By Jason Wojciechowski on July 11, 2010 at 5:00 PM
Link here. The basic idea is that Joe Girardi made a rather large mistake choosing Paul Konerko over Kevin Youkilis for the AL All-Star team.
The article illustrates the difficulty I have with the calls for there to be people proselytizing advanced analytical methods -- the simplification inherent in that exercise takes away the complexity of the measurement, and thus hides the problems with those measurements.
The lengthy discussions of UZR and the battles over where exactly to set replacement level show that none of this stuff is settled in the way that Posnanski implies it is here. There is a discussion in the comments, for instance, over the value of using single-season UZR, and one person asserts that while true talent may not emerge without three seasons of data, UZR in a season is still a measurement of what happened. This, of course, is precisely the assertion that Colin Wyers (and probably others, but Colin's work is what I know best) has been challenging for at least the last few months.
I don't expect all of Posnanski's commentators to read The Book Blog and follow these debates. I do expect someone who sees his job to be, at least in part, a popularizer of statistics to stay up with the current knowledge. If not bleeding edge, then at least cutting.
I probably wouldn't have said anything if Posnanski hadn't said this: "But I donŐt need to know HOW my iPad works to enjoy it." That's true for the iPad in most instances. It Just Works. But the Fangraphs version of WAR, because of the instability of UZR, arguably does not Just Work. And thus you do need to know how it works in order to "enjoy" it. (In this case, substitute "use to support arguments about baseball" for "enjoy".)
As long as I'm picking on Posnanski, though, I should note that he points out Baseball-Reference's different calculation of WAR from the Fangraphs version. Posnanski asks "smart baseball people" to "compromise" on this issue. If nothing else, this illustrates my point that Posnanski doesn't understand the stats enough for me to feel comfortable with him using them the way he does. WAR, as has been pointed out over and over again, is not a statistic: it is a framework. Into the framework you plug in the offensive measure you prefer, the defensive measure you prefer, the replacement level you prefer, and the positional adjustment you prefer. The various implementations of WAR around the web (I believe that StatCorner has even a third WAR, and there may be others I'm not aware of or am not thinking of at the moment; and that's even putting aside WARP, which is, I think, essentially an implementation of WAR (although it of course predates the creation/naming of WAR)) plug in the components they like. That is why the come out to different answers sometimes.