Interpreting statistical maps

MNCR is expressing some concern that the use of statistical maps may be misleading because geographic area is not correlated with population size:

It hearkens back to November 3rd, 2004, when most major newspapers contained headlines declaring President Bush's win, accompanied by a US map. This map contained little blue islands awash in a sea of red, and Mr. Bush declared a "mandate" for his policies.

Those of us who were working on the Kerry Campaign knew better than to trust a map. After all, grasslands and mountains and prairies and forests don't vote. We knew that 59,000,000 votes were cast for John Kerry, and that he lost by a much slimmer margin than that geographic map portrayed.

It is absolutely correct that maps can be (and often are) misinterpreted in this fashion. I've tried to correct for this by always offering maps of population density in addition to the other statistical maps that I present. The intent is that the reader should carefully compare the statistical map to the population density maps. For example, a very close look at the CD6 maps in my last post would have shown Wetterling's support in high population-density areas (for this district) like St. Cloud, Stillwater and part of Blaine.

(The circles aren't the same size or in exactly the same place; this was a quick and dirty freehand effort.)

A close look also shows that Wetterling did not do as well in some other high population density areas such as Andover and Elk River.

I admit that the green clusters above are not as easy to see at this level of magnification, particularly after the conversion to JPG. (That was one of the reasons I included a separate table of strong Wetterling precincts. Perhaps I should have divided the district into two parts so I could increase the magnification level.) But I will point out that I was particularly interested in showing the differences between DFL congressional support in 2004 and 2000, and the maps I offered enabled the reader to make that comparison.

MNCR suggests that cartograms might be more helpful. I like cartograms (and I would be happy if someone were to provide me with the software to make them!) But I would tend to caution against the implication that they are somehow 'better' than the statistical maps that I have presented. Cartograms serve different purposes and have different strengths and weaknesses than these maps. If we were to add them to our methodological toolbox, we would be the richer for it -- but only if we don't then deprive ourselves by throwing one of our other tools away.

posted by DK @ 3:16 PM permalink