GVC
Well-Known Member
I've already (implicitly) answered this question. The players we think of as the best players score a lot of points, grab a lot of rebounds, and dish out a lot of assists. As such, they have high PERs. You can devise any number of statistics that combine box score statistics - provided they put sufficient weight on scoring - that will return high scores for the players we think of as the best. That doesn't mean the statistic does a good job of measuring relative value generally, or measuring the value of players who play complementary, off-ball roles (since most box score stats are accumulated by players with possession of the ball). Further, even though Hollinger provides some of the constituent stats (reb%, TS%, etc.), PER is still a bit of a black box, and provides little information about how a player is utilised to attain their PER. I don't like statistics like PER or WS because I think roles and lineups - including player/skill complementarities - matter. 22 doesn't tell you anything. A couple examples:
1. Kyle Korver: Defenses have to account for Kyle Korver, even though he doesn't have the ball very much. Atlanta is a much better team when he's on the floor, as the various plus-minus statistics indicate. PER tells us he's a below average player. Nonsense.
2. Al Jefferson: I don't want to pile on here; Al Jefferson is a much better player than a lot of Jazzfanz posters think. With that said, he's a defensive sieve, who makes reads/decisions very slowly. As such, teams are limited in the style they play, and the teammates they put around Al. I like the argument that he provides consistent enough halfcourt offense to make a bad team better, but is too inflexible and defensively challenged to make a decent team great. PER both overvalues Al, and doesn't tell us why.
I've cited adjusted/real plus-minus on this site before, but even this statistic is problematic. Where box score statistics, and PER's constituent stats, give us some information about what players have done to score well, plus-minus is an absolute black box. I'm gathering a bunch of NBA data, and might use it to look at player/skill complementarities, which may provide a way to measure how different combinations of players (and their particular skill sets) contribute to effective lineups. This, along with measures of the relative scarcity of certain players/skills, may provide a better basis for measuring the value of players to a particular lineup or team (which is what really matters) and where market inefficiencies exist.
This is a last ditch effort to save my PhD, which has been an unmitigated disaster to date. I'm not nearly as smart as my mother always told me, unfortunately, and I couldn't be any less motivated/self-disciplined. Fortunately, the topic falls (comfortably) within the realm of Labor Economics - lineup/match specific capital, skill complementarities, market inefficiencies, etc. - so I may be able to turn a distraction into productive work. I probably should never have left the sheet metal shop. Oh well.
I have no idea what you mean. Who are these "numbers guys", and why/how are they driven nuts?
As previously stated, the eyeball test is problematic: We've been conditioned to see value in players who accumulate box score stats, and to ignore the contributions of players in complementary roles.
