Is N. Suh Worth the Money? | An NFL Analytics Perspective

you guys should do karate in the garage.

 

Might as well chime in as usual with my opinion on this. So they have 2 graphs, one for cap hit vs. AV (approximate value) and the other for cap hit vs. EPA+ (expected points added).

First off, I think stats like EPA (expected points added) and WPA (win probability added) etc.. form a good foundation (albeit limited) for analyzing any game that has "break points" and isn't played continuously like soccer or hockey. Why? Because you can measure it, and you're measuring the effect of different plays on your ultimate goal. For those unfamiliar with EPA/WPA, just look at the graph here:
http://www.advancedfootballanalytic...s-explained/expected-points-and-epa-explained

That graph shows the probability (across all teams) of scoring X points given that you have a 1st down and you are a certain distance away from the goal-line (they limit it to cases where the score was within 10 pts and only 1st and 3rd quarter). Because you ultimately want to calculate points/wins this is a nice starting point. The expected points (or win probability) added is just the difference between that for the whole league vs. when a particular player is playing. Of course, you can do that for any down, or any situation.

The only critique I have for EPA+ or WPA+ for defensive players is that they care only about cases where the play ended up being a net positive EPA or WPA for the team, and ignore cases where it was not. There is no mathematical reason to do this, and indeed it would make a lot of sense to see graphs showing how much a particular player HURT the team, rather than showing a player not contributing as much as another one.


OK, so EPA/WPA I think are fine starting points, but AV (approximate value) I don't think is by a long shot. Here's the relevant link:
http://www.pro-football-reference.com/blog/?p=466

One has to give huge props to the guy writing that for being transparent and honest about the (massive amount of) limitations to the stat, but that doesn't change the fact the stat has too many problems associated with it. The biggest problem is there are simply too many subjective aspects to it (it's not an objective stat). For example, they try one method of distributing value among different players on offense, they didn't like it, so they just changed it to something they liked more. There is no principle behind this stat except to make it "look" good. Essentially all the weights are made up (all derived weights are derived properly, but they are derived from made up weights just based on intuition).

So, personally I'd dismiss the graph for cap hit vs. AV but assign some credibility to the cap hit vs. EPA+.

Oh, one more note: the discussion in that original link once again doesn't suggest any good way of determining how 2 or more players impact a play, so none of this would answer questions like: what is Suh's expected value if you take into account he'll play with Wake? (I mean, I guess we'll know the answer soon!)

 

That seems like something you're banking on whether you sign $100 free agents or actively limiting yourself to not devoting a lot of money to free agency.

 

I don't really disagree with anything here, especially AV (I almost couldn't care less about awards. As he mentions, there's all kinds of biases at work there). But I think just like with EPA (I've got issues with this too but less so than AV), there's value in looking at cap hit across the league.

 

Yeah, this is a case where stat>eyeball. However, as soon as the discussion turns to the effect of Suh+Wake instead of just the effect of Suh, eyeball>stat (for now).

 

So far Suh isn't worth the money. Long season though.

I will say this. It's not just Suh but a ton of players that many have misjudged ability on if we go by just the 3 games played this season. Today's most searing memory for me is Brice McCain getting shown up over and over.