in the data set. If you want to compare the OL rating of the 02 Jets to the 98 Bengals, you can't except to say how each did relative to the other teams in the league. To compare them directly, you have to re-do the calculations with them in the same data set--and how do you choose what goes into the data set? Just the two teams? All the teams from both seasons? All the teams from 98 to 02?
Damnit, you noticed right away. Actually, I'm not concerned about that it is a relative stat that much (a relative stat is better than none at all). Besides, if you calculate every team in a given season, by definition, the average team will score close to 0. You can compare the 2002 Jets to the 1998 Bengals by saying the 02 Jets scored a -24.989 and comparing the 98 Bengals score. Actually, you should do that for every stat. Saying a guy is better because he threw for more yards in 2002 than another player in 1964 is stupid.
You have to compare each player relative to their league before comparing across era. That's the trap baseball fans fall into all of the time, thinking a .300 average in 2002 is the same as it was in 1952 or 1902.
GIOIA: And don't forget about batted passes & hurries. Those are also indications of their pass blocking. It's easy to keep track of the number of batted balls, but defining a hurry is probably more trouble than it's worth.
Finding batted balls is tougher than you would think. You can find it for the defense, but not the offense. So you can find out who did the batting but not against whom.
So without further ado, here are the new BartCopSports stats:
OLINE rating and the BartCop QB Rating (BCR).
Check it out here.