Dividing analytic attention

[Crow]

Analytics can look at team level data, lineup and sublineup level data, player level, play level and action level data. That is 5 levels. Any other ones? At each level, you can look at Factor level data and perhaps sub-factor data. And you can do this for offense and defense. And really your offense is your offense and their defense and your defense is their offense and your defense. So that would be a total of 80 or 160 different segments of data.

How to divide analytic resources to those segments? By size/ importance of impact on outcomes, available data and ability to change behavior / influence outcomes.

And then... you could also look at the performance of your analytics on those segments. How much time to give to analytics on the analytics? I dunno. But probably should be some share of total resources.

[Mike G]

Sometimes I look at correlations between player minutes and PER, ws/48, bpm, rpm, and eWins. This doesn't necessarily evaluate the metric, as it could be the coaching, or both.

But then we can check if the results -- W/L, MOV -- are better when the coach "agrees" more with PER or WS or BPM etc. This too is ambiguous, since some teams just have better players.

And in a given playoff series, it may be that teams over-perform (relative to their competition) more often when there is correlation between mpg and (say) BPM, compared to (say) Win Shares. And with many such results, we might then conjecture that BPM is the better metric.

[Crow]

What you describe is form of analytics on the analytics. I'd agree with seeing which metric a coach's minute distribution correlates most to They don't have to actively use a metric to reveal some of their weighting priorities.