[EM] Why Condorcet

[robert bristow-johnson]

Sorry to be so wordy here.  There are 3 different issues (the number  
of ranking levels and the value of ballot access criteria, procedure  
if Write-In wins, and visual effectiveness of the "pairwise defeats  
matrix") to respond to:

On Jul 7, 2010, at 9:56 PM, Dave Ketchum wrote:
> On Jul 6, 2010, at 11:31 PM, robert bristow-johnson wrote:
>> On Jul 6, 2010, at 9:31 PM, Dave Ketchum wrote:
>>
>>> Ballots:  Must support write-ins and, perhaps, 3 ranks (do not  
>>> need to rank rejects and can do equal ranking).
>>
>> i think that the number of ranking levels should be as large as the  
>> number of candidates (and there should be ballot access laws that  
>> make it difficult enough to get on the ballot that no more than  
>> maybe 5 candidates normally get on).
>
> Making getting on the ballot unreasonably difficult affects who gets  
> to run, unreasonably - lets don't.

there is a question about what is reasonable.

> No real need for a ranking level for every candidate.  Especially  
> when there are many candidates voters will happily use some equal  
> ranks and will not bother to rank those they see as not worth ranking.

well, one of the complaints of the anti-IRVers (that applied more to  
San Francisco and not to Burlington Vermont) was that, because the  
number of candidates was much larger than the number of ranks, then  
voters could not express a vote regarding every candidate.  if the  
candidates that ended up leading the race (getting to the final and  
semi-final rounds) were not any that the voter ranked (due partly to  
the too few ranking levels), the claim was that this voter was  
"disenfranchised".  i have her kill-filed, but i'm sure Kathy Dopp can  
fill you in on what this sentiment is all about.  the funny thing is  
that, while this "feature" of IRV (in SF) was a major complaint of the  
IRV opponents, what the alternative offered (the return to the  
affirmative vote for a single candidate and FPTP and/or delayed  
runoff) was even more so a "disenfranchisement" in that very same  
manner.  it is equivalent to a ranked-order ballot but with only one  
ranking level deep.

also, sometimes a voter wants to bury a candidate.  how does one vote  
negatively against a candidate when they cannot rank all the other  
candidates above that one?  it's okay that all candidates not ranked  
by a particular voters are tied for last place on their ballots if  
that voter can lift all other candidates, not deemed the absolute  
worst, above that level.

similarly to voter registration and ballot access to voters being a  
political question, i think that ballot access to candidates is also a  
political question.  reasonable people can disagree, but i see a  
future with political parties to be inevitable.  it doesn't mean that  
we should have only two viable parties nor that independent candidates  
cannot credibly run for office.

but any candidate that deserves the ink and real estate on the ballot  
for their name, should be required to demonstrate some level of  
electoral support to get there.  if the race is broad enough (a  
national election in a large country), some candidate need not be on  
the ballot for every state or district within that nation, if the  
candidate cannot satisfy the qualifying threshold (in petition  
signatures) for every district.  but whatever the size of the election  
venue, be it a city ward, entire city, legislative district,  
statewide, whatever, there should be a minimum number of signatures  
required for ballot access that is approximately proportional to the  
voter population of that venue.  that constant of proportionality  
would have to be experimentally and empirically determined, but it  
should be made high enough to reduce the candidate field to about half  
dozen meaningful candidates.

since that number would be fixed in law in advance of the race, there  
could certainly be election flukes where many more candidates satisfy  
the ballot access criteria and, in that case, the number of ranking  
levels would be much less than the number of candidates, but that  
would be a fluke.  if it turned out to become common, the ballot  
access threshold should be raised by legislative action.


>> i think Write-In can just be considered another candidate.
>
> Agreed each should be, and my way of counting provides for this.   
> Counters can hear before election day of likely need.

one thing, i don't think it would be so good if the election *law*  
pays any respect about who is a merely *reputed* viable write-in  
candidate.  i would suggest that, since Condorcet is precinct  
summable, that after the polls close, the ward or precinct officials  
transmit their pairwise subtotals up to the government seat to be  
counted, but do not yet pack up until they hear from the central  
authority that Write-In did not win.  if Write-In wins, they have to  
separate all of the ballots where Write-In is the likely leading  
candidate and all others, but just because the newspapers are saying  
that some write-in candidate is a significant figure, that should not  
be the basis of the procedure provided in law.

because, if Write-In wins, there would have to be a lot of ballots  
that have Write-In marked.  i would suggest that each precinct draw  
ballots at random from the stack, examining each Write-In name and the  
first name to get to 10 (or some other threshold) write-ins of the  
same name, that name would be identified as the likely write-in  
leader.  then they would have to separate all ballots with that name  
as Write-In from all other ballots.  then they can run the two piles  
of ballots through the counting machines separately, zeroing the  
pairwise totals between each run.  for all other candidate pairs not  
involving Write-In, they can sum the vote totals (and that should  
equal the pairwise vote totals they had before), but Write-In for the  
first pile (where Write-In is a specific person's name) would be a  
different candidate than the other Write-In.  they can report those  
totals (and the name of the primary write-in candidate that they  
separated) up to the central election authority who then reruns the  
totals for each Condorcet pair and *then* they can see if this  
specific write-in candidate continues to win.  if this Write-In  
doesn't quite win and some precincts had one name as their primary  
write-in and other precincts had a different name, the election  
officials at the central authority would have to determine (by venue- 
wide vote counts) which Write-In is the more significant candidate and  
communicate to all precincts to separate ballots according to that  
name and then run the totals again.  if that same Write-In (where all  
precincts assumed that name and separated ballots as such) doesn't  
win, someone else is the Condorcet winner or there is a cycle.

of course, if Write-In doesn't win, it doesn't matter who Write-In is,  
and all this would not be necessary.

>     Near here, a few years ago, someone's error resulted in no  
> petitions being submitted on time.  Recovery was to do entire  
> election with write-ins.

no candidates but multiple write-ins?  how would we prepare law in  
advance for that contingency?  perhaps, if the number of candidates  
gaining ballot access (having met the threshold of signatures) is  
equal to or greater than the (fixed) number of ranking levels minus  
one, then only one line for Write-In would be on the ballot.  but if  
the number of registered candidates is less than that, there would be  
that many more Write-In slots so a voter could write in more than one  
candidate and rank them against each other.  the number of ranking  
levels would remain the same.

>> the only issue is what to do if Write-In wins, then there must be  
>> some assumption first to determine who the likely write-in  
>> candidate is, hand separate those ballots from the others with a  
>> write-in (and call them two different candidates, Write-In1 and  
>> Write-In2).  and then retabulate as a Condorcet election and see if  
>> Write-In1 still wins.

oops, i guess i already said this.

>> i still think that rectangular N*N matrix is sorta useless.  it's  
>> hard to read.  each pair should be grouped together for visual  
>> inspection.
>
> Debatable.

i guess it is, but i just cannot see how this:


          M        K        W        S        H
       ------------------------------------------
   M  |          4064     4597     4570     6263
   K  | 3477              4313     3944     5515
   W  | 3664     4061              3971     5270
   S  | 2997     3576     3793              5570
   H  |  591      844     1310      721


    ... is as easy to inspect and extract meaning as this:

   M 4064
   K 3477

   M 4597     K 4313
   W 3664     W 4061

   M 4570     K 3944     W 3971
   S 2997     S 3576     S 3793

   M 6263     K 5515     W 5270     S 5570
   H  591     H  844     H 1310     H  721

the numbers are only meaningful in their respective pairs.  why not  
group the pairs together?


--

r b-j                  rbj at audioimagination.com

"Imagination is more important than knowledge."