[EM] A conversation with an English woman about IRV [WARNING: mildly obnoxious and long rant]

[Dave Ketchum]

He offers good thoughts - I will try to add a bit.

On May 4, 2011, at 4:26 PM, robert bristow-johnson wrote:
> On May 4, 2011, at 1:48 AM, matt welland wrote:
>> On Tue, 2011-05-03 at 21:38 -0400, robert bristow-johnson wrote:
>>> after IRV has been beaten up so badly because of its perceived
>>> complexity, people ask me how can i explain Condorcet in a sentence
>>> and i answer:
>>>
>>>  "If Candidate A is preferred by more voters than Candidate B, then
>>> Candidate B is not elected."
>>>
>>> it's simple and sensible and, of course, fails if there is a cycle.
>>>
>> I have
>> followed this list for years and read many explanations on  
>> Condorcet and
>> just like the description given to the English woman above none of  
>> them
>> are easy to assimilate. How the heck do you translate my rankings  
>> into
>> "if more prefer A over C ..."
>
> If more voters rank A above C on their ballots (than those who rank  
> C above A).  It translates directly to a simple, traditional  
> election where only A and C are candidates on the ballot.  It means  
> the same thing.  All the ranked ballot requires the voter to do is  
> decide how they feel about candidates B and D (relative to A and C)  
> by Election Day.  So we require voters to make up their mind about  
> the candidates by Election Day. Such an imposition.
>
Actually we often ask more effort by voters than Condorcet requires.   
Voters are required to rank only the one or more they choose to rank  
above any they choose to leave at the bottom (unranked).  If they  
choose to use only one rank level there is no more deciding to do than  
there would be in Plurality (rank only one of A/B/C/D etc) or Approval  
(rank as many as they choose among A/B/C/D etc).

>> You are asking people to have faith in
>> your fancy math and programming.

It is not fancy math - count how many voters voted A>C and how many  
voted C>A.
>
>
> Condorcet is precinct-summable.  It can be counted by hand, but is  
> much more laborious than counting First-Past-The-Post by hand.  But  
> the counting principle is simple.  We rely on computers to do  
> tabulating anyway, but what the computers are (or should be)  
> counting should be readily understood.  If there is no cycle,  
> Condorcet satisfies that.
>
And, if multiple precincts, the results from all can be summed.  The  
winner, or the data defining cycles, is contained in those results.
>
>> At the end of the day I remain
>> unconvinced that it is a sufficiently better method than Approval  
>> by any
>> metric grounded in the messy reality of imperfect humans voting for
>> other imperfect humans to be their leaders.
>
> Does not allow the voter to express a preference over two approved  
> candidates.  A Libertarian that leans Republican when forced to  
> choose in a two-party race should be able to express that he/she  
> likes the Libertarian more than the Republican (and would feel  
> forced to "approve" of both as opposed to the Dem with Approval  
> voting).

Why bother?  Worth it if it would make a difference often enough; or  
even lets voters express their desires more completely often enough  
though almost never affecting who wins.
>
>> From the perspective of US single winner elections I say the  
>> following:
>>
>> 1. Approval voting;
>>    - trivial to transition to (no over-voting), want to vote for the
>>      underdog while hedging your bet for the frontrunner, no problem
>>    - everyone gets the mechanics and the nuances of approval after a
>>      minute of explanation
>>    - very low effort to vote, avoids all the comparisons in ranking
>
>>    - minimal real world risk of unintended consequences
>>    - naturally resistant to strategic voting.
>
> No, it is not.  It requires a great deal of strategic thinking in  
> voting.  I had (virtually) a first-hand experience of that in the  
> Vermont State Senate race last year.
>
>> It's binary, what can you do?
>
> You can (and must) decide whether or not to approve of a lessor- 
> preferred (but "approved of") candidate, relative to the candidate  
> that you really want to see elected.
>
>> 2. Range voting
>>    - degree of improvement over approval is debatable, at least for
>>      today, maybe a few years from now the need will be different
>>    - significant step in complexity for the equipment, 1 bit toggle
>>      to n bit integer. I can't implement that on the current ballots
>>      used in Arizona for example.
>> 3. IRV
>>    - this one feels good to half assed thinkers and that is its
>>      greatest danger. 'nuff said.
>> 4. Condorcet
>>    - theoretically near perfect but I don't grok it and neither will
>>      99% of the populace.
>
> Grok this:  If more voters prefer Candidate A to Candidate B, then  
> Candidate B is not elected.
>
> Hard to grok?

An example may help.  With Plurality those against Bush had to vote  
for Gore or Nader.  With Approval they could vote for Gore and Nader  
as equals.  With Condorcet they could vote for Gore as better than  
Bush, and for Nader as best.
>
>>    - a bitch to implement without a computer for the UI and we all
>>      know how great it is having computers in this process
>
> Falsehood.
>
> Needs the same old ranked-ballot we had with IRV.  just a different  
> method to count them.

This matters.  Voters rank just about as with IRV (but can rank  
multiple candidates equally).

Computer counting, once learned, can be used on many computers.
>
>>    - any ranking system is way to much of a pain for the average
>>      Joe who just wants to get out of the damn polling booth and
>>      home to dinner.
>
> Falsehood.
>
> It requires less out of the voter than either Range or, because of  
> the need for tactical thinking, Approval.

Here I can vote for Gore and Nader, with one preferred over the other;  
with Approval voting for both implies equal liking.
>
>> Go try any of the example systems available
>>      on the web, I'm guessing it takes 10x the time for normal
>>      non-geniuses to articulate that they want in a ranked system vs.
>>      approval. Remember, you have an interest in the mechanics of
>>      voting and have practiced doing ranking. Everybody else
>>      will experience it as a tedious pain.
>>    - my gut tells me that Condorcet is more vulnerable to strategic
>>      twists than Approval. But that could be because I don't get it.
>
> Anyone (who has any secondary preferences) can do a ranked ballot  
> just as easily as traditional ballot.  They don't have to think to  
> themselves "how much *more* do I like Candidate A over Candidate  
> B?"  They only have to decide "If it's between A and B, I prefer A.   
> Alternately, if it's between B and C, I prefer B."  Naturally that  
> means that if it's between A and C, this voter prefers A.
>
> Even if the voter has no secondary preference, all he/she has to do  
> is marked their chosen candidate with number 1 (although any number  
> would do, since all unmarked candidates are tied for last place on  
> any single ballot).
>
> All Condorcet asks you (and the other voters) is to consider your  
> contingency vote and IF there is no cycle, Condorcet will resolve  
> the choice between any given pair of candidates exactly as a  
> traditional vote-for-one election would.  Everyone's vote has equal  
> weight (unlike it would with Range) and the election treats your  
> vote as such.
>
> UNLESS there is a cycle (which I maintain is rare), there are no  
> funny surprizes that "Gee, A clearly would be B, but when C entered  
> the race, somehow B wins."  That's what we're trying to avoid with  
> IRV, but we know, not just theoretically but in reality (Burlington  
> 2009) that IRV fails that very objective.
> --
> r b-j                  rbj at audioimagination.com
>
> "Imagination is more important than knowledge."