James Gilmour wrote:
> Bart Ingles wrote:
> > James Gilmour wrote:
> > >
> > > [...] Publishing "results"
> > > precinct by precinct is just totally irrelevant when all that matters is the
> > > city-wide totals. It is not a question of keeping them secret. Rather the
> > > question is why on earth would you want to publish such irrelevant
> > information?
> >
> > Freedom of information, public accountability, etc. One example of
> > where it was used was Florida in 2000-2001, where several news
> > organizations were able to examine the presidential ballots and
> > eventually discount claims that a selective recount would have changed
> > the outcome.
>
> I appreciate that US and UK practice are very different in these matters, but I
> still don't see how publishing irrelevant information enhances public
> accountability, etc.
In the US, I don't think it really matters why an organization would
want access to such information. The burden would be on the government
to show that the access would be harmful to society in some way, such as
damage to national security or violating someone's right to privacy.
And once such access is granted, it would be very difficult to prevent
publication due to the 1st Amendment.
> With regard to freedom of information, it is illegal in the UK to publish any
> information that could have been obtained ONLY from the inspection of a ballot
> paper. So when we comment on the voters' preferences in an STV-PR (Choice
> multiple seat) election, we are careful always to refer to the patterns of
> transfers that can be seen in the published Results Sheet, even if we have been
> privileged observers in the counting hall and have seen the actual patterns of
> preferences on many ballot papers.
Still, that may be enough to show a violation of consistency (the
original subject of this thread), or of related criteria, depending on
what is contained in the Results Sheet.
Regardless, my original intent was not to argue in favor of
post-election ballot availability, but to show that there could be
situations where an electorate is effectively partitioned, thereby
allowing a consistency violation to surface. Since you have already
agreed that a recount could be one such situation, I consider my
original point made.
> > > > In return,
> > > > approval ballots contain information not present in ranked ballots,
> > > > namely an indication of the voters' strength of preference.
> > >
> > > I don't buy that. In Approval each voter just sorts the candidates into two
> > > sets - acceptable and not acceptable. That seems to me to be LESS information
> > > than on a typical ranked ballot.
> >
> > I didn't say that approval ballots necessarily contained more
> > information than ranked ballots, I said that approval ballots contained
> > information not present in ranked ballots, and that the information on
> > an approval ballot is less prone to being used in a haphazard way.
>
> I don't want to play semantics, but surely if (as you said) "approval ballots
> contain information not present in ranked ballots", the ranked ballots must
> contain less information than the approval ballots? Or do you mean that they
> contain similar amounts of information, but that it is different? (That it is
> different is, of course, self-evident.)
If you have a box of apples and a box of tangerines, then the apple
crate contains something not present in the tangerine crate (and
vice-versa). This says nothing about the amount of produce contained in
either box-- you may have 90 kg of apples and 100 kg of tangerines, and
the statement would still be true. The comparison might be difficult in
any case-- the tangerines might be 40 percent rind, and the apples 10
percent core-- are we interested in the weight of edible fruit? Total
weight? Volume? Caloric content?
I don't think it's possible to say which of ranked and approval ballots
contain the most information. But in my view, the approval ballots
contain the more *relevant* information.
> I fail to see how the information a ranked ballot will be used in a more haphazard
> (random) way. No matter how many times ballot papers are recounted under the IRV
> rules I have used for many years, you will always obtain the same result. So
> there should be nothing haphazard about it.
I didn't say "haphazard (random)." My intended meaning was more
"haphazard (arbitrary)." IRV's elimination rule is quite arbitrary,
even if it is reproducible given the same set of ballots.
> > With ranked ballots, there is no way to determine who is acceptable or
> > not acceptable, except for the first and last preferences.
>
> Maybe, but that is not the purpose of the election. The purpose of a single seat
> election is to identify the one candidate who is most representative of those who
> vote. That seems to me to be quite different from sorting the candidates into
> "acceptable" and "not acceptable". And before anyone rushes back - yes, I do
> recognise the defects in IRV in achieving that objective.
Nobody said that the purpose of the election was to sort candidates in
to "acceptable" and "not acceptable", any more than it's the purpose of
an election to sort candidate into a strict rank order.
As for choosing the single candidate most representative of the
electorate, Approval has been shown in simulations to do this better
than IRV by a couple of objective measures. In terms of normalized
("social") utility, Approval beats IRV easily. In terms of choosing a
Condorcet winner, the two are more competitive, but Approval appears
to give the more consistent results. IRV fails both measures badly
whenever there are many candidates, or if the candidates are "bunched"
ideologically (e.g. a crowded center).
> > > If you really want information about "strength
> > > of preference" you will have to introduce some system that allows
> > each voter to
> > > weight his or her preferences as they wish. Then you must normalise those
> > > weightings if you want to ensure that each voter has one vote and
> > only one vote.
> > > And of course, in normalising the weights, you will throw away a
> > significant part
> > > of the information about the differences in the strengths of
> > preference BETWEEN
> > > voters.
> >
> > Only if you insist on finer granularity than acceptable/not acceptable.
>
> Sorry, but to me "acceptable/not acceptable" says nothing about the strengths of
> the preferences, and it was you who claimed "strength of preference" information
> was present in approval ballots. In my analysis there is no information about
> "strengths of preferences" in either Approval or IRV ballots. In Approval, the
> voter sorts the candidates into two sets - we know nothing about the 'distance'
> between those two sets. In IRV, the voter ranks the candidates (so far as he or
> she wishes) so we may know more about their relative positions - but again we know
> nothing about the 'distances' between successive preferences. (This is a
> fascinating subject, but it far removed from any voting system that we could
> implement in practice.
I generally take it for granted that "strength of preference" is
relative to an individual voter's range of preferences. The
alternative, absolute utilities, would violate even the more reasonable
interpretation of 1P1V.
With IRV, an individual's close rankings count just as much as his major
preferences. There is no assurance that a voter's second choice is
preferred over his last choice by a non-trivial amount, or even
preferred at all if the voter dislikes both equally and ranks them at
random.
With Approval, because there is a strategy cost associated with
approving candidates beside one's favorite, a voter will only approve of
an additional candidate if the preference for that additional candidate
over a less-favorite is strong enough to be worth the strategy cost.
> For the moment I'll be happy to achieve reform that will
> deal with the major defects in our voting systems and I'll leave it to others,
> long after me, to worry about differential strengths of preferences.)
I'm not sure what you consider major defects. If you are strongly in
favor of a two-party system, and want to make sure that additional
parties have no inroad into executive offices, then I suppose IRV deals
with a major defect.
To me the major defect is a vote-splitting problem which motivates
political bigwigs to decide elections ahead of time, by only supporting
candidates whom they think are most likely to defeat the major opposing
party. IRV does not deal with this defect, or does so only partially.
> > And I don't buy your interpretation of "one and only one vote"-- I don't
> > believe the phrase was ever intended to apply to voting methods other
> > than first-past-the-post, in the context of people actually voting
> > multiple times for the same candidate.
>
> This is a convenient interpretation which appears to be favoured by many advocates
> of Approval Voting. The definition I have long been familiar with is not
> conveniently truncated in this way.
How is that interpretation of 1P1V more "convenient" than yours? Do you
have a reference to a historic definition?
> > > > In computer models conducted by Merrill and others, approval voting
> > > > produced results more in line with Condorcet's method than did IRV,
> > > > especially when there are many candidates.
> > >
> > > Maybe, but that does not remove the serious defect in Approval. One
> > person, one
> > > vote is violated.
> >
> > If that's true, it's also violated in every other method except
> > Plurality. It's especially bad in IRV, where one voter may have votes
> > counted for several candidates, while another has only one choice
> > counted.
>
> There may be some problem with the rules you've seen for counting IRV ballots, but
> in the rules I've used for many years, it is just impossible for any voter to have
> his or her vote counted for several candidates and so secure more than one vote.
> With IRV, the voter has one vote. That vote is transferable, but the whole of
> that one vote counts towards the election of only one candidate. If the voter's
> most preferred candidate ceases to be a candidate, because he or she is excluded,
> then the one vote is transferred to one other candidate and counts exclusively
> towards the election of that one candidate. I think this demonstrates "one
> person, one vote" throughout the entire process.
Suppose I go to my insurance broker with the following proposition: I
would like to conditionally purchase x dollars worth of health
insurance. If I don't get sick, I would like to have the policy
cancelled and my dollars "transferred" to an auto insurance policy for
the following year. In the event that I have no auto accidents next
year, my dollars should be transferred to an annuity. Do you think my
insurance broker is likely to accept the deal?
By "transferring" my investment, I am attempting to spend dollars which
have already been spent. Likewise, a transferred vote is one which has
already been cast and counted. To say that the vote no longer exists
after the candidate loses is to redefine history.
Nevertheless, if you accept this tortured definition of "one vote", then
Approval meets it as well if you use the the following vote-counting
procedure:
1. Choose two candidates for the first round of counting.
2. Count all votes for the two candidates. If a given ballot contains
a vote for both of these candidates, disqualify the ballot from this
round as an "overvote". Thus the vote total for this round will reflect
only voters who cast a vote for exactly one of the two candidates, and
1P1V is preserved for the round.
3. Exclude this round's losing candidate, and reinstate any overvoted
ballots (since one of the two votes responsible for the overvote has now
been nullified).
4. Choose another candidate to be counted, and repeat step 2 to choose
between the winner of the previous round and the new candidate. If
there are no more candidates, then the most recent winner is the
Approval winner.
Bart
----
For more information about this list (subscribe, unsubscribe, FAQ, etc),
please see http://www.eskimo.com/~robla/em
