[EM] Robert Bristow-Johnson

[Kristofer Munsterhjelm]

David L Wetzell wrote:
> 
>     ---------- Forwarded message ----------
>     From: robert bristow-johnson <rbj at audioimagination.com
>     <mailto:rbj at audioimagination.com>>
>     To: election-methods at lists.electorama.com
>     <mailto:election-methods at lists.electorama.com>
>     Date: Thu, 24 Nov 2011 15:50:02 -0500
>     Subject: Re: [EM] More non-altruistic attacks on IRV usage.
>     On 11/24/11 2:20 PM, David L Wetzell wrote:
> 

[...]

>>>       If you're going to undercut their marketing strategy then
>>>       ethically the burden of proof is on you wrt providing a
>>>       clear-cut alternative to IRV3.

>>    Condorcet.

> b.s. 
> In a world full of low-info voters and fuzzy-choices among political 
> candidates, rankings don't have the weight that rational choice 
> theorists purport for them.  

Interestingly, Condorcet seems to deal with noise and fuzzy choices 
better than does IRV. See http://bolson.org/voting/essay.html in general 
and the graph about behavior under noise in particular. Brian Olson even 
remarked that he was surprised by the resilience of Condorcet methods 
against noise, though I have my own idea about why that is.

Of course, if you think those graphs are too synthetic, you probably 
won't be convinced by them. If you do, I'd ask if there's any way we 
could perform simulations that would convince. For instance, could we 
introduce noise to real world voting data and see how the methods behave 
as the noise is increased? Would that be good?

(My idea of why Condorcet methods resist noise is, to be short, 
something like this: Kemeny can be considered a maximum-likelihood 
estimator under a certain noise model. Other advanced Condorcet methods, 
like Ranked Pairs, can be considered so, as well, under more complex 
noise models. However, I am not aware of such a maximum-likelihood 
estimation for non-Condorcet methods, so if Condorcet methods are 
particular in this respect, then they would uniquely resist noise to the 
extent the MLE noise model is similar to real world ballot noise or 
voter uncertainty.)

>>     which Condorcet method i am not so particular about, but simplicity
>>     is good.  Schulze may be the best from a functional POV (resistance
>>     to strategy) but, while i have a lot of respect for Markus, the
>>     Schulze method appears complicated and will be a hard sell.  i also
>>     do not think that cycles will be common in governmental elections
>>     and am convinced that when a cycle rarely occurs, it will never
>>     involve more than 3 candidates in the Smith set.  given a bunch of
>>     Condorcet-compliant methods that all pick the same winner in the
>>     3-candidate Smith set, the simplest method should be the one
>>     marketed to the public and to legislators.
> 
> 
> What works best for wines among wine connoisseurs will not work best for 
> politicians among hacks. 

Well, for wine connoisseurs, you'd just use Range :-)

Many of the Condorcet properties are, I think, declared-strategy 
properties. Majority rule enforces, under honesty, what you can get in 
Range with strategy. The Condorcet criterion ensures you get, under 
honesty, what you would get in a Nash equilibrium under MJ or Range. And 
so on. If you're among honest men who know their intensity of 
preference, just use Range - you probably don't need the 
declared-strategy properties anyway.

(Although I suppose the case isn't *that* clear cut. Condorcet's jury 
theorem etc.)