[EM] Explaining PR-STV

[Raph Frank]

On Sun, Aug 30, 2009 at 4:15 PM, Kathy Dopp<kathy.dopp at gmail.com> wrote:
>> I think that would be reasonable, but some people might not like that
>> their district ends up with 1 fewer representatives.
>
> Well IRV can not elect candidates to all the seats if it follows its
> own rules for quotas.

I think people would prefer all seats filled in their district than to
have some majority failure seats.

Also, why do you have a problem with IRV not obtaining a majority, but
no problem with using plurality which also doesn't always obtain a
majority?

>> I think the non-monotonicity is not as big an issue with PR-STV when
>> the number of seats gets larger.  The more seats being filled, the
>> more accurate your polling.
>
> Well the 75 voters in Aspen who caused their favorite candidate to
> lose by ranking him first (whereas he would have won if they hadn't),
> may disagree with you, as per this oped in the Aspen Times this week:

Right, IRV is not a good single seat method.

However, the impact of the problems are lessened with PR-STV.

> Well that is where you and I disagree philosophically -- I think
> voters have a right to know that their vote helps, rather than hurts a
> candidate's chances of winning.

But in PR-STV, it is likely to help.  The issue is greater in IRV (as
polls are more accurate, so it is possible to abuse the
non-monotonicity, more easily).

>> Lots of canvassers in Ireland, when they are canvassing, will ask for
>> a first choice and if you say you are voting for another party, they
>> will ask for the 2nd chocie.
>
> So what?

It shows that in actual elections candidates want voters to give them
second choices.

In fact, I have never heard a candidate say that it would be better
for a voter not to risk it.

In real elections, ranking a candidate higher is likely to help the candidate.

After the fact analysis of a specific election might show that it
wasn't the case.

It is almost impossible to take advantage of the non-monotonicity, in
practical PR-STV elections.

However, when dealing with a single seat, it can be an issue.

> IRV is a big big step down from plurality voting due to its
> fundamental unfairness of the way it treats ballots and due to its
> removing voting rights, eviscerating election transparency and
> verifiability, huge cost increases, etc. etc.

In Ireland, PR-STV counts are done by hand in public view, and they
are considered very transparent.

> STV has all the same flaws as IRV, plus some.

But they have less negative effects.

> PLUS IRV adds nonmonotonicity, removes voting rights, adds costs and
> complexities, and reduces fairness, etc.

Well, in the 2 candidate case (plus minor candidates), then one of the
top-2 wins as currently.

Each voter still gets 1 (movable) vote.

>> However, this method means that if you want to support a party, you
>> have no choice but to vote for the party's candidates.
>
> Huh!? Does that statement make any sense?

Maybe I like a party but hate their current candidate.

With PR-STV, I can just vote for a different candidate from the same party.

> I have yet to see any method proposed that is worse than IRV/STV short
> of dictatorship.

IRV may be little better (and possibly worse) than plurality, but it
is not dictatorship.

PR-STV with reasonably large districts puts most of the power in the
hands of the people.  They get to pick which parties win and also
which candidates within the parties win.

> Absolutely false, In fact most of the methods that solve that problem
> are precinct-summable, as are both Condorcet and range and approval
> voting and Bucklin methods - and do not remove voting rights and are
> FAIR, unlike STV/IRV.

Again, I am talking *PR* methods.  The condorcet equivalents for
PR-STV are all very complex (CPO-STV and Schulze-STV).

If not for the complexity, I would much prefer those methods.
However, I think that standard-grade PR-STV is still a big improvement
over single seat election methods.

>> I am willing to accept that ranking a candidate higher increases the
>> probability of him winning,
>
> Then you are accepting a blatant falsehood because IRV/STV elicits
> nonmonotonicity more as the number of candidates running increases, as
> has been mathematically proven, and nonmonotonicity, (hurting your
> candidates' chances of winning by voting for them) frequently occurs
> in IRV/STV.

A non-monotonic method can pass my criterion.

It just has to increase the chance of the candidate winning.  The
probability is worked out when casting the vote (based on generally
available information), not when looking at the final result.

For example, imagine a (granted crazy) method where each person votes
for 1 candidate.  However, a higher last digit is a worse result (if
the other digits are equal).

So, 123,456 loses to 123,451 (as 1 is better than 6 as a last digit)
but 123,456 beats 123,000 (as the last digit rule only applies when
the other digits are tied).

If I vote for some, it will probably increase his last digit (so be
bad), but on average it is still worth voting for him (as it is
unlikely that top-2 candidates will be within 10 votes).

This is non-monotonic, but it is still worth voting for your favourite
candidate as the probability is that it will help him.

>> This is true in IRV and PR-STV.  You get 1 vote and it will be
>> assigned in whole or in parts between candidates in accordance with
>> your instructions.
>
> Misleading statement at best, as voters get *up to (and sometimes less
> down to zero)* votes per round for up to two (and sometimes less down
> to zero) candidates. Hardly could be characterized accurately as
> "equal" treatment.

They only get less than 1 if their ballot is exhausted (and thus have
given no further instructions).