[EM] Margins vs. Winning Votes

[Juho Laatu]

On Jul 28, 2005, at 07:01, Dave Ketchum wrote:

>> I think there is a trade-off between expressiveness and strategies. 
>> Rating based methods are nice since they can express so much, but 
>> they are too vulnerable to strategies and therefore unusable in most 
>> (contentious) elections. Approval, as you note, lacks expressiveness. 
>> But strategies are not that big of a problem. Pair-wise methods and 
>> IRV are somewhere in between when it comes to expressiveness. 
>> Pair-wise methods are more expressive than IRV. I hope their problems 
>> with strategies are not too bad and their expressiveness could be 
>> used in elections for the benefit of all.
>
>
> We seem to disagree as to where the tradeoff is.

Ok. I must agree that the trade-off I described is not a clear 
dominating pattern but also other (equally influential) rules and 
method specific deviations exist.

> Starting in a Plurality world, what do the other islands look like?
>      Condorcet:  The average voter, most of the time, is not 
> interested.  Bullet voting is acceptable and often adequate.  When a 
> race inspires interest, then picking out the candidates of interest 
> and lining them up according to desirability requires nothing new as 
> to kinds of thinking. Even ranking candidates as equals is no strain 
> (beyond learning to mark that on a ballot) for those with such a 
> desire.
>      IRV starts the same as Condorcet.  IRV's problem with ignoring 
> data only comes up when it messes up an election.  Seems to me we 
> argue about voters ranking candidates as equals, but that is not a big 
> deal.
>      Approval, likewise, will tolerate bullet voting.  As with 
> Condorcet, voters soon want to approve multiple candidates - and must 
> be told they CAN, ONLY, rank all approved candidates equally.
>
> For all of the above the concepts are simple to explain to voters, who 
> can move into taking more advantage at their chosen pace.

Ok. Starting from plurality most other methods look good. On the other 
hand Condorcet methods are maybe a local optimum where both 
expressiveness and freedom of strategies are quite good.

> Primaries are essential for parties in a Plurality world.  Move to a 
> different island and primaries are expendable - but the voters must 
> get past bullet voting to survive this (think of a party with two 
> competing candidates - a primary would sort this out - without a 
> primary you must get past bullet voting for a single voter to vote for 
> more than one).

Yes. Primaries as well as two rounds can make some of the simplest 
methods one step "softer".

> Then we come to ratings and the more exotic methods that get discussed 
> in EM.  How much must the average voter learn to survive the related 
> suspicions when on the same island, let alone taking advantage?

I believe ratings are not usable in most cases. The simplest strategy 
of giving your worst competitor 0 points is already too tempting to 
destroy the basic rating based methods (using ratings for tie breaking 
only is another story).

(I can see some uses for ratings in some friendly elections, e.g. when 
deciding how to invest money of a jointly owned company. But this is 
also another story. These cases are not really traditional elections in 
the competitive sense normally discussed in this mailing list.)

>>> I do not see how this example promotes margins:
>>>      Under wv the plotters achieved their goals.  NOT that I 
>>> approve, for they were only guessing as to what other voters might 
>>> do.
>>>      Under margins they claimed to prefer the LEAST liked candidate 
>>> over one they actually preferred.  Fact that their strategy elected 
>>> the least liked is not a success to celebrate - more a warning to be 
>>> cautious when considering strategy.
>> The reason why I think the example speaks in favour of margins is 
>> that the results of the strategy brought no benefits to the strategic 
>> voters (who seemed to try some sort of a burial strategy but were 
>> mistaken and failed). Therefore margins (at least in this case) 
>> encourage people to simply vote sincerely and forget strategies. 
>> (while winning votes encouraged plotters to vote strategically and 
>> also gave results to them)
>
>
> Not ready to accept this example as determining value of wv vs 
> margins. The example was one structure - other structures could read 
> differently.

Fair enough. I agree this is just one example. I think it is however 
quite possible in real life, not just one extreme case that occurs only 
on paper. All relevant examples should be listed (I tried to promote 
this view also in my first reply to James). Examples where margins 
perform better than winning votes have not been discussed too widely 
(winning votes are often seen as more strategy free by default) and 
therefore I'm happy to push forward some potential examples.

BR, Juho