[EM] IRV in action

[James Gilmour]

Dave Ketchum  Sent: 02 April 2003 18:43
<Lots CUT by JG>
>  >>True IRV weakness that I see:
>  >>      44 candidates got some first place votes and can, possibly, win -
>  >>          even with two to get started with (candidates with only one
>  >>could list them next).
>  >>
I asked:
>  > I don't understand this comment.  What would the candidates list?
>  >
Dave replied:
> Let me try for better wording:  Assuming there was only one ballot listing
> candidate X first, this ballot's second choice could have been Y, one of
> the candidates that started with only 2 first choice votes.  This would
> mean Y now has 3 votes, and possibly can advance as other weak candidates
> are discarded and their votes reassigned.

Thanks.  I now understand completely.


>  >>      Some of the above may have got only one - here the IRV counters
>  >>decide which of these die now and which, if any, get a chance to advance.
>  >>
>  >
>  > What else can you do when you encounter a tie?  In the House of Lords
> by-election
>  > election, the Returning Officer would appear to have made the decision
> to exclude
>  > the tied candidates in alphabetical order.  Our regulations for public
> elections
>  > require the casting of lots to determine which of the tied candidates
> should be
>  > excluded.
>  >
>
> What you BETTER do BEFORE encountering a tie is expect them sooner or
> later and provide rules.  That the Returning Officer apparently had no
> guiding rule above implies a failure of the rule writers to complete their
> job (being given an opportunity to write rules while holding the voted
> ballots gives an opportunity to apply bias toward getting a personally
> desired result).

We have NO evidence to suggest that the Returning Officer did not decide the rule
about ties before the count began.  The full rules do not appear to have been
published, but the news Release does use the standard UK wording to describe IRV
(which we call the "Alternative Vote").  It makes no mention of ties.

Of course, the procedure should always be specified before the count begins.  But
when I codified the IRV rules back in 1978 we were writing generic rules for use
by different organisations whom we knew had differing practices with regard to
ties.  Hence the wording that the Returning Officer should decide.  Legislation
for UK public elections specifies determination by lot, the candidate to be
excluded being the one on whom the lot falls.


>  >>      37 got no first place votes.  These all die instantly, even if one
>  >>of them got 400 second place votes - this happens because, whenever some
>  >>of their second place votes get exposed, those will be discarded as lowest
>  >>count among the remaining votes.
>  >>
>  >
>  > Are you seriously suggesting that electors would accept as the "winner" any
>  > candidate who got NO first preferences  but lots of second preferences?
>   No matter
>  > what social choice theory might say, I cannot see such a candidate ever
> being
>  > accepted in a real election for public office.
>
>
> YET, while IRV:
>        Declares a candidate with ZERO first preferences an instant loser.
>        It declares a candidate with TWO first preferences to have the same
> theoretical chance to build strength with second (etc.) preferences as any
> other candidate with two or more first preferences.
>
> I do not see either of these candidates looking much better or worse than
> the other.  If either got 400 second preferences, that one holds wide backing.

Yes, I understand and accept all of that.  But you have not answered my question.
This is a very real issue for those of us who see the advantages of Condorcet in
dealing specifically with the "everyone's second choice" problem in IRV.  No
matter how sensitive a voting system may be, it has to be acceptable to ordinary
electors who will vote in real public elections.


> Take another look at the 2002 French elections.  We know that Chirac and
> Le Pen get a bunch of first preferences and few if any second preferences.

I don't understand your statement.  There were no "second preferences" in the
French Presidential election.

>    We know that a large field of candidates each gets less first
> preferences, and loses by their rerun rules.  If the election had been
> done by ranked ballots, there could have been a candidate in the field
> that was acceptable to be winner, once you got past the candidates with
> some first preference voting but no depth:
>        IRV with luck:  the acceptable candidate survives and accumulates
> preferences until winning.
>        IRV with no luck:  the acceptable candidate gets crossed out during
> some round, and therefore loses regardless of popularity.
>        Condorcet:  here second preference counts ahead of all except first
> preference.  If X, Y, and Z each only had first preferences ahead of 1/3
> this candidates (A) second preferences, A is nearer to winning than any of
> them (A>X is twice as frequent as X>A).  Even if A had no first
> preferences, it should be more acceptable to the voters than any of X, Y,
> or Z.

I don't disagree with your analysis, but it does not address my question.

To return to the French Presidential election, had this been an exhaustive ballot
or an IRV ballot, I suggest we would have seen a progressive clustering of votes
around Chirac and around Jospin.  I suspect Jospin would then have won.  But the
French had (almost) the worst of all voting systems.

James