Bayesian Methods for Post-Election Audits

This page provides additional support and information related to the paper:
    ``A Bayesian Method For Auditing Elections''
    by Ronald L. Rivest and Emily Shen
    in Proceedings of EVT/WOTE 2012

The slides for the conference talk are here.


Related work

Philip Stark has provided some excellent tools for risk-limiting post-election audits at:
    Tools for Comparison Risk-Limiting Election Audits
In particular, his page provide strong support for randomly sampling the cast ballots in random order for a single-ballot audit. This capability may be useful in conjunction with the tools provided here, for doing Bayesian audits.


We have implemented the basic Bayes ballot-polling audit here.

This page implements an N-type audit: it uses a single prior, where all ballot types are equally likely and where all possible tallies are equally likely.

The implementation uses the Dirichlet approximation to the Polya method; this may be inaccurate if the difference between the number of cast ballots and the size of the audit sample is very small.

If a candidate has a winning probability in excess of 95%, the corresponding bar on the ``winning probabilities graph'' turns red. If this candidate is also the reported winner of the contest, then the audit may reasonably stop at this point.


We have also implemented the Bayes audit method in Python; contact the authors to obtain copies of this code.


    Ronald L. Rivest
    Emily Shen