Report from the Single-Winner Committee (draft)

Steve Eppley seppley at alumni.caltech.edu
Fri Aug 16 13:41:21 PDT 1996


The subscribers of election-methods-list at eskimo.com (a.k.a. the EM
list), after extensive discussion of voting methods which could be
used in single-winner elections, polled ourselves on the question:

  "Which single-winner voting method(s) should electoral reformers
   advocate?"

Seven subscribers posted responses.  Six ranked their recommendations
from most preferred to least preferred; the seventh listed only one
recommendation.  Some subscribers didn't post a response.

Below is a summary of the responses, followed by tallies of the
responses using several of the single-winner algorithms which were
themselves advocated by members of the EM list, followed by
definitions of the voting methods which were the subject of the 
poll.  An additional message will follow (soon, hopefully) containing 
commentary from any subscriber of EM who desires his/her comments to 
be included.

No matter how the responses were tallied, the top two recommendations 
are the same:
  1. Smith-Condorcet
  2. Condorcet
(See the definitions section at the bottom if you don't yet know how
these methods work.)  The recommendation order of the rest of the
methods depends on how the responses are tallied.

Though only seven people provided responses, the results may be
significant: the poll followed months of detailed discussion about
the pros and cons of many methods.  Most electoral reformers who have
learned a little about single-winner reforms know about and advocate
a method called Instant Runoff.  Few reformers know about Condorcet
or Smith-Condorcet, however, since "pairwise" methods haven't been
practical for large elections with many candidates until the computer 
age.  This report may serve as a wakeup call to reformers, since there 
are reasons why most of the respondents prefer Condorcet and
Smith-Condorcet more than Instant Runoff.  (Briefly, they support
majority rule and eliminate the "lesser of evils" dilemma better
than Instant Runoff.)

During the tallying a new method was identified and dubbed 
Instant-Runoff-1.  (See the definitions section for more info.)  
Two subscribers modified their responses to rank Instant-Runoff-1
more preferred than Instant Runoff (and more preferred than some
other methods, as well).  The subscribers who didn't modify their
responses made no comments about Instant-Runoff-1; their original
responses were tallied as if they ranked Instant-Runoff-1 last
(equally with any other methods they left unranked).

------------------------------------------------------------------
Message Contents:
  1. The seven responses
  2. Compact summary of responses
  3. Tallying of the responses by Condorcet, Smith-Condorcet,
       and Regular-Champion
  4. Tallying of the responses by Instant Runoff
  5. Glossary: Definitions of the single-winner voting methods


The seven responses:
Here are the responses, listed in alphabetical order by first name:

Demorep1 at aol.com, 7-8-96 22:11:04 -0700 (PDT)
  1   Demorep-1
  2   Demorep-2
  3   Instant Runoff

Donald Eric Davison, 7-4-96 11:01:33 -0700 (PDT)
  1   Instant Runoff

Hugh R. Tobin, 7-7-96 00:21:19 -0700 (PDT)
  1   Smith-Condorcet-Tobin
  2   Smith-Condorcet
  3   Smith-Random
  4   Instant Runoff
  5   Condorcet
  6   Regular-Champion
  7   Double Complement 
  8   Approval
  9   all other methods

Kevin Hornbuckle, 7-8-96 22:12:01
  1   Condorcet, Smith-Condorcet
  2   Approval
  3   Smith-Random

Mike Ossipoff, 5-26-96 23:41:56 -0700 (PDT)
  1   Condorcet
  2   Smith-Condorcet
  3   Instant-Runoff-1
  3   Approval
  4   Smith-Random
  5   -----NONE OF THE BELOW------

Rob Lanphier, 7-7-96 00:54:23 -0700 (PDT)
  1   Smith-Condorcet
  2   Condorcet
  3   Regular-Champion
  4   Smith-Condorcet using rated, not ranked, ballots 
  5   Condorcet using rated, not ranked, ballots 
  6   Instant Runoff
  7   Double Complement
  8   Approval
  9   ------NONE OF THE BELOW-------- 
 10   Plurality
 11   Smith-Random

Steve Eppley, 6-2-96 04:40:44 -0700 (PDT)
  1   Smith-Condorcet
  2   Smith-Condorcet using rated, not ranked, ballots
  3   Condorcet
  4   Condorcet using rated, not ranked, ballots
  5   ------NONE OF THE BELOW-------
  6   Instant-Runoff-1
  7   Instant Runoff
  8   Approval
  9   all the methods omitted from this response
 10   Runoff
 11   Double Complement
 12   Plurality

* *

Here's a more compact summary, using the following table of abbreviations:
  Abbr.   Method
  -----   --------------------------
  A       Approval
  C       Condorcet
  Cw      Condorcet with weighted (a.k.a. rated), not ranked, ballots 
  D1      Demorep-1 (Approved-Condorcet-MostApproved ??)
  D2      Demorep-2 (Approval of top n-1 ranked choices)
  DC      Double Complement
  IR      Instant Runoff (a.k.a. MPV, a.k.a. Hare's method)
  IR1     Instant-Runoff-1
  NOTB    [not a method: it marks one's threshold of approval/disapproval.]
  P       Plurality (a.k.a. Vote-For-Only-One, a.k.a. First Past The Post)
  R       Runoff
  RC      Regular-Champion (aka Copeland-Plurality)
  SC      Smith-Condorcet
  SCt     Smith-Condorcet-Tobin
  SCw     Smith-Condorcet with weighted (a.k.a. rated), not ranked, ballots
  SR      Smith-Random

Notes:
The symbol '>' means "is preferred more than".
The symbol '=' means "is preferred the same as".

Demorep: D1 > D2 > IR
Don:     IR
Hugh:    SCt > SC > SR > IR > C > RC > DC > A
Kevin:   C = SC > A > SR
Mike:    C > SC > IR1 > A > SR > NOTB
Rob:     SC > C > RC > SCw > Cw > IR > DC > A > NOTB > P > SR
Steve:   SC > SCw > C > Cw > NOTB > IR1 > IR > A > the_unranked > R > DC > P

* *

Tallying the responses by Condorcet, Smith-Condorcet, and 
Regular-Champion (some "pairwise" methods)

These three pairwise methods were heavily discussed and were most
favored by at least one subscriber, so the responses were tallied
using all three.  (The subscriber who favors Regular-Champion did
not provide a response for this poll.)

There are 16 choices (15 methods plus the NOTB approval/disapproval 
dividing line) included in our ballots, so that means 120 = 16*(16-1)/2 
pairings to compute.  All three pairwise tallies can make use of the 
same 16x16 array.  This section shows the array and the results
using Smith-Condorcet, Condorcet, and Regular-Champion.

The "Pairwise" Array
Notes: 
1. The number in each cell of the array shows how many voters prefer
 the choice at the row's left more than the choice at the column's
 top.  A blank cell means zero.  
2. The suffix 'L' after a number denotes that the choice at the
 column's top lost the pairing.
3. The suffix '=' after a number denotes a pairwise tie between the
 choice at the row's left and the choice at the column's top.

             IR1  IR  RC  C   Cw  SC  SCw SCt SR  A   P   DC  R   D1  D2 NOTB
  IR1         -   2   2=      1       1   2L  2   2   2L  2=  2L  2L  2L  1
  IR          4L  -   4L  3   3L  2   3L  4L  4L  5L  5L  5L  5L  4L  4L  4L
  RC          2=  1   -       2L      2L  1=  1   2   3L  3L  3L  2L  2L  2L
  C           5L  4L  5L  -   5L  1   4L  4L  4L  5L  5L  5L  5L  5L  5L  5L
  Cw          2L  2   1       -           2L  2   2   2L  2L  2L  2L  2L  2L
  SC          5L  5L  5L  3L  5L  -   5L  4L  5L  5L  5L  5L  5L  5L  5L  5L
  SCw         2L  2   1   1   2L      -   2L  2   2   2L  2L  2L  2L  2L  2L
  SCt         1   1   1=  1   1   1   1   -   1   1   2L  2L  2L  1=  1=  1
  SR          3L  3   3L  1   3L      3L  3L  -   1   4L  4L  5L  4L  4L  3L
  A           4L  2   3L      3L      3L  4L  4L  -   5L  3L  5L  5L  5L  4L
  P           1                               1       -       1=           
  DC          2=              1       1   1=  1   2   3L  -   2L  2=  2=  2L
  R                                                   1=  1   -        
  D1          1   1   1   1   1   1   1   1=  1   1   2L  2=  2L  -   1L  1
  D2          1   1   1   1   1   1   1   1=  1   1   2L  2=  2L      -   1
  NOTB        2L  1   1                   2L  2   1   2L  1   2L  2L  2L  -
             --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
 Wins:        5  13   8  14   8  15   9   2  11  12   1   3       3   2   6
 Losses:      8   2   5   1   7       6   9   4   3  13   9  14  10  11   9
 Ties:        2       2                   4           1   3   1   2   2  
 LargestLoss: 5   5   5   3   5       5   4   5   5   5   5   5   5   5   5

Smith-Condorcet defeated all other choices pairwise, so it wins when our 
responses are tallied by Smith-Condorcet, Condorcet, and Regular-Champion.  

Condorcet finishes second according to Smith-Condorcet, Condorcet,
and Regular-Champion.  Its "Condorcet score" (its largest pairwise
loss) of 3 is smaller than any but Smith-Condorcet.  Its "Copeland
score" (pairwise wins minus pairwise losses) of 12 is larger than 
any but Smith-Condorcet.  And if the Smith-Condorcet choice were 
removed from all the responses, Condorcet would defeat all the other
choices pairwise.

* *

Tallying the responses by Instant Runoff

(Note: During this tally, the question arose about how to treat a
ballot where more than one choice is listed in first place.  Two
ways were identified: give an equal fraction to each choice, or give
each choice a whole vote.  Instant Runoff gives a fraction, and the
variation which gives a whole vote has been dubbed Instant-Runoff-1.
The two EM subscribers who posted comments about this question
like Instant-Runoff-1 more than Instant Runoff; however, the 
tally shown here uses Instant Runoff.)

  Round 1:
    Eliminate D2, IR1, SR, P, R, DC, RC, NOTB, A, Cw, SCw since they 
    are the first choice of nobody:
      Demorep:  D1 > IR
      Don:      IR
      Hugh:     SCt > SC > IR > C
      Kevin:    C = SC
      Mike:     C > SC
      Rob:      SC > C > IR
      Steve:    SC > C > IR > the_unranked
  Round 2:
    Eliminate SCt, IR, and D1 since they tie for least first-ranked votes:
      Demorep:  --wasted--
      Don:      --wasted
      Hugh:     SC > C
      Kevin:    C = SC
      Mike:     C > SC
      Rob:      SC > C
      Steve:    SC > C
  Round 3:
    Eliminate C since it has fewer first-rank votes than SC:
      Demorep:  --wasted--
      Don:      --wasted
      Hugh:     SC
      Kevin:    SC
      Mike:     SC
      Rob:      SC
      Steve:    SC
Result: Smith-Condorcet finishes 1st, with 5 votes, when the
responses are tallied by Instant Runoff.

Finding the second place finisher using Instant Runoff:
Method 1: Find the method eliminated last (see above):
  C was eliminated last, so it finishes second.
Method 2: Delete the winner (SC) from the initial rankings:
  Demorep:  D1 > D2 > IR
  Don:      IR
  Hugh:     SCt > SR > IR > C > RC > DC > A
  Kevin:    C > A > SR
  Mike:     C > IR1 > A > SR > NOTB
  Rob:      C > RC > SCw > Cw > IR > DC > A > NOTB > P > SR
  Steve:    SCw > C > Cw > NOTB > IR1 > IR > A > the_unranked > R > DC > P
  Round 1:
    Eliminate D2, IR1, SR, P, R, DC, RC, NOTB, A, Cw, since they 
    are the first choice of nobody:
      Demorep:  D1 > IR
      Don:      IR
      Hugh:     SCt > IR > C
      Kevin:    C
      Mike:     C
      Rob:      C > IR
      Steve:    SCw > C > IR > the_unranked
  Round 2:
    Eliminate SCt, IR, D1, SCw since they tie for least first-ranked votes:
      Demorep:  --wasted--
      Don:      --wasted
      Hugh:     C
      Kevin:    C
      Mike:     C
      Rob:      C
      Steve:    C
Result: Condorcet finishes 2nd, with 5 votes, when the responses
are tallied by Instant Runoff.

* *

Glossary: Definitions of single-winner voting methods and terms
[This section needs to be completed before the report is posted to ER.]

Approval
Condorcet:
  This is one of the pairwise methods (see below).
  As in all pairwise methods, if there is one candidate who pairwise-
  defeated all the other candidates, then this candidate is the winner.
  If there isn't such a candidate, Condorcet elects the candidate 
  whose largest pairing-loss is the smallest (where the size of each 
  pairing-loss is the number of voters who prefer the pair-winner more 
  than the pair-loser).
  For example, suppose the ballots were:
    46 voters:  1=X   2=Y=Z
    20 voters:  1=Y   2=X=Z
    34 voters:  1=Z   2=Y   3=Z
    There are 3*(3-1)/2 = 3 pairings:
      X is pairwise-defeated by Y (46 to 54).
      Z is pairwise-defeated by X (34 to 46).
      Y is pairwise-defeated by Z (20 to 34).
    There is no candidate who pairwise-defeated all the others.
    The size of X's one pairing-loss is 54.
    The size of Y's one pairing-loss is 34.
    The size of Z's one pairing-loss is 46.
    Since Y's largest pairing-loss (34) is smaller than X's (54) and
      smaller than Z's (46), Y is the winner.

Condorcet with weighted (rated) ballots:

Demorep-1:
  A mixture of rank voting, majority yes (approval), Condorcet 
  for the majority yes candidates, limited approval tie breaker
  (limited combined choices).  
  See: Subject: Single Person Elections 6/27/96
       Date:    Fri, Jun 28, 1996 12:20 AM EDT 
Demorep-2:
  Approval with rankings (dropping the last ranked choice--such as 
  having 5 candidates, only top 4 choices of a voter would count--
  highest majority candidate wins) 
Double Complement
Instant Runoff
Instant-Runoff-1
NOTB:
  This choice is not a method.  Its insertion into a voter's rankings
  indicates that the voter approves of all the choices s/he ranked more 
  preferred than NOTB and disapproves of all the choices which s/he 
  ranked less preferred than NOTB.  Its use by some of the respondents 
  in this poll does not imply a recommendation that NOTB ought to be 
  allowable as a choice in elections for public offices; in principle, 
  good voting methods will encourage more good candidates to run (since 
  they won't face a spoiler dilemma), and if all the candidates are
  disapproved by the electorate it might create a dangerous power 
  vacuum.

Pairwise Methods:
  In all pairwise methods, the info contained in the ranked ballots 
  is used to calculate what the results would be in all the possible 
  2-candidate "runoffs", called pairings.  If there are N candidates 
  running, there are N*(N-1)/2 possible pairings.  For example, 
  suppose there are three candidates named X, Y, and Z.  The number 
  of possible pairings is 3*(3-1)/2 = 3:
    X vs. Y
    X vs. Z
    Y vs. Z
  A ranked ballot such as 1=Y,2=X,3=Z is counted as a vote preferring 
  Y more than X in the X vs Y pairing, and a vote preferring X more
  than Z in the X vs Z pairing, and a vote preferring Y more than Z
  in the Y vs Z pairing (since a voter's rankings are assumed to be 
  transitive).
  A candidate X "pairwise-defeats" another candidate Y if the number
  of voters who prefer X more than Y is greater than the number of
  voters who prefer Y more than X.
  If there is one candidate who pairwise-defeated all the other 
  candidates, this candidate is the winner in all pairwise methods.
  For example, suppose the ballots were:
    46 voters:  1=X   2=Y   3=Z
    20 voters:  1=Y   2=X=Z
    34 voters:  1=Z   2=Y   3=Z
    There are 3*(3-1)/2 = 3 pairings:
      X is pairwise-defeated by Y (46 to 54).
      Z is pairwise-defeated by X (34 to 46).
      Z is pairwise-defeated by Y (34 to 66).
    Y pairwise-defeated both X and Z, so Y is the winner.
  Each pairwise method differs in how it determines the winner when
  there isn't one candidate who pairwise-defeated all the others.

Plurality (a.k.a. vote-for-only-one, a.k.a. First Past The Post)

Rated Ballot methods:  See "Weighted Ballot methods" below.

Regular-Champion
Runoff
Smith:
  This is a part of several pairwise methods (below).  The Smith step 
  separates the candidates into two groups ("best" and "worst"): all 
  the candidates in the "best" group pairwise-defeated all the candidates
  in the "worst" group.  (To be more precise, the "best" group is the
  smallest group which can be found such that all the candidates in 
  the group pairwise-defeated all the candidates not in the group.  
  It must have at least one candidate in it, and it may have all the 
  candidates in it.)  All the candidates not in the "best" group are 
  eliminated from further consideration.
Smith-Condorcet
Smith-Condorcet with weighted (rated) ballots
Smith-Condorcet-Tobin
Smith-Random
Weighted Ballot methods

--------------- End of Report (Draft) -----------------




More information about the Election-Methods mailing list