[EM] article on preferential voting

[Abd ul-Rahman Lomax]

At 09:11 AM 12/30/2006, jure.toplak at uni-mb.si wrote:
>I have prepared a paper on the definition of preferential voting. I
>invite all of you to check it. It is posted here:
>http://rcum.uni-mb.si/~jure/preferential8_emlist.pdf

Well, it is a bit irritating to me, and perhaps to some others, that 
AV is used in the paper to refer to Alternative Vote; we routinely 
use it for Approval Voting.

Approval is mentioned in the paper, however, in my view, incorrectly 
(though perhaps merely reporting common opinion, in which case it is 
correct about an incorrect opinion.)

Specifically,

>They apply the term to "those voting methods that make use of
>a preference schedule"10 including plurality first-past-the-post 
>(FPTP) method, Concordet,
>and Borda count. The definition does not however cover approval voting.

Now, if "plurality" and FPTP use a "preference schedule," so does 
Approval. The preference schedule in FPTP consists of two classes, 
preferred and not-preferred, and the voter places the candidates into 
the classes by either voting for the candidate or abstaining. 
Generally, FPTP does not allow a voter to place more than one 
candidate into the approved class, but this is actually a detail of 
implementation having to do with ballot invalidation. That is, a 
ballot-invalidation rule is tacked onto the voting system that 
discards ballots who have approved more than the officially allowed 
number of candidates, which is generally one with single-winner and N 
with N-winner. This is an artificial restriction without, apparently, 
much explicit justification.

If, for example, a show-of-hands election were being held in a 
face-to-face meeting, there would generally be no bar to voters 
raising their hand for more than one candidate, and I am sure that it 
happens from time to time. It might even happen more than 
occasionally, we would not normally know. Enough members at a meeting 
may abstain to compensate for the "extra" votes which appear in the 
reported counts. Only by specifically observing individual voters and 
objecting if they vote more than once would we become aware of 
"overvoting." It has never become an issue as far as I know, and the 
basic reason is that there is utterly no harm in it.

Overvoting is essentially voting for every member of the approved 
class over every non-member, with more than one member allowed in the 
approved class. It is also equivalent to abstaining in the pairwise 
vote between those approved members. Many systems with more than two 
ranks allow voters to place more than one candidate at the same rank. 
This is identical to Approval Voting, the only issue is how many 
ranks are allowed. With Approval we assume there are two.

I've proposed a system called A+ where there are three ranks. The 
Approved rank is split into two ranks: Approved and Preferred. One 
may place more than one candidate into each of these ranks; however, 
I assume that normally voters would only place one candidate into 
Preferred, with any other approved candidates being placed into 
Approved. In simple A+, the Preferred rank is considered merged with 
the Approved rank, it is not used to determine the winner. However, 
it may be used, for example, to determine ballot access rights and 
public campaign funding, as well as to generate accurate poll 
information, perhaps to monitor the performance of Approval vs the 
possibility of moving to a system with more detailed ranks. It's a 
small cost in ballot complexity for a substantial gain in gathered 
information, and it answers the common objection to Approval that 
voters want to express support for their favorite while still being 
able to approve other candidates.

And then the same ballot is used for what I called A+/PW, for 
Approval Plus/Pairwise, where, once again, Approvaed and Preferred 
are considered identical in any pairwise race except for the pairwise 
race(s) between Approved and Preferred candidates. This, of course, 
is a Condorcet method with truncated ranks....

One clear advantage of Approval is that the ballot indicates Yes/No 
for each candidate. The meaning, thus, of majority failure is clear, 
the electorate has voted *against* all candidates, and, if sincere in 
this, has essentially indicated that they would prefer to see the 
office remain vacant than allow any of the candidates to be elected. 
Under Roberts Rules, with the sequential motion method of electing 
officers, which is the default, this is, indeed, exactly the decision 
that voters are making when they vote for or against each motion to 
elect So-and-So. (Of course, the problem with sequential motion is 
that one may imagine that voters will vote against a candidate, not 
because they are unwilling to accept the candidate, but because they 
prefer another. However, I would ask why these voters did not move to 
amend the motion to substitute their favorite for the one previously 
named in it? Where such amendment is allowed, which I'd think it 
would normally be, a motion to elect being like any other standard 
motion -- at least it should be -- then election by specific motion 
is fully Condorcet compatible. I think most of us have missed that.

The basic method of electing officers under Roberts Rules is 
Condorcet-compatible. Substituting methods which are not for this, 
based on the presumed inefficiency and impracticality of using 
deliberative process for elections, would seem to be a loss. Now, are 
there any methods which, as purely aggregative methods, simulate the 
deliberative process?

Yes, there are.

Approval does it. It is a standard rule where multiple conflicting 
motions are presented for simultaneous vote, that the motion with the 
largest Yes vote prevails over passing motions with a lesser level of 
approval. This is, precisely, Approval Voting. The common statement 
that Approval is not used in public elections (with a couple of 
exceptions sometimes being mentioned, such as the election of the 
Secretary General of the United Nations) is not correct. It is used, 
but not to elect officers, rather to simultaneously consider 
conflicting referenda.

And that is what an election should be, a series of motions to elect 
A, B, C, D, etc, with Yes/No votes possible. The only problem, and it 
is a severe loss, is that the rules typically prevent voters from 
voting for more than one. It is beyond me why. The assumption seems 
to be, where a reason is given or implied at all, that such votes are 
clearly mistakes, voter error, and thus, as errors, should be disregarded.

But, quite often, they are not errors, except in the sense that the 
voter did not realize that the arbitrary no-overvote rule would 
apply. A naive voter -- and many would be naive in this way -- may 
think that the votes will simply be counted. Why not?

I have been asking for some time for cogent reasons that overvotes 
should cause ballots to be discarded. What I've gotten so far, as I 
recall, is this:

(1) If overvotes are allowed, fraud become easier. I could take a 
vote for A and add a vote for B by simply making the additional mark. 
I find this argument puzzling. For I can invalidate, with the 
no-overvote rule, ballots by doing just this, thus committing fraud. 
The no-overvote rule does not prevent fraud, and, indeed, it could be 
shown that it makes fraud simpler; together with other rules that are 
essentially aimed at preventing voters from revealing their identity, 
it makes it very easy to damage and cause ballots to be discarded. 
Generally, just make a mark on the ballot. In order to prevent voters 
from identifying their vote, perhaps to interdict those who would 
coerce or buy votes from verifying compliance, we risk -- and 
probably actually enable -- other kinds of election fraud. Vote 
buying does happen, but it is quite hazardous, even charges of 
vote-buying, coupled with sufficient evidence that something that may 
have made vote-buying possible, without proof that it actually took 
place, have resulted in the overturn of an election and the 
declaration of the former loser as the winner. Buying votes can ruin 
your election, even if you would have won without doing it, and 
buying enough votes to really make a difference is difficult.

(Suarez vs. Carollo, Miami mayor race, I think it was about 1997. 
Some supporters of Suarez transported voters to a 
registration/absentee voting location and offered to pay them small 
sums. With this was made a suggested vote, but there was no 
allegation of coercion, nor, indeed that they were actually being 
paid to vote a certain way, nor even that the votes were monitored. 
There was however, evidence that some of these registrations were 
fraudulent, which was not necessarily the fault of the Suarez 
supporters. Because the total of absentee ballots resulted in Carollo 
not gaining a majority, thus automatically causing a runoff election 
to be held -- which Suarez won -- the court ultimately decided to 
discard *all* absentee ballots. Given that Suarez -- who was not 
implicated in the alleged vote-buying -- won the runoff even if 
absentee ballots were discarded, it is pretty likely that Suarez 
legitimately won the election if only fraudulent ballots could have 
been identified and discarded. But because there was no way to 
discriminate between fraudulent and non-fraudulent ballots, and the 
court ruled that absentee voting was a privilege and not a right, it 
decided to discard *all* absentee votes, thus systematically 
disenfranchising the population that must vote in that way. Which is 
quite likely not to be identical in preference to the direct voting 
population. Generally, vote-buying is illegal, and to have much 
effect, it must be a crime that is very difficult to conceal.)