APBRmetrics

Hi, I am an Assistant Video Coordinator for a NCAA Division 1 Women's Basketball Program. In the Women's game, analytics is still in it's infancy and not near what it is in the NBA, or even the NCAA Men's game. Our program through the past two years have implemented some measures of analytics, but we are always looking to better this area in order to give us a competitive advantage. Last season I kept track of On Court/Off Court Numbers for our players as well as different Lineup Plus/Minus, but as most people know on this forum, these are inexact figures and are subject to many extraneous variables such as quality of teammates and quality of opponents. I know there is plenty of information out there on building Plus Minus models, but many of them refer to the NBA's large data sets and play-by-play data from the NBA which is not so readily available at the Women's College level. I was wondering if anyone on this forum can lead me and walk me through the process of building a Plus Minus model from scratch, using the data I have or helping me find a larger data set. I'm familiar with many of the different models and was hoping to follow models in the lines of RAPM or RPM, but considering limited data I might have to consider others. Can someone assist me with this project?

It seems there is PlayByPlay for NCAA Women's basketball, like here http://www.ncaa.com/game/basketball-wom ... ay-by-play

One problem is that, because there aren't that many games in a season, RAPM might not tell you that much, unless a player is really great/terrible

RPM helps with this issue, but is obviously a lot more work

J.E. wrote:It seems there is PlayByPlay for NCAA Women's basketball, like here http://www.ncaa.com/game/basketball-wom ... ay-by-play

One problem is that, because there aren't that many games in a season, RAPM might not tell you that much, unless a player is really great/terrible

RPM helps with this issue, but is obviously a lot more work

It seems to me, Jerry, that a pure APM or even RPM may not have enough data to be of much use. Something like your Same 4 Plus/Minus could be valid in smaller datasets could it not?

Other than that, a box score metric like Box Plus/Minus is valid with smaller datasets as well, but doesn't give you as much value as RPM.

J.E. wrote:It seems there is PlayByPlay for NCAA Women's basketball, like here http://www.ncaa.com/game/basketball-wom ... ay-by-play

One problem is that, because there aren't that many games in a season, RAPM might not tell you that much, unless a player is really great/terrible

RPM helps with this issue, but is obviously a lot more work

J.E., given accessible and usable play-by-play data, what would you recommend would be a good plus/minus model to build for college basketball? Would you know how to build a RPM model?

I think the statistical techniques are published.

http://www.sloansportsconference.com/wp ... thLogo.pdf

AFAICT

Plus minus is just a counting stat.
Adjusted plus minus is a least squares regression.
Ridge adjusted plus minus uses a ridge regression instead of ordinary least squares.

micphy wrote:J.E., given accessible and usable play-by-play data, what would you recommend would be a good plus/minus model to build for college basketball? Would you know how to build a RPM model?

Just to be sure, do you mean "Real Plus Minus" by RPM? If so, since I'm developer of RPM, I should better know how to build it

J.E. wrote:

micphy wrote:J.E., given accessible and usable play-by-play data, what would you recommend would be a good plus/minus model to build for college basketball? Would you know how to build a RPM model?

Just to be sure, do you mean "Real Plus Minus" by RPM? If so, since I'm developer of RPM, I should better know how to build it

J.E., great to know! Yes, I do mean Real Plus Minus. I am a big fan of that and also the WAR values that accompany it. Would you be feeling to help me build a RPM model for our program and or even teach me how to go about the process? Would love to be able to learn how to do it.