Ed. Note: The following is a (rather technical) guest post about Value Add from its creator John Pudner. I’ve always found his rating system interesting and wanted to share these improvements with others. You can see the full ratings here.
Big Apple Buckets seems the best venue for explaining the improvements in Value Add 3.0, particularly since constructive criticism from Big Apple Buckets and its readers from the Ivy League in particular helped improve the system dramatically.
After watching Jameel Warney hit seven of nine shots to help Stony Brook to an 11th straight win on Friday, here is the improved system that makes the case Warney is an All-American.
Improvement 1. The first improvement to the results at www.valueaddbasketball.com is that we no longer distinguish between “low-”, “middle-” and “high” majors. The top 10% of all players (400 total) includes 28 Big East players, but also 26 players from other schools covered by Big Apple Buckets. Here are the non-Big East players led again by Warney, who is valuable enough to be a third team All-American at any level due to impacting Stony Brook’s average result by 8.5 points per game:
|12||Warney, Jameel||Stony Brook||4 Sr||9.77||-3.6||8.5||10%|
|32||Walker, David||Northeastern||4 Sr||10.16||-0.77||7.8||38%|
|42||Hart, Khallid||Marist||3 Jr||12||0.52||7.48|
|57||Robinson, Justin||Monmouth||3 Jr||9.49||-1.46||7.11||9%|
|96||Wyche, Trevis||St. Peter’s||3 Jr||9.32||-0.51||6.31|
|130||Henson, Corey||Wagner||2 So||9.22||-0.06||5.92|
|131||Caruso, Henry||Princeton||3 Jr||6.99||-1.62||5.92|
|173||Rhoomes, Ryan||Fordham||4 Sr||6.6||-1.44||5.53|
|176||Green, Juan’ya||Hofstra||4 Sr||7.77||0.08||5.49||4%|
2nd Improvement. Adjusting our overvaluation of steals, because we assumed steals led to more “non-steal” turnovers. The calibration improvements are outlined in a longer piece here.
3rd Improvement. The player’s offensive and defensive ratings are now a much more accurate measure of the player’s impact on the team’s efficiency rather than points. If you take Stony Brook’s 106.8 offensive efficiency rating going into Albany’s game, and subtract Warney’s Offensive Rating of 9.77, it indicates if he had not played this year Stony Brook would have had a 97.0 Offensive Rating. Stony Brook’s 95.1 Defensive efficiency Rating would increase by Warney’s -3.6 Defensive Rating (negatives are good on defense) to leave them with a 98.7 if he had not played this year.
However, the 8.5 overall Value Add is still based on the points per game, not per 100 trips like the offensive and defensive ratings. The figure is basically (offensive rating – defensive rating)*0.7 to account for 70 possessions a game, but it also includes slight adjustments for position and early in the season a adjustment based on if the player is in a large conference. We found this was necessary because big schools rest their best players a lot in blowouts – suppressing their Value Add early in the season only to have them surge past players from small conferences later in the season. We adjust early season data based on the past curve, but by the end of January phase out the adjustment to leave only the position adjustment.
4th Improvement. A player’s total points impact on his team’s margin of victory (Value Add), is not the equivalent of baseball WAR, but rather the reflection of the “domino effect” of the other top eight players on the team averaging lower efficiency due to having to play more and possess the ball more (thus drawing tougher defense and playing tired).
Yes, the WAR-like terminology explaining a player’s value “if he were replaced by a replacement player” was easier shorthand, but as readers here and elsewhere pointed out, unlike baseball that is not really what happens in basketball.
If Warney misses a game, the ninth man in the Stony Brook rotation does not get all his minutes and shots, which is why Value Add Basketball is not really the equivalent of baseball’s Wins Above Replacement (WAR).
In baseball WAR is the correct terminology because if Mike Trout is out all of his at bats go to someone else. In basketball, it is the domino effect of the ninth man getting a few minutes on the court and everyone in the rotation having to play more and do more – which we know from page 234 of Basketball on Paper lowers every player’s ability to score.”
The following table is the compilation of the average 8-player rotation of all 351 teams.
|Rotation (average numbers all 351 teams)||Poss% in game, only top 8 players used||Off Rat||Weight (Poss/ 12.5%)||Off Rat * Weight|
|Total||100.0%||86.4 above drops to at least 82 playing more||102.7|
If all 351 teams used just their top eight players, the ‘Go-to’ player averages 22.6 percent of the team’s possessions (Pos% * Min% at KenPom.com) and a 105.1 offensive rating (points per 100 possessions). As for the worse, “replacement” players – those 1200 or so are not in their 8-player rotation, their average offensive rating is 86.4, but if they had to play more minutes and take more shots due to a player being out, it would drop to 82.0 or lower.
So if the player with the 105.2 was out and and his 22.6% of possessions in the game went to one of the replacement players with an 82 or lower offensive rating, then team would score at least 4.97 fewer points per 100 trips (105.2 – 82 * 22.6% of possessions) and we would say the Offensive Value Add of the first player was 4.97.
Of course, the replacement player doesn’t really go from a “9th man” to taking 22.6% of all possessions. What roughly happens is that the No. 2 player goes from 17.9% of possessions to 22.6% of possessions, etc. starting a domino impact that makes each player less likely to be effective (see table here).
The same plays out for the No. 3 player jumping to No. 2 etc., all the way to the replacement player taking the possessions of the No. 8 player in the rotation. Stackhouse and Iverson has tremendous ability to handle a lot of possessions, but tables on other players show similar drops when they have to possess the ball more than they are used to if everyone is healthy.
This comes to about a 3.25 drop in offensive rating per player moving up one spot to take take some of the minutes and possessions created by the No. 1 player being out.
When we take 3.25 off each player’s offensive rating, the following is the actual Value Add impact of the player based on the domino effect created if he were not to play.
|Rotation||Poss% if top player OUT, Replacement player in rotation||Drop an average of 3.25 in OffRat per player when higher Poss%||Weight (Poss/ 12.5%)||Off Rat * Weight||Off Value Add (4.77 top player drops team 102.7 to 97.9)|
|3-5 Repl||5.5%||83.2||0.44||36.3||all 3-5 team Replacements = 1.1 TOTAL|
Thus the Offensive Value Add of the average top player based on this domino effect is 4.8:
- 102.7 team ORat with him in the first table
- minus 97.9 without him on the second table
- equals an impact of 4.8 (Off Value Add at www.valueaddbasketball.com)
We added the average Offensive Value Add for the impact of losing the average No. 2 player, the No. 3 player etc. Notice that not all Replacement Players have a 0.0 Value Add as would seem logical. This is because Value Add gives any player who calculates a negative Value Add as a 0.0 because you cannot say a team would get better if he were not playing since they simply do not have a better option at that point. If we ran negative Offensive Value Adds, a team might have a +0.5, +0.4, +0.2, -0.4 and -0.7 for the ninth through 13th spots on their team to total 0.0, but with all those negatives turned to 0.0 it calculates as an average of 1.1 per team for all players not in the 8-player rotation.
The defensive rating likewise measures the impact on the team’s defensive rating.
The Value Add 3.0 system makes it clear that this is not the result of replacing LSU’s Simmons offensive rating of 123.3 by Elbert Robinson’s 82.8 Offensive Rating, but by the reduced offensive ratings of Tim Quarterman, Aaron Epps, Antonio Blakeney if they had to take over games without the defense having to focus to try to stop Simmons.
In basketball it is the “Domino Effect,” not “WAR,” but does appear the Value Add database gives a similar measurement as far as determining the most valuable of the 4,000 or so Division I players on the court.