[EM] STV and weighted positional methods

[Kathy Dopp]

On Mon, Feb 2, 2009 at 8:24 AM, Raph Frank <raphfrk at gmail.com> wrote:

>
> Ok, so if there was 100 polling stations, you would pick say 5 of them and then apply to be allowed to count those votes?
>
> If the results are the same as the official count, then it is likely
> that all 100 are the same (since you picked them at random, any group trying to tamper with the results wouldn't have known which polling stations to leave alone).
>
> This covers tampering where the tampering is small but in a large number of the districts.
>
> If the fraudsters tampered a large amount but in a small number of polling station, they would have a higher chance they would get away with it.
>
> You would also need to include in your check any polling stations with suspicious results.  For example, if they don't match exit polling or if their result is an outlier when compared to all the other polling stations.

Yes. I agree with you and have been recommending the same for years
and have been helping to develop the mathematics for calculating
sample sizes, and am in the process of writing a paper on how to do
the analysis of the discrepancies found in an audit.

>
> This method could also be applied in the IRV case if they use the central office method (i.e. counts are performed in each polling station under the direction of a central office).
>
> You would have a list of the results for each round from each polling station and what candidates were eliminated.

Hmm. Have to think about that.  The mathematics is very different. And
to actually check that the overall tally is correct, you would have to
take a separate sample for each IRV/STV round from different poll
locs. It would be a very complex process and would take a lot lot more
auditing overall than in a plurality election or in an election using
any precinct-summable method.  Auditing IRV/STV in that way would take
a huge amount of effort and time and expense compared to auditing any
other voting method but I suppose is possible.  The amount of sorting
and resorting of the ballots alone would take a huge amount of time.
I'm not sure that it might be much simpler just to go ahead and do the
100% hand count.

I do not think that logically, the process you list below that checks
only one poll location, does anything to check that the overall
reported result is correct because as soon as you begin checking round
#2 the accuracy for that one poll loc depends on whether or not all
the other poll locs were accurately counted or not.  You need to
randomly sample poll locs separately for each round from the total
list of poll locs to use this method in a logically correct way.

I cannot imagine how anyone could want to make all that unnecessary
extra time, effort, and money for a system that is fundamentally
unfair like IRV/STV is when there are so many other better voting
methods.

Cheers,

Kathy


> You could then manually check a random set of the polling stations, i.e.
>
> Round 1:
>
> - Sort ballots into piles based on first choices
> - count each pile
> - make sure they match announced
>
> Round 2:
>
> - Check who was eliminated after round 1
> - Split his pile based on 2nd choices
> - count all the sub-piles
> - make sure they match the announced result
>
> and so on
>