[EM] IRV vs Condorcet - a challenge

Rob Lanphier robla at robla.net
Mon Jun 19 21:30:29 PDT 2006

Hi Dave,

I think you're right that examples like the version you constructed need
to be pointed out.  A similar breakdown occurs in the "Tennessee
example" on Wikipedia:

In both examples, the second-to-last candidate keeps picking up just
enough votes by elimination to stay in the race, knocking out the
Condorcet winner along the way.  This would happen even though the
Condorcet winner would often trounce the second-to-last candidate
pairwise.  It might be helpful to come up with an illustrative name for
that phenomenon, so that we can call it when we see it.  Suggestions:
*  "snowballing loser problem"
*  "snowballing extremist problem"
*  "snowballing outlier problem".
*  "ladder climbing loser problem"
*  "ladder climbing extremist problem"
*  "ladder climbing outlier problem".

I have more thoughts on this in a separate email.


On Mon, 2006-06-12 at 12:22 -0400, Dave Ketchum wrote:
> I see ranked choice, represented by IRV and Condorcet, as deserving 
> serious consideration.
> I offer an example emphasizing an IRV weakness - not expectable 
> frequently, but earns really ugly publicity when it happens.
> Some competing methods:
>       Plurality - pretty well agreed to be a loser.
>       PR - worth considering when a district can be constructed for 
> electing multiple members.
>       Approval - a possibility when ranked choice ballots seem too 
> difficult to implement.  I call Approval unacceptably weak for not letting 
> the voter indicate multiple levels of preference such as desired, 
> acceptable, tolerable, reject.
> Some features:
>       Permitting equal ranking for multiple candidates.  Desirable for 
> pleasing those who call Approval desirable; doable with Condorcet, though 
> there can be arguments as to counting' Not doable as practical for IRV 
> (explain how if you disagree).
>       Demanding that voters rank all candidates.  UNACCEPTABLE!  Voters 
> CANNOT be forced to do intelligent ranking among candidates they lump as 
> rejects - especially when there are many rejects.
>       Permitting voters to rank all candidates.  ESSENTIAL for at least 
> the voters' top three choices; more is better.
> I offer a vote for choosing when elections should be held:
>   8 MAR,APR
>   9 APR
> 10 MAY,APR
> 26 SEP,OCT
> 23 OCT
> 24 NOV,OCT
> Condorcet will recognize OCT as winner, over MAR,MAY,NOV,SEP,APR.
> IRV will not see OCT desirability, so will award APR after discarding 

More information about the Election-Methods mailing list