[EM] 02/02/02 - Alexander, don't get stuck in a `Time Warp':

[Forest Simmons]

On Sat, 2 Feb 2002, Adam Tarr wrote in part:

<snip>
> 
> Now that that is out of the way, I can jump to a specific example.  Now, I 
> agree that IRV works fine at removing the "spoiler" problem.  That is, as 
> long as two parties remain dominant, IRV prevents minor third parties from 
> interfering.  This is why and how it has "worked" in the single winner 
> Australian elections; it has kept the third party down.  Every voting 
> reform (except maybe Borda) does a good job of preventing the spoiler 
> problem.  The problem with IRV arises when the third party is no longer 
> guaranteed to lose.
> 
> Imagine, for a moment, that I am heading to the polls to vote for President 
> of the USA in 2000.  Only in this reality, Nader is far more popular in the 
> polls than he was... basically all three candidates are rather close, with 
> Bush slightly ahead of Nader, who is slightly ahead of Gore.  But all three 
> are very close.  The question is simple... who do I vote for if I support 
> Nader, but I like Gore more than Bush?  (obviously not an unlikely situation).
> 

<snip>

> 
> In IRV... how you vote depends on how much you trust the polls.  If you 
> assume (it seems reasonable) that the Gore voters will split between Nader 
> and Bush about evenly, then it looks like Bush will win the election, since 
> he's a little ahead of Nader in the polls.  But maybe the Gore voters will 
> break over to Nader enough to pass Bush in the second round, or maybe Nader 
> is actually ahead and the polls are off.  So you could either cross your 
> fingers and vote Nader, Gore, Bush, or you could switch your top two and 
> vote Gore, Nader, Bush.  That way, you could help beat your top choice 
> (Nader) but make sure all the Nader voters join up with Gore and beat 
> Bush.  You would be hedging your bets.  But what if Nader would have beaten 
> Bush in the runoff?  Things are close enough that it's very possible.  You 
> have to vote based on how accurate you think the polls are, and how evenly 
> you think Gore voters will split.  Not an easy decision.
> 
> Furthermore, if Bush voters think that Bush is a safe bet to make the last 
> round, but Nader is likely to lose a close race to Gore (maybe because of 
> those insincere Nader voters), they can actually make Bush win the election 
> by putting him at the bottom of their list behind Nader.  Basically, they 
> are giving their votes to Nader in the hopes that they make Gore lose in 
> the first round, get half the Gore voters in return and get Bush to win the 
> runoff.  A risky vote, and a bizarre result, but quite possible.  This is 
> the problem with IRV in close multi-candidate races: things become erratic 
> and voters have all sorts of non-intuitive options on how to "play the 
> game" to get the best result.
> 

This erratic behavior and resulting quandary of strategy is what brings IRV
up to level IV of Nurmi's scale of non-manipulability, which might be
considered desirable in a method that gives good results under sincere
voting, but should be considered a drawback in a method like IRV that
gives lousy results under sincere voting.

Approval is deemed more manipulable because its optimal strategy
is so transparent, so Approval only makes it to level II on Nurmi's scale
of non-manipulability. 

In other words, being high on Nurmi's scale could be good or bad depending
on other factors.

I'll post more on this topic under the thread "Markus and Manipulability"
that Mike recently started up, but I thought this was a good context and
example ready made for the discussion of manipulability.

Here's an analogy along these lines. You move into an apartment, and you
find that the shower valve handle is backwards.  You turn in the
direction marked "cold" and it scalds you. The direction marked "hot"
sends you freezing water.  You soon develop a very simple strategy for
defending yourself from these extremes. Nurmi would call this manipulation
of the system. The simplicity of the strategy would put it at level I.

Eventually you move to another apartment.  In your new shower it is
extremely hard to adjust the water temperature.  The valve seems to have
some built in hysteresis. It jumps from hot to cold at one setting, and
when you back off, it jumps back at a different setting.

Nurmi would say this system in highly non-manipulable (and he's right in a
certain sense), and classify it at level IV.  That's IRV.


Forest