[EM] Venzke's election simulations

[Kristofer Munsterhjelm]

robert bristow-johnson wrote:
> 
> On Jun 9, 2010, at 12:42 AM, Warren Smith wrote:
> 
>>>> 1. I think using utility=-distance
>>> is not as realistic as something like
>>> utility=1/sqrt(1+distance^2)
>>>
>>> I claim the latter is more realistic both near 0 distance
>>> and near
>>> infinite distance.
>>
>>> Why would that be? Do you mean it's more intuitive?
>>
>> --because utility is not unboundedly large.  If a candidate gets
>> further from you, utility does not get worse and worse dropping to
>> -infinity.
>> No.   Eventually the candidate as he moves away approaches the worst
>> he can be for you, which is, say, advocating your death,
> 
> :-)
> 
>> and then
>> moving the candidate twice as far away doesn't make him twice as bad
>> from your perspective, and 10X as far doesn't make him 10X worse.  It
>> only makes him a little worse.
> 
> i dunno, Warren.  maybe if the candidate advocates for starving, 
> torturing, and then killing your kids and other descendants, relatives.  
> a holocaust for your ethnic group.  then fouls the entire environment of 
> your homeland to extract resources for he and his unworthy buddies.  but 
> i agree, there might be a limit.
> 
> i'll have to confess, that i have trouble with the presumptions of these 
> simulations in the first place.  i have done simulations of physical 
> processes and communications systems (and have used all three L^1, L^2, 
> and L^inf norms) but i just am not confident of the assumptions of 
> social behavior (without first getting some empirical results from 
> actual social sampling - like getting a handle on how many voters would 
> change their vote from their favorite candidate if he/she changed her 
> position on just 1 particular issue, or 2 issues).

To some extent, I think we're going on that the artificial situations 
will be "close enough" to real ones that the results are useful, and 
thus, more generally, that election methods are robust: if they do badly 
in a constructed somewhat-real scenario, they will do badly in a real 
scenario.

This is the reasoning when I use Bayesian regret for my simulations, at 
least. Optimizing blindly for Bayesian regret would imply choosing a 
method that (among other things) may fail to pick a winner preferred by 
a majority, and it doesn't take strategy into account either 
(Range->Approval->needs strategy, and also show by that Borda, which is 
very vulnerable to strategic nomination and burial, gets a better BR 
than does Condorcet methods that resist that). However, it seems to be 
reasonably good at discriminating between good methods (Condorcet) and 
bad (IRV, Plurality, etc).

Your idea about social sampling is a good one, though. Some of this can 
be done by analyzing ballot data (e.g. how often does a Condorcet cycle 
appear in real life; but beware that strategy may be different for 
different methods), while other aspects would have to be done in real 
life, so to speak.