[Election-Methods] Fwd: [LWVTopics] IRV Voting

[Juho]

Here are some short comments also to the 40k mail.

On May 11, 2008, at 22:21 , Abd ul-Rahman Lomax wrote:

---

> Ranked methods in general make an assumption that is blatantly  
> false: that all preferences have the same strength.

I think ranked methods rather make the assumption that it is  
difficult (or unnecessary) to use more information that the rankings  
in a competitive election.

> Just to give an example of how this distorts thinking, it's very  
> common to assume that the Condorcet Criterion is practically golden  
> for election methods (even though IRV fails it). If there is a  
> candidate preferred by a majority over all other candidates, that  
> candidate should win.

In many environments (typically e.g. in politics) that is indeed a  
good rule. The majority principle is a long an accepted and sensible  
principle that can easily be extended to what we know as Condorcet  
criterion.

On the other hand this need not be the case always. For example the  
challenge of Jobst Heitzig ("method design challenge + new method  
AMP") presents another group of methods where majority decisions are  
not what we want. And also such methods can be needed in many  
environments.

And of course in some less competitive environments one could use  
e.g. sum of sincere ratings as well as the target.

There are thus many possible utility functions to approximate with  
the voting methods.

> But we routinely avoid taking that choice in small groups and  
> organizations, when the preference of the majority is small and the  
> preference of the minority is great (and there is not a large gulf  
> in numbers between the majority and minority). Thus the ideal  
> winner of an election, in situations where all voters will, after  
> the poll, agree that this was the best result, is not uncommonly  
> other than the Condorcet winner. But ranked methods can't collect  
> that information, the ballot form does not allow it.

Yes, for some uses the ranked ballots have too little information.

> One ranked method comes close: Borda count, and it does it by  
> making a rough approximation: that the preference strength between  
> two candidates is proportional to the number of candidates ranked  
> between those two by the voter.

Borda ballots may actually collect less information than some ranked  
ballots (if ties are not allowed in the Borda ballots), but it uses  
this limited information in a specific way that assumes also  
preference strengths.

---

>>> Yes.  I prefer IRV to all other systems for electing a single
>>> winner to
>>> an executive office because it elects a majority winner in a single
>>> election without the need for a separate expensive runoff election.
>>
>> Yes.
>
> Again, what I've seen with many students of election methods is  
> that they start out with this. But the incorporated assumption just  
> stated is false. IRV doesn't find a majority winner in a single  
> election, and not just rarely.

I intended to say yes to the fact that IRV eliminates the multiple  
rounds.

Use of term majority is of course quite weak here. The IRV winner is  
a majority winner when comparing the two last given alternatives and  
those voters that expressed an opinion. But there may be also other  
similar majority winner. Maybe the sentence was carefully formulated  
"a majority winner" instead of "the majority winner" :-).

---

> This is an old debate trick. Instead of stating an argument  
> directly, it is stated indirectly, as an incorporated assumption,  
> as part of an argument about something else.

Yes, that is unfortunately way too common in many election method  
related discussions.

---

>>> , and it encourages positive
>>> issue-oriented campaigns instead of mudslinging opponent-attacking
>>> campaigns.
>>
>> Probably better than plurality here but not as good argument against
>> other methods.
>
> Again, Juho swallows the argument because it sounds reasonable. It  
> hasn't done it. Show some evidence to the contrary, if you want to  
> claim this benefit.

My basic thinking here was that methods that collect also second  
preferences typically make it sensible for the candidates to try to  
attract those second preference votes from the supporters of the  
competing candidates, and therefore it doesn't make sense to attack  
those candidates (and indirectly their supporters) in whatever  
possible ways.

---

> What I've seen, though, in every forum where the question has been  
> asked, the experts have agreed that Approval makes very good sense  
> as a simple reform; quite a few have, of late, as well, come to  
> understand Range Voting and to recognize it as theoretically ideal.

I guess you mean that Approval could be considered an improvement  
when compared to plurality.

Saying that Range is theoretically ideal is a quite strong claim. I  
already noted at the beginning of this mail that there may be many  
different targets. Range might provide an "ideal" utility function  
for some use cases but not all. There are also some alternative  
target utility functions that are close to Range, e.g. ones that try  
to find some relatively even utility to all instead of caring only  
about the sum of the utilities (and potentially leaving some voters  
without any utility at all). And of course if also strategic voting  
is considered then the definition of ideal becomes different.

---

> The problem with the whole discussion of strategic voting is that  
> it's often backwards. Voters use strategic voting to improve the  
> outcome. Some Condorcet methods make strategic voting difficult.  
> But that's making it difficult for voters to improve the outcome!

Strategic voting is probably more often a result of trying to improve  
the outcome from one's own perspective (and against the interests of  
others (not necessarily intentionally but from the point of view of  
the others)) than an attempt to improve the outcome to all (possibly  
against one's own personal interests).

If you refer to the lack of ratings information as a reason why it  
may be difficult in Condorcet to improve the overall outcome, then  
one solution is to use some external poll information (or just plain  
personal guess) on what candidates would be the best for the society  
as a whole and then rank the candidates in this preference order.  
This does not give the sincere preferences of sincere Range but might  
actually yield better results (ffs), also from Range style utility  
perspective, in a competitive environment where other voters are not  
expected to generally vote for the best of the society as a whole.

> What is really lost is that ranked methods don't collect the  
> information necessary to determine the best winner, because the  
> assumptions about pairwise victories are defective.

As already indirectly noted, in a competitive environment it may make  
more sense to measure only the one-man-one-vote style preferences and  
forget the more fine grained expressions of utility (either since  
they can not be measured reliably or since majority opinions are  
considered to be the best approach to measuring the opinions of the  
(competitive) society).

Juho




	
	
		
___________________________________________________________ 
All new Yahoo! Mail "The new Interface is stunning in its simplicity and ease of use." - PC Magazine 
http://uk.docs.yahoo.com/nowyoucan.html