[EM] Methods

[Kristofer Munsterhjelm]

matt welland wrote:
> On Mon, 2011-10-17 at 20:42 +0200, Kristofer Munsterhjelm wrote:
>> matt welland wrote:
>>> Again, I think it is very, very important to note that the ranked
>>> systems actually lose or hide information relative to approval in both
>>> these cases.
>> In what manner does a ranked method hide information? Neither ranked 
>> ballot methods nor strategic Approval can distinguish between 
>> "everybody's equally good" and "everybody's equally bad".
>>
>>> Note that in the first case the results and impact of a ranked system
>>> are actually worse than the results of approval. The political pressure
>>> to converge and appeal to a broad spectrum is greater under approval
>>> than the ranked systems. The evaluation of a voting system only makes
>>> sense in the context of all the other things going on in a society. The
>>> pressure on politicians to actually meet the needs of the people is a
>>> massively important factor and ranked systems appear to wash out some of
>>> that force which is a very bad thing IMHO.
>> Again, why is that the case? In Approval, you're either in or you're 
>> out; but in ranked methods, the method can refine upon those two groups 
>> and find the better of the good (be that by broad or deep support 
>> relative to the others). If anything, this finer gradient should 
>> increase the impact, not decrease it, because the search will more often 
>> be pointed in the right direction.
> 
> A ranked system cannot give the feedback that all the candidates are
> disliked (e.g. all candidates get less than 50% approval). It also
> cannot feedback that all the candidates are essentially equivalent (all
> have very high approval).

Neither does strategic Approval. In Approval, the best simple strategy 
(if I remember correctly) is to approve the perceived frontrunner you 
prefer, as well as every candidate who you like better. In a Stalin 
election, if people were perfectly rational, the left-wingers would 
approve Stalin if the other frontrunner was Hitler.

Well, perhaps people aren't perfectly rational. However, to the degree 
they are honest, Approval can get into a contending third-party problem. 
If you have a parallel universe where Nader is nearly as popular as 
Gore, liberals would have to seriously (and strategically) think about 
whether they should approve of Gore or not - if too many approve of Gore 
*and* Nader, Nader has no chance of winning; but if too many approve of 
only Nader, Bush might win.

> Ranked systems essentially normalize the vote. I think this is a serious
> issue. A ranked system can give a false impression that there is a
> "favorite" but the truth might be that none of the candidates are
> acceptable. 

Some ranked methods can give scores, not just rankings. As a simple 
example, the Borda count gives scores - the number of points each 
candidate gets - as a result of the way it works. The Borda count isn't 
very good, but it is possible to make other, better methods give scores 
as well; and if you do so, an "equally good/equally bad" situation will 
show as one where every candidate gets nearly the same score.

As for distinguishing "equally bad" from "equally good", there are two 
ways you could do so within ranked votes. You could do it implicitly, by 
assuming that the voters approve of the candidates they rank and 
disapprove of those they don't; or you can do it explicitly by adding a 
"against all" (re-open nominations, none of the below, etc) virtual 
candidate.

> Ironically by trying to capture nuances the ranked systems have lost an
> interesting and valuable part of the voter feedback.
> 
> A voting system should never give the impression that candidates that
> are universally loathed are ok. If our candidates were Adol Hitler,
> Joseph Stalin, Pol Pot, Idi Amin, Benito Mussolini, Mao Zedong and
> Leopold II of Belgium then approval would rightly illustrate that none
> are good candidates. However a ranked system would merely indicate that
> one of them is the "condorcet" winner giving no indication that none are
> acceptable.

Here, an implicit solution would record heaps of blank votes, and an 
explicit one would show the virtual candidate to be the CW.

> I think any sane voting system *must* meet this requirement. The ability
> for the electorate to unambiguously communicate that none of the
> candidates are worthy of the post under contest. 
> 
> I don't know how to prove it but my hunch is that approval would be more
> resistant to manipulation by the so-called "one percenter" elites than
> ranked systems.

James Green-Armytage's paper seems to show Approval as one of the rules 
more vulnerable to strategic voting (see 
http://www.econ.ucsb.edu/~armytage/svn2010.pdf ). Whether or not that 
would translate into one-percenter manipulation, however, I don't know. 
I suspect that most of the rules (e.g. various Condorcet methods, 
Approval, Majority Judgement) would be sufficiently resistant. Even 
top-two seems to do well enough to break Duverger's law.