Priorities in Simulating Voting Systems

Jack Waugh

I plan to implement a simulator to help compare voting systems. It will not be useful until it can compare at least two systems. Vote on which two are worth including first.

rob

Here's my vote (these are not my favorite methods, but my votes for what you should simulate):

RCV-I (Hare) [100], RCV-I (Bottom 2 runoff) [90], Ranked Robin[70], Straight-on Condorcet[60], STAR [50], Approval [50], Score [20]

See https://www.votingtheory.org/forum/topic/129/ranked-choice-voting-and-a-noteworthy-anomaly-in-burlington-vermont-in-2009/3 for descriptions of bottom 2 runoff and straight on condorcet.

(FYI I'm not convinced you can reasonably simulated Approval or Score without doing something iterative, i.e. having the simulated voters adjust their votes after seeing how others are voting to simulate how they attempt to predict front runners after following the polls)

Jack Waugh

@rob You listed single systems, but I need a pair of systems.

In regard to strategy, I'm hoping I can come up with some quasi-reasonable ones that would use heuristics based on how the other voters feel toward the candidates, not on how they vote.

rob

@jack-waugh said in Priorities in Simulating Voting Systems:

You listed single systems, but I need a pair of systems.

So pick the top two?

@jack-waugh said in Priorities in Simulating Voting Systems:

I'm hoping I can come up with some quasi-reasonable ones that would use heuristics based on how the other voters feel toward the candidates, not on how they vote.

Well, its a problem I've thought about a good deal and I don't see how that is going to produce something that is at all realistic. How would you, for instance, simulate the way so many people who preferred Nader, chose to vote for Gore because they thought Gore could win? This doesn't really have to do with how people feel toward the candidates, but how people are able to read the news, which told them that Nader had no chance.

The key to strategy in Approval and Score and Plurality is to guess who the front runners are and then best use your vote to differentiate between them.

I mean, good luck trying, I just don't think it is possible.

@rob said in Priorities in Simulating Voting Systems:

I mean, good luck trying, I just don't think it is possible.

I also spent a few weeks trying (in the context of proportional representation) to simulate strategy. I think it's just pretty hopeless, and at this point I automatically distrust any simulation results that claim to have included strategy.

rob

@andy-dienes Yeah I think for this type of method (FPTP, approval and score) there are ways to simulate rational strategy that can be helpful, but I think it needs some sort of iterative, equilibrium-seeking behavior where voters have some degree of insight into how others will vote.

vote sim

Here I have some of the voters recast their votes in rounds, to simulate them predicting who will be front runners and voting accordingly. (and animate it so you can watch the process) Initially you see them vote naively (for instance in approval they vote for the candidates that are higher than 50% rating), and then they keep adjusting it based on who they estimate the front runners to be.

(Some people have been confused as to what I'm trying to show here. I'm not suggesting that the voting method have multiple rounds, a la instant runoff. I'm suggesting that the earlier rounds in the simulation happen "in voters' imaginations" as they are trying to guess, often based on polling data, who will be front runners.)

Obviously you don't want to voters to have perfect information, but you don't want them to have no knowledge either. So voters are randomly assigned a degree of accuracy of their predictions, a.k.a. savviness. (some know who will be the top two, some know the top three, some base their guesses on earlier rounds, etc Maybe a bit of randomness.)

I'm sure there are all kinds of things you could tweak to make it more realistic, but I think it is a plus to at least having some indication of what rational voters with partial information might do. (I see no point trying to guess what non-rational voters will do, beyond throwing a bit of randomness into the algorithm)

And I don't try to do Condorcet or IRV, which I think are incredibly hard to game and the logic for it is far from straightforward. Maybe I could do STAR but that isn't straightforward either.

Jack Waugh

@rob, I see that the comparison you would most like to explore in the simulator I am working on is RCV IRV Hare vs. RCV IRV Bottom-two Runoff.

I understand both these systems to forbid equal ranking (so this is strict ranking or you could say total order).

In regard to strategy (or tactic, which I'm not distinguishing here), for both these systems there is an obvious naïve strategy, which is to rank the candidates by diminishing affinity. That is, if I as a voter feel the most affinity toward candidate A, I rank A first, and so on. If two candidates receive equal affinity, they can be ranked in either order by a random choice. I will provide a seeded generator, so results can be reproduced.

Maybe bullet voting is useless to all voters faced with these systems, but it is easy to provide that among the options for strategy to be offered by the simulator.

Do you think of any third strategy that you would like to be able to choose for a group of voters to use in place of either of the above, for these two systems (and others that require strict ranking)?

Strategic algorithms will have available to them when they run, not only the affinities toward the candidates of the voters who are deciding how to vote, but also those of all the other voters. It's like perfectly accurate polling is available to the voter, and in answering the opinion poll, a voter cannot lie. The opinion poll reads the voter's mind.

I suppose an elaboration of the design of the simulator could add an ability to have some voters lie on opinion polls (I suspect that the MAGA crowd is doing this). But then there would have to be a way to input for a group of voters, their estimate of how different the opinion-poll results are from how they think the other voters are likely to vote, and in what direction. I haven't worked out the shape of such an input. And anyway, I don't know that such an elaboration would help us much as we probe into how voting systems differ from one another in regard to the power relations they create.

rob

@jack-waugh said in Priorities in Simulating Voting Systems:

I see that the comparison you would most like to explore in the simulator I am working on is RCV IRV Hare vs. RCV IRV Bottom-two Runoff.

Yeah I wouldn't put too much in my preferences... I personally don't know of any effective strategies for those methods that would be reliable enough for many voters to use.

@jack-waugh said in Priorities in Simulating Voting Systems:

I understand both these systems to forbid equal ranking (so this is strict ranking or you could say total order).

Personally I consider that an implementation detail, not a requirement of either method. I don't consider it all that important, though.