[EM] Real Ballot Data & Analysis

[James Gilmour]

> At 12:26 AM +0000 3/6/04, James Gilmour wrote:
> >We'll never know because we do not have access to the real ballots.
> >You'll find papers by Brian Wichman and others on the generation of 
> >"plausible election data" in the journal 'Voting matters' and 
> >possibly elsewhere.

Eric asked: 
> What do they argue?

If I have understood their methodology correctly, for a series of STV-PR elections, they worked out
the average probabilities for the successive ballot paper preferences they could determine from the
vote transfers reported on the result sheets.  They then applied those probabilities to generate the
patterns of successive preferences that could not be determined from the result sheets.  That way
they re-generated a complete set of ballot papers for each election, all marked with plausible
sequences of preferences.  When "counted" according to the rules actually used in the relevant
public elections, the ballot papers gave results identical to those originally published.  These
analysts wanted the ballot papers to count under different versions of STV-PR (eg ERS72, ERS76,
ERS97, Dáil Éireann, Meek) to see what differences might arise.

> >[snip] These realities would certainly affect the simulation of an 
> >STV-PR (multi-winner) election.
> Well, as I pointed out, there are 15 cases in which just a single 
> winner was being looked for, in which case, doesn't STV become IRV?

Agreed.

> 
> And this certainly leads me to think that collecting more single 
> winner data would be quite useful, which is why I plan to make those 
> changes to my site.

I agree.  But care must be taken in the subsequent use of such data, for the reasons Forest pointed
out in his post on this topic dated 6 March.  Indeed, I would suggest that analysing (counting) by a
different method the voting patterns of votes cast for counting under one method will not give a
very useful result.  A valid analysis would be obtained only when the same voters are asked to
complete several different ballot papers for the same election, having been given details of the
different counting methods and their effects (real and imagined).
James