Converting PER, BPM, or RPM into wins

Post #1 » by **eyriq** » Sat Mar 7, 2015 9:13 pm

I am itching to learn how this is done. B-R provides WS/48 which gives an easy Wins = (MP/48)*WS/48 option when you want to project minute allotment going forward. I know the ESPN trade machine has a PER conversion with estimated wins weighted by expected minutes, and the same is possible for Real Plus Minus or Boxscore Plus Minus, but I don't know how and would like to learn. Pretty Please!

Post #2 » by **bondom34** » Sun Mar 8, 2015 12:58 am

eyriq wrote:I am itching to learn how this is done. B-R provides WS/48 which gives an easy Wins = (MP/48)*WS/48 option when you want to project minute allotment going forward. I know the ESPN trade machine has a PER conversion with estimated wins weighted by expected minutes, and the same is possible for Real Plus Minus or Boxscore Plus Minus, but I don't know how and would like to learn. Pretty Please!

I'm sure someone more knowledgable will answer here eventually, but for a quick suggestion (and it won't take long), I'd recommend signing up for an account/posting on APBR. I think it would be based off of the stat, plus a given number of minutes played/expected to play. Thats usually how season projections are done. I'm not sure of exact ways, but I know when people do statistical projections, they "best guess" a minutes distribution for a team, then use a given metric to project the player's production in plus/minus. It would actually be a good APBR question.

blabla · Post #3 » by **blabla** » Mon Mar 9, 2015 1:23 pm

From here http://www.sports-reference.com/blog/20 ... nus-bpm-2/
"Further, BPM is scaled so that -2.0 represents a theoretical "replacement level" - thus, this concept is easily extended to permit calculations of one player's value over that theoretical threshold - that formula is [BPM - (-2.0)] * (% of minutes played), which is VORP, and interpreted as per 100 team possessions."

I *think* VORP corresponds to "individual wins", but I'm not entirely sure

andyhop · Post #4 » by **andyhop** » Tue Mar 10, 2015 11:27 am

This explains the method http://www.gotbuckets.com/ used to convert their RAPM data into wins http://www.gotbuckets.com/what-is-apm/

Chicago76 · Post #5 » by **Chicago76** » Wed Mar 11, 2015 6:30 am

eyriq wrote:I am itching to learn how this is done. B-R provides WS/48 which gives an easy Wins = (MP/48)*WS/48 option when you want to project minute allotment going forward. I know the ESPN trade machine has a PER conversion with estimated wins weighted by expected minutes, and the same is possible for Real Plus Minus or Boxscore Plus Minus, but I don't know how and would like to learn. Pretty Please!

The plus minus is pretty self explanatory, provided you know what the baseline equates to at a team level in wins, you can use player minutes to assess their contribution fairly easily.

The more difficult one is PER. For that you'd need to regress "wins" from the other metrics vs. some form of minutes weighted PER.

For example, if you know that some form of +/- suggests 7 wins by that player over the season, and their PER is 17, you'd need to determine a fit for:

(PER - baseline level that equates to X wins) x floor percent time.

vs.

those other metrics.

Post #6 » by **eyriq** » Wed Mar 11, 2015 11:37 am

Perfect guys, thanks, exactly what I was looking for.

Post #7 » by **eyriq** » Sun Mar 15, 2015 2:40 am

For weighted PER I find Win Shares = -0.745705195 + (3.36061246*Weighted PER)

PER and WS differ but are pretty close.

For weighted Boxscore Plus Minus I find Win Shares = 4.220775829 + (9.073073247*Weighted BPM)

Chicago76 · Post #8 » by **Chicago76** » Tue Mar 17, 2015 7:03 pm

eyriq wrote:For weighted PER I find Win Shares = -0.745705195 + (3.36061246*Weighted PER)

PER and WS differ but are pretty close.

For weighted Boxscore Plus Minus I find Win Shares = 4.220775829 + (9.073073247*Weighted BPM)

what does the fit look like in terms of r-squared or standard error?