[EM] Cheering for simplicity/Orphan

[John B. Hodges]

>  >The complaints against STV, as I recall, boiled down to "just like
>>IRV, STV will sometimes eliminate the wrong candidate". It is not
>>monotonic, so sometimes you get spoiler effects and perverse
>>incentives.
>>
>>The orphan method is one step more complicated than IRV/STV, but it
>>is still a "simple" method.
>>
>>SO: I am wondering what effects you would get if you applied the
>>orphan method's elimination rule to multiseat STV? How would the
>>results compare with "Sequential STV" or "CPO-STV", both of which are
>>complex and computer-dependent?  If the orphan method significantly
>>improves the performance of IRV, would it similarly reduce the
>  >complaints here against STV?
>  >----------------------------------
>  >John B. Hodges, jbhodges@  @usit.net

Some further comments. Most Condorcet-methods are "brute force" 
computationally. The first thing they do is do all possible pairwise 
comparisons. The multiseat method CPO-STV is likewise a "brute force" 
method; for an N-seat race, it first enumerates all possible 
n-candidate ensembles, then makes all possible pairwise comparisons 
between them. (It somehow deduces from the voter's ballots ranking 
individual candidates which of each pair of ensembles would be 
preferred; I'm sure it also has some method of breaking ties and 
cycles.)

By contrast, Rob LeGrand's "orphan" method does not do all possible 
pairwise comparisons; it does only a minimal number of them. It is so 
computationally efficient that it is within range of a hand count if 
necessary.

Rob's "orphan" method is IRV with a different elimination rule; 
instead of eliminating the last-place candidate, do a pairwise runoff 
of the bottom two candidates and eliminate the loser. This is 
sufficient to prevent IRV from eliminating a Condorcet-winner, which 
in turn is enough to guarantee that it will pick the Condorcet-winner 
if one exists.

I am not a computer-science major, but I have heard of the "traveling 
salesman" problem and how it is computationally very expensive to 
guarantee finding the ideal solution, to the point of being 
practically impossible for large numbers of cities. But, I have also 
heard, there are simple algorithms that will reliably get you "close" 
to the ideal solution: for example, start where you are, and go to 
the nearest city you have not yet visited; repeat until you have 
visited all cities.

CPO-STV is an awesome multiseat method, conceptually. I'm wondering 
if there is a computationally efficient way of arriving at the same 
"ideal" ensemble. My "For Dummies" guess is that the ideal ensemble 
will never include a Condorcet loser and will always include a 
Condorcet-winner if one exists. STV with Rob's "orphan" elimination 
rule would (I guess) be sufficient to do that much.
--------------------------------
>From: Markus Schulze
>Subject: Re: [EM] Cheering for simplicity/Orphan
>
>I suggest that when there are N seats then at each
>stage a plain vanilla STV count should be hold between
>the N+1 candidates with the lowest numbers of first
>preferences and the loser of this count should be
>eliminated.
>
>Markus Schulze

This is a fair multiseat analogy to Rob's single-seat elimination 
rule, but I don't immediately see any logical connection to finding 
the "ideal" ensemble (defined the same way as in CPO-STV.) I'd like 
to hear more about this idea.
-------------------------------
The reason this topic is interesting, IMHO, is that IMHO for selling 
a method to the American electorate, it would be advantageous to find 
a "unified method", one where the same basic algorithm would apply 
both to single-seat races and to multiseat Proportional 
Representation. STV has this. Approval and Condorcet have it, sort 
of, but their multiseat versions are so computationally expensive as 
to be out of serious contention. The average voter is not a CompSci 
major. They don't know how computers work, they don't WANT to know 
how computers work; computers are mysterious, and every voter knows 
they are also insecure. Methods that are computer-dependent are 
therefore suspect as to their "legitimacy", a critical necessity.

I think "Generalized Bucklin" may meet the need for a relatively 
simple version of multiseat PR for Approval. Granted it is Majority 
Choice Approval, not plain Approval, but I think Approval advocates 
should be able to live with that. The voter has the option of casting 
a "plain Approval" type of ballot, and it would have the same effect 
as under plain Approval.

Similarly I think it possible that STV with Rob's "orphan" 
elimination rule, or some other method of about the same complexity, 
could be a relatively simple version of Condorcet PR. IMHO it is 
worth looking into.

-- 
----------------------------------
John B. Hodges, jbhodges@  @usit.net
Do Justice, Love Mercy, and Be Irreverent.