[EM] [CES #4429] Looking at Condorcet

[Juho Laatu]

On 3.2.2012, at 21.45, Andy Jennings wrote:

> - If someone built a computer program that presented me pairs of candidates at a time as Kristofer suggested, that would make it somewhat easier.  I think I would still prefer to divide them into tiers first, but if I divided them into tiers first, I might not need the pairwise comparison hand-holding.  Also, suppose that I analyzed the candidates in three different policy dimensions that I consider equally important and I found that my policy preferences were:
> Foreign Policy: A>B>C
> Domestic Social Issues: B>C>A
> Domestic Economic Issues: C>A>B
> Now I prefer A to B, B to C, and C to A.  A cycle among my own personal preferences when I compare them pairwise.  Then my output ranking would depend on the order in which the pairwise questions were asked.  ??!?

A typical assumption is that the rankings of individuals are transitive. I think this is pretty much based on the assumption that people rate the candidates (unconsciously) anyway. Your three opinions were maybe 9>8>7, 9>8>7 and 9>6>3 in terms of ratings. That means that your transitive preference order is C>A>B, and you do have valid (unconscious) ratings for the candidates.

The alternative approach would be that you indeed have cyclic preferences. But that maybe means only that you may behave strangely if you base your judgements on limited information. If you consider only foreign policy and social issues, you may end up saying B>C, but if you had remembered to think also about economic issues, you would quickly change your statement.

> - If I were trying to cast an honest Score Voting Ballot, I would have to feel like there was an objective meaning for the various scores.  Then I could consider each candidate separately and give them my honest scores.  I probably wouldn't even normalize.  If I were going to normalize, then I might as well go fully-strategic and vote approval-style.

I agree. Already normalization is strategic. (Or maybe you have been explicitly requested to give min points to the worst candiate, and max points to the best, in which case you could sincerely cast a sincere (re)scaled vote.)

> - If a real election were being tabulated with Condorcet, I would vote honestly.

I agree. That is a good default strategy. (Strategic voting doesn't really make sense unless some expert that the voter trusts tells him to vote in some certain way. And also in this case the expert may well be wrong.)

> - If a real election were being tabulated with IRV, I would warn people not to vote for minor candidates.

Also this approach makes sense. This is however not a complete strategy yet. If we look at Burlington, maybe also some supporters of a major (top three) candidate should not have voted their favourite. (Sometimes voting for the minor candidates is harmless , and also a useful tool to market the minor candidates for some secondary reasons.)

> FBC is very important to me

Could one say that Condorcet methods are FBC compliant enough so that you can recommend people not to betray their favourite 1) as the defaut rule if they are not told by experts to do otherwise or 2) as the default rule that is in practice valid in all lagrge elections, where voters make independent decisions on how to vote, and where their opinions are not fixed but can change all the time?

Juho