Add Reply Create a new Topic Start a Poll

 *S3 ANY/A Leaders
Beaver
 Posted: Oct 17 2017, 12:54 AM
Quote

Extremely Rude Boi

Group:
Rookie Mentors
Member ID:
487
Member Since:
26-September 17
Total Post:
991

N/A



CODE
2060 words, typed it up at 3AM like a moron so hopefully no dumb mistakes


As ANY/A has come into vogue among NFL advanced stats nerds as a way of measuring quarterback performance I thought it would be a good entry point to start adapting some of these measures to the NSFL since it has a very straight-forward concept (yards per attempt with some added flourish) and is relevant to the most glamorous position in football. I'm going to start out with an explanation of the statistic as some background (what is it and why should it be used?), if you're not interested in that you can click here to go to the accompanying statistical analysis topic that simply has the tables and a few words. For the rest of you liars, you may proceed.

What is ANY/A?

ANY/A stands for Adjusted Net Yards per Attempt. It began as an attempt to end an age-old bar argument: Quarterback A throws for 220 yards on 30 attempts with 3 TDs and 0 INTs while Quarterback B throws for 400 yards on 50 attempts with 5 TDs and 2 INTs - who had the better performance? In order to answer this you must dovetail the various facets of quarterback play (yardage, touchdowns, interceptions, and so on) into one scale. How many more yards does a player have to throw for to make up for one fewer touchdown? How many touchdowns does a player have to throw to make up for one more interception? Most people don't systematically break it down into actual concrete and consistent values but rather use their own personal heuristics that may shift slightly from debate to debate. We ain't about that life, though.

The origins of ANY/A, like those of many advanced stats, can be found in the 1988 book The Hidden Game of Football by Bob Carroll, Pete Palmer, and John Thor in which they determined that in terms of their effect on the game touchdowns were worth about 10 yards (since revised to 20) and that interceptions were worth about -45 yards. They called their stat (which they touted as the "new passer rating") Adjusted Yards per Attempt or AY/A and in full it was as follows:

AY/A = (PassYards + 10*PassTDs - 45*INTs)/Attempts

So for the hypothetical stat lines above it would work out to:
QBA's AY/A = (220 + 10*3 - 45*0)/30
= 8.3

QBB's AY/A = (400 + 10*5 - 45*2)/50
= 7.2

Since then it has been tweaked (this 2008 article is what changed the touchdown value from 10 to 20) and the only distinguishable difference between AY/A and ANY/A is the addition of another component indicative of a quarterback's performance: sacks. Namely that sack yards subtract from the passing yards total (with the premise being that the quarterback "lost" those yards in a passing situation) and the changing of the denominator from pass attempts to drop backs (pass attempts plus sacks).

The current ANY/A formula in use around the nerdosphere is therefore as follows:

ANY/A = (PassYards + 20*PassTDs - 45*INT - SackYards)/(Attempts + Sacks)

Pro Football Reference has a great database of historical NFL ANY/A performances. Here are the best single seasons by this measure with 2004 Peyton Manning topping the list, here are the quarterbacks with the best career ANY/A with Aaron Rodgers currently topping this list, and here you can find the leaders from the 2016 season which was the first to ever pit the top 2 QBs by ANY/A in the Super Bowl. Virtually any article you read about quarterbacks on Football Outsiders, Football Prospective, the Pro Football Reference blog, or your NFL advanced stats site of choice will include mention of ANY/A these days. It's even been adapted into a team stat (a team's ANY/A compared to the ANY/A that team allows to opponents).

What will this tell us about the NSFL?

Honestly I'm not sure. At a bare minimum it gives us a rate-based way to measure quarterback performance - that is, highlighting quarterbacks who do more for their team on fewer drop backs instead of simply those who play in pass-heavy offenses. There are critical differences between the NSFL sim engine and real life, obviously, and I don't have confidence that those valuations of touchdowns and interceptions will hold up. I'm not sure whether ANY/A will overvalue or undervalue them but it strikes me as unlikely that it will translate seamlessly across. The big question to me is which of ANY/A, AY/A, and NY/A - all variants of the same basic concept - will fit this league best. For the purposes of this post I've used ANY/A but if you, dear reader, think that those touchdown and interception valuations are way off then you should expect NY/A to be more indicative of a good quarterback performance than ANY/A. You can find those numbers in the data topic.

While not its main purpose, some value of ANY/A is that it's a more predictive passing statistic than conventional stats like yards, yards per attempt, QBR, etc. That is, a good passing statistic will capture underlying indicators rather than the random fluctuations different quarterbacks see on a week-to-week and sometimes season-to-season basis and can be used to identify quarterbacks that are over-performing or under-performing their talent. For a better explanation on predictive and explanatory passing statistics than I can provide, you can check out this article.

The biggest benefit from this may not even be a direct use of ANY/A but a simple by-product as there were some interesting things in the sack data I noticed as I was collecting it. Interesting to me, at least. More on that in the previously linked accompanying statistical analysis article.

Stop milking word count and get on with it already

For the most part the components of ANY/A are easy to find in the sim index. Attempts, passing yards, touchdowns, and interceptions are all hallmark counting stats for quarterbacks and are prominently displayed and easily found. Sacks are not listed (at least as far as I know) by quarterback, which wouldn't be a problem if I could grab them from the offensive line stats but alas not every sack taken is assigned to someone.

So compared to the traditional quarterback stats, collecting sack yardage and sacks taken was a bit more involved. Since I couldn't find them listed anywhere I went through each game's play by play and recorded the lost yards for each sack (throwing out sacks taken by the two backup quarterbacks who won't qualify). This was cancerous enough to make me think about just scrapping the idea of ANY/A and simply going with AY/A but not cancerous enough for me to whip up a scraper. If I go back and do historical seasons or continue this going forward I'll definitely be using a scraper to make it easier but when I start out with a project I like to be a bit more hands on to get a feel for things.

In the end, when I was done whining I had a nice looking table like this:

user posted image

Unsurprisingly, Mike Boss of the Orange County Otters comes out on top following an absolutely stellar season. Somewhat more surprisingly is that the gap between him and his competitors closed significantly after accounting for drop backs and sack yardage. To illustrate this:

In pure yardage he was one and a half standard deviations (!!) above second place Ethan Hunt of San Jose (and 1.713 above average).
In touchdowns he was a quarter of a standard deviation above second place Chris Orosz of Yellowknife (1.213 above average).
But he was just 1/30th of a standard deviation above second place King Bronko of Arizona in ANY/A (0.954 above average).

Z-scores (standard deviations above or below average) of all qualifying quarterbacks:

user posted image

The biggest beneficiaries of these adjustments were King Bronko and Scrub Kyubee of Baltimore. Bronko rose from 4th in yardage and 3rd in touchdowns to a 2nd place in ANY/A that nearly entirely erased a 1,000 yard gap while Kyubee separated himself from a pack of quarterbacks with stats around league average into a clear 4th place position. Both of these quarterbacks had fewer than 550 drop backs (548 and 540 respectively, only Pierno had fewer among qualifying quarterbacks) which helped them a lot in a rate-based stat like this, especially in comparison to Boss's 685 drop backs.

Aside from Boss, Clifford Rove of Philadelphia was hurt the most by the adjustments. While on paper his 3,461 yards and 20 touchdowns look very serviceable, right around league average as a young player, ANY/A shows us that those numbers were buoyed by taking nearly 600 drop backs - in addition to those 24 interceptions being problematic in any analysis. He certainly has time to improve going forward, though, if he can work on mitigating those two factors. With league-average interceptions, sacks, and sack yards he can close nearly half the gap between he and Chris Orosz.

What's next?

As is usually the case with data analysis like this, more questions are asked than answered. First and foremost is the ANY/A scale. 6.3 ANY/A doesn't actually mean anything except in relation to other ANY/A numbers. What I mean by that is that in this season of the NSFL a 6.3 was spectacular but in the NFL a 6.3 is not very good: an NFL quarterback with Mike Boss's stat line in the NFL would've been ranked 17th of 32 last season, right below Philip Rivers and barely edging out Ryan Tannehill. Going forward I'd like to add the previous seasons of data to get a better idea of what an "elite" ANY/A number is for the NSFL. Was this an all-time great season by Boss or simply a great season? Because of how much more common sacks are in this sim engine we'll likely never see seasons that are typically considered elite in the NFL (9+ ANY/A) but ultimately that doesn't matter at all. With several seasons' worth of data we'll get a more complete picture of how well this statistic fits into this sim engine, too.

Further down the road I think it'd be interesting to revisit the formula and try to estimate how much touchdowns and interceptions are worth in the NSFL. As of now I don't have a plan of attack for that and am open to ideas but if it's not too complicated I think that'd be interesting to see.

Another question I have is whether sacks should even be included. They're included in the NFL formula because there's some evidence that suggests sacks are - contrary to popular belief - more on the quarterback and less on the offensive line and as such should be reflected in a measurement of the quarterback's performance. If that's not true, or even less true, in the sim than in real life then perhaps the easier AY/A is the way to go. On the other hand if that's more true in the sim than in real life then perhaps sack yards need to be adjusted to reflect their value. I'm not sure how I'd go about trying to figure that out but it's a thought at least.

Finally, I think it'd be interesting to look at the QB attributes to see what correlations there are between them and ANY/A that may be understated in comparisons with YPA, passing yards, or touchdowns. Definitely a long shot to find anything there but it's always fun to look.

Plus, there's all the stuff about sacks from the other topic. Why are nearly three quarters of all sacks either 6 or 7 yard losses? Why are there more 5 yard losses than 8 yard losses? Why are there more 11 yard losses than 8-10 yard losses combined? Is it because sacks are more prevalent in the sim so a greater percentage of them occur at the quarterback's drop back depth? Is there a scramble mechanic in the sim? There almost certainly is since there were two sacks with 0 yards lost. If so, is it working as intended? I would expect a much more even distribution of yards lost in a sim with a robust scrambling mechanic. Does it have something to do with the way we update defensive players, offensive linemen, and quarterbacks? Will we see this same type of distribution in 20 seasons? Or am I overlooking a perfectly reasonable explanation because it's nearly 4 AM and I'm tired? It's entirely possible that this issue has been hashed out and maybe even resolved, as well.


GRADED

This post has been edited by jparks98: Oct 17 2017, 01:30 PM

--------------------
Blackford Oakes - FS
Player Page || Updates

user posted image
^
BOOM™
 Posted: Oct 17 2017, 01:06 AM
Quote

N/A

Group:
Members
Member ID:
372
Member Since:
24-July 17
Total Post:
368





Excellent read and presentation. Awesome work!

--------------------
user posted image
^
JuOSu
 Posted: Oct 17 2017, 02:44 AM
Quote

N/A

Group:
Members
Member ID:
408
Member Since:
21-August 17
Total Post:
494





Great article. I did the same for the prior seasons //files.jcink.net/html/emoticons/smile.gif You should check it out.

--------------------
user posted image
^
PDXBaller
 Posted: Oct 17 2017, 05:03 AM
Quote

N/A

Group:
Head Office
Member ID:
466
Member Since:
6-September 17
Total Post:
1242





Great work! Give this man a stat money bonus!!!

--------------------
user posted image
^
daBenchwarmer
 Posted: Oct 17 2017, 07:03 AM
Quote

N/A

Group:
Members
Member ID:
49
Member Since:
23-May 17
Total Post:
514





so Ethan Hunt confirmed p. good?

--------------------
user posted image


user posted imageuser posted image
^
Beaver
 Posted: Oct 17 2017, 09:25 AM
Quote

Extremely Rude Boi

Group:
Rookie Mentors
Member ID:
487
Member Since:
26-September 17
Total Post:
991

N/A



QUOTE (JuOSu @ Oct 17 2017, 04:44 AM)
Great article. I did the same for the prior seasons //files.jcink.net/html/emoticons/smile.gif You should check it out.

Oh shit, my bad. Here I am thinking I had an original idea //files.jcink.net/html/emoticons/cool.gif . I ran a search for "ANY/A" on the site but it came up with no results despite your article clearly being titled "ANY/A." That TANY/A from your article is a good idea, I'll have to go back and add that to see how it changes things.

It's early of course but I think your theory about offensive lines depressing ANY/A is being borne out. In this post the top ANY/A performance was 5.45 which would only rank third this season. As the lines continue to get better it'll be interesting to see if we get up into the neighborhood of NFL ANY/A values.

QUOTE (daBenchwarmer @ Oct 17 2017, 09:03 AM)
so Ethan Hunt confirmed p. good?

One of 4 QBs to be better than average in Yards, TDs, INTs, sack yards, and ANY/A - consistently above average. Taking 69 sacks hurts but I think it was worth it.

--------------------
Blackford Oakes - FS
Player Page || Updates

user posted image
^
bovovovo
 Posted: Oct 17 2017, 09:30 AM
Quote

The original 3.0

Group:
NSFL GM
Member ID:
266
Member Since:
26-June 17
Total Post:
2030

N/A



Oh shoot this is awesome.

Good stuff my man http://aussielapidaryforum.com/forum/Smileys/classic/cheers.gif

--------------------

user posted image
user posted image

Player Page ... Update Page
^
daBenchwarmer
 Posted: Oct 17 2017, 09:57 AM
Quote

N/A

Group:
Members
Member ID:
49
Member Since:
23-May 17
Total Post:
514





QUOTE (Beaver @ Oct 17 2017, 12:25 PM)
One of 4 QBs to be better than average in Yards, TDs, INTs, sack yards, and ANY/A - consistently above average. Taking 69 sacks hurts but I think it was worth it.


And people are still sleeping on the guy smdh 🙄

--------------------
user posted image


user posted imageuser posted image
^
Beaver
 Posted: Oct 17 2017, 10:33 AM
Quote

Extremely Rude Boi

Group:
Rookie Mentors
Member ID:
487
Member Since:
26-September 17
Total Post:
991

N/A



Wait... this is season 3 isn't it. http://www.planetsmilies.com/smilies/confused/confused0006.gif

Could I get a title edit?

--------------------
Blackford Oakes - FS
Player Page || Updates

user posted image
^
0 User(s) are reading this topic (0 Guests and 0 Anonymous Users)
0 Members:
Share this topic:
« Next Oldest | Graded Articles | Next Newest »

View Topic Options Add Reply Fast Reply Create a new Topic Start a Poll



 


 


Mobile Version Last Visit: --