More often than not, narrowing down to smaller, selected samples to tell the "real" story that the averages are surely missing is just an exercise in wishful thinking.

I say "more often than not" because there is obviously a place for this analysis. You just want to know the smaller sample has value (e.g., clean-pocket stats). You don't want to search for subsets until you find one that confirms your priors -- because you inevitably will.

The broader averages can be full of noise too, of course! They're just *usually* more reliable. They're also less at risk of being shaped by our biases than a hand-picked subgroup of plays.