APBRmetrics

This is using statistics from the Hawks-Pacers series. Let's assume the offense Paul George is involved in is league average (around 1.07 PPP via nbawowy). He has a usage of 23 and a PPP of 1.06. It seems that with these numbers he's hurting the offense and is overall an inefficient offensive player (not including assists of course. and his usage of 23 will boost other's efficiency, but that's not the point of this exercise).

But George wasn't involved in a league average offense. In this series the Pacers scored only 1.039 PPP. When we use that instead of 1.07 PPP, George's marginal value of using a possession is actually a positive, and he uses a fair amount of possessions.

So, George was an inefficient player, but helped his team win. Put him on the Spurs though, and he would be a clear negative. This kind of gets into the whole predicting vs. explaining thing, as using league average is probably better predictively, team context better at explaining. While a player like George may not have helped a league average team on offense in this situation, he helped his team, and it feels weird to not consider that. I probably lean towards using league average, because it puts players on an even field and is better predictively, but would be interested in hearing other opinions on this topic.

This comes out in things like WARP. An average offensive player on a bad offensive team will be better than average relative to his team etc. However, to make players comparable between teams you need to measure to some common origin. However (again), thankfully, you can take your team-context value and easily create a league-context value, and vice versa.

I assume a player will have a higher efficiency when playing in a higher efficiency offense.
This may not be true in every case, but it's more likely than assuming his efficiency is constant in all offenses.

It also seems likely that as a player expends more energy on defense, his offense may suffer. And so, playing with a good defense -- such as the Pacers -- we might expect a player like George to have better offensive numbers on a lesser defensive team.

A few years ago, someone "analyzed" the 1998 Finals, and decided Michael Jordan was a drag on the Bulls' offense -- he only shot 51% or so. Never mind that this was better than the rest of his team, and considerably better than the Jazz' shooting.

Context is everything.

Jacob Frankel wrote:
So, George was an inefficient player, but helped his team win. Put him on the Spurs though, and he would be a clear negative. This kind of gets into the whole predicting vs. explaining thing, as using league average is probably better predictively, team context better at explaining. While a player like George may not have helped a league average team on offense in this situation, he helped his team, and it feels weird to not consider that. I probably lean towards using league average, because it puts players on an even field and is better predictively, but would be interested in hearing other opinions on this topic.

Put him on the Spurs and he's probably more efficient, even at the same usage, but possibly much more efficient at a lower usage, like a Kawhi Leonard.

Notice that when Monta Ellis (inefficient anywhere) went to the Bucks he actually became more inefficient, because he was surrounded by...other Bucks .

At one point, I wanted to modify ezPM to reflect this. My thought was that basically teammates "pay" each other to take shots. The more inefficient you are, the more you pay other players to take shots. The guy who takes the shots gets paid. Something like that.