I think it's ok...but I tend to lean on the on/off stuff more. Maybe because that's just my own comfort and understanding, but there's so much noise in the tracking data.. I'm just not sure it matches up with good man to man defense even over a larger sample size. My understanding is that for every shot taken, one player is assigned as the closest player no matter what the context. This can get fuzzy in different situations. For example, the majority of 3 pointers taken in the NBA are open/wide open. Now of course shooting a 3 with Wemby 8 feet away versus Sexton 8 feet away is different, but I still feel like a majority of the 3FG% defense is just random variance.
Around the basket, you kind of have the opposite effect where there is often more than one defender. So two guys can be in the area, but only one guy gets blame/credit for the make/miss. Depending on the defensive scheme, it could make a player look really under or overrated. I think comparing bigs to bigs around the basket is the most reasonable use. There's still some bleed over effects from scheme etc, but it's the most fair because most shots at the rim are contested and I think the quality of the defender is most reflected around the hoop. With perimeter players, I think it's more about taking away the shots than the percentages. I think good perimeter defense results in no shot more often than it does a bad shot.
Beyond that fouling (bad defense) is also a component that is missing. I feel like Kris Dunn has been a little disappointing on defense, but that's because he's playing Jerry Sloan defense and fouling like crazy. But there are things to pick at with every stat, so maybe I'm just being too harsh. Most of all, I'm just unsure of what to make of these numbers. It's just really hard to isolate what an individual player is doing versus everything else that's going on. I think it can be a component of a more holistic view, however.