I highly recommend reading this post before attempting to decipher what this stat is. A basic knowledge of statistics will be needed to understand how the formula was conceived, but if you are just interested in the end product, all you need is knowledge of the game. Adjusted Impact Rating is a top-down metric that measures the impact a player has on the court per 100 possessions, adjusted for strength of teammates and opponents. This statistic uses no individual player stats, but instead looks at team level statistics when the player is on the court compared to league average. A linear regression will determine the weights of the coefficients in the model in order to best fit which statistics matter the most in winning. The X-Variables will be (OnOREB%-LGOREB%), and so on for DREB%, TOV%, Opponents TOV%, TS%, and Opponents TS%. The Y-Variable will be the Net OnCourt Rating for each player (OnORTG-onDRTG). My numbers are from nbawowy.com, so credit to Evan Zamir for his site. The first season I will be examining is the most recent fully completed season, the 2013-2014 season. I haven't yet determined whether or not the model is suitable to transpose across seasons, but I hope to know soon.
The Formula for the metric is as follows (some numbers are rounded here for ease of consumption, but not in the actual calculation):
Expected Impact Rating= 2.06 + .58(OnORB%-LGORB%) + .5(onDREB%-LGDREB%) +1.33(LGTOV%-OnTOV%) + 1.2(OnOTOV%-LGTOV%) + 1.77(OnTS%-LGTS%) + 1.72(LGTS%-OnOTS%)
Adjusted Impact Rating= Expected Impact Rating - Teammate Strength Adjustment+((Minutes per game-24)*.2)
Teammate Strength Adjustment is calculated using the weighted average of Teammate Expected Impact Rating. It is weighted by ratio of time spent on court with the player whose rating is being calculated. For players who do not qualify due to insufficient minutes played, the Net OnCourt Rating (OnORTG-OnDRTG) will be used. The component of the formula designed to adjust for strength of opponent (playing against starters or bench players) is the second set of parentheses. Going by the logic of "The more minutes you play, the more likely it is you are playing against starters", I decided to set the threshold of no effect at playing half of the game. This will lower the ratings of players who played sparingly against bench units, while raising the ratings of the players playing heavy minutes, as they undoubtedly had a sizable portion of their minutes against starters. The Adjusted R-Square for the linear regression was .950441. Players must play 10% (rounding all the way from .095) of the minutes for the team to qualify.
Something to keep in mind while looking at the ratings is that players putting up scores substantially better or worse than the scores of the rest of their teammates likely means the team played as such while he was on the court, which is a good, or bad thing depending on if it is better or worse respectively. Also, these ratings are dependent on role. Patrick Beverley posts a great score, but that isn't to suggest that he could replace some other point guards with lower scores. His situation in Houston (playing next to a playmaking SG in Harden) is one that is rare, and an ideal fit for his talents. The next best fit for his talents would likely be Miami, which brings me to my next point, Collinearity. Mario Chalmers is rated higher than some may expect, but he only played 150 minutes without one of Bosh or Lebron on the floor. Playing most or all of their minutes with a superstar on the floor will likely have a positive correlation on a player's rating. Looking at players who were traded mid-season, there is a clear difference for most of the players, and it mirrored how the teams did while they were there, as should be expected. If you have any questions about how a player got his rating just leave a comment and I'll address it ASAP. I'll leave you with this:
“Someone created the box score, and he should be shot.”
― Daryl Morey
- James Patrick Oxford
The Formula for the metric is as follows (some numbers are rounded here for ease of consumption, but not in the actual calculation):
Expected Impact Rating= 2.06 + .58(OnORB%-LGORB%) + .5(onDREB%-LGDREB%) +1.33(LGTOV%-OnTOV%) + 1.2(OnOTOV%-LGTOV%) + 1.77(OnTS%-LGTS%) + 1.72(LGTS%-OnOTS%)
Adjusted Impact Rating= Expected Impact Rating - Teammate Strength Adjustment+((Minutes per game-24)*.2)
Teammate Strength Adjustment is calculated using the weighted average of Teammate Expected Impact Rating. It is weighted by ratio of time spent on court with the player whose rating is being calculated. For players who do not qualify due to insufficient minutes played, the Net OnCourt Rating (OnORTG-OnDRTG) will be used. The component of the formula designed to adjust for strength of opponent (playing against starters or bench players) is the second set of parentheses. Going by the logic of "The more minutes you play, the more likely it is you are playing against starters", I decided to set the threshold of no effect at playing half of the game. This will lower the ratings of players who played sparingly against bench units, while raising the ratings of the players playing heavy minutes, as they undoubtedly had a sizable portion of their minutes against starters. The Adjusted R-Square for the linear regression was .950441. Players must play 10% (rounding all the way from .095) of the minutes for the team to qualify.
Something to keep in mind while looking at the ratings is that players putting up scores substantially better or worse than the scores of the rest of their teammates likely means the team played as such while he was on the court, which is a good, or bad thing depending on if it is better or worse respectively. Also, these ratings are dependent on role. Patrick Beverley posts a great score, but that isn't to suggest that he could replace some other point guards with lower scores. His situation in Houston (playing next to a playmaking SG in Harden) is one that is rare, and an ideal fit for his talents. The next best fit for his talents would likely be Miami, which brings me to my next point, Collinearity. Mario Chalmers is rated higher than some may expect, but he only played 150 minutes without one of Bosh or Lebron on the floor. Playing most or all of their minutes with a superstar on the floor will likely have a positive correlation on a player's rating. Looking at players who were traded mid-season, there is a clear difference for most of the players, and it mirrored how the teams did while they were there, as should be expected. If you have any questions about how a player got his rating just leave a comment and I'll address it ASAP. I'll leave you with this:
“Someone created the box score, and he should be shot.”
― Daryl Morey
- James Patrick Oxford