To Build A Better Ballot
December 11, 2016 6:10 PM   Subscribe

No, this is not about the 2016 U.S. election. Not just that, anyway. First, I need to explain a weird glitch in our voting system…
Nicky Case Previously -who you may remember from his (and vi hart’s Previously, 2, 3, 4) “playable post” Previously about the Schelling Segregation Model Previously- has made a new playable post about different voting systems.
posted by Going To Maine (30 comments total) 29 users marked this as a favorite
This is really great!

Here was I thinking IRV was totally awesome and now it turns out that it's probably not. :(
posted by But tomorrow is another day... at 6:55 PM on December 11, 2016 [1 favorite]

I don't quite get what is the claimed "glitch" for IRV.

If you drag the voters straight up, then yes, they're getting closer to Triangle, but they are also getting closer to Hexagon. So why should I think Hexagon winning is a glitch, instead of just means that Hexagon's popularity increased more quickly? If you try it by instead moving Triangle downward, the glitch doesn't occur.
posted by RobotHero at 6:58 PM on December 11, 2016 [2 favorites]

To quote myself:

"I just don't think ranked choice voting really changes anything -- as kitten magic says, it is emotionally satisfying to be able to vote for your preferred candidate, but it seems to me that it doesn't lead to the rise of stable third parties. Likewise approval voting has not proven particularly successful or popular in the few instances it has been tried on a large scale (the Wikipedia article has some interesting discussion of its use to elect Dartmouth trustees). Ultimately I don't think the voting system really matters as long as you have single-member districts."

Australia has had ranked choice voting for a century and it still has (largely) two stable parties that aren't so dissimilar from what you find in FTFP systems like Canada and the UK -- if the voting system mattered, you might think that you would see larger differences. As far as constitutional mechanics go, I think multi-member districts and parliamentary systems more generally are much more important.
posted by crazy with stars at 7:09 PM on December 11, 2016 [9 favorites]

Very nice presentation, thanks. I would discourage ranked voting as gamey and controversially technological. My three personal favorites (not mentioned above) are as follows: 1) limited approval voting, where voters vote for only two candidates on a multi-candidate ballot; 2) party plus candidate voting, where only two candidates appear on the ballot, from two parties separately chosen in the previous election; 3) A special conference where 535 electors (US) are randomly chosen from tax records, like a jury, to attend a week-long retreat where they are presented with all presidential candidates on live television, but who get to select for everyone.

The most important thing to know about fixing voting problems is that spoilers are real and anti-socially undemocratic. In the US they appear thanks to government funding incentives at a 5% threshold. And if anyone thinks that voting for a party is weird, they may consider the idea of forcing them to democratically compete for funding rather than secretly cozy up to big donors. The run-off jungle voting recently enacted in California, where only two winners compete, is not bad at all, even considering they may be from the same party. But one election with two choices (limited approval voting) will likely do the same, if not cheaper and with least smearing and lying.
posted by Brian B. at 7:12 PM on December 11, 2016 [3 favorites]

Yeah, on closer inspection, I don't get the IRV glitch either. As the electorate changes to prefer more Triangle-y positions on issues, Hexagon picks up the Square voters who dislike Triangle a lot, and prefer Hexagon, so Hexagon wins as a compromise. That's… kinda the idea?

Moving the Triangle means the Triangle party changes its positions on issues to be more like what the voters prefer, rather than voters' preferences changing. Triangle can win by moderating its stance on "Up" issues without compromising on how "Right" they are.

This is also a static, single-election model. Say Triangle wins an election by being more centrist on Up-ist issues this time, but the electorate is naturally becoming more Up-ly over time. Triangle can take a more Up-wards position each election as the electorate moves with them.

Even if they do lose to Hexagon once, it's the trend over time that matters more than a single election win. No?
posted by But tomorrow is another day... at 7:17 PM on December 11, 2016

Okay, I have a hypothetical voting system that seems like it could be pretty cool, but unfortunately tabulating it is way harder to explain than with other voting systems.

1. Let people rank the candidates. It's okay to give two or more candidates the same rank.
2. Say there are K candidates. Imagine a regular K-simplex (a triangle-like thingy with K dimensions: line segment, triangle, tetrahedron...). Each candidate is a vertex of that K-simplex. Each ranking defines a *region* of that K-simplex, in which any given point's distances to each vertex correspond to the ranking. Treat each ranking as if it were located at the centroid of that region.
3. The candidate assigned to the vertex which is closest to the average of all ballots' positions is the winner.

Now, you'll notice in the linked example that the given regions' centroids lie on a hexagon. That's because the hexagon is the 3-permutahedron; fully-filled out ballots' positions correspond to permutahedron vertices. But partial ballots don't, hence the centroid-of-your-ranking's-region-finding.

I think this ends up having a Borda-like flavor, but with a different (and I think, more sensible) point schedule.
posted by Jpfed at 7:17 PM on December 11, 2016 [2 favorites]

Actually, ignore that. I think the point schedule is equivalent to Borda. Poop!
posted by Jpfed at 7:30 PM on December 11, 2016 [3 favorites]

Australia has had ranked choice voting for a century and it still has (largely) two stable parties that aren't so dissimilar from what you find in FTFP systems like Canada and the UK -- if the voting system mattered, you might think that you would see larger differences.

Well, kinda, but also not. One 'party', the Coalition, is a coalition (hence the name) made up of the Liberal Party, the Queensland Liberal/National Party (LNP), and The Nationals (and the NT Country Liberals, ostensibly). Without the Nationals being in lockstep with the Liberals, they wouldn't be able to form government in their own right. And sometimes (but not often) the Nationals don't vote the party line.

You can see the 2016 election results here:

In the last couple of Federal elections, we've had it on a knife edge where the balance of power was basically one or two people, so the (multiple) third parties became really important in terms of which legislation actually gets passed. We've had parties that were quite powerful implode completely (the Australian Democrats, Palmer United Party) and others come into positions of influence suddenly (Palmer United Party, One Nation). They don't rule outright, but they do influence which compromises get made and which legislation gets passed.

The parties also aren't monolithic, so rather than having a point position on issues, it's more of a region or a smear. The conservative faction of the (nominally left-oriented) Labor party are a lot closer to the left-leaning faction of the (nominally right-oriented) Coalition than either of them would like to admit.

There's a lot of wheeling and dealing as legislation issues are negotiated, both within parties and between them, which isn't really captured in this simulation.
posted by But tomorrow is another day... at 7:57 PM on December 11, 2016 [6 favorites]

Your mission, should you choose to accept it: use Sandbox Mode to construct a system of candidates and voting groups such that each voting system causes a different result. Good luck!
posted by jsnlxndrlv at 8:30 PM on December 11, 2016 [1 favorite]

But tomorrow is another day...: "They don't rule outright, but they do influence which compromises get made and which legislation gets passed."

Thanks for the correction and fleshing out of my not-so-great knowledge of Australian politics, but I'm not sure that smaller but important third parties really arise from ranked choice voting. The UK and Canada also have third parties that often are vital for the balance of power, including the larger parties of the Liberals and the Lib Dems but also the Bloc Quebecois, the SNP, etc. Do you think that the choice of voting system has had an important effect on the rise and fall of the Australian Democrats, One Nation, etc.?
posted by crazy with stars at 8:35 PM on December 11, 2016

The problems with IRV are more obvious when you consider how voters will vote strategically (which is impossible to prevent and thus virtually guaranteed to happen).

Here's a rundown of the many problems with IRV.

For me as a Canadian, my top priority is improving the viability of parties that aren't the Liberals or the Conservatives, but unfortunately IRV is very likely to lead to 2-party domination and that's why I'm opposed to it.
posted by mr. manager at 9:08 PM on December 11, 2016 [4 favorites]

I don't think it's clearly causative, more one influence among many.

I think IRV makes people more comfortable voting for a party that more closely reflects their actual views, rather than holding their nose and voting strategically for the "least worst" option. Thus people will vote Green knowing that the Labor candidate is more likely to win, without any fear their vote is a spoiler, because Labor is still a closer representation to their point of view than the other alternatives. The same goes for One Nation; nazis can vote for the truly white-supremacist authoritarian they want and be okay with an only slightly less fascist candidate from the Nationals actually winning.

You'd think that the Greens and Labor would be able to form a coalition, given their broadly similar positions on many issues. It would make them a far bigger threat to the Lib/Nat Coalition as a combined force and yet it doesn't happen. Instead, Labor hates the Greens, and "Labor-Greens Alliance" is used by the Coalition as a perjorative, and Labor perversely agrees that it is one and shies away from the label.

Labor seems to spend almost as much time and energy campaigning against the Greens as it does against the Coalition, and I don't pretend to really understand why. Labor kindof makes out that the Greens are 'stealing' their voters a la Nader, but with IRV that really isn't true.

I suspect it's a far more complex intermingling of emotions and history than a pure artefact of the voting system.

Humans are weird.
posted by But tomorrow is another day... at 9:11 PM on December 11, 2016 [2 favorites]

I don't quite get what is the claimed "glitch" for IRV

Displaying things with spatial models doesn't help in this particular case. The usual way the nonmonotonicity glitch is described is more like:

Suppose there's an election going on where Voter's preferences are that A is better than B is better than C, and that if the election happens right now the winner will be B. But right before Voter goes into the voting booth, he or she changes their mind and votes B in first place and A and second. Under some circumstances, Voter ranking B higher can cause B to lose when they would have won.
posted by ROU_Xenophobe at 9:25 PM on December 11, 2016 [5 favorites]

I love the little widgets. I always have to look up the voting method I prefer, which is here called Score Voting. William Poundstone calls it Range Voting.

Theorists worry hard about strategic voting, but I'd like to see more worry about how the voting method creates the parties/candidates/ideologies. The first diagrams on the page bothered me because the implied voter model-- that most people are in the center-- is precisely what we don't have, or no longer have, in the US. He corrects this later.

But the thing is, it isn't some weird accident that we have two parties and two ideological lumps of voters. The first causes the second. People don't choose their political opinions in a philosophical void; they align themselves with and against the factions they find in their country. In a FPTP system, you generally have two parties pretty evenly splitting the vote. (The UK is mega-weird this year; the US fits exactly.)

The author does acknowledge the effect at one point, but the widgets don't. They let you switch voting systems with the existing candidates and voter-clumps in place. I'd be curious what kind of clumping occurs with the different systems.
posted by zompist at 10:13 PM on December 11, 2016 [2 favorites]

Wouldn't the ideological polarization thing have as much to do with non-proportional, winner-take-all representation as it does with plurality voting? Parliamentary systems are sounding better and better tbh, though we still have to fix gerrymandering (and ameliorate some of the anti-urban structural bias peculiar to the USA...).
posted by en forme de poire at 11:05 PM on December 11, 2016 [1 favorite]

Proportional multi-candidate districts sound great until you realise that local issues immediately lose almost all traction with any large political party because they're guaranteed at least some representation from the district.

This isn't as big an issue if you live in a city with enough population to contain at least one voting district, but in a rural area as a voter you're stuck.
posted by zymil at 1:58 AM on December 12, 2016 [1 favorite]

The UK and Canada also have third parties that often are vital for the balance of power...

The UK does? The a Lib Dems have been in coalition literally once in the last several decades, that being the only coalition government the UK has had in that time. That's not exactly often.
posted by Dysk at 3:15 AM on December 12, 2016

Lib Dems have been in coalition literally once in the last several decades

Don't forget the minority Labour government and Lib-Lab pact of the 1970s, though.
posted by rory at 4:37 AM on December 12, 2016

If you drag the voters straight up, then yes, they're getting closer to Triangle, but they are also getting closer to Hexagon. So why should I think Hexagon winning is a glitch, instead of just means that Hexagon's popularity increased more quickly?

But if you take the spatial/mathematical metaphor being used here seriously, this is simply not true. The strange behavior occurs even if you move the voters directly towards Triangle, which should by definition mean that Triangle is gaining more popularity than Hexagon (though Hexagon is also gaining), and yet that move causes Triangle to lose. You can even move the voters "northeast" slightly away from Hexagon and still get the same effect.

I think a more intuitive way of thinking of the problem is that in the initial setup, Hexagon is getting screwed. A clear majority of the voters prefer Hexagon to Triangle, but because Square manages to squeak into second place in first-place preferences, Hexagon is eliminated and Triangle wins.

(Dammit, now I'm going to have TMBG's Particle Man stuck in my head all day)
posted by firechicago at 5:02 AM on December 12, 2016 [1 favorite]

Hey I have Opinions on IRV. I think I can use the article's tool to show bigger problems with it than the article did.

IRV tells people it's safe to vote their heart first, but it's not. The elimination process is capricious and produces weird discontinuities.

1. Clear consensus candidate rejected. Triangle is alllllmost everyone's first or second pick, the winner by all other non-FPTP methods. I think intuitively triangle looks like the best winner, but triangle can't win in IRV, and in fact looks like the biggest loser. The final outcome, 61-53 pentagon over hexagon, sounds like a solid win but seems bizarrely arbitrary given the situation.

2. IRV doesn't actually fix the spoiler problem. In this setup voters are polarized and one side is split between a moderate and an extremist. (I don't think this is a wildly unrealistic situation.) The IRV result seems okay—triangle wins 62-54 which matches intuition and agrees with all the other non-FPTP methods. However, if you move hexagon just a tiny bit towards the center it throws the vote to square in a decisive 66-50 victory over hexagon. Two voters switching from triangle to hexagon resulted in a 12 point swing towards their least favorite candidate.
posted by fleacircus at 5:54 AM on December 12, 2016 [2 favorites]

What confuses me is that changing from FPTP serves two very different goals.

The first is simply to facilitate third-party protest votes by reducing the odds they'll be spoilers, and (when you use a system like California's jungle primary followed by top two) increase choice between candidates of a single dominant party.

The second is actually to increase the representation of third parties in government or opposition. This is a very different thing, and frankly, hard to justify in practice. The need for the two major parties to recruit coalition parties doesn't seem to provide for better governance of proportional representation countries, but instead seems to work mainly to enable minor parties to have their special interest bed feathered out of proportion to their numbers, or to perpetuate a few dominant personalities in the minor parties retaining ministerial office despite swings in electoral sentiment.
posted by MattD at 5:59 AM on December 12, 2016 [1 favorite]

This is as much of an artifact of representing political opinions as a 2D position, but between each pair of candidates there is a dividing line where people on one side prefer candidate A and on the other side prefer candidate B. So now I see that "glitch" shows up when the voters are situated right where those lines intersect.
posted by RobotHero at 6:02 AM on December 12, 2016

IRV was tried here in Burlington, Vermont where we actually do have a relatively stable three party situation (with a currently in control Democratic party holding center between Progressives and Republicans).

The second election showed exactly all the problems in can lead to, including results showing that three of the five candidates would have won, simply depending on which voting system was used. Plus, people did indeed find it confusing, and obviously frustrating. After trying it for two elections the idea was repealed.

It is really hard for me to imagine this idea working better in other less politically and civically engaged communities I have worked within.
posted by meinvt at 7:05 AM on December 12, 2016 [2 favorites]

Hexagon's supporters like how he tackles problems from more angles

* snorts * haha
posted by numaner at 7:29 AM on December 12, 2016

Displaying things with spatial models doesn't help in this particular case.

By the way, if you break their instructions you can see this a little bit more easily. Instead of moving the constellation of voters straight up, move it up and to the right. You'll have to do this by eyeball, but try to follow a circle with Hexagon at the center, so the constellation of voters isn't net moving any closer to Hexagon. You can still get the result to switch from Triangle to Hexagon.
posted by ROU_Xenophobe at 8:50 AM on December 12, 2016

To see what's going on, you have to imagine those dividing lines, half-way between each pair of candidates.

So don't think it in terms of getting closer or further to Hexagon or Triangle but think about when you cross those dividing lines.
posted by RobotHero at 11:19 AM on December 12, 2016

The problems with IRV are more obvious when you consider how voters will vote strategically
Strangely, approval and range (score) voting can actually become better when you consider how voters will vote strategically. Look at the author's argument against Approval and Range voting but then ask yourself, what will happen in the real world? Click on "Approval" and you see that two thirds of the voters who prefer square to triangle disapprove of both of them; this causes triangle to win. In the real world, when voters are presented with two candidates they hate, they are indeed less likely to vote, but not by nearly that magnitude.

In general, if voters' have good knowledge of each other's preferences, the strategic Ranged or Approval vote is:

1. Consider the two most popular (front runner) candidates
2. Vote approval (or maximum allowable score) for the one of those two you prefer, and for any third (fourth, etc) candidates who you would prefer over both.
3. Vote disapproval (or minimum allowable score) for the other front runner, and for any third etc. candidates who you would prefer over neither.

The great aspect of this fact is that, if a Condorcet winner exists, you get the Condorcet winner.

The poor aspect of this fact is that even if it improves the results to give you a Condorcet winner in some cases, it also forfeits your chance to do better than a Condorcet winner. All the nice "reduction of Bayesian regret" results of Range voting go out the window - to cast a strategic vote, you have to throw away most of the preference information you would otherwise have be conveying to the voting system, which then naturally is unable to best fulfill everyone's preferences.

The debatable aspect of this fact is that strategic votes become much more powerful at deciding between the front runner candidates. If my honest score vote would be "99 John Jackson, 98 Jack Johnson, 0 Hitler", and the Jack Johnson supporters likewise agree in their heart of hearts that "John Jackson is as bad as Hitler!" is merely awful rhetoric, we are all nevertheless motivated to vote as if our opponents are as bad as Hitler, because for anyone who does so their vote counts 99 times as much. No longer would Greens and Libertarians be the only people whose votes don't count: by making tactical voting effectively mandatory we would nearly disenfranchise anyone who is scrupulously honest and/or just bad at game theory.

The scary aspect of this fact is that the above algorithm for tactical voting depends on the assumption that voters all have a pretty good idea of what all other voters are choosing, but that idea becomes much less stable when you move away from plurality voting. Do you think pollsters were doing a reasonable job at predicting the results last month? Imagine what kind of job they'll be doing when there are far more candidates in the mix with a chance of winning and when voters' strategic poll responses will be changing iteratively with each previous poll's results.

If the above assumption isn't a good one, then I don't know what will happen. With Condorcet, if voters try to vote strategically based on imperfect information then the worst candidate can be elected. I don't think I've ever seen any similarly subtle pathologies that approval voting might exhibit, but I haven't seen any disproofs of their existence either.
posted by roystgnr at 11:50 AM on December 12, 2016 [3 favorites]

we are all nevertheless motivated to vote as if our opponents are as bad as Hitler, because for anyone who does so their vote counts 99 times as much.

In that case it devolves down to Approval Voting, which is demonstrably equal to or better than FPTP.

Score voting does make it possible for enthusiasm to win over numbers - say if Trump voters all vote T 10 and C 0, and Clinton voters vote Trump 0 but give a range of scores from 2 to 9. And that actually sounds like a really likely scenario. I'm not totally sure that's a glitch; maybe enthusiasm should win over numbers.

Has anyone worked up a sandbox model of this year's election? This is my try so far. But I'm having a lot of trouble placing Donald Trump on the economic axis, and I'm not sure how to place the voters.
posted by Rainbo Vagrant at 2:49 PM on December 12, 2016

Approval voting fails because it essentially encourages clones to appear on the ballot, then uses the ballot to vote up too many people in order to vote down too many others. It is completely insincere past a point where one cannot say they are undecided between two candidates, which brings us to the point of voting. Voting should use the majority rule, and supply reliable polling information about voter preferences, while not encouraging clones and spoilers to appear on the same ballot, and not encouraging strategic gaming. All ranked voting falls short in non-obvious ways, and the required technology to calculate the outcome fails where voting should never fail, which is to inspire confidence in the selection in order to avoid violence.
posted by Brian B. at 3:14 PM on December 12, 2016

The various systems all still result in a binary decision on who takes office. While we're complicating voting, we might as well add computers in and ask why voters are choosing that option and get more data there. A yes vote "because your opponent is a horrible person" would count the same as a yes "because I support your work fighting poverty", but if enough people vote the second yes, then a candidate's mandate is more clear.
posted by fragmede at 3:31 AM on December 13, 2016 [1 favorite]

« Older GOTY 2016   |   Vexillology Newer »

This thread has been archived and is closed to new comments