LiP responded to my last entry about the "is Paul Krugman partisan or not" debate. Apparently I've hit home to some extent, as he does describe it as a "plausible case" that Krugman may simply be pointing out what others are afraid to, but points to other columnists such as Frank Rich and Michael Kinsley as being left-leaning but not nearly as partisan, willing to attack Democrats and Republicans alike. Leaving aside the actual worthiness of attacking both, one of the reasons that Krugman is considered so dangerous (and so interesting) is because he isn't a journalist, and therefore isn't overly worried about building the sort of superficial "fairness" between centrists and right-wing extremists that Eric Alterman aptly described as "working the ref".

It's also important to remember that Krugman's most powerful (and theoretically partisan) attacks are on economic issues, as economics is a subject that he knows much better than almost any of the other major newspaper columnists right now. His purely political articles are often considered somewhat weaker, although he's been getting better, and I personally believe it's because he's exploiting the liberal blogosphere's able work on such issues. (If his critics can do it, why not him?)

Getting back to LiP, though, it's more interesting looking at what wasn't discussed, which was my objections to his methods themselves (as opposed to questions of Krugman's partisanship.) It's not just about the environment in which the comments themselves are made, but about the very notion of trying to derive meaningful data from what amounts to word-counting. It's relatively simplistic: it merely counts the number of references in a column to a Democrat or a a Republican, assigns them a binary "plus or minus" for partisanship, and creates an eventual "partisanship index" derived from the numbers of "pluses and minuses".

As should be immediately obvious, this is woefully deficient. Operationalizing the concept of "partisanship" is tricky enough in the first place, but LiP's methodology systematically strips the inherently qualitative elements from columns. It misses the multiplying effects that things like sentence structure, tone, theme, and choice of words can have on the "partisanship" of a column. it punishes and rewards columnists based on merely the number of partisan references, leaving the severity of their references unremarked and untouched. This sort of methodology would be unacceptable when dealing with survey data, let alone the analysis of political texts; there's simply too much left out, and this raises serious questions as to the applicability of LiP's "index". It's like those income statistics that lump together people that make $100,000 a year and those who make $5 million when counting the number of "wealthy people" in the United States and the effects on them... it produces numbers that are simply untrustworthy.

(At the very least, there should be a weighted scale for the partisan remarks. The act of judging these things, however, is in-and-of itself such a qualitative act that it renders the whole quantification process pretty much useless.)

Politics isn't economics- it's hard to quantify, harder to quantify well, and extremely easy to quantify badly, leading to results that are either misleading, banal, or horrifically wrong. This is why the "Perestroikans" have been so bitterly criticizing the formalism in American political science, and why many of political theorists and researchers outside the United States tend to avoid quantification altogether. (Many Canadians tend to take a particularly dim view of the whole enterprise, for example.) Even your most doctrinaire APSA formalist, however, would likely look askance upon LiP. It simply leaves too much out.