[EM] IRV vs Plurality (Dave Ketchum)

[Terry Bouricius]

Abd has repeated an erroneous statement about Nicolaus Tideman's 
assessment of voting methods. Abd wrote:

<snip>"FairVote has really poisoned the air, citing Tideman, for example, 
when criticising other methods but never mentioning that Tideman considers 
IRV unacceptable." <snip>

Tideman does NOT consider IRV unacceptable (If you ask him, I believe he 
will say he favors it in real world implementations). To set the record 
straight, I will paste the rebuttal I sent to this list when Warren Smith 
made the same mistake last year.

Terry Bouricius


On page 238 [in his book _Collective Decisions and Voting"_]Tideman has a
chart with five categories summarizing his analysis of methods...
First is "Not supportable" which includes
Borda, Range, Dodgson, Copeland, Coombs and Est. centrality.

The next category is "Arguably inferior to maxmin" which includes
Condorcet, Simp. Dodgson, Nanson, Bucklin, Black, Young, and Wt. 
Condorcet.

The third category is "Supportable if ranking is infeasible" which 
includes
Plurality, Approval, and Two-ballot majority.

The fourth category is "Supportable if a matrix is uncalculable" which 
includes
only Alternative vote [IRV]

The last category is "Supportable if a matrix of majorities is calculable" 
which includes
Maxmin, Ranked Pairs, Schulze, Alt. Schwartz and Alt. Smith.

Warren [now Abd] is assuming that a matrix is always "calculable" and thus 
the
supportable category that includes only IRV is in fact null. However, that
is not what Tideman is arguing (or why would he create the category if it
was always empty)? Elsewhere he discusses the practical limitations of
voting methods used for public elections including ease of voter
acceptance and argues that a hypothetical improvement of a system that
requires complexities such as matrices may be impractical in large scale
elections. He writes on page 240 "If it is feasible to require voters to
rank options, then much more sophisticated processing is possible.
However, it is conceivable that it would be feasible to require voters to
rank options but not feasible to require vote-processors to produce a
matrix of majorities. In this event the Alternative vote is supportable."


----- Original Message ----- 
From: "Abd ul-Rahman Lomax" <abd at lomaxdesign.com>
To: "robert bristow-johnson" <rbj at audioimagination.com>; "EM Methods" 
<election-methods at lists.electorama.com>
Sent: Wednesday, January 20, 2010 11:54 AM
Subject: Re: [EM] IRV vs Plurality (Dave Ketchum)


At 02:31 PM 1/16/2010, robert bristow-johnson wrote:
>and FairVote.org will point to experts that strongly advocate IRV.
>big fat hairy deel.

They will. Anecdotal. Look, a lot of experts follow the EM list. Have
been for a long time. Very few who aren't politically committed in
some way, such as Terry Bouricius, who is actually more of a
politician than an expert on voting systems, will defend IRV here. I
can only think of Chris Benham, who is from Australia, and we all
know what that means. He's upside-down.

(Seriously, Chris is quite knowledgeable, but he's pretty isolated.
IRV has been considered unacceptable by experts since the 19th
century, I found criticism from political scientists from that
century. It was not adopted in Australia because it was an ideal
system, it was adopted because it saved the day for a party in power,
eliminating the spoiler effect, defanging minor parties and thus
eventually killing them.)

>whether it's here or on the Burlington blog or longer ago at the
>Fairvote site (that i have since gotten tired of), i have never
>appealed to authority in evaluating or advocating any method.

It's just a fact, and it's relevant, whether you appeal to it or not.
FairVote has really poisoned the air, citing Tideman, for example,
when criticising other methods but never mentioning that Tideman
considers IRV unacceptable.


>>I'm thinking over the opposition material. It might look rabid to
>>someone who isn't aware of the problems,
>
>but i think i *am* aware of the salient problems.

You are only aware of some of the problems, not enough to realize
that IRV may actually be worse than plurality in some situations.
Which happen to be most of the implementations, i.e., nonpartisan 
elections.

>   but despite that,
>i support the goals that we had in adopting IRV.  i still don't
>support "correcting" IRV by reverting back to FPTP, given that is the
>choice presented (the reverted rules would include a delayed-runoff
>for less than 40%).

The only problem with the 40% is that it's the wrong margin to be
looking for. Or it's inadequate. 40%/39% is very different, runoff
required, than 40%/30%, which is probably an unrecoverable margin.
Depends. In particular, it depends on data which isn't on the primary
ballot. Hence, make the primary ballot a preferential ballot, but
simply use a better method to analyze it.

Or, at least, allow additional approvals in the primary. Count All
the Votes. Look, Robert, if you can't get behind that slogan, you
will continue to irritate both the election integrity experts and the
voting system experts....

>   and "voting security" concerns have been
>persuasive.  even with IRV, with a reasonably small number of
>credible candidates (and assuming the worse case, that "write-in" is
>always the same person, without yet checking), there are a finite
>number of ranking permutations, and there can be a ballot for each.
>continued below...

I'm going to cut to the chase, here. Robert, you have failed to
understand what happens in the first counting. All votes in the first
rank must be counted and reported, period. Even write-in votes must
be counted, but they can generally be reported as a sum, and only if
the sum is large enough that it is all for one candidate, would the
instruction have to go back to the precinct to break down and
separately report the category. It goes back to all precincts
reporting any write-in votes! Normally, that would be all of them.

You cannot tell which votes should be reported, in advance, based
solely on local precinct data. It's quite possible that the winner
has no votes in the local precinct at all, is just unpopular there.
Are you saying that dark horses should be ignored, that write-in
candidates should be ignored?

So, consider Burlington. Three major candidates. How many minor on
the ballot? Three? That's six, plus write-in. You must report all
combinations, but you may collapse where there are empty ranks,
because, with the IRV counting methods, A>(blank)>B is equivalent to
A>B is equivalent to (blank)>A>B.

"Credible candidate," quite simply, must be ignored by the method.
Data for any combination of 7 candidates including Write-in, which is
lumped tentatively, must be tabulated into all of the relevant
patterns (after collapse of equivalent ones as described). So with
RCV, three-rank, you would have:

three candidate combinations:
7*6*5 = 210.

two candidate combinations:
7*6 = 42.

bullet votes or equivalent:
7

Total 258 piles. Now, go to San Francisco. 23 candidates on the
ballot, plus write-ins. I'm not even going to do the math. Simpler to
just transmit the raw ballot images, with computers. Or the counting
is done at each precinct, centrally directed, and it is far more
complex than precinct summable methods. Approval, very simple.
Bucklin all ranks can be counted and reported as sums. Range the
same. IRV. Arrggh. No. Real life elections in the U.S.? Major delays.
When an error is made, it can require going back and starting over!

All this for a method that produces, for all practical purposes,
exactly the same results as plurality in nonpartisan elections? You
have to be kidding. FairVote sold a pig in a poke. With the help of
people like you. In the beginning, it was quite understandable, and
the data to understand the uselessness of IRV in nonpartisan
elections didn't exist. I believe I discovered that in 2008, one
could look at Talk for Wikipedia articles related to instant runoff
voting. I never saw it asserted before. I didn't expect it. Did
anyone? Perhaps someone can find a prior reference.

Now, you mentioned in the remainder, what I deleted, late ballots.
IRV provides many opportunities for these late ballots to shift a
result. Basically, if you are going to use IRV, counting before all
the ballots are in is setting up a possibility or even probability of
extra work, which means extra expense. Actual IRV expense figures are
showing up and it isn't a pretty picture. IRV was oversold that way, as 
well.

With summable methods, late ballots can simply be counted and
reported and the results can be summed. It is not the only argument
against IRV, by a long shot, but it's important.

----
Election-Methods mailing list - see http://electorama.com/em for list info