Wednesday, July 9, 2008

Nearest neighbors

A puzzling bit of nearest-neighbor analysis that determines how "similar" states are on an aggregate of politically relevant metrics. I don't really know what's screwing up the results, but my guess is that the sheer number of variables (19) might have something to do with it. Obviously the usefulness of these things depends on their purpose, but for electoral purposes e.g. NC and SC (the most similar states as per the survey) are known to be very different. For one thing SC is 28% black vs. 20% for NC; for another, NC is 22% upscale (per capita income above avg) vs. SC 13%. Seems like a more useful treatment would have more criteria like "% of people in socioeconomic bloc X" and fewer like "average x" and fewer still like "latitude and longitude"(!).

No comments: