Metric correlation with minutes

[Crow]

I imagine this topic has come up before but I am not immediately remembering the findings & discussion. Using Knarsu's file for this season I checked the correlation of minutes to BPM, RPM and more for the top 350ish on minutes. Overall BPM and RPM were real close though the correlation was a moderate .54. DRE was only .45. WS only .37. PER pulled a .46. Knarsu's SHAC just .37.Looking at offensive & defensive splits if BPM and RPM, the offensive only splits were mildly higher whereas the defensive splits were ridiculously weak at 0.08 and 0.02. Coaches are doing modestly or modestly well at putting better players and better offensive players on the court but they are showing almost no touch at fielding better defensive players. Not surprising, but quantified and put out there for consideration of coaching / managerial practice, on average, in this supposed analytic influenced age.

I've suggested doing this by team to get to implicit style of player valuation and maybe I'll get around to it later. Unless someone else wants to do it before then.

[Crow]

A simple straight blend of overall BPM & RPM did a little better than either alone.

My newish custom blend of offensive & defensive splits of these metrics did very poorly... because it gives a heavy weight to DRPM, which coaches don't, hence the poor correlation, regardless of how well it measures true impact.

[Crow]

Ok I did one, the Thunder, at least for the top 11 on minutes and still on team at end of season. RPM has a very high .88 correlation for this team. BPM at .74. My custom split blend did very well at .83. Of the 4 separate splits DBPM was the clear highest and the other 3 were close. SHAC was terrible and DRE the worst.

So Presti & Donovan were perhaps consciously or unconsciously inline with RPM. Decidedly not focused on shooting over-achievers or restrictive of underachievers. Perhaps driven heavily by individual defensive box score stats.

[Crow]

I had a glitch, so here is the revised analysis:

Of the 18 teams I checked, about equal numbers had a higher correlation to BPM and to RPM. Team to team there was quite a range in each of the correlations.

Teams were split pretty evenly on degree of correlation for SHAC, relatively high, medium low and very low.

Mavs, Raptors were very low on BPM. Suns, Spurs and Clips also lower than most others.

Raptors, Suns, Spurs and Clips among the lowest on RPM. Most on both lists.

The highest on BPM & RPM blend included Thunder, Utah, Houston and Indiana. Cavs, Celtics, Charlotte and Orlando were also high.

Just focusing on these 3 metrics, there was a lot of diversity of implicit style of player evaluation or team & results.

The eastern teams checked leaned a bit more frequently higher on BPM than RPM. The west mostly were more frequently higher on RPM.

The Warriors weren't particularly high on either separately or combined.

[Mike G]

Correlations change quite a bit depending on whether you check top 12, top 10, etc.
You might just check players with >500 minutes or some other cutoff. Or use different cutoffs and average them.

[Crow]

Nets very low on RPM correlation with minutes, negative on BPM, barely positive on the BPM & RPM blend. Very power on SHAC. Not correlating with any of these analytics. But correlating with a overhyped bad pwrforming youngster and a Russian national with a terrible RPM and contract.

[Mike G]

In general, better teams will have better correlations between player minutes and proficiency.
Better stats would correlate more strongly with better teams and their minutes distribution.

You could weigh the stat-minute correlations by the teams' Win%, to rate the stats.
To rate the teams' use of stats, check their correlations with MPG rather than total minutes. Some guys are used less because they were injured. Others are overused to fill up the minutes, to increase their market value, or just to see what will happen / not quite "tanking."

[DSMok1]

FYI, the reason defense correlates very little with minutes is because the spread on defensive skill is lower than on offense. In other words--you can find capable defense in tons of players. It's much, much harder to find capable offense. A replacement-level player typically has close to league average defensive ability, but far below average offensive ability.

[Crow]

The original correlation includes all teams. There was a minutes cutoff. I'll have to go back and check if it was 100 minutes or whatever.

Within teams there aren't many players so the correlations could be bouncy but checking the patterns may gave some value.

Defensive performance may be flatter but flat around decent / acceptable or is it typical trash? Being poorly understood and not as highly valued or selected, teams may be not have discovered defensive talent to a degree anywhere near offensive talent. I can't think of a convincing at face value reason why defensive talent is actually flatter. Easy to see flatter in execution.

[jgoldstein34]

I think defensive metrics are flatter because far more of defense is team based than offense so it's not likely to appear in an individuals ratings.

[nbacouchside]

BPM and RPM both include MPG components. DRE does not. That explains the difference in correlation.

[Crow]

At least part of it.