[OT] Voting systems [was Wrong ad on www.linux.org]

[Henry Spencer]

On Fri, 2 Jul 2004 cbbrowne-HInyCGIudOg at public.gmane.org wrote:
> > > Simply, in each riding, you pick a candidate to win with probability
> > > proportional to the number of votes he/she received.
> 
> Unfortunately, it leaves two problems:
>  1.  It doesn't permit any way of reacting to a need for a recount;

You don't really need recounts.  The point of a recount is to verify the
decision in a very close race, but in this scheme, all a recount can do is
change the odds very slightly, which is very unlikely to make a difference
to the result.  You can simply live with that. 

>  2.  It mandates having a _very_ good way to choose randomly.
>      I have read _Seminumerical Algorithms_, which is a pretty
>      canonical text on the matter; understanding that the method
>      is good requires knowing mathematics that to the average Joe
>      is pretty much majick.

You're a computer guy thinking in terms of computer methods.  The key here
is *don't use a computer*.  In any case, you're not after a pseudorandom
number, which is what Knuth is agonizing over; real randomness is what's
called for here. 

Say in your district, 34,000 people voted Liberal, 27,000 voted NDP, and
9,000 voted for assorted fringe parties :-), making a total of 70,000. 
Use one of those drum-full-of-numbered-balls machines they use (or used
to, last I looked) for the 6/49 lottery, except load it with a couple of
hundred balls with numbers 0-9.  The first five out give you a number from
0 to 99,999.  If it's 70,000 or higher, toss those balls back in and try
again.  If it's 0-33,999, the Liberals win.  If it's 34,000-60,999, the
NDP wins.  And likewise for the fringe parties.  The machine is easily
audited, and you do the actual selection with the TV cameras live.

                                                          Henry Spencer
                                                       henry-lqW1N6Cllo0sV2N9l4h3zg at public.gmane.org

--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml