There is very little directly measurable about individual perimeter defense, so it ends up being mostly inferential. Which is fine so far as it goes. But those metrics (RAPM etc) aren’t measuring defensive ability, they are estimating effectiveness in role. 1/x https://twitter.com/drewww6/status/1380666859956142080">https://twitter.com/drewww6/s...

For a player like Simmons for whom versatility is an important component of his defensive value, he’s being shifted between “roles” much more than a player with more narrowly defined strengths. 2/x

He’s going to be better/more effective in some roles than others, but because he’s *good enough* in less ideal situations, he’s going to be exposed to those situations more than a less versatile player. 3/x

To simplify it down, suppose he’s a +4 defender in a weakside help role but only +1 as an onball stopper. Meanwhile a player he being compared to is +3 weakside but -3 as a stopper. 4/x

Simmons might only play 1/3 of his possessions in his “best” role because his team needs him as a stopper. Meanwhile the comparison player would never get used as a stopper because he sucks at it, so plays 100% as weakside 5/x

In this (ludicrously simplified and hypothetical) example, Simmons “averages” +2 defense while the other guy averages +3, meaning the other guy might *appear* better by not being able to competently fill other roles. But the varied deployment means that edge is illusory /end