[EM] Why to mix methods...

Tue Sep 24 07:32:33 PDT 2002

Craig, one of the method on the ESD list could select
A and E.

It mixes approval distinction from approval voting, sequential
elimination from IRV and a Condorcet
method (ranked pairs with relative margins).

It is "K) ranked pairs using sequential elimination, residual approval
weights and relative margins" or a title alike.
In this case you would need an approval limit (all ballots are
fully ranked), standard elimination process until
only two candidates remain is not enough.

By the way, anybody interested to vote for selecting the best
single-winner and multiple-winners methods can join at
http://groups.yahoo.com/group/Electoral_systems_designers/

I hope it helps,

Steph.

---------- Forwarded message ----------

> Date: Tue, 24 Sep 2002 11:41:07 +1200
> From: Craig Carey <research at ijs.co.nz>
> Reply-To: single-transferable-vote at yahoogroups.com
> To: single-transferable-vote at yahoogroups.com
> Subject: Sequential STV, would a faster CPU show it up as better?
>
> Is this method interesting or not?.
>
> http://www.electoral-reform.org.uk/publications/votingmatters/P4.HTM
>
>  >The name 'Sequential STV'
>  >
>  >From now on the name Sequential STV will be used to mean this new
>  > version.
> ...
>
> The method described in the paper is a new version with the same
> name.
>
>  >
>  >Suppose, however, that the voters for A, B, C and D had all put in E
>  >as second preference to give (the example used in reference 1).
>  >
>  >104· AEBCD
>  >103· BECDA
>  >102· CEDBA
>  >101· DEBCA
>  > ·3· EABCD
>  > ·3· EBCDA
>  > ·3· ECDBA
>  > ·3· EDCBA
>  >
> Meek STV decides in this way (I checked it)
>     1. E is excluded
>     2. D is excluded
>     3. B is elected
>     4. C is elected
>     5. A loses
>
> It picked B and C.
>
>  >This evidently makes E a very much stronger candidate, for if any one
>  >of A, B, C or D had not stood, E would have been the first elected,
>  >but plain STV takes no notice, electing BC just as before. Sequential
>  >STV chooses BC as probables but then tests BCD, where BC stay as
>  >probables and D goes to the end of the queue, followed by BCE where BE
>  >become the new probables and C goes to the end of the queue. It then
>  >tests BEA and BED, BE winning each time. There is no need to test BEC
>  >again as that result is already known, so BE are elected.
>  >
>
> They say that Sequential STV elects B and E.
>
> I would hope that A and E win.
> So far my IFPP method seems to resemble the Alternative Vote provided
> it is transferring votes away from a candidate with a lot less votes.
> So E should win. In the 2nd step, there may be quotas (not known
> quotas). But it seems to me that quotas won't be moving against
> candidate A. To get serious, it would have to be reduced to a 4 candidate
> example.
>
> After the transfer of E's votes:
>
>   107· ABC
>   106· BCD
>   105· CDB
>   101· DBC
>    ·3· DCB
>
> It looks like quotas for winners or loser maybe would make A win or would
> fail to make A lose.
>
> ---
>
> The Sequential STV's worst case example showing it to be just failing
> truncation resistance could be searched for.
>
> But instead of making do with existing optimizing algorithms, it is my
> opinions it is better to abandon research into STV until a good
> optimizing/minimizing algorithm is created.
>
> Here the article continues suggesting that research was being done.
> I am sure it can't be done if they never created a new optimizing
> method.
>
>  >Real voting patterns
>  >
>  >In 43 real elections held on file, the sequential method merely
>  >confirmed the original result in 38 of them, and replaced just 1
>  >candidate in 3 more of them. In only 2 cases were loops found, making
>  >it necessary to do more than the easy part of the method.
>  >
>  >Timings
>  >
>  >Some timings were made on an 11-year old PC with a 386 chip. In a real
>
> The testing method was not beyond being questioned too.
>
>  >election with 10 candidates for 6 seats and 841 voters, simple STV
>  >took 11 seconds. Sequential STV made no change in those elected and
>  >took 23 seconds.
>  >
>  >In a much more difficult case with 30 candidates for 15 seats and 563
>  >voters, simple STV took 1 minute 6 seconds. Sequential STV found 1
>  >candidate to be definitely replaced and 3 others who were in a loop
>  >for the final seat. It took a total of 18 minutes 30 seconds.
>  >
>  >Should it be used?
>  >
>  >With this new version, should it be recommended for practical use?
>  >That depends upon whether the user is willing to abandon the principle
>  >that it should be impossible for a voter to upset earlier preferences
>  >by using later preferences. Many people regard that principle as very
>  >important, but reducing the frequency of premature exclusions is
>  >important too. We know that it is impossible to devise a perfect
>  >scheme, and it is all a question of which faults are the most
>  >important to avoid.
>
> That says that the checking was light, and it was only a check for whether
> the method was different, rather than say, (1) a search for how its
> behaves at its worst [they can come to this mailing list and explain],
> or a comparison showing that it is better.
>
> Meek and plain STV agreed on the wrong answer.
>
> Sequential STV is close to be able to be checked and it presumably
> it is rejected since Gazeley alone told us it is not truncation resistant.
>
> So it is rejected it though it could be better.
>
> I ask Brian: is Voting Matters going to continue to show that STV methods
> are only different.  Surely a paper can't reject Meek STV and plain STV
> by saying the replacement method is better when the replacement method is
> very probably (to quite possibly) getting the winners wrong too.
>
> ----
>
> I am sure that the British are not actually making progress since lacking
> a numerical optimizing method. Maybe Brian Wichmann could mention that in
> the next editorial of the ERS journal. I should not be expected to do
> since behind the problem of no optimizer are 2 other problems:
>
> 1 * sticking with "stages of the count";
>
> 2 * once that is rejected, then inability to guess a good method; and
>
> 3 * need for a numerical optimizer (again) to check for "one man one vote"
>     and some stronger power<=1 equal suffrage rules.
>
> ----
>
> I ported QEPCAD and SACLIB to Windows 2000 MinGW and cygwin. My code
> does not run but at least now the code compiles.
> URL: http://www.ijs.co.nz/polytopes.htm
>
> ------------------------ Yahoo! Groups Sponsor ---------------------~-->
> Home Selling? Try Us!
> http://us.click.yahoo.com/QrPZMC/iTmEAA/ySSFAA/YgSolB/TM
> ---------------------------------------------------------------------~->
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em