Andrew Johnson introduces his NBA Draft Model
Note: This article was written by Andrew Johnson.
Earlier this year, I started working on a statistic-based NBA draft model. Right from the beginning, I decided to include international players and D-League prospects in the model. It was an easy decision given that last year, fourteen players out of the sixty drafted came from outside of the NCAA, with thirteen international players and one from the D-League, and there have been at least ten international players taken in each of the last three years, yet most draft stats models are blind to 20% of players taken.
I developed and tested the model against draft data over the last twelve years, finally arriving at the point where the model beats the results of actual GM’s drafting at the time, using only data that was publicly available at the time. To support the model, I also developed both an aging and a league adjustment for European players.
The models are scored against a box score metric called Alternate Win Score (AWS) that has been shown in another study to do well at predicting success even when the context around a player changes, which is particularly relevant in predicting success in the draft. The two models I am using give two rating scores to judge players, the Projected-AWS, or P-AWS, and a percentage likelihood of being a starting quality player (% Player rating). You can read more of the gory details on the development process for model here, if you are so inclined.
Running a stat-based model usually gives you a pretty different top twenty than the consensus picks. The good thing about a stat-based model is that it is not influenced by the opinions of other writers; it has no idea who Chad Ford is so the model can give an independent judgement. The key then is to use the model to recognize possibly undervalued (or overvalued) prospects, while being aware of the model’s limitations.
Here’s the top 20 in both P-AWS and %Player. Currently, I have three European players in the top twenty, Exum not included since there are no relevant statistics to score. I highlighted some of the biggest movers compared to the scouting consensus:
| Name | Team | Position | P-AWS | P-AWS Rank |
% Player | % Player Rank |
| Jordan Adams | UCLA | SG | 6.55 | 1 | 0.88 | 2 |
| Kyle Anderson | UCLA | ????? | 6.3 | 2 | 0.88 | 1 |
| Clint Capela | Chalon (France) |
PF | 6.29 | 3 | 0.86 | 3 |
| Bobby Portis | Arkansas | PF | 6.11 | 4 | 0.83 | 4 |
| Aaron Gordon | Arizona | PF/SF | 6.04 | 5 | 0.83 | 5 |
| Noah Vonleh | Indiana | PF | 5.99 | 6 | 0.81 | 7 |
| Jabari Parker | Duke | SF | 5.85 | 7 | 0.77 | 11 |
| Tyler Ennis | Syracuse | PG | 5.84 | 8 | 0.83 | 6 |
| Marcus Smart | Oklahoma State |
PG | 5.78 | 9 | 0.8 | 8 |
| Branden Dawson | Michigan State | SF | 5.69 | 10 | 0.78 | 10 |
| Mario Hezonja | Barcelona (Spain) |
SG | 5.67 | 11 | 0.79 | 9 |
| Joel Embiid | Kansas | C | 5.59 | 12 | 0.76 | 12 |
| Jarnell Stokes | Tennessee | PF | 5.57 | 13 | 0.75 | 13 |
| Montrezl Harrell | Louisville | PF | 5.5 | 14 | 0.73 | 16 |
| Damien Inglis | Roanne (France) |
SF | 5.38 | 15 | 0.75 | 14 |
| Jerian Grant | Notre Dame | SG | 5.38 | 16 | 0.74 | 15 |
| Gary Harris | Michigan State | SG | 5.29 | 17 | 0.7 | 20 |
| Willie Cauley | Kentucky | C | 5.23 | 18 | 0.71 | 17 |
| Rondae Hollis-Jefferson |
Arizona | SF | 5.22 | 19 | 0.7 | 18 |
| Julius Randle | Kentucky | PF | 5.17 | 20 | 0.67 | 23 |
There are a couple of points to take from this. Jordan Adams is, I think, tremendously undervalued. He puts up great numbers on offense, has a high steal rate and is only a half-year older than Jabari Parker and actually younger than Joel Embiid. Embiid, by the way, would be top ten in the model if he had been able to play more minutes, which is partially related to his back injury.
Swiss big man Clint Capela is an athletic power forward playing against grown men in France at nineteen and putting up very efficient numbers, which is why he is the highest European player in both models.
Add The Sports Daily to your Google News Feed!
