Thursday, August 24, 2006

Keith Law is a Cool Dude

Not that this has that much to do with this post.

OK, I have now emailed Keith Law at his email address over at baseballthinkfactory.org, since he didn't answer my question about EqA vs. wOBA during his webchat (once again, insider only, sorry) this afternoon (although he did post my suggestion that, apropos of his recommendation of "Chronicle of a Death Foretold," he should read "Song of Solomon" by Toni Morrison). However, I also found this article on baseballprospectus.com.

To sum it up since it is pretty complicated they say that this stat is superior to others (not specifically wOBA) because it more accurately estimates the number of runs created by a given hitter than any other statistic over the history of baseball. Thus, it is accurate now and it is also more accurate over long periods of time. For example, it is not only more accurate than other similarly accurate statistics for data from 2005, it is also more accurate for data from seasons that occured a long time ago, say 1894. They say that they convert this to an average so that the non-sabremetrically inclined fan can feel comfortable with it since it looks like a batting average, but, to me, this just obscures what is so usefull about this statistic: the fact that it is an incredibly accurate way to describe how many runs a hitter produced by himself, independently of the performance of others on his team.

2 comments:

Theo Von Hohenheim said...

That article was good, but I am still not convinced that wOBA can't do the same thing. If run expectancy is determined in the relevant year (granted this might be quite hard, as the necessary data simply may not exist going back) it should coume out appropriately adjusted . . . Basically, I'd like to see the RMS error in wOBA.

The Fabulous Galdstoner said...

I totally agree but at least we have a possible idea of how they came up with the weighting. It looks to me like it was most likely trial and error. They probably just adjusted the weighting to see what would produce a better correlation over a variety of time periods. At least that's what makes the most sense to me.