[EM] Dectecting Clone Sets

[Juho]

How should we continue from this? Should we divide the seats  
proportionally at top level to the top level branches of the tree, and  
then repeat the process towards the leaves.

What are the main differences between the tree if derived from the  
ballots using clone analysis vs. if given by the candidates (before  
the election)?

There might be also richer connections than in a tree (e.g. based on  
political views and based on geography). How about allowing the  
neighbours to inherit the votes in any direction based on the derived  
"cloneness" relationships?

Juho



On Nov 16, 2009, at 9:44 PM, fsimmons at pcc.edu wrote:

> Here's a suggestion for detecting clone sets based on Range Ballots:
>
> Define the distance between two candidates as the square root of the  
> sum (over the ballots) of the squared
> diffference of their respective ratings.
>
> If the ballots are approval style, this becomes the square root of  
> the number of ballots on which just one of
> the two candidates is approved.
>
> Use these distances to do a cluster analysis of the candidates.
>
> The tree structure of the clusters gives the tree structure of the  
> clone sets.
>
> If you have a clone dependent method, like Copeland, that you would  
> like to "de-clone," you might find this
> clone set detectiion method to be useful.
> ----
> Election-Methods mailing list - see http://electorama.com/em for  
> list info