What is the true impact of a player?
Player Impact Stats: Innovative insight by applying RAPM to other stats beyond points
In my previous post, I took a look at how NCAA play-by-play stint data can give more accurate advanced stats. This post will take that a step further by applying RAPM to other stats beyond just points. Doing so unlocks unique insight which may not show up in traditional advanced stats. We can use this innovative data to better understand the impact that players have on team performance and playing style.
Player Impact Stats
By performing RAPM-like calculations, we are able to control for opposing players and teammates on the floor, isolating the true impact of an individual player. The primary usage of the RAPM framework is to determine how much each player impacts team and opponent points per possession (ORAPM and DRAPM). However, the amount of interesting and valuable information we can learn by applying RAPM to other box score counting stats (ast, blk, reb, etc.) is limitless. For example, we can better understand how well a player attacks the rim (e.g. blocks against, personal fouls drawn) or how a player impacts the game on defense (e.g. opponent pace impact, defensive eFG%). These insights can be used by NBA draft models to better project NBA potential.
Below you can find descriptions of all the innovative stats I am currently calculating. Let me know if you have additional ideas!
| Stat | Description | What measuring? |
|---|---|---|
| aFGP | aDFGP | Adjusted (Defensive) Field Goal Percentage | Player impact to team/opponent FG% |
| aFG3P | aDFG3P | Adjusted (Defensive) Three Point Percentage | Player impact to team/opponent 3P% |
| aEFG | aDEFG | Adjusted (Defensive) Effective Field Goal Percentage | Player impact to team/opponent EFG% |
| a3PAr | aD3PAr | Adjusted (Defensive) Three Point Attempt Rate | Player impact to team/opponent 3PAr |
| aFTr | aDFTr | Adjusted (Defensive) Free Throw Rate | Player impact to team/opponent FTr |
| aORB | aDRB | Adjusted Offensive|Defensive Rebounds | Player impact to team ORB%/DRB% |
| aAST | aASTA | Adjusted Assists (Against) | Player impact to team/opponent AST% |
| aSTL | aSTLA | Adjusted Steals (Against) | Player impact to team/opponent STL% |
| aBLK | aBLKA | Adjusted Blocks (Against) | Player impact to team/opponent BLK% |
| aTOV | aTOVA | Adjusted Turnovers (Against) | Player impact to team/opponent TOV% |
| aPF | aPFD | Adjusted Personal Fouls (Drawn) | Player impact to team/opponent PF% |
| aOPACE | aDPACE | Adjusted Offensive|Defensive Pace | Player impact to team/opponent PACE |
Bayesian Prior
Similar to Bayesian RAPM, I incorporated Bayesian priors to improve reliability, stability, and general performance. The RAPM-style calculations being performed all include an offensive and defensive component (e.g. ORB|DRB, BLK|BLKA, 3PAr|D3PAr). Many of the components have an obvious prior that you can assign for each player (e.g. ORB%, DRB%, 3PAr), however others don't have readily available statistics (e.g. BLKA, D3PAr). Typically, each RAPM calculation is able to utilize a prior for at least one of the two components. PACE is the only calculation which doesn't utilize any prior. Below you can see a list of the priors being used (coefficients based on results from corresponding "vanilla" RAPM-style calculation).
| Stat | Prior | Coefficient | Intercept |
|---|---|---|---|
| aFGP | FGP% | 0.029391 | -0.011547 |
| aFG3P | 3P% | 0.01857 | -0.00603 |
| aEFG | eFG% | 0.037284 | -0.017261 |
| a3PAr | 3PAr% | 4.907318 | -1.696262 |
| aFTr | FTr% | 3.729186 | -1.297644 |
| aORB | ORB% | 0.083874 | -0.394939 |
| aDRB | DRB% | 0.028404 | -0.262161 |
| aAST | AST% | 0.024444 | -0.295840 |
| aSTL | STL% | 0.206034 | -0.372796 |
| aBLK | BLK% | 0.080668 | -0.128338 |
| aTOV | TOV% | 0.041269 | -0.827779 |
| aPF | PF% | 0.058135 | -0.353120 |
Results
Below you can view players with the highest and lowest values across the different stats (going back to 2010). These results are currently filtered only to players who were selected in a NBA draft. Leading up to the 2019 draft, I will hope to share similar insight on 2019 draftees from these innovative stats - follow me on Twitter to hear more. Lastly, I am very excited to incorporate these innovative stats into my NBA draft model (results coming soon!).
| Name | College | Year | aFGP | FG% |
|---|---|---|---|---|
| TJ Leaf | UCLA | 2017 | 0.035 | 0.617 |
| Georges Niang | Iowa State | 2013 | 0.033 | 0.515 |
| Georges Niang | Iowa State | 2016 | 0.033 | 0.546 |
| Donte DiVincenzo | Villanova | 2018 | 0.031 | 0.481 |
| Jakob Poeltl | Utah | 2015 | 0.031 | 0.681 |
| Will Barton | Memphis | 2012 | 0.031 | 0.509 |
| Larry Nance | Wyoming | 2014 | 0.03 | 0.544 |
| T.J. Warren | NC State | 2013 | 0.03 | 0.622 |
| Justise Winslow | Duke | 2015 | 0.03 | 0.486 |
| Anthony Davis | Kentucky | 2012 | 0.029 | 0.623 |
| Marcus Morris | Kansas | 2011 | 0.029 | 0.57 |
| Doug McDermott | Creighton | 2012 | 0.029 | 0.601 |
| Rondae Hollis-Jefferson | Arizona | 2015 | 0.029 | 0.502 |
| Justin Patton | Creighton | 2017 | 0.028 | 0.676 |
| Robert Williams | Texas A&M | 2018 | 0.028 | 0.632 |
| ... | ||||
| Kentavious Caldwell-Pope | Georgia | 2012 | -0.007 | 0.396 |
| Sindarius Thornwell | South Carolina | 2015 | -0.007 | 0.34 |
| George King | Colorado | 2014 | -0.007 | 0.282 |
| Tony Carr | Penn State | 2017 | -0.008 | 0.377 |
| Melvin Frazier | Tulane | 2016 | -0.008 | 0.401 |
| Nate Wolters | South Dakota State | 2010 | -0.008 | 0.381 |
| Isaiah Cousins | Oklahoma | 2013 | -0.009 | 0.279 |
| Josh Richardson | Tennessee | 2012 | -0.009 | 0.353 |
| Josh Selby | Kansas | 2011 | -0.009 | 0.373 |
| Grant Jerrett | Arizona | 2013 | -0.01 | 0.409 |
| Frank Kaminsky | Wisconsin | 2012 | -0.01 | 0.411 |
| Johnny OBryant | LSU | 2012 | -0.013 | 0.399 |
| George King | Colorado | 2016 | -0.013 | 0.446 |
| Abdel Nader | Northern Illinois | 2012 | -0.014 | 0.337 |
| Abdel Nader | Northern Illinois | 2013 | -0.018 | 0.337 |