payitforward wrote:Bigwig wrote:payitforward wrote:Players are as good as their numbers. Fans, however, always think their good players are *better than* their numbers -- fans of every team. It's natural.
It's also easy -- Jangles watches John Wall more than he watches Mike Conley. So he can explain away turnovers as someone else's fault, for example. Conley's TOs are just numbers, so they can't be explained away.
The real takeaway from Nivek's analysis is that John Wall, 24 and entering his prime, is one of the top 10 (not 5) point guards in the game. That's quite good, and since he has been an improving player pretty consistently over the last 3 years he may wind up one of the top 5 for some stretch of years.
I agree with your main point, that fans tend to be biased towards the players they watch regularly and root for. But I don't see why Nivek's method for aggregating the various statistics would be more valid for measuring which player is better (whatever that means in the context of a team sport) than some other method. In the end, it's still an opinion.
No, it's not just "an opinion." I'm not arguing for Kev's PPA methodology, because it isn't published, so I don't know how it's being used, but in principle you can certainly go a long way to determine whether particular methods measure more/less accurately which players are better than others.
The place to start is agglomerate all a particular method's results (taking PT into account) for all the players on every team. Then list from highest to lowest result. If you don't get a very strong correlation with a list of those teams from best to worst win-loss record, you can disregard that method.
To take an example, the correlation for PER is 80+% (I don't remember exactly -- sorry), whereas the correlation for WP48 is 94%, and Kevin claims his method produces about the same high correlation.
Once you have that team aggregate, the question arises whether the method being used allocates correctly among players. There are a lot of ways to check for that as well. It's too much stuff to write about it here. But the key point is that, no, this is not a matter of one person's "opinion" vs. another person's different "opinion."
The question is subjective, so of course your answer amounts to an opinion. And I would guess that a lot of people think they have good reasons for their opinions.
The procedure you outlined sounds reasonable enough, but it doesn't really capture my own notion of "better" very well. You mentioned one important potential flaw in your last paragraph. But I'll argue on your turf for a moment: do you think there could be a different aggregating stat that weighs Wall's strengths in defense and assists more heavily, and still correlates highly with win-loss record?
Also, wouldn't the definition of correlation affect the answer? For example, in a four-team league, suppose Team A has the best record, Team B next, Team C next, with Team D last. Suppose my aggregating stat predicts a D, B, C, A ordering, and yours predicts a B, A, D, C ordering. Does one stat clearly correlate better than the other?