RAPM 'error of estimation'

[J.E.]

Re-hashing an old topic. From now on I'll show some measure of confidence in each player's RAPM estimate. Those interested in the methodology can read up on it in the link above. The short version is that I'm simulating NBA seasons (thousands) with players that have a fictitious skill level (different skill level each simulation run), then compute RAPM on the simulated season and finally compute RMSE for 'ficititious skill vs. RAPM estimate'. It's not really a 'standard error' but it should give people a ball-parkish number of how much confidence I have in the estimates.

The 'error estimates', I'd say, behave as one would expect them to. Players with low possessions have the highest error, so more possessions usually means less error. Switching teams or playing for a team with higher roster turnover also generally reduces the error

You can find the 'error of estimation' on the standard rating page

[colts18]

Is this error listed the margin of error (2 stdev)?

[DSMok1]

So you used the actual matchup file? I can't quite tell if this accounts for any error associated with multicollinearity.

[J.E.]

So you used the actual matchup file? I can't quite tell if this accounts for any error associated with multicollinearity.

I used the actual matchupfile to fill the X matrix of the regression, but I didn't use actual points scored to fill Y.

As for accounting for error associated with multicolinearity, as I said, players who changed teams or played for teams with high turnover rate generally have lower error. The 5 players with lowest error are Iguodala (DEN+GSW), Belinelli (NOH+CHI+SAS), Sessions (CLE+LAL+CHA), Harden (OKC+Houston), Ellis (GSW+MIL+DAL)

Is this error listed the margin of error (2 stdev)?

No, just 1 stdev