I don't know if anything would come of a thorough analysis, but at first glance there seems to be a pattern that holds almost everywhere even down to the precinct level. Unfortunately they don't let you download all the data at once, as far as I can find. You have to go county by county and precinct by precinct. Maybe someone more computer savvy can find a way to grab all the data (which would be a good idea anyway, before they decide to "fix" it).
The algorithmic document seems to find the same pattern even in Iowa, where the votes were counted by hand, with witnesses, at the precinct level, and reported by the precincts in real time. Many locations had first-hand reports from Ron Paul supporters to verify the count, but only one of those turned out to disagree with the officially-reported totals AFAIK. If the same pattern occurred under those conditions, as with machine-counted ballots, then maybe it is demographic after all.
Want DP delivered to your inbox daily? Subscribe here: