Sunday, February 12, 2017

Executive Summary of What Value Add Ratings Mean, Followed by History of Versions

The executive summary for understanding the player ratings at is as follows. If a player must miss a game, you subtract AdjO, AdjD and AdjEM from the team's ratings at Pomeroy, and you subtract his Value Add Version 5.0 Pts/Game from his Sagarin total to estimate the team's ratings if he were not there. 

History of Versions

Since the inception of Value Add was announced on in 2011 and then to a national audience through Sports Illustrated, several updates have been issued. This blog intends to give a basic overview of the system. At the highest level, there are three factors that go into determining a player's value; 

1) How many fewer points would a team likely average if the player in question missed a game (we do not give any player a negative rating even if his offense is worse than a typical replacement player, he just gets a 0.00). This number is the individual equivalent of the AdjO team rating at 

2) How many more points would an opponent likely score if the same player was missing from a game (on defense, the player wants a negative rating but can get a positive rating if poor defense allows more points). This number is the individual equivalent of the AdjD team rating at, and if you subtract this number from the AdjO in #1 above, you get the individual equivalent of the AdjEM team rating on

3) What adjustments do you need to make to that figure to account for the comparison to what a replacement player would give you playing the same position (PG, SG, SF, PF or C), for the fewer possessions a player will get playing on a great team that often rests players in blowout wins, and to adjust this figure to a points per game total that is more easily understandable by fans and corresponds to the point total of each team as calculated by Sagarin's Basketball Rankings.

As outlined in the first paragraph, a player must miss a game, you subtract AdjO, AdjD and AdjEM from the team's ratings at Pomeroy, and you subtract his Value Add Version 5.0 Pts/Game from his Sagarin total to estimate the team's ratings if he were not there. 

To be even more precise, the actual impact on the team's margin of victory or defeat is equal to the player's Value Add Pts/Gm MINUS that same total for the seventh best player on the team, meaning the impact of losing a 7th, 8th or 9th player on a team is usually not work any adjustment.
Here is an outline of the version of Value Add Basketball since it's public introduction in 2011.

Value Add Basketball 1.0

Value Add 1.0 was released in My of 2011, constructive critics like Basketball Prospectus took a good look at some tweaks they would make to the system. Some tweaks were made over the course of three years, but we did not formally note a Version 1.1, 1.2, 1.3 etc.

Value Add Basketball 2.0

In 2014, we updated the calculations in a Value Add 2.0 update. This not only took into account calibrations for peer criticisms but also for the scoring exploding from 1.004 to 1.043 points per trip due to freedom of movement rules. Among other things, these changes temporarily removed the need for a big position adjustment for point guards - since they could easily get to the basket rather than having to throw the ball into big players for the easy buckets.

Value Add Basketball 3.0

Value Add 3.0 was introduced in 2015 to pinpoint even more accurately each players offensive and defensive value, as it reflected specifically how much a given player impacted his team's Pomeroy AdjO (adjusted offense) and AdjD (adjusted defense) of his team. This breakthrough meant that if a player had an offensive rating of 5.0 and a defensive rating of -2.0, and his team had a AdjO of 100 then if they did not have the player's 5.0 for the season then his team's AdjO would project to drop by those five points to 95 - in other words the team would start scoring 95 points per 100 trips down the court rather than the 100 points per 100 trips they were averaging. Likewise if the team had an AdjD of 100 after benefiting from the player's defense holding it down two points with the -2.0, then if he left the team the AdjD would be expected to increase to 102.

Value Add Basketball 4.0 - Introduced then Disregarded

We experimented with the likely impact of adding projected possessions to players in big programs that would likely have played more for a small program. To sum up, this turned out to be a bad experiment and we completely disregarded the v4.0 and temporarily went back to v3.0, and began working on v5.0.

Value Add Basketball 5.0

The adjustment to a per game point value rather than per 100 trips made the main rating much easier to understand. We also go into greater detail in posts below on the adjustments made for players stuck on the bench for extra time due to playing for great teams and the position adjustment.

See this link for a deeper explanation of Version 5.0.


Value Add 1.0 Overview: Anthony Davis added 7.29% to Kentucky’s scoring with his offense, and took away -5.06% from opponent’s scoring in the first season after the introduction of the system, so his total impact on the score was 12.35%, the highest Value Add in the country. Value Add 1.0 therefore calculated that if Kentucky would have lost a game 69-70 with a typical fourth or fifth man off the bench playing instead of Davis, then with Davis they win 74-66.
Version 2.0 Overview: As we tested the system we found we were under valuing players and it appeared the actual figure was close to the POINTS rather than the PERCENT OF POINTS added by a player.
Version 3.0 Overview: For various reasons, it became important to set the level of the "replacement player" lower - thus meaning we assumed a player further down the bench. Therefore a solid mid-major/high-major starter (pegged as the 100th best player at either SG, SF, PF or C) is now likely worth about 4.0 rather than 3.0 points per game and a player would need to be higher than 10.0 Value Add to be a candidate for an All-American team. The fact that the best teams in the country now calculate as close to 50 points better than the worst teams is consistent with other team calculations.
See this link for a deeper explanation of Version 5.0.

1st of 3 Components - Offense

Value Add 1.0 Offense. The Offensive component was first explained in this post, as we can measure with great precision how many points a player ad to his team’s score. Perhaps the most important breakthrough of this system was that the level of every defense faced is measured, so a player must put up much better stats against a mediocre team than a top-level team to get the same stats. 
Value Add 2.0 Offense. This was the other benefit of this system is that it led to a pretty reliable projection tool as players' main improvement came between their freshman and sophomore seasons.
Value Add 3.0 Offense. The main offensive adjustment increases the value rating, but as we continue to look for small tweaks in measuring the value of each stat in an era of increased freedom of movement and a 30-second shot clock Value Add 3.0 also added a control factor so that if the sum of a team's players is dramatically above or below the team value at, then each players offensive value add is scaled to correct the discrepancy. 
See this link for a deeper explanation of Version 5.0.

2nd of 3 Components - Defense

Value Add 1.0 Defense. The Defensive component is not quite as precise, but even some in the analytics arena admit to me that noone else matches the measurement of the impact of a college player's defense better than Value Add. This system measures a player’s ability to block shots, steal the ball, grab defensive rebounds - all of which goes into other systems. HOWEVER, the key component is it measures every trip down the court and what percent of the trips result in a basket or miss when there is no blocked shot or steal.
Value Add 2.0 Defense. Two adjustments were necessary to the defensive rating. First, I had assumed a player who had a lot of steals also tended to force more turnovers in addition to those steals, and in studying the last few years this does NOT appear to be the case. Secondly, the system was built assuming scoring would always be very close to 1.00 points per trip, and when scoring exploded to 1.04 in 2014 as suddenly the vast majority of players with a lot of minutes looked like bad defenders. We "patched" this system in 2.0 on the fly in 2014.
Value Add 3.0 Defense. Now the defensive adjustment has been adjusted so that a decent defensive player in any season will be 1.5 points a game better on defense than a replacement player no matter how many points are being score per trip or in the average game.

See this link for a deeper explanation of Version 5.0.

3rd of 3 Components - Position Adjustment

Value Add 1.0 Position. The original Point Guard/Perimeter Defensive Rating (PG/Per) redistributed a small percent of the credit from post players who do not turn the ball over as much because they do not have to dribble as often and who grab more defensive rebounds because they do not have to play defense on the perimeter.  I wrote that after extensive study, this figure was determined the most accurate way to fairly adjust ratings based on position, as explained in this post.
Value Add 2.0 Position. And the bottom line is that the peer review on this system was terrible. Others in the analytical arena hated this approach as a way to throw subjective evaluations into what was otherwise an objective system worthy of serious consideration. While I would still like to reward a player like UVa's Malcolm Brogdon extra credit for the subjective evaluations of him running the team on offense and defense, it was just a non-starter. Therefore we changed the system to simply adjust the final rating so that the 100th best player at each position would be worth 3.0 points per game and the average point guard 3.5.
Value Add 3.0 Position. We did modify further - still based on the top 100 players at each of the five positions - so that the 100th best at each position will usually be around a 4.0 Value Add and the average point guard at 4.5 Value Add. Point guards usually are more valuable - and it is because they must run the show. However, as rules were changed to allow more and more freedom of movement, the overall stat productivity moved from the front line (you needed to lob it inside once physically manhandled at the perimeter a few years ago but now you can drive by them). But this is all relative - if your opponents are now getting a lot more from their guards and you are only getting a little more, then your guards are far less valuable.

See this link for a deeper explanation of Version 5.0.

In summary:

I want to thank those NBA officials who met with me to talk about the NBA Indicators developed in conjunction with Rob Lowe, particularly those who asked me to explain why the numbers showed Jimmy Butler and Jae Crowder would be so good at the NBA-level prior to those drafts. 
I must thank Sports IllustratedESPNNBC Sports and especially Fox Sports for their praise of the system as well as all of the other outlets who have covered Value Add.  
In addition, thanks to school sites (e.g. Kentucky,ArizonaNC StateBaylor) and League sites for the Big Ten, Summit, Horizon and Patriot Leagues for their coverage, and the many Sports Athletic Departments who email and offer corrections on rosters. It is hard to track 4,000+ players a year!

No comments:

Post a Comment