Monday, February 22, 2021

How to Adjust KenPom Ratings Based on Injured Player's Value Add

UConn is a 3-point favorite against Georgetown tonight at www.kenpom.com, due to a calculation we estimate below this table. However, www.valueaddbasketball.com calculates that UConn is 4 points better when Bouknight plays and 4 points worse when he does not play (he has played 8 times and was injured for 8 of UConn's first 16 games). Since now Georgetown player makes the table, that is the only adjustment we would make to the expected score - Bouknight is expected to play tonight and therefore we adjust UConn to a 7-point favorite instead of a three point favorite.

Obviously in a one game sample they might win by three, or by more than seven, or lose, but this gives us a player adjustment for any player who is worth at least 4.0 points per game and has missed at least three games this season. If any other player not on this list misses a game, then simply go to www.valueaddbasketball.com and subtract his Value Add from his team's expected score to get an updated expected margin of victory, but for these players you improve their team projection by the "Playing" column or reduce their team projection by the "Not Playing" column.

Top 2% PlayersTeamPlayingNot playingPer GameRankPlayedValue AddRank
James Bouknight #2Connecticut4-47.68468 of 163.84448
Tyrece Radford #23Virginia Tech2-67.615014 of 185.85136
Sharife Cooper #2Auburn3-47.355512 of 233.87443
Kevin McCullar #15Texas Tech3-46.807512 of 213.78459
Iverson Molinar #1Mississippi St.1-66.768320 of 235.63159
Michael Nuga #1Kent St.2-56.758412 of 184.5320
Top 5% PlayersTeamPlayingNot playingPer GameRankGamesPer SeasonRank
Matt Mitchell #11San Diego St.1-66.619217 of 205.51170
Ian DuBose #11Wake Forest4-26.531006 of 162.42909
Nijel Pack #24Kansas St.1-56.4810519 of 235.4179
Sasha Stefanovic #55Purdue1-56.4710720 of 235.39181
AJ Green #4Northern Iowa5-16.231183 of 200.931777
Obadiah Noel #11UMass Lowell1-56.1412415 of 194.72273
Jordan Bruner #2Alabama2-45.9214214 of 233.7479
JT Shumate #32Toledo2-45.8415217 of 244.17377
Devon Daniels #24N.C. State2-45.6017412 of 193.5529
Jason Preston #0Ohio1-45.5817612 of 164.29348
Shanquan Hemphill #4Drake1-55.5318318 of 214.61291
DeAndre Williams #12Memphis2-35.5018811 of 183.44548
KJ Walton #1Ball St.1-55.4819015 of 184.57306
Matt Bradley #20California2-45.4519317 of 243.89436
Keyontae Johnson #11Florida4-15.382034 of 171.251547
Ricky Lindo #4George Washington4-15.382043 of 131.251542
N'Faly Dante #1Oregon4-25.372066 of 181.791233
Top 10% PlayersTeamPlayingNot playingPer GameRankGamesPer SeasonRank
Chad Baker #44Duquesne1-45.2721711 of 144.05399
Anthony Polite #2Florida St.1-45.2521812 of 164.04403
Marcus Shaver #0Boise St.1-45.1123418 of 214.26357
Keion Brooks #12Kentucky2-34.9726112 of 212.76759
DJ Funderburk #0N.C. State1-44.9026915 of 193.77466
Chris Smith #5UCLA3-24.842828 of 261.861200
Jalen Johnson #1Duke1-34.8228513 of 183.44549
Christian Guess #23Samford2-34.732959 of 162.63820
Cam Mack #32Prairie View A&M1-34.683068 of 113.34577
John Stansbury #5Delaware St.3-14.643143 of 121.161607
Mikael Jantunen #20Utah1-44.6331715 of 193.56514
Daejon Davis #1Stanford3-24.6231810 of 232.011106
Justin Smith #0Arkansas1-44.6232018 of 223.85447
Scottie Lewis #23Florida1-44.6232113 of 173.55516
Jabari Walker #12Colorado1-34.5434318 of 243.49533
Johnny Juzang #3UCLA1-44.4835217 of 213.73475
Dallas Walton #13Colorado1-34.4635418 of 243.43553
Bo Hodges #1Butler3-14.453556 of 191.391459
John Meeks #12Bucknell2-24.443572 of 42.221001
Brevin Galloway #55Charleston3-14.403644 of 161.11656
Oscar Tshiebwe #34West Virginia2-24.3936510 of 212.091072
Eric Williams #50Oregon1-34.3636814 of 183.35575
Joe Quintana #2Loyola Marymount1-44.3437215 of 183.62500
Jalen Gibbs #2Mount St. Mary's3-14.323816 of 181.441417
Geo Baker #0Rutgers1-44.2140017 of 203.51523
Elijah Childs #10Bradley1-44.2040920 of 233.5528
Charlie Moore #11DePaul1-34.2041011 of 143.23608
Jemarl Baker #3Arizona2-24.2041112 of 232.211009
Other Players 4.0+TeamPlayingNot Per GameRankGamesPer SeasonRank
Aljaz Kunc #4Washington St.1-34.1841520 of 243.48536
Jordan Cintron #20Niagara1-34.1841615 of 183.48538
Arkel Lamar #14UMKC3-14.174195 of 151.391458
Isaac Bonton #10Washington St.1-34.0843619 of 243.14628
CJ Fleming #25Bellarmine1-34.0444513 of 163.37572

Long Mathematical Explanation

For any mathematicians who want to understand the math, here is a basic rundown.

Value Add Basketball calculates each player's value above a replacement player to his team during the season. As an example, James Bouknight's Value Add tells us he has been worth almost 4 points a game - 3.84 - to UConn in their first 16 games.

Ken Pomeroy's rankings tell us how much better or worse a team is than average per 100 trips down the court. UConn's +17.43 means they would be an average team by just more than 17 points per 100 possessions, but since there are close to 70 possessions in an average game you can multiply the KenPom ranking by 0.7 and +17.43 x .7 = +12.20. UConn is 12.20 points per game better than the average team in a given game.

UConn's next opponent, Georgetown, has an +8.90 rating which is multiplied by .7 tells us Georgetown is 6.23 points better than the average team.

UConn's 12.20 minus Georgetown's 6.23 indicates UConn is almost exactly 6 points better than Georgetown, but since Georgetown will be the home team and Pomeroy calculates Georgetown's home court advantage is 3.1, UConn is a 3 point favorite in the game.

His actual formula's for determining the predicted margin of victory in each game likely have some tweaks, and if two teams are playing at a fast pace or slow pace then the margin would increase or shrink due to more or less than 70 possessions, but that is the basic idea when you reverse engineer.

The only variable this does not capture is if a key player is missing from one of the teams, or if a team's rating was held down because a player was missing for some games, but they are now actually better than their Pomeroy rating because he is back.

Thus the example of Bouknight, who conveniently played in exactly as many games as he missed during UConn's first 16 games. 

We can figure out how many points Bouknight was actually worth in the games he played through the following formula:

Value Add listed divided by Games Played by Player multiplied by Games Played by Team equals Actual Per Game Value Add when able to play.

Bouknight's rating at www.valueaddbasketball.com is 3.84, so we divide that by the eight games he played to date to get 0.48, then multiply that figure by the 16 games UConn has played to get 7.68.

This tells us that Bouknight has been worth and additional 7.68 points to UConn in the eight games he played (so turns a 4-point loss into a 4-point win). This means he is actually the 46th best player out of 4,183, so just misses the top one percent of all players. The reason the Value listed in the database is 3.84 is that in the eight UConn games when he was out injured, he was worth 0.00 points, so over the course of the season he has improved UConn 3.84 points per game UConn plays. He calculates as the 448th best player in the country, but if he had missed not games he was on pace to be the 46th best player in the country.

The next best player held back by injury is Tyrece Radford of Virginia Tech, whose 5.85 Value Add Rating would actually be 7.61 for the games in which he played.

We will go a little deeper into the math below this table, but 

Virginia Tech is projected to beat Georgia Tech by 2, so if Radford missed the game we subtract 6 points (5.85) and Virginia Tech is a 4 point underdog.

However, you also adjust up for a team that is at full strength, but the math is a little trickier. Below are all they key players who have missed at least three games, with their actual per game Value Add listed first, and then the listed Value Add next. You need to subtract the second figure from the first to see how much better than the kenpom rating the team is with the player now on the court.

For Bouknight, 7.68 minus 3.84 = 3.84, so UConn is actually four points better than their kenpom rating with him on the court and instead of a 3 point favorite against Georgetown they become a 7 point favorite at full strength. 

For Radford, 7.61 minus 5.85 = 1.76, so Virginia Tech is actually 2 points better than their kenpom rating with Radford back. Instead of a 2-point favorite Virginia Tech is a 4-point favorite against Georgia Tech.

The following are the only players whose injuries have made a substantial difference to factor in their adjusted kenpom rating. In looking at a match-up you need to see if the opponent likewise has a key player who has missed games. Neither of their opponents, Georgetown or Georgia Tech, has a player on the list below and therefore we can make the adjustment based on Bouknight and Radford being on the court.


No comments:

Post a Comment