Re: [wsbpel-reqts] Approval Voting

[Diane Jordan <drj@us.ibm.com>]

I've read the site - it makes very good
points about anomalies of STV/IRV.  After thinking about the approval
voting approach, I'm uncomfortable telling our members to treat all choices
that they could live with equally in a vote - given the strengths of opinions
on the TC, I'm not sure this would work well.  

This web site goes into some depth explaining
alternate voting methods and recommends "condorcet" voting  which
allows for ranking of choices much like the STV/IRV but eliminates some
of the oddities of the vote transfer with that method.  It is a bit
more complex to figure out the results as it takes each two way vote preference
into account but the good news is they provide a link to a site which can
do the calculations.    
To the voter it would appear the same
as STV/IRV - you'd get to indicate your preferences in a 1st, 2nd, 3rd
(or on indefinately - you can rank as many options as exist or cap the
choices at 3 out of x).  

I'm ok with using this method.  Let
me know what you think.     
Regards, Diane
IBM  Emerging Internet Software Standards
drj@us.ibm.com
(919)254-7221 or 8-444-7221, Mobile: 919-624-5123, Fax 845-491-5709

"Yaron Y. Goland" <ygoland@bea.com>

10/06/2004 05:06 PM

Please respond to
ygoland

To	bpel rqmts <wsbpel-reqts@lists.oasis-open.org>
cc
Subject	[wsbpel-reqts] Approval Voting

Here is an excerpt from the mail I sent to Martin
on Monday about 
approval voting:

Slashdot had an article last week on different voting techniques so I 
read up.

One of the most interesting things I found was the Arrow Theorem by 
Professor Kenneth Arrow. He won a Nobel prize for the theorem which 
proved that all voting systems MUST be unfair in some way.

I then read up on IRV and found that it has a really nasty behavior 
(besides being incredibly complex). It violates the monotonicity 
principle. In a nutshell this says that voting for a choice you want 
shouldn't make that choice lose and not voting for the choice you want
should help that choice win. That may seem obvious but it turns out that
IRV violates it. See <http://www.electionmethods.org/evaluation.html#MC>
  for an example.

I then discovered another voting system that is trivial to run, can be
used with Kavi as is and doesn't violate the monotonicity principle. 
It's called approval voting. <http://en.wikipedia.org/wiki/Approval_voting>

The way approval voting works is that if you have N choices then you 
open a ballot and let people put a check next to each choice they can 
live with. Then you add up the results and which ever choice got the 
most votes wins.

Kavi can already do this today. In fact, just about all voting systems
can do this today. This is just a 'select N entries from M choices' vote
where N = M. Diane actually used this feature in our IRV ballot for the
BPEL spec name when she let people select 2 out of the N choices so they
could indicate a preference for spec name and version number from the 
same list.

So if we use approval voting then we can run the whole thing on a single
ballot, no fancy math, nothing. Just add up the votes and which ever 
choice got the most wins.

Of course approval voting also has a flaw (remember the Arrow theorem).
Imagine that we have a vote with 10 people and three choices, A, B and
C. 6 people strongly support A. 2 people strongly support B and 2 people
strongly support C. However, 7 people weakly support B and 3 people 
weakly support C. Since people can vote for multiple options this means
that someone who strongly supports A and weakly supports B but doesn't
support C would cast a vote for A and B but not C. In that case the 
voting total will be:

A - 6 votes
B - 9 votes
C - 5 votes

So even though over 50% of the group strongly supports A, A still loses.

The reason is that approval voting detects consensus. It explicitly 
looks for the option that the largest number of people can live with.

I actually think the 'flaw' in approval voting is a feature in our case.
Standards bodies are supposed to be run on consensus and approval voting
is excellent at finding that consensus.

As such, given that it is simple to run, doesn't violate the 
monotonicity principle and is well designed to detect consensus doesn't
it make sense for us to switch from IRV to approval voting?

                
Yaron