A while ago I did the following thread:
http://www.finheaven.com/showthread...-was-Ryan-Tannehill-Responsible-for&highlight=
...which was based on the following formula for wins added by NFL quarterbacks:
http://www.advancednflstats.com/2007/08/qb-passer-rating.html
The result was unappealing in my opinion and in the opinions of some others here in the forum (available in the thread linked above).
So I decided to do the same thing from another angle, using different statistics.
The following is based on regression model similar to the one above (in red font in the quote), though instead it utilizes the following four statistics, each of which is strongly correlated with winning in the NFL:
1) DVOA (explained here: http://www.footballoutsiders.com/info/methods)
2) WPA (explained here: http://www.advancednflstats.com/2010/01/win-probability-added-wpa-explained.html)
3) Net YPA, which is derived from subtracting sack yards from passing yards, adding sacks to pass attempts, and dividing the former by the latter.
4) The percentage of pass attempts targeted at least 20 yards downfield, which is surprisingly strongly correlated with winning, and which adds a "degree of difficulty" component, if you will, to the equation. In other words, quarterbacks who attempt a larger percentage of their passes downfield (defined as 20+ yards) are likely taking on a more difficult task than those who attempt a smaller such percentage of downfield passes.
The idea here was to develop as complete a picture of quarterback play as possible, and the variables above reflect overall play on a play-by-play basis, taking into account a vast array of other variables (i.e., DVOA), clutch play (i.e., WPA), time-honored measures of individual quarterback play (i.e., net YPA), and "degree of difficulty" if you will (deep pass attempt percentage).
The resulting regression model accounts for 66% of the variance in team win percentage, and each of the above four variables is statistically significant at a p-value less than 0.05.
Using the regression coefficients from the model, we get the following results for NFL quarterbacks for the 2013 regular season:
I'd be interested in comments (or questions) from the members of the forum. :up:
http://www.finheaven.com/showthread...-was-Ryan-Tannehill-Responsible-for&highlight=
...which was based on the following formula for wins added by NFL quarterbacks:
The above was taken from the following webpage:Advanced NFL Stats said:
http://www.advancednflstats.com/2007/08/qb-passer-rating.html
The result was unappealing in my opinion and in the opinions of some others here in the forum (available in the thread linked above).
So I decided to do the same thing from another angle, using different statistics.
The following is based on regression model similar to the one above (in red font in the quote), though instead it utilizes the following four statistics, each of which is strongly correlated with winning in the NFL:
1) DVOA (explained here: http://www.footballoutsiders.com/info/methods)
2) WPA (explained here: http://www.advancednflstats.com/2010/01/win-probability-added-wpa-explained.html)
3) Net YPA, which is derived from subtracting sack yards from passing yards, adding sacks to pass attempts, and dividing the former by the latter.
4) The percentage of pass attempts targeted at least 20 yards downfield, which is surprisingly strongly correlated with winning, and which adds a "degree of difficulty" component, if you will, to the equation. In other words, quarterbacks who attempt a larger percentage of their passes downfield (defined as 20+ yards) are likely taking on a more difficult task than those who attempt a smaller such percentage of downfield passes.
The idea here was to develop as complete a picture of quarterback play as possible, and the variables above reflect overall play on a play-by-play basis, taking into account a vast array of other variables (i.e., DVOA), clutch play (i.e., WPA), time-honored measures of individual quarterback play (i.e., net YPA), and "degree of difficulty" if you will (deep pass attempt percentage).
The resulting regression model accounts for 66% of the variance in team win percentage, and each of the above four variables is statistically significant at a p-value less than 0.05.
Using the regression coefficients from the model, we get the following results for NFL quarterbacks for the 2013 regular season:
| QB | WINS ADDED |
| Peyton Manning | 9.42 |
| Nick Foles | 8.47 |
| Philip Rivers | 7.66 |
| Drew Brees | 6.66 |
| Aaron Rodgers | 6.56 |
| Colin Kaepernick | 5.11 |
| Russell Wilson | 5.10 |
| Tony Romo | 3.98 |
| Matt Ryan | 3.92 |
| Tom Brady | 3.88 |
| Jay Cutler | 3.69 |
| Ben Roethlisberger | 3.43 |
| Matthew Stafford | 3.13 |
| Andy Dalton | 2.98 |
| Andrew Luck | 2.85 |
| Carson Palmer | 2.85 |
| Sam Bradford | 2.73 |
| Cam Newton | 2.66 |
| Ryan Fitzpatrick | 1.68 |
| Matt Cassel | 1.63 |
| Alex Smith | 1.10 |
| Mike Glennon | 0.88 |
| Ryan Tannehill | 0.57 |
| Jason Campbell | 0.18 |
| Robert Griffin III | 0.04 |
| Chad Henne | -0.34 |
| Joe Flacco | -0.38 |
| Matt Schaub | -0.60 |
| EJ Manuel | -0.84 |
| Eli Manning | -1.01 |
| Case Keenum | -1.14 |
| Geno Smith | -1.37 |
| Brandon Weeden | -3.47 |
| LEAGUE AVERAGE | 2.48 |
| LEAGUE STANDARD DEVIATION | 3.07 |
| TANNEHILL PERCENTILE RANK | 31.2 |
I'd be interested in comments (or questions) from the members of the forum. :up:
