[EM] Testing 1 2 3

[James Gilmour]

> The major problem [with STV] was that the vote set for a 3 party 3 seat  
> election with party preferences:

> Could be the  1st count vote distribution:
> Which gives parties A, B and C one seat each. 
> Or
> Which gives party A two seats and C one seat.
> 
> The model takes two extreme situations one in which the first 
> preference vote of a party is equally divided between all the 
> candidates of that party, another in which one candidate of 
> each party has all the first preference votes. It then 
> calculates the result under both sets of assumptions and if 
> they are different decides the result using a random number.

This is probably an extreme illustration of the benefits of "vote management".

I know it is common in modelling exercises to assume that all electors vote "the party ticket", but
that is not what a surprising number of real electors do in real elections.  I have looked only at
the "ballot papers" for the Meath constituency of the 2002 Dáil Éireann STV-PR election, but the
voting patterns have some major lessons for party managers  -  and voting system modellers.

There were 14 candidates for 5 seats and some 64,00 voters.  The two largest parties (FF and FG)
both put up 3 candidates; all the other parties put up only 1; there were also some non-party
independents.  FF received the most first preferences. About half of those first preferences went to
one candidate.  But of those who voted "1" for that candidate, only 44% voted "1", "2", "3" for FF's
three candidates.  35% of those voters were "lost" at the second preference and a further 21% were
lost at the third preference.  Of course, many of those voters "came back" to FF at a subsequent
preference, but it does show that the "party ticket" is a gross over-simplification as a model for
real voter behaviour.

James