I’ve released RotoValue 2021 NBA individual projections. RotoValue projections take into account a player’s NBA statistics, age, position, and their depth chart position (if any) to predict statistics totals. The model regresses all players to a per-position league average, so centers will tend to have more blocks, and point guards more assists, than other players. For players with no NBA stats, the league averages greatly impact their per-minute projection data, but playing time ranges are set by where they are on the team’s depth chart. So a rookie projected to start may get about 25 minutes per game of roughly league average performance (I also will use what preseason data exists, although this year’s is especially short, with just 2-4 games per team).
In addition, I have another variant, Adjusted projections (called AdjRotoValue) that uses the same per-minute projections as RotoValue, but adjusts playing time to ensure a team’s total aggregate minutes are neither too high nor too low. The adjustment will either add playing time to players (tending to give better ones more time, until their per-game minutes reach a cap based on their depth chart position), or remove playing time from players (starting with poorer performing ones) until the team’s total minutes are reasonable.
This year I dealt with an additional complication: the NBA schedule has not been fully released, which impacts displaying projections. By default, Search pages showing projections prorate the projections to the team’s remaining schedule, which normally is the full season. But this year, the NBA has only released the schedule until the All-Star break, and, making matters worse, the schedule is not evenly balanced: most teams have 37 games on the schedule, but the Cavaliers, Celtics, Clippers, Hornets, Kings, and Pacers are scheduled for 38. To address this, I’ve now added a “Raw Projection” choice to the Show dropdown. When that is chosen, the page will not adjust projections based on schedule games.
So instead of showing data prorated to the 37/38 game schedule, clicking that option will show the projections assuming this year’s full 72 game schedule. Unfortunately the same code which does prorating for schedule also handles prorating players with known injuries, so this will tend to overrate players who will miss the first few weeks of the season. But that’s still overall probably the better choice, as injured players who come back (and remain healthy) are probably better approximated by their raw full season projection than by having them miss many games in basically just half a season.
I will rerun the model once more after today’s preseason games (the last on the schedule) are over. While that shouldn’t change things much, it would have the largest impact on rookies who have no other NBA statistics.