Reply #6: A suggested analysis. [View All]

I just thought I would point out that, with the lower ticket information, it is possible to use a second method to get a corroborating figure to back up other analysis which are based on the number of third party votes. Here's how:


Take two races at a time, let's work with the president/senate race as an example. Find all the precincts where there was no crawl in either race, which obviously includes precincts which did not share polling places. However, as an example, it also includes some mixed precincts (not very many, but a few):

"132","1612","CLEVE02L","r0(jfk-d,rn-dq,mp-o,mb-o,gwb-r)","r1(gvv-r,edf-d)",
"132","1617","CLEVE02Q","r0(jfk-d,rn-dq,mp-o,mb-o,gwb-r)","r1(gvv-r,edf-d)",

Calculate the ratio of votes received by kerry over those received by fingerhut, and the ratio of votes received by bush over the votes received by voinovich. If there is only one candidate in the race, calculate the other race's numbers against that candidate. In the races for justices, figure out the party alignment based on the vote totals.

With the numbers you get from all these calculations, find an average ratio, and a margin of error.

Then do the rest of the pairs of races. You end up with a table of ratios and margins of error between each set of races. Throw out as many of the ratios with large MOEs as you can, and still cover all the precincts for the next step.

Now, use the ratios for the races that had no crawls to calculate the likely votes for the other races. In some precincts, there will be only the unopposed candidates to work with.

The final results should give an idea of which candidates were hurt and by how much, across the whole ticket, with a margin of error to boot. Getting the MOE calcs right, though, would require a statistician, which I am not.

But I'll help a credidentialed statistician do data reorganization as needed, if they would like to tackle it.