[EM] A possible "compromise" incentive in most methods

[Kristofer Munsterhjelm]

Say we have a country where elections are done by random ballot. One of 
the voters knows someone that he thinks would be very suited for the 
position (MP, president, elector, etc.). The person in question may have 
impressed the voter with her skills, be someone who takes an unusual 
interest in politics and whose ideas seem well-founded, but is not known 
to the public at large.

Then in a random ballot method, the voter can vote for the person he 
respects, reasoning that if his (the voter's) ballot is chosen, then 
there's no strategy incentive, so his favorite will be elected. And if 
his ballot *isn't* chosen, then it doesn't matter who he named as a 
favorite.

On the other hand, in a deterministic method, a voter would know that 
this person is not well known to the public, hasn't done the marketing 
required or doesn't have the connections or power to get brand 
recognition. Therefore, voting for her is pointless. There's kind of a 
"compromise" incentive to vote for candidates who people have heard of, 
simply from coordination failure.

Then it's *possible* that the random ballot outcome is better than the 
deterministic outcome in expectation, because getting brand recognition, 
having the power required to become a candidate, etc. narrows down the 
field of candidates and excludes people who, in retrospect, would be 
good picks.

And if that's possible, then it's possible that for a sufficiently large 
assembly, filling this assembly using random ballot would be better than 
using a nominally proportional method.

I'm not saying it's a certain thing, and the variance may still be a 
killer, but it's an interesting thought.

Every deterministic method needs a filtering phase, an "invisible 
primary" if you will, that will reduce the number of candidates so that 
it's viable to focus on them. And that filtering phase could produce 
various forms of bias. What other filtering systems may exist than 
straight out party campaigning? And could they produce better outcomes? 
Could a somewhat nondeterministic method have "acceptable" variance and 
still be better than deterministic ones?


As an alternate filtering system, I'm reminded of Fred Gohlke's 
Practical Democracy, which essentially involves groups of three people, 
randomly picked, debating and selecting one among their number for the 
next level. Then three randomly picked people on the second level choose 
a candidate for the third, and so on.

Say we, instead of electing them outright, stop at a point where there's 
a reasonable number of candidates to do a proper election. This 
filtering method would be less reliant on overt marketing, but would, on 
the other hand, reward people who are ambitious in another way: having 
the endurance to go through the initial rounds and defend their 
positions clearly, and not doubt those positions.

Would it be better, would it be worse? It's hard to say. But it is 
interesting to consider the wider perspective.

-km