Help with calculating WARP (from basketball-prospectus)

[AcrossTheCourt]

Yeah I fixed some things and it's a little closer. EDIT: Here's a link to a spreadsheet with the calculations. A couple things I'm still shaky on.

Link.

Notes:
The usage seems too high.
The formula for TmOReb% seems off. If a guy never gets an offensive rebound the TmOReb is still league average. So I used a factor of 4/5 on the last term.
Teammate FGA% was low. I used a factor of .44 on the last term (LgFTA/LgPoss.)
Are regressed assisted field goals used or actual totals?

Thanks for the help. After tweaking the numbers it's pretty close.

[Kevin Pelton]

All of your assumptions continue to look correct. I think the problem with possessions (and thus usage) is you're subtracting .375*points created instead of .375*assisted field goals made.

[AcrossTheCourt]

All of your assumptions continue to look correct. I think the problem with possessions (and thus usage) is you're subtracting .375*points created instead of .375*assisted field goals made.

Nope I checked. So there are two kinds of usages: normal usage where LeBron is a 0.321 player. And the other where you incorporate assists and assisted field goals where his usage is 0.335.

Sorry I got busy too but I took another crack at it. I updated the numbers. Some of the league totals were off because I was summing player totals (I guess some guys got counted twice but I thought I fixed that.)

Full details:

Points created: 1683+0.75*387-0.75*205.9 (205.9 is the regressed assisted field goals) = 1818.9

Possessions: 1169+0.44*502+213 + 0.375*387- 0.375*205.9 = 1670.81

Offensive rating: PC/Pos*100 = 108.9

Usage: Pos/2326/( (5212+0.44*1598+971)/(16040/5)) = 0.335

TmOffRatMinor: Usage*OffRat+(1-Usage)*(Usage+100*.25*(Usage-0.2)) = 99.9

TmOReb%: (94/2326)/((1897+686)/(16040/5))+0.26935*(1-(94/2326)/((1897+686)/(16040/5))) = 0.3062
(That's way too high, and even if your individual OReb% is 0 your team will be an average offensive rebounding team. I want to place 4/5 as a product of the last term.)

Player RebOpp%: ((1169-621)+0.56*(502-387))/(1169+0.44*502+213) = 0.382

Teammate OffRat: 102.52+0.25*(0.335-0.2)*100 = 105.9

Teammate FGA%: (1 -( (27620/208413)*( 1+(0.5*(0.2-0.335))))-44472/208413) = 0.663

Teammate eFG%: (105.9/100-33465/208413*0.56)/2 = 0.4845
"For the teammates, we must start with their Offensive Rating. From this, we subtract the league average for points from free throws per possession for points from shots from the field. In the denominator, we must take out possessions that end in turnovers (which is affected by the usage rate) and those that end in free-throw attempts. This gives us points from shot attempts -- effective field-goal percentage."

The denominator isn't explained too well. How do I adjust turnovers for usage rate? Here's an alternate attempt for eFG% below:
=((91.5+100*0.25*(0.335-0.2))/100-33465/208412.7)/(1-27620/208412.7*(1-0.25*(0.335-0.2))-0.44*44472/208412.7)/2
=0.5061

Teammate FG%: 0.4845/( ((72218+0.5*12693)/161225) / (72218/161225) ) = 0.4454

Teammate RebOpp%: (1-0.4454)*0.663+0.56*(44472-33465)/208412.7 = 0.3973

TmReb Opp%: 0.382*0.335+0.3973*(1-0.335) = 0.392

TmOff Rat: 99.5+0.306*0.392*1.121*0.8775*100 = 111.3

3A% = 0.093
Tm 3A% = 0.1496

TmOff Rat3: (111.3+7*(0.093-0.1496))/0.96 = 115.6

According to the 2013 season preview I found, the real mark is 115.6. But I think this is a coincidence. I tried it with other places and was off by 1 or 2 offensive efficiency points. I mostly have problems with Teammate eFG% and team OReb%.

League points:
190594

League possessions (simple FGA+.44*FTA+TO method):
27620

LegOffRtgMinor:
91.5

LgOffRtg:
102.52

[Kevin Pelton]

Well, this was useful. I discovered part of the discrepancy is that I was still using an older estimate of assisted field goals to calculate the possessions. That's why we were off there. You were right and I was wrong.

I think the other place we're differing right now is that when I go to Team Offensive Rating, I go back to the traditional value of possessions used. The alternate value, with assists and assisted field goals, I only use to calculate points per possession.

[AcrossTheCourt]

Well, this was useful. I discovered part of the discrepancy is that I was still using an older estimate of assisted field goals to calculate the possessions. That's why we were off there. You were right and I was wrong.

I think the other place we're differing right now is that when I go to Team Offensive Rating, I go back to the traditional value of possessions used. The alternate value, with assists and assisted field goals, I only use to calculate points per possession.

I'm not using that either.

Season totals:
PTS: 190594
FGA: 161225
FTA: 44472
TOV: 27620 (no team turnovers)

League Pos: 161225+.44*44472+27620 = 208412.68

I also still have questions on calculating Teammate eFG%, which was a total guess, and TmOReb%, which seems too high.

Ultimately, what I wanted was to calculate the WARP of every player in 2012 because I wanted to compare WARP to the other metrics via retrodiction (predicting 2013 team wins.)

[Kevin Pelton]

Hmm. That seemed to be the most obvious explanation for why the Team Rating Minor was different.

On the team offensive rebound percentage, that should definitely be 4/5.

[AcrossTheCourt]

Hmm. That seemed to be the most obvious explanation for why the Team Rating Minor was different.

On the team offensive rebound percentage, that should definitely be 4/5.

Changing it to 4/5 changes LeBron's rating to 113.5. I'm still not entirely sure what to do with the Teammate eFG% calculation, particularly the denominator. For another player, Haslem's rating is 102.9 and Battier is at 106.3.

[Kevin Pelton]

I want to make sure we agree on the previous steps before getting to the Teammate eFG% since that's going to affect it.

[AcrossTheCourt]

What part is the problem? The team ratings? I double-checked the totals. I'm not including team turnovers.

Here's another example with Haslem:
PC:
344.2

Pos:
441.5

OffRat:
78.0

Usage:
0.129

TmOffRatMinor:
88.2

TmOReb%:
0.2874

Player RebOpp%:
0.4690

Teammate OffRat:
100.8

At which step is there a problem?

[Kevin Pelton]

We're good through Team OReb%, where I have .299.

[AcrossTheCourt]

0.299 for Haslem? What about LeBron? How are you calculating OffReb%?

Refer back to LeBron's calculation:
TmOReb%: (94/2326)/((1897+686)/(16040/5))+0.2696*(1-(94/2326)/((1897+686)/(16040/5)))*4/5 = 0.2550

I calculated LeBron's OReb% as 0.0502 and the league OReb% as 0.2696. Both of these agree with b-ref's calculations.

[Kevin Pelton]

OK. So the problem with multiplying 4/5 of the league rebound percentage is that the denominator is now too small (it's available rebounds minus player offensive/defensive rebounds, instead of all of them). So my solution was apparently to multiply by (4*League%)/(5-League%). I'm not exactly sure what my derivation was there but it solves the problem.

[AcrossTheCourt]

Well, 4*LgOReb%/(5-LgOReb%) wasn't working, but 4/(5-LgOReb%) brings me to 0.299 for Haslem. LeBron's at 0.2667 now.

[Kevin Pelton]

Then we should -- should -- be good through TeamOffRatMinor: 88.2 for Haslem and 99.1 for James.

[AcrossTheCourt]

Okay I have that for Haslem but for James I'm at 99.52, but I'm at 99.1 if the usage I use is without the assists modification (i.e. using usage=.321 instead of 0.335.)