[EM] STV - the transferrable part is OK (fair), the sequential round elimination is not

[Kristofer Munsterhjelm]

James Gilmour wrote:
> Kathy Dopp  > Sent: Friday, October 30, 2009 4:45 PM
>> A fair proportional multiseat STV representation system could be made
>> by eliminating STV's elimination rounds but using the rank choices to
>> transfer partial votes to a 2nd choice candidate in cases where more
>> voters than needed for the threshold for each candidate voted for the
>> same 1st choice candidate.
> 
> If the rankings were limited in this artificial way, the
> proportionality obtained would be poor, and very poor in some
> circumstances.

I agree with what you're saying there. However...

>> If the rank choices were limited to a 1st choice and a 2nd choice
>> candidate only, unlike Fairytale Vote's IRV/STV method this method
>> would would be monotonic and precinct-summable (and so be OK to
>> manually audit and countable in the precincts) using an n x n matrix
>> where n is the number of candidates running for office.
> 
> As has been explained many times, it is not possible to devise a
> voting system that simultaneously meets all the "desirable
> criteria". Voting systems that comply with 'monotonicity' fail 'later
> no harm'. As has also been explained, monotonicity is of no
> importance whatsoever in public elections because it cannot be
> exploited either by the candidates or by the voters. In contrast,
> failure to comply with 'later no harm' opens the way for undesirable
> strategic and tactical voting. Also, compliance with 'later no
> harm' does seem to be important to real voters.

Untrue. DAC and DSC meet monotonicity and either LNHelp or LNHarm. The 
thing which you can't have is both LNHelp and LNHarm, as well as 
monotonicity[1].

As for monotonicity itself: IMHO, it's not a strategy issue, but rather 
an issue of the method being in conflict with itself. Rather like, say, 
Condorcet Loser for Condorcet methods: the method claims some property 
is desirable, but also some times elects those that would lose when 
ranked according to that property. In the case of monotonicity, the 
method elects X due to support of X, but further support of X causes X 
not to be elected, and so the inconsistency is that the support both 
helps and harms.

Finally, the Schulze method (as used by Wikimedia, Debian, and others) 
fails later-no-harm, something it must do since it's a Condorcet method. 
This fact doesn't seem to have upset the voters much.

>> In other words, for a multi-seat election where we want proportional
>> representation, limit voters' choices to a 1st and 2nd choice and
>> count all voters' 1st choices and transfer excess votes to the voters'
>> 2nd choices and you're done - no rounds and no transfers of already
>> transferred votes.
> 
> If you cannot eliminate candidates and transfer their votes in
> accordance with the voters' instructions, you cannot obtain 
> proportional representation (or only very poor PR).

In my previous post, I gave an example of a method that doesn't use 
elimination. Schulze STV (but probably not CPO-STV) is another. 
Technically, my "Setwise Highest Average" method doesn't use elimination 
either, but you could argue that its use of Sainte-Laguë on the 
coalitions serves the same effect.

Of course, if your point of view is that the voters' ballots are like 
the punchcards to the program - explicit instructions to the method 
itself about which candidates should be eliminated and in what order, 
then the above fails. By that reasoning, only the voters' intended 
method fits (be it STV, Bucklin with winner elimination, CPO-STV, whatever).

-

[1] Actually, I'm not even sure about this. Woodall's impossibility 
theorems listed in Voting Matters #6 say only that mutual majority (he 
calls it Majority), LNHelp and LNHarm implies nonmonotonicity. Perhaps a 
method can have both LNHs as well as monotonicity if it gives up on 
mutual majority. Plurality would be such a method (technically 
speaking), though it is a really bad one.