[EM] Second (and higher)-order methods?

[robert bristow-johnson]

On 4/30/12 10:50 PM, Paul Kislanko wrote:
> I think you misunderstood.
>
> If I have to rank the 31 baskin-robbins flavors 1 to 31, it is difficult.
> I'd get tired of the exercise after about the 5th choice and whatever you
> infer from my 6th-31st choice is based upon unreliable data.
>
> If you ask me successively whether I like flavor A better than flavor B for
> all 465 pairs it is easy for me to pick one or the other.
>
> All I'm suggesting is that if you ask me to rank them 1-31 and then try and
> infer from that how I WOULD HAVE put 1's and 0's into a pairwise matrix had
> I made pairwise-choices you'll probably get it wrong because I didn't
> construct the 1-31 list that way. If you asked me the 465 separate
> questions, you wouldn't have to infer anything.
>

this is different from the email sent to me, Paul.  but i'll try to 
answer both.

the premise that i am not yet giving up on is that pick any particular 
voter, that rational voter has a favorite candidate.  (i am not sure 
that a rational has a favorite ice cream flavor.)  now take that 
favorite off the candidate list and *force* that voter to consider the 
remainder, that voter will again have a favorite candidate (unless 
he/she hates all of the remaining candidates equally and would not 
bother to vote if his/her original favorite is removed from the list).  
take both the 1st and 2nd favorite offa the list and force the voter to 
consider the remainder and that voter may have another favorite (that's 
#3).  i really do not see how a rational voter would prefer #3 over 
their original favorite candidate.

-- 

r b-j                  rbj at audioimagination.com

"Imagination is more important than knowledge."