Since the recount in Florida commenced two days ago, the margin for Bush has decreased by over
1400 votes. Out of approximately six million votes cast, this seems like a very small number,
representing a shift of little more than 2 votes per 10,000. Thus one might be inclined to think that
the shift probably arose purely by chance, and that any further recounts will have statistical flukes
of a similar magnitude.
We shall show, however, that under benchmark assumptions about the error-generating process,
1400 is actually a very large number. It is so large, in fact, that it strongly suggests that the errors
in the original count were not purely random. This is not to say that these errors were necessarily
malicious or intentional; it could be, for example, that the errors in each locality were unbiased but
that more errors tended to occur in counties where Gore was ahead. But the data do suggest that
the change in the vote count was not the result of purely random errors in the ballot-counting process.
Read the complete paper (PDF—10kb)