[Election-Methods] MMPO: the best 'transitional' method?

[Abd ul-Rahman Lomax]

At 10:38 AM 9/3/2007, Diego Renato wrote:
>As a newbie in this list, I have no preference about the best voting 
>method. I am aware that instinctively Condorcet criterion is 
>desirable if consensus does not exist, but approval or range can 
>produce good results too.

It is important to understand that the Condorcet Criterion is 
seriously flawed, and it has nothing to do with consensus. If you've 
got consensus, the Condorcet Criterion will be satisfied, 
unconditionally .... but the problem is that the Criterion completely 
neglects preference strength, so a maximally weak, trivial preference 
of a majority can outweigh a maximally strong preference of a 
minority that is almost half the voters, resulting in an outcome that 
the *entire electorate* would reject if asked in those terms.

You think A and B are just about equal, but A is ever so slightly 
better than B to you, so you rank A over B, having no idea how the 
other voters feel, and you and those like you are, barely, a 
majority. Unbeknownst to you, the B voters have knowledge that they 
will die if B is elected -- or at least they think so!

Now, if this were a group of friends, and the truth came out, how do 
you think they would decide? If they were not friends, but simply 
other human beings, how would they decide? Would we put almost half 
the people in fear of their lives for a benefit so small that we 
really didn't care, the election of B was quite acceptable to us, 
almost indistinguishable?

So the Condorcet Criterion, in an extreme case, could require the 
election of a *horrible* option.

Now, other things being equal, and, in particular, if we have and 
cannot get information about preference strength, then the Condorcet 
Criterion makes sense. And if we are only thinking about ranked 
ballots, which, by their nature, do not collect preference strength 
information, then, of course, we want to see that Criterion 
satisfied. Which is why we oppose Instant Runoff Voting; if we are 
going to have a pure ranked ballot, why not insist on the Condorect 
Criterion? Why allow a candidate to win who would be beaten by 
another candidate? (only if there were a cycle would we permit it, 
i.e., there is no candidate who remains unbeaten in this way.)

>However, based in Bucklin experiences in USA, I think that any 
>method that violates later-no-harm (except asset voting) is likely 
>to provide incentive to bullet vote and became a costly version of 
>plurality. If later-no-harm is indispensable for a transitional 
>method, MMPO seems the best alternative because it is nearly 
>Condorcet-efficient and still easy to understand.

We have been suggesting a new look at the Bucklin experience. I think 
we've been lied to about the reason it was dropped. That reason makes 
no sense. I don't have time to detail the reasons, but I consider it 
a near-certainty that Bucklin was dropped because it was *working* -- 
or there was a danger it would work -- and so those threatened by a 
more equitable result acted to dump it, and they gave an excuse which 
too many have accepted as valid without really thinking about it.

It only takes a few percent of votes to deal with the spoiler effect. 
The ten percent usage of second ranks in Bucklin, purportedly too low 
to justify it, would be plenty to remove the spoiler problem. It's a 
phony reason to drop it, particularly because the extra votes cause 
very little extra counting hassle. They only need to be counted when 
needed, and, if the conditions were as have been claimed, that's ten 
percent more counting, in an election where no candidate gets a 
majority first choice. If it fails to produce a majority winner, it 
fails, but it has not made things worse, and quite possibly it has 
made them better, the plurality winner will be a more fair one.

No, we've been had.