Models & Projections
As a quick refresher, several years ago I moved to 5 models, which could each help us gain different perspectives into why prospects might succeed or fail. Here is a quick one sentence description of each model:
Original model: The base model I’ve always done with all statistics and scouting variables dating back to 2005
No impact stats model: Eliminates variables like Box Score Plus Minus (BPM), Win Shares, and Net Rating – as the impact stats are very predictive, but can hide some of the metrics driving them which are interesting for us to learn
Positional Model: This model evaluates each position separately, which allows the algorithm to “discover” metrics that are very important for say a PG, but not for PF
Humble: Adds actual draft order and “mock draft order” (average of ESPN / Ringer / Athletic) as a variable to see if there’s additional wisdom scouts have that is not emerging in the other draft metrics and also to discover what metrics scouts maybe underrating
Position Humble: Evaluates each position individually, but also incorporates draft order as an additional variable
How to interpret the scores
- 12.0 and up: Superstar potential
- 8.0-11.9: Plus starter with all star potential
- 6.0-7.9: 6th man / solid starter potential
- 4.0-5.9: Solid bench contributor
- Below 4.0: Lucky to be in the league