dww@stl.stc.co.uk (David Wright) (11/20/89)
Here is a 'worked example' of how the STV system for voting for new news
groups would work. Despite the fact that people have been using STV in
elections for at least 75 years, most people don't know much about it.
For voting, it really is "as simple as 1,2,3", and you don't need to read
all this - just read the "25 line proposal" that I posted earlier.
But if you want to know how an STV ballot for a new news group might
be counted and reported to the net, read on.
By the way, this shows a special case of STV in which we want to elect a
single winner. STV is often used to elect a group or committee; in that
case the working is similar to that shown here, but the surplus vote of
winning candidates is also transferred, and the quota which a candidate has
to reach in order to be elected is 1/(N+1) +1, e.g. 1/4 +1 for 3 places,
rather than 1/2 +1 as it is below.
Asssume that during the discussion period on a group for discussing rabbits
the following names have come up:
rec.pets.bunnies
rec.rabbits
rec.pets.rabbits
sci.coneys
and that someone has mailed the moderator of news.announce.newgroups
requesting that the names talk.rabbit and comp.society.rabbit
be added, as the original proposer had ignored his request to include them.
The rules say the moderator *may* add extra names; this allows him to use
his discretion, but in practice it does little or no harm to include extra
names even if they get little support; the few who vote for them will have
their later preferences counted instead.
So the call for votes would list the following options:
A comp.society.rabbit
B rec.pets.bunnies
C rec.pets.rabbits
D rec.rabbits
E sci.coneys
F talk.rabbit
N no
I have used alphabetic order here - the rules don't require it, but it
might be a good convention to follow.
As suggested by the rules, most voters put their first preference in the
subject line. This makes it easy for the vote-taker(s) to file each vote
as it comes in into a folder of that name, only having to read the votes
of people who have forgotten to edit the subject line. People could set
up a script to do this stage automatically if they wanted, so long as
voters who had not got the subject line right were not disenfranchised.
So the voters would vote for their preference in order, e.g. one might send:
Subject: Rab vote: rec.rabbits
1 rec.rabbits
2 rec.pets.rabbits
3 rec.pets.bunnies
4 no
5 talk.rabbit
An alternative (currently being trialed in the {sci|comp}.groupware vote)
would be to ask voters to code their complete preference in the subject
line, e.g. as DCBNF-- for the above case. This would be counted in the
same way. It might make vote counting a little easier, at the risk of
being harder for voters. No doubt various people will write software
for automatic counting a 1 2 3 vote; in the meantime semi-manual methods
with mail folders or even paper should suffice.
So the first stage count is made, and produces the following result:
STAGE 1
A comp.society.rabbit 1
B rec.pets.bunnies 123
C rec.pets.rabbits 115
D rec.rabbits 98
E sci.coneys 7
F talk.rabbit 35
N no 32
Total 411
Quota 206 Quota is smallest integer > Total/2
Total excluding NO 379
Margin positive over NO 347
No choice has got 'quota' votes, so we go on to the next stage. As there
are 7 possibilities, there will be at most 6 stages (though it is unusual
to have to go through all of them). Rather than take this stage by
stage, let us look at the final position, then I'll comment on each stage.
STAGE 1 STAGE 2 STAGE 3 STAGE 4 STAGE 5 STAGE 6
Elim. A Elim. E Elim. F Elim. N Elim. D
A comp.society.rabbit 1 -------------------------------------------
B rec.pets.bunnies 123 123 123 130 131 161
C rec.pets.rabbits 115 115 115 128 138 208
D rec.rabbits 98 98 98 100 108 ---
E sci.coneys 7 7 ---------------------------------
F talk.rabbit 35 35 35 -----------------------
N NO 32 32 39 49 -------------
Non-transferrable 0 1 1 4 34 42
Total effective votes 411 410 410 407 377 369
Quota 206 206 206 204 189 185
Total excluding NO 379 379 372 362 377 369
Positive margin over NO 347 346 332 313 377 369
Winning choice: rec.pets.rabbits
=============================================================================
This is how it happened:
Stage 2: comp.society.rabbit is transferred.
Unfortunately this ballot did not express any other preference,
so it cannot be transferred, and plays no more part in the count.
Stage 3: sci.coneys is transferred.
All these people would rather have NO group at all than any of
the others (unlikely in practice, of course).
Stage 4: talk.rabbit is transferred.
Note that without the previous transfer, we'd have transferred NO
This vote splits 3 ways. Also three foolish (or care-less)
voters have not expressed any more valid preferences, so their
votes are non-transferrable, and the quota changes.
Stage 5: the NO vote is transferred!
Understandably quite a few of these expressed no further
preference, but for those who did, they are counted.
This means that the NO votes won't be able to prevent the group
passing now - but if they were ever going to, they would have
done so by this stage anyway.
Stage 6: the rec.rabbits vote is transferred.
At last we have reached the quota. Most rec.rabbits votes
transfer to rec.pets.rabbits, and rec.pets.rabbits is THE WINNER.
Pity, I'd got to like the idea of rec.pets.bunnies myself :-)
You'll see that the final result in this case is not the one that was ahead
on first preferences - but it is the one that most people are happy with,
and thus the one most likely to be acceptable to the net. Incidentally,
votes don't normally go through all the (n-1) stages, unless there are two
or three choices which are of similar popularity, as in the above case.
A note to those who might actually do one of these counts: it helps to have
two columns per stage, the first showing the added votes from the transfer
(+7 or whatever) and the second the new totals. Use a wide spreadsheet,
table or sheet of paper! As most people will read this on an 80-character
screen, I've just shown the totals columns.
Now to publish the result - that's the table above - plus list the voters.
Remember that they must be listed against their first preference. Assuming
some clever awk script writing, or a lot of manual typing, this can be
done in tabular form. But if you only list people's first preferences, you
should offer the full voting results by mail to anyone who wants to see or
double-check them.
Here are the first few lines of what might have been the voters list for the
above example:
A comp.society.rabbit
A B C D E F N
1 - - - - - - ab@somewhere.net (Adam Bloggs)
B rec.pets.bunnies
A B C D E F N
- 1 3 4 2 5 6 xm@some.company.com (Xavier Moore)
7 1 2 3 5 6 4 FMB09K@frxy89.BITNET (Franco Bolliver)
............. and so on for the 123 rec.pets.bunnies ballots
and so on for C D E F N
For those who missed the article this a follow up to, in which I proposed
rules for deciding new news groups by STV, here it is:
>
> Subject: STV new group proposal in 25 lines
>
> Following the discussion period, the proposal shall be put to the vote, and
> shall include the names suggested during discussion, plus 'no group'. The
> news.announce.newgroups moderator may add any that have been mailed to him.
>
> Voters will mail their votes by putting the names in preferred order;
> 1 first.preference
> 2 second.preference ... (including 'no' if that is one of their preferences)
> and should (not must) state their first preference in the subject line.
>
> Voters' first preferences are counted. If no choice gets over 50% of the
> votes, the choice having the least votes is eliminated, and those votes are
> transferred to the voters' next preference. Repeat until either
> -> the total of all other choices less 'no's is under 100 => vote fails
> -> one name obtains over 50% of the vote (the quota) => that name passes
>
> After the count, the results shall be posted to news.groups, showing the
> count at each stage, and listing voters by their first preferences. The
> list may also show their other preferences in any convenient tabular form.
>
> After 5 days the group will be created, unless irregularities have been found.
>
> For group proposals where discussion leads to agreement on a single name,
> the above procedure will work exactly as the present system - name vs. 'no'.
>
> Regards, David Wright STL, London Road, Harlow, Essex CM17 9NA, UK
> dww@stl.stc.co.uk <or> ...uunet!mcsun!ukc!stl!dww <or> PSI%234237100122::DWW
> "The principle of proportional representation is very simple; election
> requires a quota of votes, not a majority, and votes that would otherwise be
> wasted on a candidate who does not need them are utilised by transfer, on
> those voters' instructions, to other candidates" (Enid Lakeman, ERS, 1971)
ERS is the Electoral Reform Society of G.B. and Ireland. Enid Lakeman was
its Director for many years; now in her mid 80's, she is still campaigning
for fair electoral systems.
DISCLAIMER: I like rabbits. I have kept rabbits. But I have no intention
of proposing a rabbit news group on the net. I don't even read rec.pets.
The subject of the above discussion is purely by way of an example. Yes I
know *you* realised that, but someone somewhere out there probably didn't :-)
Regards, David Wright STL, London Road, Harlow, Essex CM17 9NA, UK
dww@stl.stc.co.uk <or> ...uunet!mcsun!ukc!stl!dww <or> PSI%234237100122::DWWbbc@nysa.rice.edu (Benjamin Chase) (11/26/89)
In article <2440@stl.stc.co.uk> dww@stl.stc.co.uk (David Wright) writes: >Stage 5: the NO vote is transferred! The NO vote should not ever be transferred. The people voting no don't have any other choices, and yet must be heard. If the NO category is on the bottom of the heap, choose the second least popular candidate, and redistribute those votes. -- Ben Chase <bbc@rice.edu>, Rice University, Houston, Texas (First one up against the wall when the fish police arrived.)