DIAMOND: Identifying Comparable
Players In The WNBA And The NBA
July 2019
Introduction
Basketball fans are much more familiar with NBA players than their WNBA counterparts, and, inspired by FiveThirtyEight’s CARMELO NBA Player Projections*, I sought out to create a model that identifies WNBA players who are statistically similar to a given NBA player, and vice versa.
*CARMELO stands for Career-Arc Regression Model Estimator with Local Optimization.
FiveThirtyEight
CARMELO Projections
FiveThirtyEight’s model displays percentile ranks (left column), career projections (top right column), and a list of comparable players (bottom right column). Along with each player comparison comes a similarity score, which quantifies how statistically similar two players are, with a higher score indicating a higher degree of similarity.
This example shows Portland Trail Blazer Carmelo Anthony, who the model’s former acronym was named after.
FiveThirtyEight doesn’t publish its computational methods, so I tried my best to recreate their model that outputs a similarity score. My model, which also compares two players, begins with a similarity score of zero, and for each statistical category, each player’s production is subtracted from the other’s. For example, two players who score 15 and 20 points per game respectively will yield an index of 5; this means that, when all differences are summed, the two players with the lowest similarity score are the most similar.
Certain statistics have multipliers to account for their varying importances; for example, two players who score the same amount of points but play different positions or shoot at drastically different efficiencies shouldn’t be compared, so points are weighted less. Another modification was to normalize all the data, which accounts for the varying ranges of every statistical category; points may be in the 10-30 range, whereas steals are in the 0-3 range.
DIAMOND DeShields
To pay homage to FiveThirtyEight’s CARMELO projections, I tested my original model – prior to adjusting weights – with Carmelo Anthony to identify his WNBA counterpart: Diamond DeShields. This comparison yielded a relatively high DIAMOND rating (similarity score) of 9.06. Thus, I named my model DIAMOND, which stands for Design for Identifying Analogous Members Of the NBA Distribution.
As shown above in the two percentile comparison graphs, my model inputs 10 total features: points, field goal percentage, three-point percentage, free throw percentage, effective field goal percentage, rebounds, assists, steals, turnovers, and blocks.
Case Study
Seattle Storm forward Natasha Howard’s closest NBA comparison is Los Angeles Lakers forward Anthony Davis, yielding a DIAMOND rating of 6.16. Evident by the percentile comparison graphs to the left, these two players’ statistics follow nearly identical trends, both posting high numbers in rebounds, turnovers, steals, points, and blocks.