[EM] Scoring (was Re: OpenSTV 2.1.0 released)

[Juho Laatu]

On 19.9.2012, at 20.26, Michael Ossipoff wrote:

> Juho--
> 
> This thread is demonstrating something that I spoke of earlier: There
> are an unlimited number of things that different people can ask for
> from voting systems, just as there are infinitely-many ways to count
> rank ballots.
> 
> It couldn't be any more obvious, could it, that there's just no way
> that agreement can ever be achieved, in such discussions, regarding
> which method to support as an actual proposal.

Well, maybe the experts are unable to agree, but practical people that seek practical methods for practical problems might be more sensible :-).

> 
> But that's ok. EM's discussion needn't be toward the goal of practical
> agreement for support of an actual proposal. EM's goal needn't have
> anything to do with actual social improvement.
> 
> Discussion of "social choice methods" can be (and usually is) entirely
> divorced from any social practicality, use, or value.

EM discussions may cover also theoretical (= not practical) topics. But I agree that discussions that try to find widely agreed conclusions on practical real-life situations are often surprisingly difficult.

> 
> There's no law that says that you have to care about such things.
> 
> I just want you to know that I'm not criticizing you about that.
> 
> You said:
> 
>> For example FBC is an important criterion, but I can accept methods that do
>> not meet it, but that are good enough in the sense that they allow voters to
>> rank their favourite always first, as a safe enough rule of thumb. I don't
>> like methods that fail FBC in the sense that voters often have to betray
>> their favourite, or if voters have to decide whether to betray or not based
>> on some complex analysis. In the same way many other criteria can be met
>> "well enough".
> 
> That's what I used to say. There are a few problems with that.
> 
> You can certainly be forgiven for not knowing what's important to
> voters in this country. No doubt each country is different in that
> regard.
> 
> But understand that that means that what you say might not be
> applicable to this country. And, from what I've heard, some other
> Plurality countries have a very similar habit of lesser-evil voting.
> 
> So, in fact, could it be that what you're saying is applicable only to
> countries that don't use Plurality for their main political elections?

When I wrote that I was thinking about single-winner elections that genuinely elect from multiple candidates. But I think it covers also plurality and two-party systems. The idea that there are third candidates but that are never elected, and that can act as spoilers does not fly very well. If we want to have a two-party system, there are also better election methods for that purpose than plain plurality. Or one approach could be also to have only two parties and two candidates (=> meets FBC).

> 
> As I said, I used to say what you said above. That was before I
> observed a progressive lesser-evil Democrat-voter voting in a
> Condorcet Internet poll, for a presidential election. Yes, I've
> mentioned this before.
> 
> She's a progressive, and preferred the policies of Nader to those of
> the Democrats. But she felt that Nader couldn't win, and that, because
> only a Democrat can beat the Republicans, the one and only goal is to
> maximize the probability of a Democrat winning instead of a
> Republican. We've been over this.
> 
> So she ranked all of the Democrats over Nader. I couldn't tell her
> that she needn't do that, because it was optimal strategy, given her
> assumptions and her goal.
> 
> You see, that's what you're missing. It's what you were missing before, too.

But wasn't that more a problem of the voter than a problem of the method? She betrayed her favourite although there was maybe no need to do so. She just didn't believe that sincere ranking was "a safe enough rule of thumb", or at least a better strategy than the one that she used.

> 
> If you believe that the winner must necessarily be a Democrat or a
> Republican, if you believe that only a Democrat can beat the
> Republicans, then you also believe that maximizing your expectation
> and optimizing the outcome must mean maximizing the win-probability of
> a Democrat.
> 
> In an election with a progressive (whose policies you prefer best), a
> Democrat, and some Republicans, your optimal strategy, in unimproved
> Condorcet, is to rank the Democrat _alone_ in 1st place.
> 
> When we discussed this before, I told why that is. The reason hasn't
> changed since then.

I'm not quite convinced.

> 
> So it isn't a matter of "How likely is it that this method will show
> its FBC failure?". Instead, it's a matter of "Does (can) this method
> fail FBC?"

Maybe the question is if (rational) voters converge towards voting sincerely or towards (some opinion groups) burying their favourite.

> 
> Maybe unimproved Condorcet won't often show its FBC failure.
> Irrelevant. By the beliefs, assumptions, and goal that I spoke of
> earlier--the beliefs, assumptions and goal of the fully-devoted
> Democrat lesser-evil voter, the mere possibility of unimproved
> Condorcet showing an FBC failure means that favorite-burial is the one
> and only optimal strategy in unimproved Condorcet.

Again, not quite convinced.

> 
> It's as simple as that. That isn't in doubt.
> 
> But there's good news: Not every Condorcet method fails FBC.
> 
> Improved Condorcet doesn't fail FBC.
> 
> And yes, Improved Condorcet is a Condorcet method. I've told, at EM,
> why Improved Condorcet passes the Condorcet Criterion, when that
> Condorcet Criterion defines "beats" so as to interpret equal top
> ranking in a way that is consistent with the intentions, preferences
> and wishes of an equal top ranking voter.
> 
> I've defined and described ICT and Symmetrical ICT here at EM. I've
> extensively discussed their properties.
> 
> And I've asked this question:
> 
> Does anyone think that unimproved Condorcet can be defended in a
> comparison with ICT and Symmetrical ICT?

I tend to think that all common Condorcet methods are likely to perform quite well in typical large public elections (i.e. in elections where there are very many different opinions, it is not easy to guess how people will vote, it is not easy to control the voters etc.).

Juho


> 
> Mike Ossipoff