Including team lead in APM/RAPM calculations?
Posted: Wed Jun 01, 2016 8:42 pm
Hi all,
I was learning to make APM using the guide at countthebasket.com, and had the thought to include team lead as an extra independent variable, since it's semi-common knowledge that teams play somewhat better than usual when facing a deficit. I calculated the coefficient to be 0.75 +/- 0.05 per 100 possessions for the 2007-2008, which is pretty significant. It means that (for instance) facing a 20 point deficit, the trailing team plays 15 points per 100 possessions better than if the game were tied, approximately the efficiency difference between the warriors and the kings this year.
Has anyone else looked into this kind of thing in greater detail? I poked around a bit more and found that the effect is indeed linear to good approximation, and doesn't change too dramatically depending on time left, although I'm sure it must depend on time left in some way. In case anyone's curious, here are my top-25 in APM from the model that accounts for team lead, and from the model that does not account for team lead: http://puu.sh/pdaxe.txt. It shuffles some players around, and I'm not sure if the result is better or worse. It also appears to shrink the coefficients somewhat, which is probably a good thing. What do you guys think?
I was learning to make APM using the guide at countthebasket.com, and had the thought to include team lead as an extra independent variable, since it's semi-common knowledge that teams play somewhat better than usual when facing a deficit. I calculated the coefficient to be 0.75 +/- 0.05 per 100 possessions for the 2007-2008, which is pretty significant. It means that (for instance) facing a 20 point deficit, the trailing team plays 15 points per 100 possessions better than if the game were tied, approximately the efficiency difference between the warriors and the kings this year.
Has anyone else looked into this kind of thing in greater detail? I poked around a bit more and found that the effect is indeed linear to good approximation, and doesn't change too dramatically depending on time left, although I'm sure it must depend on time left in some way. In case anyone's curious, here are my top-25 in APM from the model that accounts for team lead, and from the model that does not account for team lead: http://puu.sh/pdaxe.txt. It shuffles some players around, and I'm not sure if the result is better or worse. It also appears to shrink the coefficients somewhat, which is probably a good thing. What do you guys think?