[EM] IRV inconsistency

[LAYTON Craig]

Richard wrote:

>It may be sufficient for some, but I would think at least a few voters
would
>demand more complete information about an election that will determine
>many aspects of their lives for the next several years. Exactly where do
the
>numbers come from in each step? If there were 100000 A votes in the first
>round, and A got eliminated, how many were ABC votes and how many
>were ACB votes? This can be determined by the difference between first
>and second round votes for B and C. But how many of the B votes were
>BAC votes and how many of the C votes were CAB votes? If this
>information were published, it might reveal how undemocratic the result is.
>It might show that A is more popular than the winner. Of course, we
>wouldn't want anybody to find out IRV's dirty little secret, would we?

People don't realise that there is relevant information that the
step-by-step results do not reveal, and I have not heard calls for more
complete information to be published.  But even if people wanted more
complete information, it would not be available, because it is not recorded.
The difficulties that some of the list members have had in developing
programs that run IRV stems from the fact that they are trying to record all
of the information on the ballot - this is not how IRV is counted.  In
practice, IRV is counted thus; Add up all first preference votes.  Eliminate
all candidates but the top two, and do a pairwise count.

In some cases you can't eliminate all but the top two (ie where the total of
first preference votes for candidates other than the top two is more than
the number of first preference votes for the no. 2 candidate), but this
actually happens in a minority of cases (I don't have figures, but I would
guess about 20% or 30% of the time).

It isn't just that the information that makes IRV look bad isn't published,
it doesn't exist [in a Orwellian sense]

Craig