DYAR Question

[Smelly McUgly]

I don't know if people use DYAR or read FO's weekly DYAR rankings, but if someone does, can you chime in on how I should look at DYAR? It seems like a big influence on it is "successful plays" and "failure plays," but those are tied into basically how "effective" the gain of yardage was on a particular play. So, for example, if Marshawn Lynch gains seven rushing yards on third-and-nine, the value of that (which likely leads to a punt) is less than if Lynch gains seven rushing yards on first-and-ten (because the success rate of getting a first down on second-and-three before a team has to punt is fairly high, so Lynch was successful at likely gaining a first down for his team eventually).

OK, if I have all that explanation right...isn't this just basically judging how much each particular play moved the team's win probability up or down? Couldn't I just read a WP chart and get a general idea of the value of each player based on that? What makes DYAR different than just a cursory aggregation of a player's effect on his team's win probability?

 

[Perfundle]

There are several differences that I can think of.

1. The D part of DYAR. Unlike every other unit-specific ranking system that I know, FO takes opponent adjustments into consideration (although not yet for this current season; I believe it kicks in around week 5). So Lynch running for 5 yards against Jacksonville last year is worth less than running for 5 yards against Arizona.

2. DYAR rounds off extremely large chunks of yardage. It would consider a 50-yard touchdown and a 90-yard touchdown pretty much the same, because both involve a breakdown in the defense and the main factor in how long the TD ends up is the starting field position. WPA does not do this, because the expected points starting at your own 10 are clearly different than the expected points starting at midfield.

3. DYAR does not treat a QB run the same as a HB run, whereas WPA would. The reason is that the average QB run is more successful than the average HB run, because QBs usually only run when they think they can pick up good yardage.

4. DYAR cares only about the fumble itself, not who recovered it. The reason is that who recovers the fumble is mostly luck, whereas forcing fumbles is a skill and being fumble-prone is a trait.

5. DYAR doesn't care about the yardage obtained from running back a fumble or interception, because a lot of that is also luck.

6. DYAR throws out hail-mary type plays.

7. DYAR weighs the time and score of the game far less than WPA does. In other words, garbage time stats are not given almost no weight like they would be in WPA, and last-minute clutch play is treated much the same as play at the start of the game. Instead of seeing what effect a play has on the current game, each play is compared to all plays run with roughly similar point differential and game clock situations.

8. I'm not sure if WPA does this too, but DYAR takes into account stuff like DPI and intentional grounding.

I'm sure there are many other differences, but the one common factor between these is that they were all chosen because they were found to be more predictive of future performance.

 

[Smelly McUgly]

Perfundle, great response, especially since I posted this on the wrong board. I especially forgot the most obvious difference - the adjustments based on who the player was playing against.

Thanks for taking the time out to respond like this.

 

[RolandDeschain]

Nice work, Perfundle. Smelly, I highly recommend reading this page - it's kind of lengthy, but it details how they do what they do...And why it has SO much credibility and overall accuracy:

http://www.footballoutsiders.com/info/methods

The only thing that doesn't have a "DVOA level" of accuracy are their offensive line rankings. I emailed them last year with some questions about how they rank O-lines, and it's kind of simplistic compared to how they calculate DVOA for other stuff, so take those specifically with a grain of salt. Pretty much everything else they rank and rate is gold, though.

As Perfundle said, though - a HUGE part of their entire system is predicated on how good or bad your opposition is, and win/loss record is almost irrelevant as far as how they do it. They know there are elite players on crap teams and adjust accordingly, so if a great corner on a crap defense is covering a great receiver, a TD catch by that receiver's worth a lot more than say, a second TD catch by that receiver later on in the same game but against a mediocre corner.

They don't start calculating opponents until after week four, though. They need a month's baseline to start shooting up the ladder in terms of accuracy. Don't read much into their DVOA ratings before week five's numbers come out.