[EM] Condorcet strategy and anti-strategy measures

[James Green-Armytage]

Dear Mike, and all election methods fans,

Mike wrote:
>First, I don't believe that Condorcet wv has a serious strategy problem. 
>We've often discussed the defensive truncation strategy. Here's how I
>often 
>word it: The only way you can steal the election from other voters is if 
>they were trying to help you. Doesn't that make you proud of yourself?

	I do believe that Condorcet wv had a serious strategy problem.
	While it is unknown how likely a successful execution of the burial
strategy would be in a public election, I believe that if it did occur, it
would produce an *extremely* negative reaction towards both the result and
the election method itself. 
	Imagine if Kerry beat Bush by a few hundred thousand votes, and yet Bush
actually won the election because most of the Bush voters strategically
ranked Nader second, creating a fake Nader > Kerry victory to knock Kerry
out of the running.

	Something like this, for example.
44%: Bush > Nader > Kerry
5%: Bush > Kerry > Nader
38%: Kerry > Nader > Bush
5%: Kerry > Bush > Nader
8%: Nader > Kerry > Bush
Bush : Kerry = 49 : 51
Bush : Nader = 54 : 46
Kerry : Nader = 48 : 52

	If this was a single-balloting wv Condorcet election, and Bush won (as
above), there would be an enormous outrage, and it would be entirely
justified.
	In that sort of situation, you would really want to hold a second
balloting. In between ballotings Nader would be given the opportunity to
drop out of the race, and hopefully he would do it (based on the idea that
significant number of the Bush > Nader > Kerry votes are strategic). Then
you would have a straight contest between Bush and Kerry, which Kerry
should win.
	Is an example just like this very likely? No, I guess it's probably
pretty unlikely. (However, I think that there are other kinds of examples
where successful burying strategies are a bit easier and more plausible...
I think that I've provided some of these examples already...) But anyway
it is *possible*, and the possibility of such a disaster justifies at
least one more balloting.
	And yes, Mike, I *do* think that just one more balloting in the event of
a majority rule cycle would be a great help. 
	Let me put it this way: Without a second balloting, I would not
necessarily feel comfortable about recommending Condorcet for public
elections instead of something like IRV. 
	Not because IRV is a good election method (it isn't, really), or because
the Condorcet principle is unimportant (I think that it's extremely
important), but because a successful burial strategy or an out-of-control
chicken game in single-balloting Condorcet would just undermine the
democratic process too severely, and I wouldn't feel comfortable in
opening up the possibility for that. 
	With a second balloting, I would say that Condorcet is vastly superior to
IRV and pretty much any other public election method around.

	Mike, in your analysis of Condorcet strategy, you seem to be assuming
more coordination, more communication, and more time for people to respond
to each other's strategies than what I think is necessarily realistic. In
a previous e-mail on this topic, you made many statements similar to the
following:

>To the extent that there's any perceived danger of offensive
>order-reversal, defensive truncation will be employed as a
>countermeasure. 

	The thing is, the people who would be employing the "countermeasure" will
not know whether the strategy they are attempting to counter has taken
place, until after the election! (And even then they might not be sure!)
And of course, at that point, it's very much too late to change whether or
not you truncated.
	As you know, there are various examples where if voters truncate against
a strategic threat that in fact does not exist to begin with, it can
result in the election of the car crash candidate in itself. So voters
can't always take lightly the decision to truncate against a candidate who
may or may not be supported by strategic buryers.
	You seem to assume that a strategy of Dean voters to bury Kerry would
involve someone in the Dean high command calling up various supporters and
telling them to vote against their conscience. Then of course, the Kerry
high command would find out about it, and call up the Dean high command
and say "Hey! We know what you're up to! So if you don't tell your guys to
vote sincerely, we'll tell our guys to truncate against Dean!" In which
case it would be fairly cut and dry. 
	But in reality it doesn't have to happen like that. Voters for Dean and
Kerry can all become *aware* of the fact that if they bury their second
favorite, that it might help elect their favorite, and then again it might
help elect their least favorite. Then, what happens on election day
doesn't have to be determined by any coordinated strategy efforts. Before
the vote no one will necessarily know who will try to bury and who won't.
People who do intend to bury might decide not to mention it if they happen
to be polled.
	What it may come down to is: To what extent do voters on each side have a
propensity to gamble with burial strategies? To what extent do voters on
each side expect voters on the other side to engage in burial strategies.
	The result of such a situation is unpredictable even if you know the
sincere preference rankings, and that unpredictability is very dangerous.
It is quite possible that successful burial will take place without anyone
on the other side stopping it. It is also possible that a combination of
mutual burial and mutual truncation (due to either aggressive or paranoid
voting) could lead to the election of the car crash candidate. Both
possibilities are chaotic and fundamentally undemocratic.
	The problem is just that there is no time to respond to the use of
strategy. The purpose of extra rounds is precisely to provide the time and
opportunity for such a response.

Mike also wrote:
>James A. also suggested some many-balloting (more than 2 ballotings) 
>enhancements, but it seems to me that it would be much more difficult to 
>propose to the public something that can need more than 2 ballotings.

	That is quite true. I think that 2 balloting wv Condorcet (2nd balloting
only in the event of a cycle, and with the option for candidates to
withdraw *in between* ballotings) is an extremely good proposal for public
elections. This is what I would propose for a public election if anyone
seriously asked me now.
	The multiple balloting thing would be nice, I think, but I agree that it
is very unwieldy for very large-scale public use. 
	I think that in a two-round procedure, candidate withdrawal would play a
more important role in effectively neutralizing strategy. The best way to
assure that no abuses would be effective in the second round would be for
a set of candidates to come to some sort of agreement whereby all but one
of them drop out of the race before the second balloting, so that the
opposition would not be able to use one of them to 'impale' the other one
with a huge artificial defeat and leave the legitimate defeat of their own
candidate as the defeat of least magnitude. There is still plenty of
opportunity for cooperation to break down during the negotiation stage
here, but at least it isn't possible for a strategic maneuver to steal the
election completely by surprise, as it is in the single-balloting version.
	
	Although the two-round version is probably good for public elections,
when you are voting in a reasonably small group, one where the cost of
repeated elections is very minimal (for example a legislature, a club,
etc.), I would definitely definitely recommend the multiple-round
procedure over the two-round procedure. That is, take a Condorcet vote,
and then follow it with a yes-or-no vote on the winner, whether the winner
is a pure CW or a completed winner. This repeats until some option has won
its yes-or-no vote, or until everyone gives up in frustration.
	This procedure has very real benefits over the 2 ballot procedure, and so
it should always be used when the cost of additional ballotings is very
low.

	Going back to the two-ballot procedure, I still think that the second
balloting should be a Condorcet balloting. I prefer a Condorcet balloting
to an approval balloting here for all the reasons that I prefer Condorcet
to approval in general.

	And oh, Mike, my name is James Green-Armytage. "Green" isn't my middle
name, but is rather a part of my surname. Hence "James A." is not an
appropriate abbreviation for my name. If you feel the need to abbreviate
it, I suggest "James G-A". There's nothing at all wrong with the name
"James Armytage", except for the fact that it's not my name. (Other than
that, I rather like it.) Green-Armytage was my father's surname, and also
the surname of his father, and so on.

best,
James