This is a short follow up to my previous post concerning QBR. A quick summary of that post for those that do not want to wade through that lengthy and somewhat dry argument:
QBR is driven by EPA. While EPA is a good descriptive and predictive statistic for overall team performance, trying to assign portions of EPA to individual players introduces significant variance in the resulting value ascribed to a quarterback's performance. (In other words, similar performance can produce wildly different ratings based on game situations.) This problem is compounded by the subjective nature of the process used to divide EPA among players, which is highly vulnerable to either intended or unintended bias. Lastly, the QBR process applies a multiplier based on game situations which can add further variance to ratings of similar performances. I concluded, to no one's surprise, that QBR was simply not a very good descriptive statistic.
Field Guller Tokyo Slim posted a comment that had the ring of truth to it concerning QBR as a marketing tool. I believe that Tokyo Slim is largely correct in that were QBR to become the de facto statistic for rating quarterback performance, then as a proprietary stat that requires ESPN employees to calculate, ESPN would, in effect, own a monopoly on rating quarterbacks. Given the incredible popularity of the NFL, this would indeed constitute a very powerful marketing tool.
ESPN is now so heavily invested in QBR and so obviously waging war to have it become the preeminent QB rating statistic, it is likely that they are no longer objective concerning its validity.
However, it seems probable to me that QBR was originally conceived of in a more honest intellectual light and that ESPN wished to construct a statistic that was more accurate in assessing overall quarterback value than the traditional Passer Rating. Some interesting work was being done with EPA and they decided that was the future. I have made my case that that decision was a mistake and will not belabor it. However, it was not a dishonest endeavor, nor a blatant marketing tool, in its conception, in my humble opinion.
Some of the flaws in Passer Rating (PR) that it can be surmised ESPN wished to address are the failure of PR to give value to the running ability of a quarterback, the disproportionate value ascribed to touchdown passes, and PR's failure to capture situational (i.e. clutch) performance.
As for clutch performance, after analyzing what I conclude to be major flaws in QBR, I believe that trying to include a method for analyzing performance under situational stress in a statistic is a fool's errand.
As for the other two issues, a simple revision of PR is enough to give what I believe is a more accurate quantitative assessment of quarterback value than either of the existing statistics.
I believe that the pass accuracy portion of PR should be left intact, as accuracy and decision making are perhaps the two most important traits a quarterback can possess. However, by replacing passing yards with total yards passing and running and adding the number of rushing attempts to the number of pass attempts, the yardage per attempt portion of PR is easily adjusted to include rushing numbers.
Interceptions can be replaced by total turnovers per combined rushing and passing attempts.
In order to address the touchdown to attempt ratio that probably over estimates the value of a touchdown pass, I suggest replacing touchdowns with explosive plays. I used 10 yard runs and 25 yard passes because this data was readily available.
I also suggest using a multiplier for rushing yardage based on the idea that rushing yards are more valuable than passing yards. I believe this is especially true of quarterback rushing because a dual threat quarterback stresses the defense in ways a traditional pocket passer does not and, in theory, creates yards for the running game that will not be attributed to the quarterback otherwise. I used a very modest 1.2 multiplier.
I also divided the resulting numerator by 9.5 instead of 6 in order to to have the resulting value lie within a 0 to 100 range.
I wish to stress the non-proprietary nature of this system. I do not think Passer Rating is broken, but that it merely needs minor changes to its parameters. I invite any and all to fine tune this idea to create an easy to use, non-proprietary statistic that is of greater value than either QBR or PR.
In honor of Marshawn Lynch, I dub it Awesomer Passer Rating (APR).
Using the parameters described above, I compiled the APR of the top 26 quarterbacks in the NFL in 2014. I will just leave that here:
Player |
APR |
74.1249 |
|
72.62785 |
|
Ben Roethlisberger |
67.14416 |
66.34823 |
|
Peyton Manning |
63.87774 |
62.87417 |
|
61.94725 |
|
63.64013 |
|
63.54103 |
|
60.0962 |
|
59.67676 |
|
58.43947 |
|
59.22711 |
|
58.31872 |
|
58.48869 |
|
58.86833 |
|
56.62098 |
|
58.03815 |
|
58.78789 |
|
57.32161 |
|
57.70853 |
|
56.37198 |
|
55.0396 |
|
54.90867 |
|
53.27062 |
|
49.0228 |