[E-voting] The VVAT Debate - Technical or Not ?
Dr J Pelan
J.Pelan at gatsby.ucl.ac.uk
Tue Mar 9 14:03:52 GMT 2004
On Tue, 9 Mar 2004, Michael McMahon wrote:
> If an accountant audits the books of a company, would the shareholders
> be happy if the accountant only checked a sample of the records?
That depends on the question that the shareholders are asking. In
examining the paper-trail we are asking "Have the electronic copies of the
votes been altered so as to change the outcome of the election?".
The premise here is that we have a large number of electronic copies of
votes of which we can ask any sort of question because we can get a rapid
answer from a computer. On the other hand, we have a paper-trail which we
assume is difficult to process(*) in its entirety in a timely fashion, so
we prefer just to look at a section of it.
On Mon, Mar 08, 2004, Michael McMahon wrote:
> I'm also sceptical about sampling in general as a technique for finding
> errors and instances of fraud.
Being sceptical is good but it does mean that the people attempting to
persuade you have to work harder ;-) I don't have the time to do a proper
job of this but I can throw out a few hints about how statistics can help.
I accept that some real numbers injected here might blow some of the
argument away as the thresholds of detection disappear into the 'noise'.
Let's be clear, statistical analysis won't detect everything but it
*might* just be sufficient to answer the question as outlined above. At
the very least it can help us answer the related questions; "How accurate
is our auditing system?" and "How accurate do we need to be?".
Every day, large companies like credit-card issuers and mobile telephone
companies use statistics to spot fraud by looking for anomalous activity.
Now, they may call it 'pattern recognition', 'data mining' or 'neural
networks' but it's just statistics by another name. So is there an
equivalent to 'anomalous activity' in voting with STV ? There might be.
An STV process is an opinion poll with multiple questions. The answers to
those questions will correlate strongly within each vote because, for
instance, a person will vote 1,2,3 for the same party or they might have a
tendency to vote left rather than right etc. Some individuals will buck
all trends but there will nonetheless be lots of valid trends in the data.
We can, using statistical analysis, extract these trends from the
electronic records and then turn to some of the paper ballots to see if
the trends are similarly reflected there. This is the sort of thing that
can be achieved. It's not a panacea but it can be a great help.
(*) If we used machines to process the paper-trail then why not just
go with my preferred solution of dropping e-voting altogether
and going with manual ballots with post-processing by machines ?
Machines can check for inadvertently spoilt votes pre-ballot box.
More information about the E-voting