[EM] Margins vs. Winning Votes

[Juho Laatu]

On Jul 28, 2005, at 06:05, James Green-Armytage wrote:

>> I think the correct way forward would be to write those examples down
>> and then see what we have and estimate then relative vulnerability (of
>> winning votes, margins and pair-wise methods in general) to 
>> strategies.
>
> 	Seems a bit too anecdotal. It's hard to quantify vulnerability. A
> computer simulation might somewhat interesting numerical data, but the
> result depends on what kind of assumptions you put into it. So far, I
> think that intuitive reasoning is a more useful way to approach this 
> topic
> than any quantitative method that I can think of.

One must try. Assumptions too must be openly explained and justified. I 
think this is all part of the good old scientific method. I agree that 
intuitive reasoning, with justification, is better than e.g. trying to 
create numeric formulas to justify opinions.

(Btw, one area where numeric values appear to be misused is criteria. 
Sometimes it appears as if the number of criteria invented is used as a 
means of justification to defend ones opinions. :-) Maybe this is not 
exactly the case, but anyway I'd like to see more intuitive justified 
reasoning to summarise the findings. The criteria appear to be 
intuitive at the first sight, but since the studied issues are complex, 
it is also easy to provide several criteria supporting one viewpoint 
and forgetting some other problems in their definitions.)

> 	Right... it's a difficult issue. In my opinion, cardinal pairwise 
> (and to
> a slightly lesser extent, approval pairwise) does a better job than WV 
> at
> allowing voters to vote sincerely without having to worry much about
> strategy. But these methods are perhaps too complicated for a first
> generation proposal.

I agree with you analysis. They may also be complicated for a start 
(MinMax is simple btw ;-). I haven't been yet able to produce any good 
analysis of your new methods and so I'm for the time being unable to 
comment on their possible success. I'm planning to check also their 
margins style variants - one day.

> 	How complex is the counterstrategy in WV? Basically, if you prefer 
> X>Y,
> and you don't trust the Y>X voters to refrain from trying to bury X, 
> you
> shouldn't rank Y above any other candidate. To be honest, I don't know 
> how
> well this would work in practice, i.e. whether voters would be 
> comfortable
> using the counterstrategy, how often successful incursions might occur,
> and how often strategic truncation might obscure a sincere Condorcet
> winner. However, I would much rather implement WV than margins, as I
> consider the latter to be truly unstable.

I just hope this wouldn't lead to bullet voting, just to be sure. 
Better if strategies are not present and sincere voting is the main 
rule. The scenario where majority of voters truncate heavily (just in 
case) of course reduces Condorcet methods to something close to 
plurality.

The instability of margins is still ffs for me. I know your web page 
and examples there. Maybe you'll continue providing good justifications 
for comparisons between methods there.

> 	On the subject of public salability, one could argue that the
> Smith-efficient Condorcet methods with WV or margins are too complex 
> for a
> first generation proposal. It depends to whom the proposal is being 
> made,
> and for what purpose. If a Smith-efficient WV method is too complex 
> for a
> given function, I am happy to start with approval or IRV (preferably
> ER-IRV, and CWO-IRV as a very ambitious proposal), in the hopes that
> smaller steps will eventually lead to greater ones. I'm also 
> interested in
> non-binding ranked ballot direct democracy systems that uses a pairwise
> tally; I expect that this would help make the concept of a Condorcet
> winner more concrete and intuitive to the general public.

I'm pretty much on he same line. I have also promoted the "least 
additional votes" method, i.e. MinMax(margins). One reason behind that 
is its simplicity and understandability. I think understanding the idea 
of ranking candidates is quite easy, understanding that the one that 
would beat any other candidate is also quite easy, and understanding 
that the one that needs least additional votes to beat any other 
candidate is also one quite easy explanation that would help people 
seeing through the most complex issues (like Arrow's theorem, loops 
etc.).

Different countries, different political systems and different 
elections have different level of readiness for Condorcet methods. 
Their use might also start within parties, companies and associations. 
That could pave the way for making them more acceptable in general. And 
that would also serve as a testing ground for evaluating the 
probability of elections turning into strategic battles vs. turning out 
to be valuable sincere democratic tools.

> 	I find it fairly realistic that some moderate Republicans (B voters)
> would not want to rank the more extreme Republican.

True. In the group of examples I presented the claim that ranking own 
party delegates is more probable and ranking other party delegates is 
less probable, but of course there are many reasons for ranking and not 
ranking.

>> I should also calculate how easy/difficult it is to apply the 
>> different
>> strategies (number of votes needed, risks etc.) but that's too much 
>> for
>> now and I leave that for further study.
>
> 	Again, I doubt that quantitative analysis will yield a satisfying 
> result,
> but there is no harm in trying.

Ok. I should not use word "calculate" but talk about estimating and 
justifying. I think I agree with your thinking that at this phase exact 
formulas are not a credible way forward. Constructive discussions and 
trying to gain better understanding is.

> 	This may be obvious to you, but I should mention it to avoid 
> confusion.
> If B>A voters are worried about the A>B voters trying to bury B, their
> truncation counterstrategy in WV is only useful if they are able to 
> make
> it clear to the A voters that they will use it. That is, the 
> information
> should come out in pre-election polls, or something like that. If it
> doesn't, then its deterrent value is lost, and its value to the B 
> voters
> is questionable. Of course, the B voters can't wait to see what the A
> voters do before they decide to use the counterstrategy.

Yes, that seems to be the way this game should be played. Of course 
ending up at that level of plotting in real elections would make at 
least me dislike the used voting method. I thus hope that strategy 
risks will be eliminated already at an earlier stage (i.e. before the 
regular voters have to face them).



Taking into account the number of times I have now mentioned the 
interest to avoid voters having to consider complex strategic issues 
and to allow them the wonderful freedom of voting sincerely, maybe I 
should formulate that as a criterion. Requiring that a voting method is 
in practice free of strategic concerns from individual voter point of 
view (maybe assuming that there is no central control of strategies) is 
of course quite strict. In plurality strategic voting is very common 
and people have learned to live with it (="don't vote for Nader, vote 
for the Democrat candidate"). In Condorcet the problem is more severe 
since different strategic settings are more complex. One would need 
experts to tell how each voter should vote. For this reason more 
complex methods like Condorcet should also be more free from 
strategies. Therefore the criterion should maybe be formulated so that 
whatever strategies are needed, an average voter must be able to 
understand them (without extensive training and without any specific 
interest in voting methods). If this is not the case, then we are at 
least close to having an election method that people may with good 
justification call corrupt.

The basic strategies of plurality may still meet this criterion. People 
don't like giving up their first choice but at least they (in average) 
understand why it is sometimes sensible to do so. Condorcet methods may 
have serious problems meeting this criterion if the strategies that we 
have discussed are something that may often/sometimes have influence on 
the outcome of the elections.

BR, Juho