[EM] IRV in Burlington VT

[Abd ul-Rahman Lomax]

At 12:20 PM 1/10/2010, Dave Ketchum wrote:
>IRV is of interest ONLY to those considering using that - hopefully no
>one.
>
>Condorcet would be of interest.  If ballot format permitted assigning
>multiple ranks to any candidate, educating its voters as to this
>problem would be of interest.  Otherwise, how could Condorcet be
>expected to have more trouble than Range?

There isn't necessarily a problem. In my view, if voter intention 
isn't clear, and considering the various configurations that can make 
it unclear, the interpretation for each configuration that is *most 
likely* to fairly represent voter intention should be followed. That 
could, in fact, be a complex algorithm, but it would be important for 
it to be determined in advance, for it to be part of the rules, or it 
could and would be gamed and contested when it matters.

Example: 2000, overvotes for Gore and Buchanan. Clearly the voter 
intended to vote either for Gore or Buchanan or both. (Some voters 
*do* apparently intend to vote for more than one, misunderstanding 
the rules, but that particular combination isn't terribly likely!)

In my view, we should count the votes as writ. Thus both Gore and 
Buchanan would get one vote. The ultimate effect of this is that 
either one vote actually counts to create a winner, or none. This 
rule would, of course, implement Approval Voting. Notice, also, that 
if the voter actually intended to vote for only one of the two, the 
effect of counting both is to make the ballot moot. Yet the vote 
still counts as to any contest in which the top two are either Gore 
or Buchanan.

But, it may be argued, what if the voter actually intended to vote 
for Buchanan, and would be distressed by the vote counting for Gore over Bush?

Sure. That's possible. However, it's unlikely. How unlikely? Well, 
Buchanan was not a widely-supported candidate. Even neglecting the 
butterfly ballot problem, it's much more likely that the voter 
supported Gore than that the voter supported Buchanan. I.e., if one 
vote was a mistake, probably it was the Buchanan vote, just on 
general principles. In fact, we know that these overvotes were due to 
poor ballot design. So, it's possible, the vote counts might be 
modified by the vote percentages on the fully-clear ballots. 
Constitutionally, though, it would be simpler to just count them all, 
letting the voters be responsible for how they actually voted.

A few voters might intend to invalidate their ballot. They made a 
mistake, they don't have time to go back and get another ballot, so 
they add an overvote. But they would need to know a better way to do 
this: vote for all the candidates, or all but the most-disliked 
candidates, depends on the nature of the original error. Or write 
VOID on the ballot, or as a write-in. Lots of ways to do it....

Count All the Votes. Simple concept, eh?

With ranked choice voting other more complex ballots, such as Range 
ballots, new kinds of errors become possible, but there is normally 
some reasonable interpretation that can be applied that preserves 
*likely* voter intention.

I've described one, in Bucklin. What if the voting equipment (such as 
a plain paper ballot and a pencil) permits a voter to rank a 
candidate in more than one rank? Obviously, if we count more than one 
of these votes, we will be giving the voter unfair power. We must 
discard all but one of them. Should we discount all? I'd claim not, 
for then we would be neglecting what is clearly support for a 
candidate. There are two obviously possible approaches:

Count the vote at the highest rank found.

Count the vote at the lowest rank found.

In some cases it might seem necessary to count at the highest: the 
most obvious is where the voter doesn't vote for any other candidate, 
and the method allows overvoting. If it allows overvoting, it's 
possible that the voter's vote could be decisive or prevent a 
decision in the first round. Clearly the voter has a preference and 
has expressed it. The vote should be counted in the first round.

If the voter has voted for other candidates at the same rank, we 
can't tell which one the voter actually preferred, if any, so the 
ballot should become, as to the contest between those candidates, 
moot, best expressed by counting the vote for both, thus abstaining 
from that pairwise election but participating in all the rest.

Another approach would be to count these votes separately, and add 
them in, in an intermediate process. So if the voter votes for a 
candidate in first and second rank, the first rank votes would be 
counted without such votes, and if no victor appeared, then the 
additional votes would be added in as if it were an intermediate 
rank. The same could, in fact, be done with "overvotes," i.e., voters 
who vote for more than one candidate, but I'd greatly prefer, I 
suspect, not to treat those "overvotes" as anything other than 
ordinary votes, expressions of support for more than one candidate 
with preference not expressed between them.

There are more complex algorithms possible, but, hey, I'd be happy to 
simple get a Count All the Votes principle across. Approval, Range, 
and Bucklin or other hybrids.