Statistical control of electronic voting

Begeleider: Richard Gill

Recently methods have been developed in the US for verifying election
outcomes when electronic voting methods have been introduced. The methods
assume that a secure back-up is available, for instance in paper form, of
all votes cast. The idea is to take a sample of votes, verify that they
have all been processed correctly. If that was the case, the election
outcome is accepted. If a small number of errors are found, the sample is
extended. If a large number of errors are found, the electronic count is
declared invalid and a re-count is performed.

If the outcome was neither accepted nor rejected, but instead the sample
was enlarged, the process is repeated, and this continues indefinitely:
ultimately either the outcome is accepted or rejected or all votes have
been recounted.

The idea is to choose sample sizes and acceptance/rejection criteria so
that a false outcome is only accepted with at most a small risk, say 1%.
The closer the outcome of the election, the more likely it is that a large
sample is taken before coming to a conclusion.

Now in the US situation there are typically just one or two important
candidates for an important position (e.g., Governor of California). Quite
a lot of votes would have to be changed to change who is the winner. In
the Netherlands, “tweede kamer”, the outcome is not a single winner, but
it is a distribution of seats in parliament. Possibly only a small number
of votes have to be changed in order to change the number of seats which
two or more parties get.

The aim of the project is (a) study the Dutch seat-distribution algorithm
and develop a method, given a certain counts (votes per party), to figure
out what is the “margin”: the smallest number of votes which it is needed
to change in order to change the resulting seat-distribution; and (b) find
good sampling plans for the Dutch situation.

The research is commissioned by “Adviescommissie Onderzoek Elektronisch
Stemmen in het Stemlokaal”, Afdeling Bestuurlijke Inrichting & Democratie,
Ministerie van Binnenlandse Zaken en Koninkrijksrelaties.

Security and Privacy (2012) vol. 10, issue 5, pp. 42 - 49, published by IEEE
A Gentle Introduction to Risk-limiting Audits
Mark Lindeman and Philip B. Stark

http://www.stat.berkeley.edu/~stark/Vote/index.htm
http://www.math.leidenuniv.nl/~gill/audit_voting.pdf