[EM] Debian leader 2006

[raphfrk at netscape.net]

From: Warren Smith <wds at math.temple.edu>
> I suspect RobL's method for converting ranked ballots to range and 
approval
> ballots tends to be distortionary and favor better known candidates 
as opposed
> to better candidates.   (I.e. unknowns maybe tend to be left unranked
> by the voters).  Or I just converted the data wrong.

A better way would probably be to assume voters vote strategically.
Perhaps an iterative method

e.g.

1) All voters approve of the candidate at their approval threshold and 
above.

2) This threshold is set at their first choice initially.

3) Run count

4) Any ballot which doesn't approve of one of the top 2 approved 
candidates has
its approval threshold moved down 1 rank.

5) If the threshold on any ballot changed due to step 4, goto step 3)

6) The candidate with the highest approval "wins"


In fact, I wonder if that might be a good way to count ranked ballots, 
rather than using
IRV or condorcet.  Has something like that been suggested before ?

Also, if voters are allowed to skip ranks, then compromising could be 
incorporated
into a ballot.

For example:

A: 1
B: 2
C: 5

would quickly compromise to get B elected, but would wait longer before
approving of C

OTOH,

A: 1
B: 5
C: 6

would represent a voter who wants A to be elected and won't compromise
unless necessary.

A highest ranking would probably be necessary so that people don't rank

A: 1
B: 1,000,000

> Anyhow, I have a question.  Look at the Instant Runoff results in 
detail.

> It looks to me like the electowidget printout is erroneous i.e prints 
out the
> wrong guys in each round.

Yeah, looks like you are right.  For example, the lowest candidate in 
round 1
is Jonathan (Ted) Walther

He has 2 votes and is eliminated.  This means that the votes after 
elimination
should be almost identical for all the other candidates (2 candidates 
just get
1 more vote).

However, he isn't actually eliminated, but his 2 votes are transfered 
and his
name is put beside the votes for Jeroen van Wolffelaar who shouldn't 
have
been removed.

Round 1:

1) Bill Allombert   36
2) Steve McIntyre   93
3) None Of The Above    14
4) Ari Pollak   7
5) Andreas Schuldei 50
6) Anthony Towns    128
7) Jonathan (Ted) Walther   2
8) Jeroen van Wolffelaar    43

Round 2:
1) Bill Allombert   37
2) Steve McIntyre   93
3) None Of The Above    14
4) Ari Pollak   7
5) Andreas Schuldei 50
6) Anthony Towns    128
7) Jonathan (Ted) Walther   44

What seems to have happened is that he didn't transfer the names.

Line 1 before and after is Bill Allombert

Line 7 before and after is Jonathan (Ted) Walther

If he had eliminated the name Jonathan (Ted) Walther, rather than
just removing the bottom name, the totals would be correct.

This means that Bill Allombert and Steve McIntyre will make it
to the final round no matter what votes they get as they are the
top 2 candidates listed.

It does seem to pick whichever one of them has the highest score,
but I am not sure if that is in fact in any way related to the votes
that they actually obtained.