APBRmetrics

Using minutes played so far, RPM has the Warriors as by far the best team in the playoffs. With Love's injury, the Warriors should be overwhelming favorites to win it all. They are balanced on both sides of the ball. 3rd best offense and best defense in the playoffs. RPM has the Clippers and Spurs as dead even.

Team Off Def Total
GSW 9.3 8.8 18.1
CLE 10.5 3.0 13.5
SAS 4.8 6.9 11.7
LAC 9.8 1.6 11.5
ATL 5.3 3.1 8.5
CHI 4.3 3.6 7.9
HOU 3.8 3.2 7.0
MEM 1.7 5.3 7.0
WAS 1.9 3.9 5.8
POR 4.2 1.4 5.6
DAL 5.4 -0.1 5.3
NOP 4.7 0.0 4.7
MIL -2.8 6.6 3.8
TOR 5.5 -2.8 2.7
BOS 2.4 0.3 2.7
BRK 1.2 -2.7 -1.5

Bucks are the most defensive biased team by raw regular season net efficiency and RPM by playoff minutes used here. More Middleton and Henson minutes pushed this even farther than late in season and much more than earlier in season. Griz a very, very distant second and rare defensive biased team by RPM (only 4 of the 16), Spurs 3rd.

Raptors the most offensive biased by RPM, which apparently they see no need for. Gave up 2nd most raw points per game in playoffs at 110.0. I guess you could see it from the raw regular season efficiency numbers too if you compared both to league average (rather than just taking the net). Wizards were not favored by net efficiency or SRS from regular season but by minutes played RPM they were on paper and in fact. Clips a very close second on RPM offensive bias, Cavs 3rd.

Average RPM bias for the group is about plus 1 on offense. Of the 4 first round winners in the books, 3 were more defensive biased than that roughly estimated plus 1 average. 3 more defensively biased are up and 1 tied. Has any major media reported on this trend favoring the greater than average defensively biased (by this method or other)? I haven't heard or seen it yet. Would be good to look at past years as well of course.

For whatever reason, Warriors just don't match up well vs. SAS. OTOH, the Spurs having to play tough first round and maybe second against Houston could help.

Wonder what could be found if you or someone else able used machine learning process to study fit of overall RAPM findings by team to opponent team level data subsets or "team-type" subsets (by 4 factors, coaching types or other criteria).

Basically, the big question in analytics to me is how to get past the "linear" regime that we are operating in? We all know that there are complicated interactions between teammates and between players on opposing teams that RAPM/RPM alone is not going to solve. Is anyone working on this?

We all know that there are complicated interactions between teammates and between players on opposing teams that RAPM/RPM alone is not going to solve.

what/which complicated interactions might you be alluding to here?...

for example simulation on offense accounts for things like how individual player touches are affected (increased/decreased) by the touches of specific teammates, and on defense things like how individual offensive player FG%s are affected by the defensive FG% ratings of various defenders, even on defensive switches...

Evan, has logistic regression and its variants (multinomial bayesian?) been tried, tried adequately? Are you calling for more of that or beyond?

Hou over LAC the only RPM upset in first 14 series.