The graph below was prepared by frequent F&V propagator Rici Lake. It shows the dispersion of votes among the three main presidential candidates in Mexico’s 2006 election, at the level of the polling place.
You may view a larger version in a new window by clicking the image. Better yet, go to Rici’s website and download the PDF, which also has graphs of the House and Senate elections, and for each state. Fascinating stuff, and thanks to Rici for the permission to post the graph and link here at F&V.
From Rici’s correspondence with me, here is a quick quide to interpretation:
I only used the votes for the three main parties, so all points are normalized to add up to 1; they can then be placed on an equilateral triangle where each component is the distance from the point to one side of the triangle. I plotted every acta (taking the data from the official IFE results, which still have a few errors in them, but not enough to alter the results significantly), weighting the acta by the size of the corresponding casilla. Each plotted point represents the sum of the weights of the corresponding actas, normalised so that the 99 percentile point is solid black.
As you can see, the main plot essentially has four modal points (or clouds), one at about <55, 27, 27> (PAN, APM, PBT); one at about <30,17, 50>; one at around <20, 7, 73> and a faint one (at the bottom) at around <5, 37, 58>. Arguably, the first and second clouds show three-party contention, although it is clearer in the first cloud. The third cloud is PRD vs. PAN with PRD in the strong majority (possibly showing PRI->PAN strategic voting?) and the fourth cloud is mostly Tabasco.
This is clearer if you look at the dispersion maps per state. For example, you can see very clear two-party races (everything clusters to one side of the triangle) in Chihuahua, DF, Mexico state, Nuevo LeÃ³n and Tabasco.
For the images mentioned in the last paragraph, go to the PDF, linked above.
UPDATE: There was an interesting follow-up to the above in the comments at Andrew Gelman’s blog.