[EM] Thoughts on a nomination simulation

[Kristofer Munsterhjelm]

Kevin Venzke wrote:
> Hello,
> 
> The last thing I did with my simulation is check whether on average a
> candidate would prefer to have withdrawn (considering the results of
> thousands of trials of one position) than stand, with the assumption that
> they care what happens when they lose. (I'm not sure that's actually a
> good assumption: It would be better to assume the voters are the ones who
> care, and don't support a candidate who spoils the election.)
> 
> I got odd results. It could very well be a bug. But for example I found
> that (sincere) FPP had very few scenarios where a candidate would prefer
> to exit the race. Maybe it's because I had filtered out uncompetitive
> elections. But, even if FPP can handle some three-way races doesn't mean
> that we can score FPP based on them, with the assumption that they will
> occur.
> 
> That seems like a big problem with my simulation, that there are always
> three candidates, and no check for incentives for more or fewer to be
> nominated.

I think that a nomination simulation would have to be more complex, to 
take feedback into account. Candidates would position themselves 
somewhere in opinion space, then move closer to the winners depending on 
the outcome of the simulation (and possibly decide to drop out if this 
would elect a candidate closer to their position).

Even so, the simulation would fail to catch certain aspects of the 
election cycle itself. Consider a two party state under FPTP. In a pure 
opinion-space analysis, the two parties would converge on a common point 
(the "center") in an effort to eat into each others' voters, yet in 
reality that doesn't seem to happen - the Republican and Democratic 
parties appeal to different voters.

Changes in voter sentiment might be able to handle some of that problem; 
by having voters change their opinions between elections, candidates 
know not to get too specialized (because it takes time to move about in 
opinion space). That would also limit stagnation in even advanced 
systems: if you have a Condorcet method and a party places itself at the 
(static) median voter, the game is over and all the other parties can 
just as well go home.

There are other effects as well: Parties and candidates might also slide 
into corruption unless checked by competition. One could model that by a 
candidate wanting to both be elected and to be placed at a certain point 
in opinion space (individual corruption), or by candidates being 
attracted towards a certain area in opinion space (coordinated 
corruption, e.g. by lobbying). Candidates may be of use (as opposition), 
even if not elected - not sure how to model that; and the candidates, 
particularly organized ones, may choose to employ strategy if doing so 
is feasible (as the New York parties did under STV) - I'm not sure how 
to model that, either.