*Author: *Don Armstrong <[EMAIL PROTECTED]>
*Date: *2007-05-19 22:012007-05-20 05:01 -700UTC
*To: *debian-policy <[email protected]>
*Subject: *Re: Range Voting - the simpler better alternative to Condorcet
voting
On Sat, 19 May 2007, CLAY S wrote:* *
*> My name is Clay Shentrup and I am an Ubuntu user - so I have much
respect *
*> for your endeavors and your hard work on them. I write to discuss a
simple *
*> improvement that could be had with your elections, merely by changing
to a *
*> better and simpler voting method: Range Voting.*
Warren Smith's copious arguments to the contrary, it's not entirely
clear that Range Voting is superior to or even simpler than
concordecet voting.
According the social utility efficiency figures, it is substantially better
- in all 720 different combinations of the 5 "knobs" he chose to tweak; from
0% strategy, to 100% strategic voters, few candidates, to many candidates,
ignorant voters or informed voters. And Condorcet suffers much worse from
strategic voting - especially considering things like the DH3 pathology.
Range Voting handles strategic voting quite gracefully, such that its large
advantage over Condorcet voting actually _increases_ with a more strategic
electorate.
As for simplicity, it is pretty easy to show that Range Voting is
objectively much simpler than Condorcet voting. Imagine the ease of writing
a computer program to get the average of a set of ratings, as opposed to
doing any type of Condorcet method. It is a demonstrable fact that the
Range Voting program will be shorter.
And for voters, I do think that Range Voting is much simpler, and in some
sense this is provable, especially the more candidates there are. Consider
the process of going through a list of, say movies on IMDB.com, and rating
them. This is rather simple, and the number of operations increases
_linearly_ with increasing options ("candidates"). But in order to rank
them properly, you are essentially doing a sort, making a very large number
of comparisons, which increases exponentially with increasing candidates.
Think about the number of comparisons required for a quicksort, and then
realize that what most people do is actually more like a bubble sort - far
less efficient. And Range Voting even allows voters to abstain for a
candidate they do not know much about, without helping or hurting his
average (it is not affected, but he must achieve a quorum of total points to
be legitimate, regardless of his average). In Condorcet, you are
effectively forced to rank all candidates, because any you do not rank must
be treated as "tied for last place".
In some sense it does come down to personal preference as to what people
"perceive" to be simple; but in any case, I don't think that its worth it to
sacrifice the quality of the candidates elected in order to make an election
_marginally_ simpler. And I actually did exit polling using Range Voting in
Texas last November - http://RangeVoting.org/Beaumont.html - and found that
people had an easy time of it.
Also we must understand that the fundamental axiom of Condorcet voting is
WRONG. That is what is so ironic about the differentiating factor in
Condorcet methods - how they handle cyclic ambiguities. The fact that such
an ambiguity can exist in the _first place_ proves that the axiom "if A is
preferred by a majority over B, then A is a better winner" is wrong.
Therefore the entire basis of Condorcet voting, that it aims for
majoritarianism instead of social utility efficiency, is flawed.
I definitely think it's admirable that Debian has used an election method so
sophisticated - Condorcet voting is definitely worlds ahead of plurality or
Instant Runoff Voting, in most regards. But I cannot see why Debian members
would not want to upgrade to a substantially better, and arguably simpler,
method.
*> Dr. Smith also has done some analysis of Debian's elections, which
represent *
*> a wonderful dataset for scientific study. *
*> http://rangevoting.org/Debian2003.html*
His analysis is rather scanty, unfortunatly, and fails to provide any
real reasoning why we should even consider switching to Range Voting
especially as the conversion from the balots cast to RV balots is
entirely arbitrary.
That page is not meant to be a strong case for Range Voting's superiority to
Condorcet voting, and the conversion was just for theoretical purposes, to
allow a comparison of how many different voting methods would have called
the election. The strong case for RV's superiority to Condorcet was
provided in the links I gave about social utility efficiency, the DH3
pathology, etc.
Finally, if you are really interested in even being able to test RV in
Debian, you need to actually write the code to implement RV in
devotee.
That would be trivial. I'm primarily concerned with convincing you that
Range Voting is the superior choice. And I really truly believe that the
evidence is pretty staggering.
Regards,
Clay Shentrup
The Center for Range Voting
