APBRmetrics

Mike G wrote:Hmm, well you could scale player rate vs league average rate, then scale team and opponents' rate to league rate, and what you'd end up with is player rate vs opponent rate. I just skip the redundancy.

In the example I gave, Dumars' raw Pts/36 was 19.3; adjustments boost that to 21.4.
This is equivalent to his being not the 20th best scorer of 1991, but about 12th.

I've adjusted players' Sco rates only to that of their opponents; assists to Tm+Opp points; rebounds also to both, but more to opponent rates.

OK - makes sense - since you are scaling just sco, reb, & assists. If you were trying to scale all the minutiae (steals, blocks, etc) - and you just looked at opponents or team + opponents - THEN I think some of the scaled results may be a little funky (ie - overvaluing a steals guy who happens to play on a team that gets low steals AND rarely has the ball stolen relative to the league).

The way I do my ratings - it's easier for me to adjust to league first - THEN figure out each player's statistical share of their own team (thus adjusting to pace and statisical fingerprint of that team - like your '91 Nuggets example), then adjust the players's rating based off the team rating and their playing time. After that (getting a per minute and final overall rating for a player) - I can figure out how the final rating broke down in each skillset quite easily. This way I don't have to have opponent totals outside of points against (for the team rating).

Like I've said before - I think you and I have similar ideas on how to get to our ratings (somewhat similar results) - but how we realize these ratings is quite different.

I've never seen the need to scale Steals, Blocks, or Turnovers to any kind of pace, opponent FGA, etc. A fast pace generally means quicker shots and fewer passes, thus less chance to steal or block per possession. There may be correlations, but they don't seem to be much at all.
If anything, a steal (or turnover) is less likely when a shot is going up more quickly. Turnovers happen when teams are trying too hard for a high-% shot, not when they're jacking up at the first opportunity.

If we look at the 100 fastest-paced teams with a W% between .40 and .60 since 1990, we see the median TO/G is 14.1, median Stl/G are 7.2, median blk/G 4.3

Of the 100 slowest-paced teams in that W% range, median TO are 14.6, Stl are 7.5, Blk are 4.75
http://bkref.com/tiny/n8WEA
(Just sort by the column headers, and look at lines 50-51)

Actually, it looks like negative correlations; but choose different intervals, and in general you'll find little or no correlation. Check opponent averages, as well.