As an approach to the problem of answering the question:....
"How Many Machine-counted Ballot Batches do we need to Hand-count ("Audit"), in order to be reasonably sure we would catch any error in the machine-counting process?"
....we now have a straightforward way of figuring that out.
Citizen activists have cooperated to develop a tool for that estimate.
This can be used in three ways: As background to write Law or Procedures; as the basis for establishing simplified rules; or as the active guidance during the processing of ballots, to decide "when to quit" hand-counting.
(With thanks to Kathy Dopp, the Brennan Center, Ronald Rivest and Marian Beddill.)
There are five independent variables which need to be determined or estimated:
Download the spreadsheet (it's a small file):
- The whole set - "the NUMBER" - of ballot-batches (precincts, machine-runs, etc) for which the jurisdiction is holding the election.
- The Margin between winning and second-place Candidates, in the initial election results. (How close is the election?)
- Your desired Probability or Certainty Level, for finding one or more corrupt counts. How sure do you want to be? Like, 90% sure? ...or 95%? ...or 99% sure? (higher numbers mean more work.)
- Maximum amount of vote switching, per vote count, that would not be immediately recognized as suspicious. If a fraudster shifted 50% of the votes in one set, would it be noticed by ordinary caution? Clearly yes. But what if they only shifted 15%? Or 3%? In plain english, for this measure of amount of shift, how bold can we guess that a fraudster would be in THIS jurisdiction on THIS race? Another element is how many batches (machines, etc) he would try to fiddle with?
- So, what you need to know, is how many batches should be pulled for a hand-counted double-check? Those factors above, determine how much effort is worth doing. Since the formulas don't work backwards, you have to input a guess, and this model will tell you if "YES", it was good enough, or "NO", it was not.
HowManyToAudit.ods (OpenOffice Calc)
and the version posted by Kathy Dopp, at her website: electionarchive.org
HowManyToAudit.xls (Dopp- Excel97)
(we are each tweaking it, headed to settle on the same version.)