Thursday, June 28, 2007

How well does your voting system work?

Yee pictures show off graphically how various election mechanisms–e.g., plurality and instant runoff voting, as well as a collection of other esoteric methods–actually respond to the will of the populace. It’s interesting to see what happens in the case of ties and when the vote is split for several mechanisms (see the result for plurality above). It’s a good reminder about how problematic plurality is, but it’s also a good wake-up call for IRV supporters who think that ties won’t be a problem.


Anonymous said...

Ties won't be a problem in any instant runoff voting election with very many votes cast. A percentage tie doesn't trigger a tie. you have to have a tie in the actual number of votes. There are thousands of IRV elections to point to where ties are not an issue.

Nathan Herring said...

Perhaps I should have been more specific. Close races (near ties) are the problem, and they needn’t be between the “major” contestants to have very few votes difference cause a disproportionately large swing in results. The result is randomization in the close races, and the Yee pictures demonstrate that–check out the equidistant 4-opponent election and the random-distance 14-opponent elections. In the latter, it demonstrates IRV’s preference for extreme candidates (note that the centrists lose in almost all elections). Furthermore, there are some decent arguments for why IRV makes it hard for third parties to become serious contenders, which for me is another strike against it.

Anonymous said...

There's no question that third parties should not want IRV, but instead Range Voting. And Range Voting has a much higher social utility efficiency, meaning it will statistically make all types of voters more satisfied with election outcomes.

And if you want proportional representation, Reweighted Range Voting is way better than the antiquated STV system.

Rob Richie said...

There's lots of data from people like Nic Tideman that shows IRV in real elections (as opposed to theroetical ones with theoretical distributions of voters) almost always elects the Condorcet winner -- the "centrist" candidate who is strongest against each opponent. When it doesn't, it's typically because the centrist candidate has weak first choice support. So the "wild swings" theory is pretty out there as a concept in the real world.

That also helps explain why IRV doesn't typically boost third parties to victry. Third parties generally lose winner-take-all elections becasue they don't have as much support as major parties. Even after being well-esablished with proportional representation, for example, third parties almost never wint he single-member district seats elected in German's half-district, half-PR system.

What IRV does very clearly is give people a fair chance to support them and those candidates a fair chance to prove their level of real support. At times that ability to get sincere support helps third parties and independents win (1990 Irish presidential race, 2006 Burington mayoral race, 2000 London mayoral race), but typically isn't enough.

The range voting folks haven't presented any evidence that lack of electoral success for third parties in Australia's IRV elections is tied to any kind of spoiler dynamic -- and I doubt they can find that evidence. In fact, the major parties are good enough for most people, which is why rates of voter satisfication with their government are so high. Where voters are more dissatisfed, IRV might change results -- but always will change the ability to have better, more wide-ranging debate in campaigns.

Nathan Herring said...


Coming from the software industry, especially with a background in software security, I have a whole different viewpoint of actual results vs. theoretical results. In the computer world, we shipped code for decades before people started to take advantage of flaws we wrote (inadverently, mind you) into our software. We didn’t even forsee certain types of attacks back then. Well, now that we’re releasing security packs with multiple fixes every month (and have huge teams to support that security fix pipeline), we see the value of programming with security in mind.

What you’re saying about elections is entirely analogous. We might not have data of (m)any real elections that have these problems. That doesn’t mean we won’t. It’s even worse to have identified results (having explored the entire range of results programmatically), found examples that indicate the system has failed, and then dismiss them out of hand because they’re theoretical and we haven’t seen them in actual elections. The industry would have our necks if Microsoft’s policy was to ship known security holes because “they’re just theoretical.”

Furthermore, if we finally wrangle people into accepting a system like IRV and then find problems that have jumped from theoretical to demonstrated-in-real-world-elections, you’ll be happy with the month-long response time of Microsoft’s security teams in comparison with the years it will take to identify and make a fix to voting. And we’ll have to live with the consequences the whole time.

Steve Chessin said...

I'm a software engineer with over 30 years of experience. I'm also President of Californians for Electoral Reform, and have been involved in electoral politics (both candidates and ballot measures) for over 20 years.

There's a saying that, to a small boy with a hammer, everything looks like a nail. The analogy that Nathan Herring tries to make between electoral systems and software security just doesn't apply. Even the pretty mathematical pictures don't apply, because they don't -- they can't -- take into account that candidates will change their behavior so as to take advantage of the electoral system itself. Since Range Voting can elect someone who is rated in the middle by everyone, it will encourage candidates to be bland and inoffensive, with no strong stands on anything. IRV, on the other hand, requires successful candidates to be the first choice of a significant number of voters, so will encourage candidates to stake out clearly defined positions.

(The pretty pictures also assume a gaussian distribution of voters in a two-dimensional space, when reality is more like a multi-modal distribution in a many-dimensional space, but I'm trying to keep this at the level of a lay person. Did I mention that I also have a Ph.D in physics? :-)

I challenge the Range Voting advocates to get involved in electoral politics, to choose a jurisdiction where IRV is not being used nor under consideration (there are many of them), and get that jurisdiction to switch to Range Voting. Then we can have a true real-world side-by-side comparison of the results of Range Voting with the results of IRV, including both intended and unintended consequences. It will also be a more productive use of their energy, instead of just attacking IRV whenever and wherever its advocates make progress.

Nathan Herring said...


I concur that Yee pictures do not show everything; the real model is indeed more complex than can be displayed in a two-dimensional picture. However, being able to visualize results even with a uniform Gaussian distribution is better than not being able to visualize results at all.

I however, don't follow you on two points. The first is your mention that the analogy between electoral systems and software security "just doesn't apply". Unfortunately, you failed to mention any reasons why that was the case, so I am somewhat at a loss to understand. I would think that the ability to hack a computer system would be seriously analogous to potential electioneers trying to elect their candidate, either within a particular election mechanism such as IRV, or without (e.g., tampering with votes, or preloading vote machines with predetermined ballot counts). Even without the direct "hacking" aspect of computer science, there's still the area of computer science that tries to describe how to take advantage or "game" systems -- trying to hoard bandwidth by not playing by the rules of TCP-throttling, for example. Perhaps you'd do me the favor of giving some explanation as to what you had meant.

The second issue is that the pictures "can't" take into account that candidates will change their behavior so as to take advantage of the system. On the one hand, every individual picture is the result of a static snapshot of a political scenario. You're right in that there's no dynamism there. However, any smart candidate would run "scenarios" to figure out how they did, and to me, looking at various Yee pictures for a given election mechanism tells me (or a potential candidate) what they would have to do to maximize their chance for election (e.g., become more "extreme" as a candidate in an IRV-based election). But just like opinion polling, it's not actually what's happening, but rather a tool that approximates what's happening in theoretically useful ways.

I am not particularly a Range Voting advocate; the topic of my blog merely is being hosted on their site. However, I find it frustrating that IRV proponents are basically asking their non-status-quo opponents to acquiesce, or rather, spend their energy "fighting the good fight" against the status quo, plurality voting. Whereas I understand not wanting to expend unnecessary energy that could be used to broaden the campaign to support alternative methods, I think it's misguided in three ways: (a) I should think that IRV would stand up to competitors on its merits (and not just based on historic evidence, seeing as that a number of these systems have never been enacted). (b) These areas where IRV is making progress are not just areas where people have become enraptured with IRV, but rather have seen the problems with their voting systems and are looking for an alternative; that IRV is the most visible alternative neither means it's the best, nor that the IRV contenders should be automatically ceded the contest because "they've made progress". (I don't mean to say that IRV folks have not driven results in some areas; it is clear that they have. What I mean is that in many of those areas, people were just waiting to be driven.) Lastly, with (c), I'll reiterate that while enabling a system takes a fair amount of time, deciding to change the system if/when it is realized that it is broken will take significantly more time, and furthermore may just cause a relapse. My argument, specifically against IRV, is that, whereas it appears to fix things, deluding the less informed of its proponents into thinking that viable third-party politics is now possible, it will do the reverse. It will have current-day results that mimic plurality with the benefit of having shown support for some "throw away" third-party candidate, but when any third-party candidate gets sufficiently successful and the races become close, we will start seeing skewed results that don't well represent the populace's wishes (to wit, many of the Yee pictures). The backlash against alternative systems in general will be large, and the default result will be a desire to return to plurality voting, a return we'd all wish to avoid.