Lorraine,
Others with more expertise will undoubtedly weigh in, but in brief,
you can statistically allocate over or undervotes according to other
criteria that you can attach to the votes. The thing to remember is
that while the statewide vote is very close, the totals vary more in
lower levels of aggregation (counties, precincts, etc).
This means that, if you were to remove overvotes according to the
statewide candidate totals then yes, in a close election, removal +
margin of error will make no difference.
But overvotes may not be randomly distributed across the lower levels.
Suppose that we know that there are 100 overvotes in County X. And
suppose that the candidate vote totals in County X are 60% for
candidate A and 40% for candidate B. Then you could remove the 60
votes from A and 40 from B, with appropriate margin of error (go here:
http://electionlawblog.org/ for some postings that explain how to
allocate error in these estimates).
Then suppose in another county Y, there are the same 100 overvotes,
yet Candidate A received 80% of the vote. Thus, for this county,
Candidate A loses 80 votes (+- standard error).
See how it works? If you really wanted to get elaborate, it may even
be possible to identify other characteristics of the over and under
votes and make the statistical reallocation more precise. You may
have over/under votes by precinct; or you may know more things about
the individual from their registration record (such as their party of
registration), or etc.
----
Paul Gronke
Department of Political Science
Reed College
http://earlyvote.blogspot.com
On 5/3/05, Lorraine Minnite <lcm25@columbia.edu> wrote:
I am not clear about how statistical estimation really solves the
problem of distributing (over or under) votes when the voter's choice is
not known, since statistical estimation always has a margin of error.
In the Washington state case, where the difference between the
candidates apppears to be so small, how can statistical estimation
improve a hand count of the vote? Doesn't it simply allow us to
quantify the uncertainty? I would assume that there is likewise a
"margin of error" when it comes to hand counting, we just don't know
what it is. But we do or should be able to figure out a margin of error
when it comes to statistical estimation. Can anyone with more
statistical know-how than me weigh in?
Frank Askin wrote:
RE: Statistical distribution of disputed ballots. Does anyone know whether in Bush v. Gore, the Democrats ever raised an argument in regard to the overvotes on the butterfly ballots (where voters punched for both Buchanan and Gore) that they should have been statistically distributed according to the percentage of vote each received in the appropriate voting jurisdiction?
That had always seemed to me the appropriate way to make sure that the will of the voters was effectuated without disfranchising them. I know I attempted to transmit that argument at the time to several of the Democratic lawyers involved, but never learned whether the theory was actually tested.
Prof. Frank Askin
Constitutional Litigation Clinic
Rutgers Law School/Newark
(973) 353-5687