APBRmetrics

[Ryoga Hibiki]

I have to say I've never read JH's books, so I'm basing everything on what I read on the internet and my understanding of the formula as I
found it here

- I'm not a big fan of the idea of having a one for all stat based on boxscores. The numbers we have give us way too few informations to allow us to group them without considering the context. Whatever way we chose for weighting the different factors we have available is going to remove informations instead of adding some, preventing us from really understanding where the numbers are coming from and the real impact they have on the only thing that really matters, contribution to wins. Easy example, I can have two players average the same steals, but seeing them playing I can understand one is cheating and hurting his team, while the other is doing that in the flow of the game and leaves no hole in his defence. Once you put everything in a melting pot like PER you lose any possibility to make further reasoning.

- If I'm understanding the formula properly, it weights every stat in the boxscore using the equivalent points. It might look very scientific at first, but in this way you're ignoring all the hidden effects that are actually correlated with that stat. For instance, a block is usually an indicator for strong inside defence and can change the other team offensive plan, while PER rewards it for less than a possession. I also don't like how fouls are evaluated, in this way it's basically rewarding guys who don't play defence.

- I really think this method overrates scorers, badly, it's not a case that once you isolate the single elements of uPER 70% of point are scoting related, for the top50. The reason it happens, imo, is because the system is not consistent and doesn't apply to made shots the same way it's applied to rebounds, for instance: every made shot causes a change of possession, so there's an opportunity cost related and we should think that without that shot the team would have gotten a VOP out of that possession, shouldn't we? In this way taking dozens of shot with a 0.45 ts% is gonna help PER while hurting the team.

- PER doesn't consider positions, and that's a problem because a center is supposed to rebound more than a guard, it doesn't give the same competitive advantage. Also a weak rebounding center is going to be a liability no matter how good a of a scorer he is. Check this Kaman vs Krstic comparison, who had a better per40min boxscore?

- It's still not clear to me waht PER is actually measuring. I mean, what should I think comparing a guy with PER 22 to a guy with PER 17? Should I conclude the first is clearly better unless the second is just a defensive monster? Or the difference is not enough to reach any conclusion? This is obviously Chris Paul vs Deron Williams, I'm missing what knowing their PER really could add to the discussion. The standard error imo is so high that I really don't know how to use it properly and what's its scope.

- There's actually one thing I like in PER, is the idea of normalizing it to competion making it possible to use it for players in different eras.

What do you think? What kind of weight you actually put in PER?

[deepak_e]

There are lots of limitations to boxscore stats, which you've covered. PER is a way of summarizing, not increasing, the information that's available in those stats. That makes it pretty useful, but it should be one of many tools used by the statistical analyst.