New Statistical Analyses?

New to me, anyway . . . from someone at Harvard? at:
http://elections.fas.harvard.edu/

Gotta love Harvard! :)

I'm a poli sci / history / law person and don't have a clue about complicated math (sometimes I even get lost balancing my checkbook).
So what did it say?:shrug:

wonder what the bottom line is?

They dug into the data to try to find pairs of counties/precincts that were similar according to a few factors (race, party affiliation, a couple of other things--the intro isn't too technical), but which crucially differed by voting machine type: optical scanner vs DRE. They ran a few statistical tests to make sure that the baseline factors were comparable in their effects.

Then they can compare DRE to Optical scanning results for these pairs.

The found no statistically significant difference between optical scanning and DRE machines: similar precincts/counties yielded the same # of Kerry votes, so they conclude the voting method was immaterial.

They allow that it may be possible to extrapolate these results to some other precincts/counties, but only if those units aren't too different from those investigated. They don't extrapolate. They explicitly state they have nothing to say about Palm Beach or Miami-Dade because those units are too far removed from the data.

I can't evaluate the math (at least not yet), but it seems like a reasonable approach, provided you can find comparable units.

Check uscountvotes.org, it may have been discussed already by the authors of the study it criticizes.