Deconstructing a Model >APM<

[hasslinghoff]

thought it was quite informative:

http://arturogalletti.wordpress.com/201 ... g-a-model/

edit: i would also recommend reading some of the comments below the article.

[floppymoose]

Statistical +- is crap and I don't know of anyone who takes it seriously. (At least the way Rosenbaum does it. I think mystic has something he calls statistical plus minus, but I don't know the details of his.) I had hoped to read an article about some better form of APM.

The other problem with the article is that it does not dive into explanatory power to wins. WP has the problem that it correlates well with wins at the team level, and this is assumed to mean that it is explaining things at the player level. But that assumption is false.

The models need to be tested in a different way. For instance, do the various lineups perform as expected over the season? WP fails that test, while a decent APM metric will do much better there. The same is true for anticipating the impact of player trades.

[EvanZ]

That article was highly flawed. I ended up contacting Aaron Barzilai about it, and he responded and set Arturo straight.

[floppymoose]

Somehow I missed the entire comment thread after the article when I read it the first time. I see my comments above are already covered ground there. Glad to see I wasn't the only one who caught his confusion.