How Passer Rating Works

[theogt]

It is emphatically not true. The far simpler (pass yards + 20*(pass TD) - 45*(interceptions thrown) - sack yards)/(passing attempts + sacks) is superior.

There are other options as well.

Poke around Pro Football Reference, Football Outsiders, Advanced NFL Stats a bit.

I have no idea why anyone would call this "far simpler". At a glance, this appears to be a highly flawed formula. The variables are almost certainly not comparably adjusted and would therefore result in a greater amount of outliers in a small sample size.

I also have no idea why anyone would penalize a QB for sack yardage, though I almost certainly can guaranty that including sack yardage would make a higher correlating stat than passer rating.

 

[percyhoward]

Aikman was a perfect example of a QB that the scheme hurt his QB rating. Long passes and we ran the ball with Emmitt inside the 10 yard line. Meanwhile QB's like Chris Miller could achieve a better QB rating thru their scheme. Do I believe the Falcons were more effective passing the ball than Dallas in the early 90's?

I agree with the greater point about schemes, but not to the extent of a Chris Miller. Miller didn't even come close to Aikman in passer rating.

League Rank (5 best seasons)
Aikman: 2nd, 3rd, 3rd, 5th, 6th
Miller: 2nd, 12th, 12th, 13th, 20th

The time Miller finished 2nd, he only played 8 games.

 

[AbeBeta]

Not a passer rating per se, this is actually just the formula for adjusted net yards per attempt.

In the largest sample I've found (22 seasons) comparing the two stats, the difference in the correlation coefficients is .04, with ANY/A edging out passer rating. (This is more or less the result in studies with smaller samples too).

How significant is that .04 in practical use when comparing stats to measure the Cowboys' rankings with Romo? Not very.
Cowboys Ranks in Passer Rating, 2006-12
Passer Rtg: 8th, 3rd, 13th, 6th, 12th, 4th, 9th 7.9 avg.
Adj Net YPA: 7th, 3rd, 12th, 4th, 11th, 5th, 11th 7.6 avg.

Whichever method you use, during the Romo era, Dallas has ranked between 7th and 8th, and closer to 8th.

A .04 difference when you are talking about moving from .51 to .55 is substantial.

However, I am pretty sure these analyses are crap. You can't run the sort of analyses they are on data like wins - wins are "count" variables that do not follow a normal distribution. Their analyses are meaningless -- sadly this is what happens when someone takes a stat course in college -- they learn a bunch of tools but don't really know the limits of the computer programs they employ

 

[percyhoward]

A .04 difference when you are talking about moving from .51 to .55 is substantial.

They might as well be the same stat.

Dallas' average rank during Romo's time at QB is 7.6 (8th) according to one metric and 7.9 (8th) according to the other.
Dallas' defense's average rank during that time is 20.6 (21st) according to one metric and 19.7 (20th) according to the other.

Rankings
ANY/A Differential 2012 (Passer Rating Differential 2012)
1. Den (3)
2. Sea (2)
3. SF (4)
4. GB (1)
5. Atl (5)
6. Was (6)
7. NE (7)
8. Pit (10)
9. Chi (9)
10. Car (19)
11. Cin (11)
12. Hou (8)
13. Bal (12)

23. Ind (27)
27. Min (24)

Good luck finding any evidence of ANY/A's superiority there.

Passer rating's top 8 were all playoff teams, 10 of the top 12 were playoff teams, Indy and Minny don't follow the pattern.
ANY/A's top 7 were all playoff teams, 9 of the top 12 were playoff teams, Indy and Minny don't follow the pattern.

They're basically the same stat.

 

[CyberB0b]

The problem with the formula is that it favors more WCO and dink and dunk passing offenses.

Too bad that is an absolutely flawed assertion. Passer rating takes into account yards per attempt.

 

[AbeBeta]

They might as well be the same stat.

Dallas' average rank during Romo's time at QB is 7.6 (8th) according to one metric and 7.9 (8th) according to the other.
Dallas' defense's average rank during that time is 20.6 (21st) according to one metric and 19.7 (20th) according to the other.

Rankings
ANY/A Differential 2012 (Passer Rating Differential 2012)
1. Den (3)
2. Sea (2)
3. SF (4)
4. GB (1)
5. Atl (5)
6. Was (6)
7. NE (7)
8. Pit (10)
9. Chi (9)
10. Car (19)
11. Cin (11)
12. Hou (8)
13. Bal (12)

23. Ind (27)
27. Min (24)

Good luck finding any evidence of ANY/A's superiority there.

Passer rating's top 8 were all playoff teams, 10 of the top 12 were playoff teams, Indy and Minny don't follow the pattern.
ANY/A's top 7 were all playoff teams, 9 of the top 12 were playoff teams, Indy and Minny don't follow the pattern.

They're basically the same stat.

That is totally incorrect. 26% explained variance vs 29%. That is substantial.

 

[jobberone]

Too bad that is an absolutely flawed assertion. Passer rating takes into account yards per attempt.

Your completion rating goes up more and also if you can avoid penalties and TO then your O does become more efficient. Look at NE % drives resulting in scores (some of which will be TDs and increase rating). NE scored 48+% of the time it got the ball which is VERY good. So dink and dunk does help some. Look at Brady's YPA and then Romo's. Big difference. I find the NE offense BORING but you can't deny the results. Maybe Romo should throw short more.

 

[dwmyers]

I have no idea why anyone would call this "far simpler". At a glance, this appears to be a highly flawed formula. The variables are almost certainly not comparably adjusted and would therefore result in a greater amount of outliers in a small sample size.

I also have no idea why anyone would penalize a QB for sack yardage, though I almost certainly can guaranty that including sack yardage would make a higher correlating stat than passer rating.

Actually, a good chunk of a whole book was written about the superiority of Tricericon's formula, called "The Hidden Game of Football" --> http://www.amazon.com/The-Hidden-Game-Football-Carroll/dp/0446514144

That's where the entire class of adjusted yards per attempt formulas were introduced. Course, for entirely arbitrary reasons, they set the TD term to 10. Later, PFR, in one of the worst articles I've ever read, reset it to 20.

But as you can derive those formulas from actual expected points curves, how factual those formulas are can be tested with real data.

The NFL passer rating can be reduced to a AYA style formula, so yes, it too can be compared to real data.

http://codeandfootball.wordpress.co...ions-and-yardage-truly-independent-variables/

http://codeandfootball.wordpress.com/2011/09/03/the-nfl-passer-rating-as-a-scoring-model/

http://codeandfootball.wordpress.co...ard-per-attempt-to-a-nfl-style-passer-rating/

This is where I actually test the validity of a version of the model Tricericon used..

http://codeandfootball.wordpress.com/2011/10/21/the-valid-range-of-a-linearized-scoring-model/

This tag covers most of what I've written on the subject..

http://codeandfootball.wordpress.com/tag/scoring-model/

 

[AbeBeta]

That is totally incorrect. 26% explained variance vs 29%. That is substantial.

26% to 30% .... mistyped

 

[AbeBeta]

Actually, a good chunk of a whole book was written about the superiority of Tricericon's formula, called "The Hidden Game of Football" --> http://www.amazon.com/The-Hidden-Game-Football-Carroll/dp/0446514144

That's where the entire class of adjusted yards per attempt formulas were introduced. Course, for entirely arbitrary reasons, they set the TD term to 10. Later, PFR, in one of the worst articles I've ever read, reset it to 20.

But as you can derive those formulas from actual expected points curves, how factual those formulas are can be tested with real data.

The NFL passer rating can be reduced to a AYA style formula, so yes, it too can be compared to real data.

http://codeandfootball.wordpress.co...ions-and-yardage-truly-independent-variables/

http://codeandfootball.wordpress.com/2011/09/03/the-nfl-passer-rating-as-a-scoring-model/

http://codeandfootball.wordpress.co...ard-per-attempt-to-a-nfl-style-passer-rating/

This is where I actually test the validity of a version of the model Tricericon used..

http://codeandfootball.wordpress.com/2011/10/21/the-valid-range-of-a-linearized-scoring-model/

This tag covers most of what I've written on the subject..

http://codeandfootball.wordpress.com/tag/scoring-model/

Myers knows what he is talking about!

 

[dwmyers]

They're basically the same stat.

Provably related:

http://codeandfootball.wordpress.co...ions-and-yardage-truly-independent-variables/

http://codeandfootball.wordpress.co...ard-per-attempt-to-a-nfl-style-passer-rating/

 

[TwoDeep3]

Win the Super Bowl and all these stats and ratings mean zip.

 

[Eskimo]

Win the Super Bowl and all these stats and ratings mean zip.

Fascinating insight.

 

[theogt]

That's where the entire class of adjusted yards per attempt formulas were introduced. Course, for entirely arbitrary reasons, they set the TD term to 10. Later, PFR, in one of the worst articles I've ever read, reset it to 20.

This is precisely the problem with this formula. When it was initially created, someone set the multipliers arbitrarily -- most likely to get the formula to spit out whatever they thought was the "right" result. Then someone else came along and said "no, I like this result better, so I'll change the multiplier". Instead, each variable should have a multiplier so that the variables are directly comparable, not because someone decides what the result should be.

 

[Yakuza Rich]

I agree with the greater point about schemes, but not to the extent of a Chris Miller. Miller didn't even come close to Aikman in passer rating.

League Rank (5 best seasons)
Aikman: 2nd, 3rd, 3rd, 5th, 6th
Miller: 2nd, 12th, 12th, 13th, 20th

The time Miller finished 2nd, he only played 8 games.

1992

Chris Miller QB Rating: 90.7

Troy Aikman QB Rating: 89.5





YR

 

[percyhoward]

That is totally incorrect. 26% explained variance vs 29%. That is substantial.

You saw the results when the two stats were put into practical use and compared for the 2012 season, and before that with Dallas' offensive and defensive ranks over the last seven seasons.

If this substantial mathematical difference had any corresponding substantial impact on either one of those sets of rankings, I'd like you to explain where.

 

[percyhoward]

1992

Chris Miller QB Rating: 90.7

Troy Aikman QB Rating: 89.5





YR

Well yeah, Rich that would be that one year that I said Miller only played 8 games. 1992 isn't "the early 90's" any more than 8 games are 16.

 

[percyhoward]

Provably related:

They should be. When put into use (for the practical purpose of ranking players and teams) and compared, they give you virtually the same results.

 

[AbeBeta]

You saw the results when the two stats were put into practical use and compared for the 2012 season, and before that with Dallas' offensive and defensive ranks over the last seven seasons.

If this substantial mathematical difference had any corresponding substantial impact on either one of those sets of rankings, I'd like you to explain where.

Percy - you cannot take a single example and claim it proves your point.

Also the idea PR being related to offensive ranking as being some sort of evidence is preposterous. Of course it is related! It takes many of the same statistics into account in the computation.

Also, you cited correlations with winning -- offensive rank and defensive rank factor into winning but are NOT the same variable.

 

[Super_Kazuya]

Passer rating is kind of interesting in that it is not terribly meaningful and certainly not very useful in a play to play or even game analysis, but if you stretch it out to an entire season and look at the rankings, it usually measures up pretty well with reality.