Prompted by a reader comment, I did some research on North Carolina's instant runoff voting pilot program, and I am simply stunned. It's no secret that I'm not an IRV fan; it fails to fix the "spoiler" problem, it fails to help third party candidates, and it fails in a host of other ways. But what they're doing in North Carolina goes beyond even that. They've actually managed to create a system that is even worse than IRV.
Attack of the Clones
One of the few things IRV actually has going in its favor, and one of the few ways it actually improves over plurality, is that it is immune to "cloning". To abuse cloning under plurality, you prop up candidates who are ideologically identical to your opponent; i.e., clones. All else being equal, this will lead to voters splitting their ballots among the clones, hopefully (for you) more or less equally. So, instead of you losing 40% to 60%, if you introduce a clone, you'd win 40% to 35% and 25%. In practice, things are a bit more complicated, and usually results in, for example, Republicans paying for Green Party ads to shave a couple points off the Democrat in a close election (or Democrats paying for Libertarians).
IRV actually fixes the cloning problem, because 100% of the clone-first voters will have your original opponent as their second choice. So, when the clone is eliminated, you're back to losing 40% to 60%, where you should have been all along.
But North Carolina added two short-cuts to the IRV process. One, they limited a ballot to listing at most three candidates. This re-introduces the threat of cloning, because if there are four ideologically similar candidates on the ballot, a voter can't list them all, so the incentive to list one of the major two parties highly is even stronger than regular IRV. Two, rather than eliminating just one candidate at a time, as is normally done for IRV, North Carolina has decided that they will eliminate, all at once, all except the top-two. This makes the already vanishingly-small odds for a third-party candidate under IRV an order of magnitude worse, as there's only one opportunity for second- and third-choices to cascade together.
Consider an election where two "third" party candidates each have 17% support, and lets say they are effectively clones of each other, with the voters for each listing the other as their second choice (even if all they agree on is "anyone but the Republicrats"). The two major parties evenly split the remaining votes at 33% each. Even though the third-party coalition has a 34% total—more than either major party candidate has—both third party candidates are eliminated immediately. Even though under normal IRV, only one of them would have, and then one of the major party candidates would have been. Every third party effectively becomes a spoiler for every other third party, while the major party candidates are insulated.
What this amounts to is a gentleman's agreement between the two major parties whereby they agree not to prop-up third party candidates as spoilers (even if it's cost effective, it just feels like cheating I guess), while simultaneously making it harder for a coalition of third parties to act together against them (which is unlikely under IRV, but still more-likely than it was under plurality). And thanks to IRV's inability to eliminate the spoiler effect, they can continue to have a virtual guarantee that their duopoly will continue.
As much as I rail against it, I want to state that, by the metric of Bayesian regret, IRV is an improvement over plurality (at least, as long as some voters vote honestly rather than strategically.) This is probably also an improvement (I haven't run the simulation) but certainly less of one than normal IRV.
It Can Always Get Worse
Even if this systematically-assured lack of change weren't enough, NC still has to deal with all the normal problems of IRV: difficulty counting, costly machines and software, inability to centrally count, etc.. But there's one thing that's even worse: North Carolina wants to use IRV in multi-seat elections.
First, a brief history lesson: IRV was invented as a single-winner analog to a multi-winner method called "single transferable vote". It's actually been pretty successful and effective. But NC is not using STV, they're using a home-grown and absolutely insane system that is so convoluted I can't even begin to analyze it (mostly because I start crying before I can finish reading the description).
A New Hope
There is hope, though. This is just a pilot. If NC is open to the idea of trying new voting methods, maybe they can be convinced to give score voting a try. Heck, maybe they could be convinced to try actual instant runoff voting and single transferable vote too, instead of these abominations. Almost anything looks good when compared to plurality, but time is running out on this opportunity; lets get some real options out there for North Carolinians to look at, and let them make an informed decision.