[EM] RE : Ranked Preference benefits

[Juho]

45: L>>C>R
20: C>>R>L
35: R>>C>L

On Nov 5, 2006, at 10:34 , Chris Benham wrote:

>> In this case Ranked Preferences would elect L.
>
> Imagine that this is the first election after the abandonment of   
> FPP, and you have just unveiled
> this result in front of an audience of  supporters of  Condorcet  
> methods, IRV and even DSC.

To the Condorcet supporters I'd tell that the L and R supporters  
clearly said that they do not want C. Some of them may appreciate the  
fact that from Condorcet point of view L can be seen as the second  
best candidate (since R loses badly to C).

IRV supporters are more tricky since all the votes of the example  
looked like IRV votes (= A>>>B>>C>D style votes) but they didn't lead  
to the expected result. Maybe I should ask them to learn to handle  
the >> in a new way (put it in different place). However if C and R  
team up the method easily elects C and not R (since A supporters find  
C better than R).

I'm not familiar enough with the DSC state of mind to tell right now  
how to handle DSC supporters best. (I'll study more.)

I have to say however that when studying different alternative ways  
of voting in this example I was not fully satisfied with the balance  
of the Ranked Preferences Method. The variant that I presented has  
the tied at top/bottom rules to allow voters to use the stronger  
preferences and trust that the method takes care that their vote is  
used in best possible way to push up/down the top/bottom candidates.  
It however seems that the position of A supporters was quite strong  
when compared to the other groups, and it seemed that in some cases  
the C supporters' votes worked against their interest by pushing R  
down. I'll try to see if some tuning is needed / if some other  
variants perform better than the described one.

> They ask you "what is so special about this election profile that  
> you can justify electing the
> FPP winner? There is nothing remarkable about it, just that all the  
> voters really want to elect their
> favourites."  What do you say to them?

The FPP part is maybe not that difficult. I guess the intent of the  
various alternatives to FPP is not to not elect the FPP winner. If  
each party still has only one candidate (maybe that was the case in  
the example) then it is quite common that the other methods agree  
with FPP on the winner. And people are quite used to the FPP winners.

> I  think it is just impossible to justify collecting the  
> information that tells us who the Majority Loser
> is and then electing that candidate.

Do you mean that the Ranked Preferences Method would intentionally do  
so?

> DMC(Ranking)

I like Condorcet methods as well as their enhancements with approval  
cutoff, but note that in this study my goal is to study if one could  
add some more expressiveness to them, and achieve something positive  
with the new expression power, and still maintain sufficiently good  
strategic properties. The three candidate examples are good in  
revealing weaknesses of the ranked preferences methods but I hope  
that in elections with more candidates the benefits of several  
preference strengths could be demonstrated. If not, then I'm happy to  
jump back to some less expressive methods.

>> I hope the readers of this list will point out any potential  
>> weaknesses.
> Unfortunately the fact that your method is so complicated and  such  
> a daunting chore to operate makes
> this less likely.

Ok. Sorry for not providing a counting service for it. I agree that  
without software the calculations may kill the interest of otherwise  
interested people.

I like simplicity (like simple Condorcet methods, e.g. minmax) but in  
this case I couldn't make my point with less material.

>> In principle I think all the candidates should be compared  
>> simultaneously against each others.
> This seems to be just an aesthetic prejudice of yours that you  
> don't justify.

No, no. Or maybe I didn't justify my claims well enough. I'll try  
again, although explaining this part is not so easy.

Nice aesthetics sometimes correlate with good solutions. But I also  
tend to fight against some aesthetics that don't provide good  
results. In this case I start from comparison of IRV and Condorcet.  
Condorcet compares all alternatives in one round and declares the  
winner. Many people on this list feel that IRV doesn't do as good job  
when sequentially eliminating the weakest candidates. Sometimes one  
may eliminate e.g. the Condorcet winner. This is exactly the pattern  
that I refer to. A tournament like serial competition where the  
weakest ones are dropped is in a way entertaining to the watchers but  
there is the risk of introducing randomness and heuristics in the  
method (since in most cases there is no theoretical basis on why the  
one by one elimination needs to be done). I can't prove that, so you  
just need to trust me :-). I'd say this is more common sense than  
aesthetics (although common sense can be aesthetic to watch :-).

There are also some aesthetics that I don't like. Often Condorcet  
methods are justified by reasoning that includes forcing the cycles  
in the preferences into a linear order. That makes sense for human  
brains that are used to handling physical objects for which linear  
order makes sense and cycles do not. I however prefer to see cycles  
as cycles and not try to forcefully open them. This means that I like  
ideas like ignoring certain expressed preference relations in order  
to achieve a linear order, or Smith set, or Condorcet loser  
criterion, less than many others. 3D aesthetics thus do not work well  
in models that are cyclic by nature.

> mono-raise

No comments on meeting that criterion at this moment. I'll come back  
if I find something useful later.

Juho Laatu

Send instant messages to your online friends http://uk.messenger.yahoo.com