[EM] STV: who could ask for anything more?

[James Green-Armytage]

John Hodges wrote:

>My question is this: what is there about these two methods that does not 
>satisfy? People on this list are busy inventing voting methods that 
>require calculations many times more complex, even one that required 
>a few billion calculations for each voter's ballot. Granted that 
>computation is cheaper nowadays, there remains the problem of 
>legitimacy.
 
>What is so desperately bad about Party List and STV that 
>would lead you to consider such complex alternatives?

I (James) reply:

Dear John,
	Okay, that's a fair question. First let me talk about party list vs. STV,
then I will talk about STV vs. CPO-STV.

	Personally, I wouldn't lump STV and party list together on the same
level. I would argue that STV is a real leap forward from party list in a
number of ways.
	For one thing, there is the fact that many countries use a 'closed' list,
which means that parties determine the order of the list rather than
voters. For any reasonably popular party, the first couple candidates on
the list are pretty much assured to gets seats no matter how things go on
election day, and I am among the people who think that this is screwy.
(One thing that has been said in defense of closed lists is that sometimes
women and minorities do better in closed list systems than open lists.
This is worth consideration, but I think that there are better ways for
women and minorities to gain office that don't have the same drawbacks in
terms of accountability.)
	Anyway, some countries use an open list, which allows voters to vote for
candidates on the list, and then orders the list by who gets the most
votes. This should increase the accountability of seat-holders, but there
are still some unsettled difficulties here.
	For example, lets say you figure that your party will get only three or
four seats in the district. Let's also say that your favorite candidate on
the party list is fairly obscure, and more likely to come in eighth or
ninth on the list. If you have a strong opinion about the differences
between candidates higher up on the list, then you might consider voting
for one of them, instead, probably someone who is likely but not certain
to come in at third or fourth place. If you are doing that, then you are
voting strategically, and nobody will know that you actually supported
that poor old guy who was your initial first choice. 
	And there is a chance that this kind of thing can have a self-fulfilling
role, as we suspect may happen in our own system. Lots of people might
like a candidate, but assume he doesn't have a chance, and hence
strategically vote for someone else... and lo, the perception becomes a
reality, although the possibility remains that if everyone voted sincerely
he might have won
	Alternately, your favorite candidate on the list might be very popular,
and almost certain to get a seat. So, if you vote for him, you might be
wasting your vote again. Maybe you have a strong opinion about some of the
candidates that are closer to the margin. Should you vote for one of them,
or your real favorite? It makes sense to vote for a marginal candidate,
but if everyone does that, yikes!, the marvelously popular candidate
actually ends up with not enough votes and doesn't get a seat.
	This kind of stuff is messy, and STV solves the problem very efficiently,
in my opinion. A vote for a failed candidate is transferred. A vote for an
over-popular candidate is transferred, or transferred in part. 
	Also, I am still a little bit uneasy about how party list really
reinforces the whole sense of party identity and party unity. Some people
value that, I'm sure, but I am much happier seeing candidates and
politicians as individuals, rather than part of a collective ideological
identity.
	There are some countries where candidates *have* to run as part of a
political party, i.e. can't run as an independent. I'm thinking of Italy,
here, I think, but I imagine there are more. 
	Actually, that has always seemed to me to be a fatal flaw in the Mixed
Member Proportional (MMP) system, which is otherwise somewhat attractive:
it basically seems to assume that members of the same party are roughly
interchangeable. 
	One weird ramification of MMP is this: If you were allied to a party, but
you were allowed to run as an independent, and you think you would win
your district anyway, then as I see it, it would be in the interest of
your party for you to run as an independent, then just vote with them all
the time anyway once you got into office. That is, if you ran as an
independent, then when the extra members were added to assure party
proportionality, it would register your (undeclared) party as having one
less member, and hence they might stand to gain an extra compensatory
seat...
	Anyway, STV is much nicer in terms of a politics of individuals rather
than only parties. Using STV, nobody has to run with a party, and also it
is simple for voters to vote across party lines, which is often difficult
or impossible using party list systems.

	Okay, now let me try to get to the question that you actually asked,
namely, why is any other system other than STV worth the added complexity?
	First, let me point out that there are several forms of STV (as there are
several forms of party list PR). And I don't just mean one main form and
the rest sort of newfangled branches off of it. STV isn't, properly
speaking, a specific voting system in itself, but rather a family of
voting systems that share common principles. Okay, now I'm lifting words
directly from Nic Tideman. So, I would encourage you again to have a look
at that paper, because he goes into the various forms of STV over the last
century and a half, and the reasoning behind the changes from previous
systems. The address again is www.econ.vt.edu/tideman/rmt.pdf 
	Anyway, I will talk about my own reasoning for thinking that CPO-STV
might be a serious improvement over Newland-Britain (fractional transfer)
STV. (Which I guess can be assumed to be the standard for purposes here,
although in actual U.S. city elections (such as Cambridge, I think...),
randomized versions of STV have usually been used.)
	
	What are the flaws of STV? Yes, when reduced to single seat elections,
STV is precisely IRV, or "The Alternative Vote" (AV). If you don't see any
advantage of Condorcet over IRV, then you will almost certainly not prefer
CPO-STV to Newland-Britain STV. Basically, the arguments for CPO-STV are
the same arguments as for Condorcet, although perhaps on a more subtle or
small-scale level. 
	In single seat elections (IRV), the somewhat erratic nature of the
elimination order can violently throw the entire outcome to one side or
another of the actual median / majority. However, using STV with a
substantially larger district magnitude, I do not think that the
elimination order troubles have such a global effect. That is, the overall
outcome will, I think, be fairly well proportional to the voters, roughly
speaking, of course. I don't claim to be really sure about this point, and
if someone can argue me out of it, okay. 
	But what is fairly clear to me is that, if the elimination order doesn't
produce crazy distortions in the overall outcome, it can still produce
IRV-type distortions in races for individual seats.

	With IRV we have the problem of the "second-order spoiler effect", that
is a candidate X (who could be a Condorcet winner) is foiled by similar
candidate Y who enters the race, has more top-choice votes at some point
or another, causing X to get eliminated, and then Y proceeds to get
eliminated as well, throwing the race to someone else altogether who had
less support. The classic example of this for me has always looked
something like this:
	
41:Bush
4: Bush, Gore
	(45 total Bush first)
4: Gore, Bush
5: Gore
17: Gore, Nader
	(26 total Gore first)
25: Nader, Gore
4: Nader
	(29 total Nader first)

	Anyway, this dynamic provides a kind of incentive for anyone who
seriously dislikes Bush to dutifully vote for Gore first even if they like
Nader better.
	Using STV, I can imagine situations where very similar things happen in
races for single seats. For example, lets say that in such and such a
multi-member district the Green party has just barely enough support for
one seat. Okay, great. But lets also say that some Greens are a little bit
dissatisfied with the incumbent Green. They'd still prefer him to any
non-Green running, but not everything is quite right with him. Maybe he
has gone waffly on one or two key issues, or something. Maybe there are a
couple other people who are just a little bit brighter, or nicer, or
listen better, or have better plans. Should any of these candidates run to
challenge him, and if they do, should voters who sincerely prefer them
rank them first?
	In this sort of case we basically have the same sorts of pitfalls as in
IRV. Maybe one other candidate will run, and get more first choice votes,
causing the incumbent to get eliminated, but the challenger will get
eliminated too later on. Why? Maybe because he didn't get as many
transferred surplus votes or transferred votes from non-Green eliminated
candidates as the incumbent would have. Maybe because some of the voters
for the incumbent Green were stubborn and didn't rank the challenger Green.
	Or maybe two or three challengers could run, and one could be the
Condorcet winner among the Green electorate but get eliminated due to low
top-choice votes at some point, still giving a seat to a Green, but not to
the one who was most preferred overall.
	Or, alternately, let's say that the Mexican population of a district
generally vote together, and have just enough for one seat. Let's there
are three candidates roughly similar to my Bush Gore Nader scenario, and
the Mexican Gore candidate gets eliminated early on throwing the race to
the Mexican Bush candidate despite the fact that Mexican Gore would have
beat him 51-45 in a pairwise competition.

	Anyway, the point is that there is just a sort of fuzziness, imprecision
at the edges of the system. The method is not entirely precise, there is
still a good amount of chance, and some pretty real incentives to keep too
many similar candidates from running for the same seat.
	My own meager experiments with CPO-STV shows that it produces good
Condorcet-efficient results in these little one-seat subelectorates, in
the midst of a larger election. So I am thinking that it might really be
the best system for really precise proportional representation, the one
which offers the most responsiveness, highest standards, etc. 
	If so, then this is a pretty big deal, in my opinion. It's a pretty high
distinction for something to be the best proportional system in existence,
even if no governments use it. 
	Also, if you have computers that can handle it, and voters that can trust
it, then why settle for a system that is known to be less precise when you
are making really important political decisions? Why not go for the system
that allows for more competition, higher standards, less insincere voting,
and greater voter enthusiasm those who are actually elected.
	I think of CPO-STV as an attempt to take the good principles of STV to
their logical conclusion, to be what STV has been meant to be all along.
	Does it succeed in this? Once again, it is hard for me to test such a
complex system, but I would be very interested if someone has an idea of
how to evaluate it.
	
	To be honest, I'd be pretty pleased if this country started going over to
Newland-Britain or random STV within my lifetime. I agree that CPO-STV is
a little much to ask for now, when we are still messing around with
plurality, the electoral college, and all of that. 
	But, in the very long run, in the utopian future that I imagine we might
reach if we someday stop fucking ourselves up so badly, I think that
CPO-STV will be a jolly good idea then. And also, some other, more
openminded organizations than the federal government might be interested
in such things in the meantime.
	After all, there are two questions, right? Theory and practice, what
would be best and what can be achieved. Personally, I am interested in
both of these questions. Even if the US will never ever use Condorcet's
method to elect the president, the fundamental question of "what is the
best way to amalgamate conflicting preferences into a single decision" is
still fascinating to me in its own right, because it touches on the
question of what democracy really means, or could mean, and to what extent
the ideals of democracy are possible. Which in turn touches on the idea of
whether humans will ever be capable of making really intelligent group
decisions, or whether we will be doomed to eternal public stupidity.
	For good or ill, I have never really been good at keeping my thoughts
within the boundaries of what is practical in my own particular historical
situation. Yet I imagine that sometimes it is possible to find things
outside that boundary that have relevance for things inside. 
	Political considerations aside, it is still interesting to me: Is CPO-STV
in fact the superior method of proportional representation? I'm not sure,
what do you think?
	

John Hodges also wrote:
>maker of voting 
>machines in the U.S. was selling machines that (a) left no paper 
>trail (b) were vulnerable to hacking. I think it's important that a 
>"hand count" remain possible; so the voting method must be one that 
>can be done by hand if necessary, both the counting and the 
>calculating. 

I (James) reply:
	Hmm. Well. Counting for CPO-STV can theoretically be done by hand.
Actually, the counting process would be the same as any other rank ballot
method.(That is, counting the raw data of number of voters for each
expressed preference ranking.)
	As for calculating the results by hand, there is no way, unless it is a
very small election.
	I wouldn't much want to do a Newland-Britain STV computation by hand
either, though. The randomized version of STV can be calculated by hand,
although it is a pain.
	I think that with any ranked system, the election commission should make
the raw vote publicly known in addition to the results of the final
computation. Whether it is IRV, Condorcet, or STV, it is really important
for people to have access to this information so they can analyse and
understand it in different ways, rather than just having the result fed to
them and being otherwise in the dark.
	If you do this, then people can take the raw information (which can be
hand-counted if necessary), feed it into their own computers, and see if
it comes out right. Personally, I think that this is sufficient to keep
someone from getting away with hacking into the system and producing a
fake result.