[EM] IRV vs Condorcet Voting Methods

[Dave Ketchum]

These two voting methods inspire a lot of hot words, so I am scattering my 
2-cents worth on many lists.

First, some incidentals:
       Plurality voting is the one many of us want to escape from - being
able to select only one candidate is not satisfactory.
       Approval voting - some argue for this, saying existing voting
machines can be set up for this.  While Approval allows indicating more
than one candidate as being acceptable, it fails to let the voter indicate
preference among them.
       PR (proportional representation) can be used to elect legislatures
and similar boards - not discussed since this post is concerned only with
single-seat voting, as is needed for mayors and governors, and can be used
for legislatures if the laws/rules specify this rather than PR.

Voting machines capable of supporting more complex methods.  Most
jurisdictions need newer machines than they have been using, and capable
machines should cost little more than their cousins with less capability. 
I argue STRENUOUSLY for two features:
       They SHALL NOT be connected to external signals such as the
internet.  Being connected during voting means secrecy could be destroyed
- and this could easily be suspected by voters even if it was not done.
Being connected during setup means you are not certain you have a true
copy of the setup information.
       They SHALL NOT have any secrets.  Those tempted to corruption like
and need secrets.  Builders like secrets, but do not believably have any
valuable ones for a task as simple as controlling voting via computer (if
they really have something to protect, use copyright or other method to 
protect the builders).  Those buying should be required to evaluate the 
programs for suitability; citizen groups should be allowed to verify via 
their own evaluation.

Complex and exotic methods can be dreamed up and promoted by those into
debating methods - most should be rejected for difficulty in explaining to
and using by average voters,
---------------------

IRV and Condorcet - the purpose of this post.  For these, ballots are
identical and, in most cases, both identify the same candidates as
winners.  Because of the votes that result in different winners, I argue
for Condorcet.

For both of these each voter, in preparing their ballot, can choose among:
       Vote for just one, as in Plurality.  Certainly suitable to vote for
just A or B if one of them is your choice and one of them is expected to win.
       Vote their true preference followed by least of evils, as in Z, B
or Z, A.  I see this as satisfying the average voter.
       Vote for more, up to all candidates, in order of their preference.
Certainly permitted, and sometimes useful - perhaps A, B, and C are
believable winners and you LIKE Z and DISLIKE C - so vote Z, A, B or Z, B, A.

Condorcet looks at whole ballots, and considers  which is best liked in
each possible pair of candidates.  Thus, for the pair A & B, if a voter 
lists A before B, that counts for A; if a voter lists only A or B, that 
counts for listed candidate.  In doing this, Condorcet considers only the 
pair being evaluated and their relative position, but not their position 
relative to any other candidates.
       Usually one candidate will be declared winner of the race for 
winning when paired with each other candidate.
       When approaching a tie, Condorcet can have cycles of 3 or more
candidates, as in A>B, B>C, and C>A.  Here a formula must identify the
strongest pair and award winner accordingly.

IRV looks only at each voter's first choice.  Candidate with least first
choice votes loses and ballots for this candidate are edited to exclude
reference to loser.  Repeat until one candidate has a majority of
remaining ballots (when and if all of the votes on a ballot are edited
out, that ballot cannot take part in remaining processing).
       Usually the best liked candidate will be declared winner as less
liked candidates are excluded as losers.
       When approaching a tie IRV, by only looking at first choices, can
reject candidates that are truly best liked.  Try the following ballots:
       40 CC - these conservative voters agree on a candidate.
       21 LM, LC - these liberal voters LIKE LM's money campaign.
       20 LW, LC - these liberal voters LIKE LW's wages campaign.
       19 LC - these liberal voters care only for LC's common approach.

Here IRV will first discard the 19 LC-only votes as losers.
        Then the 20 LW, LC become LC-only - and lose.
        Then the 40 CC win over the 21 LM, LC.
        With the same ballots, Condorcet would have seen the 60 LC as
winner over the 40 CC.

So, IRV does not have the same spoiler problem as Plurality, but it has
enough of a spoiler problem that I keep promoting Condorcet.

Some would do IRV because it has many vocal backers.  I argue against
following the vocal group into, sooner or later, getting caught backing a
failing method - better to work a little harder and getting it right.

Another argument against IRV is instability.  Look at the above example
and think of a couple absentee votes for LC turning up.  Then LW discards
first, leaving 41 LC; followed by discarding LM to leave 62 LC - correctly
labelled winner.  Condorcet can change winner per absentee ballots, but
only if it was already close to a tie, apparent from the vote counts.

Then there are strategy temptations - in the IRV world voters can be
tempted to vote to try to beat the system, rather than per their true 
preferences - Condorcet offers very little temptation of this sort.
-- 
     davek at clarityconnect.com    http://www.clarityconnect.com/webpages3/davek
      Dave Ketchum    108 Halstead Ave, Owego, NY  13827-1708    607-687-5026
                 Do to no one what you would not want done to you.
                       If you want peace, work for justice.


----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em