Good question. I looked up his plays and it turns out Spiller had a large number of negative EPA plays in very low leverage situations. He has a net of -19 EPA when the WPA swings were only between -.01 and +.01.
So how do we interpret those two numbers next to each other? In other words, is it sort of like BABIP in baseball, where an artificially high/low number likely isn't sustainable?
It will very probably regress. The way to really interpret them is that WPA is very heavily dependent on circumstance, most of which are non-repeating. EPA is less dependent on circumstance and is more consistent/predictable. Still, both are retrospective numbers that capture the past.
If you're looking for predictive numbers, I would synthesize something between SR and EPA.
Random question: how does CJ Spiller have a 0.17 WPA (good for 19th among RB), but with a -12.5 EPA (12th worst)?
ReplyDeleteThe only other example I could find of numbers close to those just from quickly skimming the list was Ben Tate (0.27, -8.9).
Good question. I looked up his plays and it turns out Spiller had a large number of negative EPA plays in very low leverage situations. He has a net of -19 EPA when the WPA swings were only between -.01 and +.01.
ReplyDeleteSo how do we interpret those two numbers next to each other? In other words, is it sort of like BABIP in baseball, where an artificially high/low number likely isn't sustainable?
ReplyDeleteIt will very probably regress. The way to really interpret them is that WPA is very heavily dependent on circumstance, most of which are non-repeating. EPA is less dependent on circumstance and is more consistent/predictable. Still, both are retrospective numbers that capture the past.
ReplyDeleteIf you're looking for predictive numbers, I would synthesize something between SR and EPA.
Gotcha. So EPA is probably more representative of his real contributions then, yes?
ReplyDeleteI wouldn't say 'real' because EPA contains low-leverage 'trash' plays. But it's more stable from game to game and year to year.
ReplyDelete> If you're looking for predictive numbers, I would synthesize something
ReplyDelete> between SR and EPA.
Do you know if anyone has looked at median EPA/P (or something similar) as a player stat?