After thinking about this study, I'd like to clarify my answer a bit.
First, you have to consider two different types of recount results, so the analysis can't be done statewide but must be done county by county.
If the recount produces ADDITIONAL total votes (as occurred in those counties where the machine rejected ballots), then you would EXPECT those extra ballots to break in the same ratio as before.So if one candidate had 60,000 and the other had 40,000 but 1,000 new votes were lumped in, you would now expect the leader to have 60,600 and the trailer to have 40,400 or a net gain of 200 for the leader. This is consistent with what we saw in a number of counties where ADDITIONAL votes were found.
If the recount produces no ADDITIONAL votes but merely a redistribution of the votes then you would EXPECT the net change to be positive for the TRAILING candidate.
For example, if 60,000 were for Gore and 40,000 were for Bush, and 1,000 votes were in error then you would expect 60% of the error votes to be Gore votes that are truly Bush votes and 40% of the error votes to be Bush votes that are truly Gore votes, so Bush would gain 600 of Gore's previous votes while losing 400 of his previous votes...net gain of 200 for Bush and of course a net loss of 200 for Gore. Totals would then be 59,800 Gore and 40,200 Bush.
In both of the examples above, departures that are statistically significant can be attributed to some external force, BUT IT IS IMPOSSIBLE TO DETERMINE WHAT THAT FORCE IS without further inspection.
The cancellation of gains with losses would only be expected in those counties where the vote distribution was nearly 50-50 to begin with.
TG |
