Reply #61: After giving it some more thought

[Time for change]

I realized that non-parametric tests of statistical significant would be more appropriate for testing the association between county size and discrepancy between the Gore and Kerry vote. This is because the county size data is not normally distributed -- in fact it is nowhere close to being normally distributed.

The non-parametric tests that I used to test the statistical significance between Triad and non-Triad counties indicated borderline statistical significance (p=.05 using the Wilcoxan rank sum test, and p=.06 using Mantel Hantzle Chi Square). This still could have been confounded by county size, but I don't know of a way to test for this using non-parametric statistics.

One interesting thing about this is that the whole correlation between county size and Gore/Kerry discrepancy was attributed to three huge counties (Franklin, Hamilton, and Coyahoga). These three counties accounted for 29% of the Ohio vote, none of them used the Triad machines, and all of them resulted in large gains for Kerry compared to Gore (88,076 votes).

Perhaps the reason for the large Kerry gains over the Gore vote in the very large counties had to do with the huge increases in voter registration in Ohio. Did this take place disproportionately in some very large counties? If so, that could explain above paragraph.