[EM] Re: [RangeVoting] further reply to Parr... et al... 2-party dominance recap

[Adam Tarr]

I've copied the EM list on this second message.

On 8/13/05, warren_d_smith31 <wds at euclid.math.temple.edu> wrote:

> Let me return to the (more important) question of 2-party domination being
> caused by Condorcet methods, which is a question we presumably
> can analyse fully within the simpler 3-candidate case, thank heavens.
> 
> I have been unfairly (I think) disparaged by Parr.
> Let me recap what my original proof (which started this thread) accomplished.
> 
> It showed in a large-random-electorate mathematical model, with at least
> 25% probability there will exist some subset of voters, who could
> not elect their favorite "3rd party candidate"
> no matter how they changed their votes, but who could by ranking
> him dishonestly second (and NOT co-equal first - that will not suffice),

You are correct that co-equal first will not suffice when defeat
strength is measured by margin of victory, but co-equal first will
suffice when defeat strength is measured by winning votes.  I am a
winning votes advocate, and this is one of the main reasons.

For the rest of the message I will refer to this strategy as "favorite
burial".  By this, I simply mean equal rankings at the top of your
ballot.  I consider this to be not especially distasteful, and after
all, this is the same strategy that is used in approval and range
voting.  But even if you don't believe this, most of the following
comments still apply.

> force the election of the least-worst of the two major-party candidates.
> (If they voted honestly, they'd instead get the worst.)
> This holds for any Condorcet-type election method whatever.
> 
> Furthermore,  100% of the time, if those voters
> BELIEVE their favorite third-party canddt has much less than 25% chance
> of winning,  then they can change their votes to rank him dishonestly 2nd,
> and this decision will in *expectation* help them -
> since it helps 25% of the time, and the remaining 75% of the time,
> they believe it only hurts them with a much smaller than 25% probability.

This is only true if they also believe that the electorate is
uniformly distributed about all preferences, since only in that
condition does favorite burial help them 25% of the time.  This seems
to be an absurd set of assumptions - they know they have virtually no
chance of winning, but they know nothing about the electorate.

In reality, if you know enough about polling to know that your
favorite can't win, then you likely also know whether there is any
sort of cycle in the preferences of the electorate, and if so, which
way it cycles.  Only if a cycle exists and rotates a specific way does
favorite burial help.

> In the USA, third party candidates for reasonably high office
> DO experimentally have a much smaller
> than 25% chance, in fact <1% chance, of winning.  Therefore, this
> belief is entirely rational. 

Another entirely rational belief in the USA would be that (for
example) the Greens do not have pairwise victories over EITHER the
Democrats OR Republicans.  As such, the Greens are not involved in a
cycle, so favorite-burial has a <1% chance of hurting them, but a <1%
chance of helping, either.

 Therefore, by individually acting rationally-strategically
> to improve the world by their own measures, Condorcet voters guarantee
> that 3rd party candidates can never be elected.  This is the
> same kind of vicious self-reinforcing cycle as under plurality.

No, your argument ONLY applies if we don't know whether there is a
cycle, and what way it points.  If there are even marginally accurate
polls, then voters will know going in whether or not favorite-burial
is required or not.  When it is not, which includes any time there is
not a cycle, then sincere voting is a perfectly rational strategy.

> Therefore, we expect under Condorcet (or IRV, which I also proved
> by a parallel argument), 

Actually, there are simple IRV examples, that do not involve cycles,
where full order-reversal is needed.  A simple example (names of
candidates are from the perspective of the backers of "favorite", of
course):

45 Worst>Compromise>Favorite
10 Compromise>Worst>Favorite
10 Compromise>Favorite>Worst
25 Favorite>Compromise>Worst

This example is obviously reasonable - it's just a linear political
spectrum example.  In this case, some of the Favorite>Compromise>Worst
voters must reverse to Compromise>Favorite>Worst in order to prevent
Worst from winning.

In my opinion, it is these sort of incentives that have driven
countries using IRV toward a 2-party system - not anything related to
cyclic preferences.


> 2-party domination. I'm really quite surprised anybody possibly could now 
> argue with this. But there are some very stubborn individuals on the EM list, 
> Parr among them, who apparently feel obliged to try.

This Parr fellow sounds like quite a bastard.  I hope I never meet him.

> In contrast to all this:
> with RANGE VOTING, even if 100% of all the voters
> in the universe are convinced Nader has almost zero chance,
> but 53% of them think he is objectively better than Bush and Gore, then, if all act
> rationally strategically, Nader would win.  Guaranteed.

Sincere preferences

53% Nader>Gore>>>>>>>Bush
10% Gore>Nader>>>>>>>Bush
10% Gore>>>>>>>Nader>Bush
27% Bush>>>>>>>Gore>Nader

Strategic votes

63% Nader 99, Gore 99, Bush 0
10% Gore 99, Nader 5, Bush 0
27% Bush 99, Gore 0, Nader 0

Gore wins.  Let me know if I missed something.

Finally, in all these methods (Condorcet, approval, range), a
significant effect is simply that voters have the ability to express
more preferences.  This will affect polling questions, and it will
affect the way results are perceived, and as such will give third
parties more of a voice.  I hesitate to use this argument to bolster
IRV, since any close multi-candidate race is such a nightmare in IRV,
cycle or no cycle.

-Adam