[E-voting] The VVAT Debate - Technical or Not ?

Michael McMahon michaelmcmahon at vodafone.ie
Mon Mar 8 19:36:50 GMT 2004

Opinion-polls are one thing, but you can't take a sample of  individual
ballots from an election, because there is no way to correlate the paper
with the electronic record.

All you could do would be take a sample of  elections at the constituency
level, but the number of constituencies is too small (42 ?) to be 

I'm also sceptical about sampling in general as a technique for finding 
and instances of fraud.


Dr J Pelan wrote:

>On Mon, 8 Mar 2004, Michael McMahon wrote:
>>But, what is an acceptable chance? A 50% chance means checking 50% of
>>constituencies, or ballots whatever. A 90% chance means checking 90%
>I'll try to give a quick answer here and perhaps a longer answer later if
>time permits and demand requires it. I am not a statistician and have an
>embarrassingly woeful recall for that subject.
>If we take the example of an opinion poll, it seeks to determine the
>opinion of a very large number of people by asking questions of
>('sampling') a much smaller, more manageable number of people. This
>smaller group must be selected randomly from the larger set, i.e. without
>any bias, if they are to be truly representative.
>There are two important factors to describe the accuracy of the poll - the
>'margin of error' and the 'confidence level'. These are mathematical
>functions of the sample size and the overall distribution (of opinions).
>To put some figures to this and without going into any detail, a typical
>opinion poll might sample 1000 people. The resulting margin of error is
>+/- 3% (i.e.  1/sqrt(1000)) and by assuming a particular distribution type
>we get a confidence level of 95%. This means that 95% of the time the poll
>result will be within +/- 3% of the true result. 
>The beauty of statistics is that this applies to any sufficiently large
>population be it 1, 10 or 100 million people. So survey groups in the
>Ireland, the UK and the USA can all limit their questions to one thousand
>people and still have that +/- 3% result accuracy. Note that it is more
>meaningful to talk about absolute sample sizes rather than sampling
>'percentages' of whole populations, so checks wouldn't be of the nature of
>'10%' but more by absolute numbers like 1000, 2000 and 10000.
>By obtaining the opinions of a relatively small number of people we can
>achieve a result that applies - on the balance of probability - to the
>whole, sampled population and which may be accurate enough for our needs.  
>A similar principle applies in checking the validity of electronic voting
>records by comparision with a subset of the verified paper trail.
>John P.

More information about the E-voting mailing list