[EM] IRV vs Plurality

[Juho]

On Jan 26, 2010, at 1:12 AM, Abd ul-Rahman Lomax wrote:

> At 04:05 PM 1/25/2010, Juho wrote:
>> I reply to myself since I want to present one possible simple method
>> that combines Condorcet and added weight to first preferences
>> (something that IRV offers in its own peculiar way).
>>
>> Let's add an approval cutoff in the Condorcet ballots.
>
> Highly recommended. I also highly recommend explicit approval cutoff  
> on Range ballots, it then becomes possible to sense majority  
> approval, which is, in practice, quite important in the general  
> election case, allowing any method to be used with a standard  
> majority-of-the-votes approval requirement for any democratic  
> decision.

There are many (working) uses for an approval cutoff in ranked  
ballots. But on the other hand they may add complexity and confusion  
and not add anything essential. => Careful consideration needed.

>
>> The first
>> approach could be to accept only winners that have some agreed amount
>> of approvals. But I'll skip that approach and propose something
>> softer.
>
> I.e., normally, a majority. Or sometimes lower than that, or some  
> margin of victory requirement that, ideally, is successful in  
> predicting majority approval in a runoff election.
>
>> A clear approval cutoff sounds too black and white to me
>> (unless there is already some agreed level of approval that must be
>> met).
>
> I've proposed a midrange approval cutoff as standard in range  
> ballots because of the simplicity, and it dovetails with the concept  
> of equality of expression as to positive and negative opinion, and  
> allow discrimination of a "lesser-of-two-evils" vote from a "both  
> acceptable" vote. If our goal is broad public satisfaction as to  
> results, making decisions based on the lesser of two evils isn't a  
> good idea, unless, of course, it is the lesser of two evils itself,  
> with no better choice being possible!
>
>> The proposal is simply to add some more strength to opinions that
>> cross the approval cutoff.
>
> Bucklin does that, basically, by only considering approval votes,  
> but it sets up a declining approval cutoff, typically in three  
> batches, loosely named as Favorite, Preferred, and Approved. I've  
> suggested that in a runoff voting situation, majority required,  
> "Approved" has a very specific meaning: it means "I would prefer to  
> see this candidate elected over holding a runoff election." Voters,  
> then, by what candidates they choose to approve given their overall  
> understanding of election possibilities, will sincerely vote this.  
> It makes no sense not to.

Bucklin is a bit simpler.

>
>> Ballot A>B>>C>D would be counted as 1 point
>> to pairwise comparisons A>B and C>D but some higher number of points
>> (e.g. 1.5) to comparisons A>C, A>D, B>C and B>D. This would introduce
>> some approval style strategic opportunities in the method but basic
>> ranking would stay as sincere as it was. I don't believe the approval
>> related strategic problems would be as bad in this method as in
>> Approval itself.
>
> I suggest looking at Bucklin. In Oklahoma Bucklin, which was  
> declared unconstitutional before ever being used, fractional votes  
> were assigned to lower preferences, making this the first attempted  
> Range method in the U.S. It is a shame that the framers of the law  
> place mandatory ranking in it, in a misguided attempt to push for  
> majorities, for that was the basis for rejection, not the multiple- 
> vote aspect of Bucklin or the fractional votes.
>
>> Counting only first preferences would not be a good approach since it
>> should be possible to vote e.g. X>Y>>.... when X is my favourite
>> candidate of my favourite party and Y is the strongest candidate of  
>> my
>> favourite party.
>
> Bucklin allows this vote, easily. But parties would generally be  
> advised to avoid multiple candidates running in the same election.  
> It is a bad idea for many reasons, if they already know -- and they  
> should know -- which candidate is the strongest. It isn't just vote  
> splitting, it is splitting up campaign funding, much of which is  
> devoted to generating name and affinity to name.

It depends on the election if multiple candidates per party is  
practical. Also scenarios with e.g. two left wing parties (minor and  
major) should work.

>
> Bucklin is ranked approval voting. Not for public elections yet, but  
> for theoretical consideration, I've proposed Range/Bucklin, where  
> the method simulates a series of repeated elections (not "runoff  
> elections" with reduced candidate sets, a basically bad idea, unless  
> very good selection of the candidate set is used that would include  
> any condorcet winner and any range winner, for starters) with  
> declining approval cutoff. The voter controls the level at which the  
> approval votes are cast.
>
> This method then seeks to find majority approval, one step at a  
> time. It thus provides limited later-no-harm protection for the  
> voter, it only brings in lower preference votes at a point where  
> it's clear that a majority can't be found without them. And the  
> method might terminate at midrange, or perhaps a bit below midrange,  
> and this would, of course, affect voter strategy.
>
> But something else becomes possible. The method sets up an incentive  
> for the voter to vote preferences accurately, I believe. If it does,  
> the method will vote intelligently in the voter's interest. Given  
> that, the range votes themselves, in toto, should be quite useful.  
> I'd suggest that whenever the range winner is different from the  
> Bucklin winner, as defined, and if the difference is significant  
> (which should be precisely defined, of course), the range winner  
> would be included in any runoff election.
>
> Analysing methods like this could be quite complex, but the first  
> part (the Bucklin election) simulates what would happen in a real  
> series of repeated elections, missing only the additional advantages  
> of closer examination by the electorate, it is still based on a  
> snapshot on election day. The range analysis predicts two things:  
> overall real satisfaction including all ratings, and election  
> turnout in a runoff. Weak preferences don't encourage voters to turn  
> out and vote, so real preference strength is tested.
>
> This latter phenomenon has generally been overlooked in considering  
> the effects and implications of runoff voting, and it probably  
> improves the quality of results according to absolute differential  
> utility summation.
>
>> This is just a simple method to demonstrate that if one wants to put
>> some extra weight on first preferences also Condorcet methods could  
>> be
>> modified to cover such needs. There may be different requirements on
>> what kind of first preferences or "core support" should be given
>> additional weight. Depending on that definition also other kind of
>> Condorcet variants could be developed.
>
> By definition, range methods put extra weight on first preferences,  
> if the voter chooses to express the first preference exclusively, as  
> does Bucklin, at least in the first round.

Range is also designed to elect candidates with no core support, e.g.  
one that gets 60% of the points from all voters while all others have  
only limited amount of strong support and no support from the rest.  
IRV puts always main weight on first preferences (among the remaining  
candidates).

>
>> This method could be used to reduce the chances of candidates from
>> minor parties (with no strong "core support") to win. I'm not saying
>> that that is a general target, but if someone wants to set that  
>> target
>> then this type of approach could be used. The point is that Condorcet
>> could also emulate, approximate or even improve some of the  
>> properties
>> of IRV if needed.
>
> Sure. And allowing equal ranking would improve IRV, and keeping the  
> majority requirement for election, under some conditions, will also  
> improve IRV, probably greatly. (In real runoff elections with top  
> two runoff, which will almost always imitate IRV if a majority of  
> votes cast are required, and looking only at nonpartisan elections,  
> the runner-up goes on to win the runoff, whereas with IRV, it almost  
> never happens. Real runoff elections test voter preferences, and  
> weak preferences don't result in votes, as a general principle.  
> Thus, I expect, if voter turnout is considered, we can expect real  
> runoffs to pick the social utility maximizer of the candidates  
> eligible to receive votes in the runoff.)
>
> Note that the form of IRV that Robert's Rules *actually* describes  
> (contrary to FairVote propaganda repeated all over the place)  
> requires a majority of votes or the election is repeated, without  
> eliminations. I believe new nominations are even required, but they  
> should certainly be allowed even if existing candidates were  
> automatically nominated, or some were automatically eliminated but  
> could be restored by some process.
>
> Hey, what if any candidate could get on the runoff ballot if the  
> collection of such candidates guaranteed, by deposit, the payment of  
> the runoff election costs.... Or signatures of registered voters in  
> lieu of the payment or as part of it?
>
> The idea would be to guarantee that a truly supported candidate  
> could keep in the race. If the candidate wins .... the public  
> treasury covers the payment. Not an idea thoroughly worked out, to  
> be sure.....

I guess many countries must have some rules like this (= allow all  
candidates with sufficient support to run).

Juho