I’m pretty sure if you showed that list to the majority of people who work in the NBA, they wouldn’t give it a second thought either.

Also, comparing team chemistry to fairy dust and astrology just shows an astonishing ignorance of how NBA teams work.

When it turns into a disaster, they say one of two things: "That's what the numbers said," or, "Just think how bad off we'd be if we hadn't done that."

In this case, I don't think the problem is with the model. The problem is with the input data.

Valid statistical outputs require measurable, repeatable events.

Track and field has a lot of those. So track statistics are almost perfect. Baseball has fewer, but still a lot, so the statistics are pretty good.

Basketball has hardly any because the sport is one of free-flowing interaction between multiple players at a time.

A statistical method that misses an NBA team's win total by 14, and then explains it by saying some guys played better, is not a model on which I'd want to build my franchise.

Basketball talent evaluation is, and is likely to remain, more art than science.

And to me, "team chemistry" is far from fairy dust or voodoo or whatever. It's simply a shorthand expression of the well-documented human psychological trait that a player will try harder the more he cares about letting his teammates down.

At the level of the NBA, where everybody with a uniform is one of the 500 or so best basketball players on the planet, even tiny differences in motivation can be absolutely critical.

I think it’s pretty distasteful how Berri likes to take shots at Hollinger (the guy whose system every GM in the NBA uses to analyze players). Hollinger’s system isn’t perfect, but it’s not totally wrong either. And like I said, everyone uses PER among other statistical methods. But the fact that PER is there means it works (at least in the minds of those who matter and are making the decisions).

No statistical model will ever be perfect because statistical models of this nature are simply attempts to explain observations. Some things are inexplicable. Are defense, rebounding, points, steals, blocks, etc the only way to track games? Do the guys on the bench matter, even when they’re on the bench? How about practice methods or the practice squad? How about trainers? What about the crowd? People are products of their complete environment, and no statistical model will ever account for the entirety of the universe, they simply strive to explain a pool of data under hand-picked parameters. Unbiased?

If it’s 94% accurate, then it’s 6% inaccurate and thus not truly “right?” I like this model a lot, but it’s not the best one I’ve seen. It can be used in conjunction with other models and it’s a useful tool, but this is not the be-all-end-all of advanced basketball stats.