[EM] Probability of losing a sincere Cond. Winner

[Adam H Tarr]

>With my scenario, I would need to track and convince 11% of the 18%
>potential truncaters in order to steal the strongly supported CW with
>(wv) (more than half).
>
>With your scenario, you would need to track and convince 47% of the 49%
>potential truncaters in order to steal the poorly supported CW with (rm)
>
>and 48% of the 49% potential truncaters (almost all of them) in order to
>steal the poorly supported CW with (margins).

Well, I had cooked things to give Ralph just about the minimum support to 
make the example work.  I could just as easily cook it the other way, 
just by strengthening Ralph's faction:

49%: George>Al>Ralph
6%: Al>George>Ralph
6%: Al>Ralph>George
39%: Ralph>Al>George

Note that Al is still the CW and still has a majority over every 
candidate -- I haven't changed the fundamentals of the situation at all.

Now, only 25% of the 49% in the George>Al>Ralph faction need to truncate 
to foul up the results.  Thats comparable (favorably comparable actually) 
to the 11% of 18% that you cited.  If I really wanted to skew the results 
here, I could set it up so you only needed around 7% of the George 
faction to truncate.  Again, I could do this without fundamentally 
changing the nature of the election; Al would still be a CW with majority 
support over the candidates on either side.  So really, we're just 
talking about an artifact of the example I chose.

The more important issue is that it's very easy for these problems to 
crop up in margins, and it's easy to recognize them when they do, and 
it's easy to foul up the results just by truncating (which is a very 
natural thing for the voter to do).  In the winning votes examples, the 
problems only show up when there's a lot of truncation to start with, and 
then they only show up in a pretty fractured race where it would probably 
be hard to tell whether truncation will help or hurt you.

-Adam

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em