[EM] kemeny

[Richard Fobes]

I finally have time to respond to Warren Smith's comments (repeated 
below) about VoteFair Ranking, which is based on the Condorcet-Kemeny 
method.  In a separate message I'll respond to his more-recently stated 
concerns about computation time.

VoteFair party ranking, which can be used to limit the number of 
candidates per political party, would not be used in all elections.  It 
is intended for use in elections where the winner becomes quite 
powerful, such as in U.S. congressional elections, U.S. presidential 
races, U.S. state governorships, provincial-governor elections, and 
parliamentary elections.  It would not be used for less-important 
elections, such as for the mayor of a town.  Of course it does not apply 
to voting situations that do not involve political parties.

In high-stakes elections the two (or three) parties that have the most 
"popularity" according to VoteFair party ranking would be allowed two 
candidates each, some smaller parties would be allowed one candidate 
each, independent candidates would be allowed if they collect enough 
petition signatures, and write-in candidates would also be accommodated.

Based on one of Warren's comments, apparently I need to clarify that 
VoteFair party ranking can be used to determine who gets to run in a 
race, but it does not influence who wins.  This means that it meets the 
fairness criteria that "candidates are treated equally and the winner is 
deduced purely from the votes".  Expressed another way, VoteFair party 
ranking is in the category of signature-collection requirements because 
it is simply a way to ensure that an election does not have lots of 
"nobody" candidates, and does not have multiple candidates from very 
small political parties.

Remember that this issue will become significant when the risk of vote 
splitting disappears and every political party -- including small ones 
-- may be tempted to enter lots of candidates in a single important 
election.

My un-named (and not-yet-announced) Kemeny-like ranking method would be 
used in situations where fast estimations are useful, such as ranking a 
list of 1000 movies, with updated results after each person casts their 
ballot.  It is not intended for use in elections (where a worst-case 
calculation time of perhaps one hour would be acceptable).  In other 
words, I did not create it for use in elections.  Yet if it were used in 
elections, the results would be fairer than IRV, and probably -- 
although I don't yet know for sure -- about as fair as most other 
Condorcet methods.  I'm not proposing it as a substitute, I'm just 
mentioning that it's useful for quickly (and fully) ranking hundreds or 
thousands of choices.

Warren seems to have assumed that I wrote "Ending The Hidden Unfairness 
In U.S.  Elections" before I had written software to implement what I 
described.  Actually I did not start writing that book until I had 
several years of experience calculating VoteFair popularity ranking 
results (which are mathematically equivalent to the Condorcet-Kemeny 
method) for real surveys and polls.

Looking further back, I first developed VoteFair popularity ranking 
while writing my earlier book titled "The Creative Problem Solver's 
Toolbox".  In that book I very briefly described the method and included 
two illustrations showing how it works.  At the time I published that 
book, which was in 1993, I had only limited experience calculating those 
results.  Much later, when I published my voting book in 2006, I had 
lots and lots of real-world experience calculating VoteFair popularity 
ranking results.

Admittedly, while writing my voting book I developed VoteFair 
representation ranking.  Yet I wrote the software to implement that 
method before publishing the book.  Since then it has performed exactly 
as I expected.  Specifically VoteFair representation ranking identifies 
the winner of a second equivalent seat by reducing the influence of the 
voters who are already well-represented by the winner of the first seat. 
  Unlike STV, it includes an extra step that prevents a strategic voter 
who helps elects the first winner from also excessively influencing who 
wins the second seat.

Admittedly I have not yet used VoteFair party ranking.  As I said, it is 
designed for especially important elections, and it will take awhile 
before advanced voting methods are used at that level.  Yet it is very 
simple, so I don't expect it to produce unfair results.

Warren wrote: "And frankly, it is now obvious to me that you never in 
your life wrote and tested a good software package for carrying out 
Kemeny elections, and nor has anybody else, and it probably cannot be done."

The thousands of elections, surveys, and polls that have been hosted at 
VoteFair.org over the past decade or so demonstrate that 
Condorcet-Kemeny calculations are very useful -- and practical -- in 
real voting situations.

Yes, Condorcet-Kemeny calculations do run into a brick wall if every 
sequence score is calculated.  And yes there are occasions when circular 
ambiguity arises (at any ranking level) and the 
calculate-every-sequence-score approach is not practical.  Yet, for 
reasons I will explain in my reply to Warren's other message about 
calculation time, this brick wall is not as solid as it appears.

Richard Fobes


On 9/12/2011 12:52 PM, Warren Smith wrote:
 > ...
 > I think Kemeny is pretty easy to define.  It also seems to have a 
pretty nice
 > combination of abstract properties.  On the other hand, "Fobes 
defining several
 > different election methods, of which Kemeny is just one, and trying
 > (very inadequately) to define when to use one and when to use the other,
 > to thus, in net, define a kludgy hybrid method" --
 > which is your current stance as indicated in this post by you -- 
constitutes
 > an election method which is NOT "easy to define," by any stretch of
 > the imagination,
 > and which does NOT have a nice combination of abstract properties.  It
 > therefore
 > has virtually no interest.
 > ...
 > I' m glad to see you are finally implicitly admitting that Kemeny is 
flawed, by
 > e.g, inventing your undisclosed method.  I remind you that on a
 > previous occasion
 > you claimed to me in email that you could easily solve elections with
 > 100s of candidates using integer programming
 > and knew of no instance where said solver would have trouble.
 > That claim was, we can now see, bunk.  I resent it.
 >
 >> By the way, if a real election is likely to involve 20, or  50, or more
 >> choices, then I recommend using VoteFair Ranking as described in my
 >> book, "Ending the Hidden Unfairness in U.S. Elections".  VoteFair
 >> Ranking uses "VoteFair party ranking" to identify which political
 >> parties deserve to have two candidates in the main election, and which
 >> parties are not popular enough to justify having any candidates.  This
 >> limitation is not for the purpose of reducing calculation time, but
 >> rather for the purpose of giving voters a reasonable number of
 >> candidates to keep track of, without distractions from
 >> cannot-possibly-win candidates.
 >
 > And I guess somebody who runs independent of a political party, or who
 > runs from a previously unpopular party, never deserves anything?
 >
 > Normally, election methods obey these "fairness" criteria: all
 > candidates are treated equally and
 > the winner is deduced purely from the votes, and unaffected by
 > renaming the candidates.
 > You are here, for perhaps the first time ever, recommending a voting
 > method now which intentionally treats candidates unequally and
 > intentionally disobeys that kind of fairness.  Yet you have the gall
 > to call this
 > "VoteFair Ranking."
 >
 > When you stoop to such desperate measures, it constitutes an admission
 > that your techniques are busted.  And frankly, it is now obvious to 
me that you
 > never in your life wrote and tested a good software package
 > for carrying out Kemeny elections, and nor has anybody else, and it 
probably
 > cannot be done.
 >
 > Which causes me to wonder about your priorities.  If you are going to
 > go to the trouble
 > to write an entire book boosting an election method, don't you think
 > you ought to try it out
 > a lot first?  Had you done so, you would have realized you were unable to
 > write a program for reliably finding Kemeny winners, and then you 
could have
 > avoided writing the book and all the rest of that waste of time for 
us all.
 >
 > In short, test out your world-shaking idea first using simulations; then
 > write book second.   I realize this approach is a stunningly new one 
for you,
 > but for a lot of the rest of us, it's just the standard obvious approach.