[EM] Real IRV Election, Disputable Result

James Gilmour jgilmour at globalnet.co.uk
Tue Mar 14 11:26:53 PST 2006

> Dave Ketchum Sent: Tuesday, March 14, 2006 5:48 PM
> With IRV, tie-rank votes need to either be prohibited or the exact way of 
> accounting for them defined (counting them with each tied candidate 
> holding the same rank encourages taking advantage of such 
> voting).

In fact, "tie-rank votes" make no sense in IRV if you recognise the origins of IRV and the purpose of the preferences in
IRV.  The idea of "tie-rank votes" in IRV has been brought in by those whose approach is based in social choice theory.

IRV is an "instant" recording of what the voter would do in a sequence of exhaustive ballots (exhaustive run-off) where
only one candidate is excluded at each round (ie "exhaustive" as opposed to "top-two run-off").  There is no possibility
of casting a "tie-rank vote" in an exhaustive ballot, so from this perspective, "tie-ranking" has no place in IRV.

The second and lower preferences marked in IRV are contingency choices.  Your second preference comes into play ONLY in
the event that your first choice has been excluded.  So from this perspective also, "tie-rank votes" do not fit with the
purpose of IRV.

Of course, you can modify IRV to allow "tie-rank votes", but then it is no longer "IRV", but a different voting system.
I suspect that if you want a voting system more in tune with the social choice approach (the reason for introducing
"tie-rank votes"), it would be better to start with something other than IRV because IRV comes from a different stable.

>  With Condorcet, tie-rank should be normal, for voters can like it 
> and it gives no special advantage to users with the round-robin counting used.

Tie-ranking certainly fits better with Condorcet, and as you say, it will not affect the result because it is the
equivalent of the voter saying "count me out for this round" when the head-to-head is between two candidates that voter
ranks equally and so makes no call.  You would, however, have to be sure that "tie-rank votes" also had no effect on the
various methods that have been suggested for breaking Condorcet cycles when they occur.

> BTW - IRV claims to getting a majority winner need understanding.  Try 
> voters who do not rank every candidate and could vote 25A  30B  35C.  IRV 
> will discard the 25A and declare C the winner among the 65 votes still 
> being counted.  Agreed C properly wins, but total votes were 90.

This is a very important point.  Some promoters of IRV don't make this clear and some critics of IRV misuse this
information.  There is nothing to hide here, nor is it a cause for criticism.  If IRV is seen for what it really is, an
"instant" recording of a sequence of fully exhaustive ballots, the difference between the initial 90 votes and the final
65 votes in the example above is nothing more than those voters who said (for what every reason) "count me out  -  I
have no further preference to express  - I am content to leave the decision to those who do have preferences relating to
the continuing candidates".  Just as in an exhaustive ballot, the eventual winner is the candidate who has a majority of
the votes cast AT THAT STAGE.  If large numbers of voters away at home for the later rounds of the exhaustive ballot,
that is their choice.  They could have participated, but didn't.  So in IRV, the voters can mark many preferences or
very few  -  that's their choice.  Of course, the result sheet should always make clear the total numbers of votes cast
and the complete pattern of transfers at every stage.

James Gilmour

More information about the Election-Methods mailing list