First, a reposting for Jobst:<br>
CIBR stands for "Correlated Instant Borda Runoff," and is a tweak of Baldwin to
solve the clone problem. <br>
<br>
Individual ballots are scored according to the Borda count, and then
all possible candidate pairs are ranked according to correlation.
The Borda loser of highest correlated pair is eliminated, and the next
round proceeds with the remaining candidates. <br>
<br>
Under this method, clones can be seen as a special case of correlated
pairs (they're perfectly correlated), and so the Borda loser of a clone
pair is eliminated immediately, before they can spoil the election. <br>
<br>
Now to answer Forest's questions: <br>
On 5/25/05, <b class="gmail_sendername">Simmons, Forest</b> <<a href="mailto:simmonfo@up.edu">simmonfo@up.edu</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
This looks promising. I like this kind of creativity.<br><br>Three Questions:<br><br>1. Exactly how do you define correlation?<br><br>2. Do you re-calculate the correlations after each elimination?<br><br>3. What about clone triplets, quintuplets, etc.?
<br><br>Forest<br><br></blockquote></div><br>
I have to admit that I struggled with how to define correlation. <br>
I looked at the standard formulas, and decided that I was either not smart<br>
or patient enough to determine how to apply them to ballots. So for my <br>
working model, I came up with a custom function that works as follows: <br>
<br>
For each possible candidate pair on each ballot, I calculate the Borda value<br>
of the rank difference between the two candidates to get their "team score". <br>
For example, if candidate "A" is in position 1 & candidate "C" is in position 3,<br>
their rank difference is 2. With 5 candidates, the Borda value of position 2 is 3,<br>
so that's the "AC" team score for that ballot.<br>
<br>
The team scores across all ballots are then summed, and the team with the <br>
highest score is the most correlated. <br>
<br>
Let's work through a full election involving a clone (BC): <br>
51:A>B>C<br>
49:B>C>A<br>
<br>
This results in the following first round counts: <br>
Ballots
Borda
Value
Team Points<br>
A B
C
AB AC BC <br>
51:A>B>C
102 51
0
102 51 102 <br>
49:B>C>A 0
98
49
49 98 98<br>
-------------------
----------------------<br>
Totals:
102 149
49
151 149 200<br>
<br>
BC has the most team points, so the Borda loser between them,<br>
C, is eliminated. <br>
<br>
In the next round, we recalculate the correlations (I think this is <br>
necessary to compensate for clone triplets & multiples), so the <br>
tally goes: <br>
<br>
Ballots
Borda
Value
Team Points<br>
A B
AB<br>
51:A>B
51
0
51<br>
49:B>A
0 49
49<br>
-------------------
--------------<br>
Totals:
51
49
100<br>
<br>
<br>
Since there's only one team left, it's Borda loser, B, is eliminated <br>
and A wins the election.<br>
<br>
-Ken<br>
<br>