[kwlug-disc] What is all this about systemd?

[Hubert Chathi]

On Wed, 19 Nov 2014 18:49:19 -0500, William Park <opengeometry at yahoo.ca> said:

> You just lost me.  I don't understand why you need 2D table for vote.
> I thought voting is choosing one of N options.

Hmm.  Yeah, and this is why, despite Condorcet being all sorts of
awesome, it will not be accepted by the general public for a long time.
People look at the table and get all confused.  Although the graph that
you get is nice.

For the sake of those completely unfamiliar with Debian's voting system,
I'll start at the beginning.  For those who already know some of this,
feel free to skip over the parts you know, or correct me if I say
something wrong.

Most people are familiar with the voting system generally referred to as
First Past the Post (even though that term is a nonsensical description
of the system), and more formally referred to as Plurality voting.  It's
the voting system used in most of our general elections in Canada.
Basically, each voter selects only one choice out of the available
choices, and the choice with the most votes wins.

Despite being simple and easy to understand and to execute, it is a
truly terrible system.  Many people are familiar with the term "vote
splitting", which is an effect of this system.  It also encourages
strategic voting (as a result of the vote splitting effect) -- that is,
voting contrary to what your actual preferences are.  To cite a recent
example, and without making any claims to whether it actually made a
difference to the results, Olivia Chow probably received fewer votes
than she would have otherwise received, due to people voting for John
Tory in order to keep Doug Ford out.  Closer to home, Ken Seiling likely
received more votes than he otherwise would have, due to people voting
against Jay Aissa.  The American two-party system is largely a result of
First Past the Post, and it's somewhat of a mild miracle that we have
more than two major political parties, and even more of a miracle that
we once had five major political parties at the federal level.

A mildly better voting system is known as approval voting, in which
voters select all the options that are acceptable, and the option with
the most votes wins.  This method somewhat alleviates the issues of vote
splitting and strategic voting, but still has some obvious problems.

After this, there is a class of voting systems known as ranked voting,
in which voters rank the options from most preferred to least
preferred.  The difference between the systems lies in how the rankings
are handled.

One method that may be familiar to some is known as Alternative Vote, or
Instant Runoff Voting.  This method is used by most of our major
political parties in their leadership races.  Basically, all the
first-place votes are counted up, and if one option gets the majority of
votes, then that option wins.  If no option gets the majority of votes,
then the option with the least number of votes is dropped, and for any
ballots that ranked that option first, their second choice (if any)
becomes their new first choice, their third choice (if any) becomes
their new second choice, and so forth, and the process is repeated.

There are several other ranked voting methods, but I only mention that
one because some may be familiar with it due to its use by some of the
political parties.

The method used by Debian is referred to as Condorcet, though Condorcet
itself can refer to multiple methods.  The specific method that Debian
uses is called the Schulze Method[1].  Basically, the way that it
works, as I mentioned in my last email, is that each vote encodes the
voter's preference between any two given options.  For example, suppose
there is a vote with three options, A, B, and C.  A vote of "2,1,3"
means that the voter has ranked option B first, A second, and C last.
This is interpreted to mean: the voter prefers B over A, they prefer B
over C, and they prefer A over C.

The votes are then tallied using this interpretation.  That is, the
votes that prefer A over B are counted, and compared with the votes that
prefer B over A.  That's what the 2D table is: a count of the number of
votes that prefer one option over another.  So to see if A is more
preferable to B, or vice versa, you see which number is bigger.  Each
pair of options is considered, and so this is essentially performing
simultaneous pairwise voting on all possible options.

If there is any option that beats out every other option (that is, in
each pairwise comparison, that option is preferred over the other), then
that option is the winner.  That property is common to all the Condorcet
methods.  The difference between the Condorcet methods is what happens
when there is no such option, but I'm not going to get into that,
because for one thing, I'm not very familiar with exactly what happens,
and for another thing, I don't think that has ever happened in a Debian
vote.  At least, I am not aware of it actually happening.

Compared to First Past the Post, the Condorcet methods more or less
eliminate vote splitting and strategic voting.  There is virtually no
reason to vote contrary to what your actual preferences are, and there
are no problems with having many similar options on the ballot[2].

[1] https://en.wikipedia.org/wiki/Schulze_method
[2] https://en.wikipedia.org/wiki/Independence_of_clones_criterion

So, in a nutshell, that's the system that Debian uses.  Let me know if
there is still any confusion.