My Model Monday: Predicting NBA Awards

  1. Now that the NBA season is halfway completed, I trained a model to predict the major individual award winners – MVP, Defensive Player of the Year, and Rookie of the Year, using statistics from the first half of the season. The modeled probabilities do not reflect who we at Model 284 think should win the award; instead, they indicate the probability of a given player winning the award based on the statistics from the players who the voters have chosen in the past. 

MVP

James Harden is clearly the front-runner for MVP through the first half of the season. He is putting up ridiculous offensive numbers and has led the Rockets to a 42 – 13 record so far. If Harden wins the MVP this year, then all of Sam Presti’s OG big three – Harden, Westbrook, and Durant, will have won the award within the past five years. It would have been fun to keep that core together eh? There is a big drop off after Harden, but the model also liked three different players from the Warriors, which is further evidence of how insanely talented that team is. LeBron James also makes the top ten but is not as high on the list due to his team’s declining win-loss record. The strongest predictors of winning the MVP are team winning percentage, the percent of games that the player has played in, minutes per game, points per game, win shares per 48 minutes, blocks per game, assists per game, rebounds per game, and turnovers per game.

PlayerPts / GAstTRBFG pctWin Prob
James Harden31.29.15.00.450.52
Stephen Curry27.56.45.10.490.10
Kevin Durant25.95.66.80.520.09
Chris Paul19.28.55.70.460.07
Giannis Antetokounmpo28.04.710.40.540.04
Klay Thompson20.32.64.00.490.04
LeBron James26.38.78.00.540.03
DeMar DeRozan24.45.14.00.470.03
Kyrie Irving24.95.03.70.470.02
Andre Drummond15.03.815.40.540.01

Defensive Player of the Year

Defensive Player of the Year is the most difficult of the three major awards to model. There are fewer defensive statistics out there, and team success has been less of a factor in the voting for this award. The most important variables for predicting this award are percent of games played in, fouls per game, blocks per game, and rebounds per game. The model is high on Anthony Davis this year, who is putting up a solid 2.1 blocks and 10.6 rebounds per game. Clearly, the model is biased toward big men. This is because there aren’t many statistics out there that quantify how well a guard/wing player can move his feet on defense or close out on three-point shooters.

PlayerBLKSTLTRBWin Prob
Anthony Davis2.11.210.60.10
Andre Drummond1.51.615.40.08
Dwight Howard1.70.712.60.08
Karl-Anthony Towns1.40.812.20.05
LeBron James1.11.68.00.04
DeAndre Jordan1.00.414.90.04
Marc Gasol1.50.78.60.04
Kristaps Porzingis2.40.86.70.03
Kevin Durant2.00.86.80.03
DeMarcus Cousins1.61.612.90.03

Rookie of the Year

Our model for Rookie of the Year has been the most effective of the three. In out-of-sample predictions, it has correctly chosen the winner 77% of the time. It has also correctly predicted the winner in 13 of the past 14 years. In 2018, the model really likes former number one pick Ben Simmons, who has flashed versatility and play-making ability on offense all season long. Hot-shooting Donovan Mitchell is second with a 19.5% chance of winning the award. Although Lonzo Ball and Kyle Kuzma are both long shots for the award, the model likes Ball slightly more as he has a significant edge in assists and rebounds. Minutes per game, points per game, rebounds per game, and PER are the most significant variables of Rookie of the Year voting.

PlayerPts / GAstTRBFG pctWin Prob
Ben Simmons16.77.27.80.530.59
Donovan Mitchell19.33.53.30.450.19
Jayson Tatum13.91.45.40.490.07
Lauri Markkanen15.31.37.70.430.05
Lonzo Ball10.27.17.10.360.03
Kyle Kuzma15.71.85.70.460.02
John Collins10.41.06.90.590.02
Tyrone Wallace11.32.53.50.470.01
Bogdan Bogdanovic11.23.12.70.470.01
Dennis Smith14.64.83.90.390.01

Below are the model’s predicted winners since 1980. We used leave-one-out cross-validation to generate these predictions for each year. The players colored in green denote correct predictions.