The promise of RCV is no more spoilers, no more splitters, no more ‘throwing your vote away’. If you don’t get your first choice (your honest first choice), you might get your second choice, or third.
But, there’s more than one way to do it. Mathematicians have been debating how to do it for hundreds of years and continue to propose new methods to theorize about. In my reading of the field I think there is a consensus pretty-good option, and consensus that a bunch of other methods proposed over the years have flaws and should be archived as oddities.
To say it in a modern way, if you care about what your news feed shows you or what search results show you or what your credit score or insurance bills are: the algorithm matters. I studied Computer Science and to me an election is an algorithm that turns data (votes) into a decision of who is the winner. Different algorithms pick different winners.
Two Algorithms
Here’s two algorithms, both of which work on a ranked choice ballot, (voting for candidates 1st, 2nd, 3rd, etc.)1.
For each ballot, add a point for A preferred over B: the 1st choice gets one point against the 2nd choice, 3rd, 4th, and so on; and the 2nd choice gets one point against the 3rd, 4th, etc. If A is preferred over B more than B is preferred over A, we say “A defeats B”. Usually one candidate is undefeated and the winner, but it is possible to have a cycle where A defeats B, B defeats C, and C defeats A. But maybe 1000 people prefer A to B, 999 prefer B to C, and only 990 prefer C to A, then A should be the winner.
2.
Count all the first place votes. If a candidate has more than 50% of the votes, they win. Until there is a winner, disqualifying the lowest scoring candidate and move votes to each ballot’s highest currently active candidate.
The first algorithm was popularized by a French mathematician in the 1700s and is called “Condorcet’s Method” but I like to think of it as a “Virtual Round Robin Tournament” with a contest between every pair of candidates. The second algorithm has been called “Instant Runoff Voting” (IRV) because it holds a pick-one election, but then instantly rearranges the vote for a ‘runoff’ if no candidate got over 50% of the vote.
If more people want A than B, A should win. Let’s call that the ‘majoritarian principle of democracy’. Condorcet’s method is built around it and implements it as best as possible (with its built in tie-breaker details). IRV can fail this. It happened in 2009 in the election of Mayor of Burlington, Vermont; and then they repealed IRV there.
Currently IRV is getting more press than Condorcet. I think this is because it's slightly easier to explain (see the lengths of my paragraphs above), but this simplicity is not a good thing, it's a blind spot. Condorcet considers everyone's whole ballot at once; IRV only considers a tiny part of each ballot; and that leads IRV to make mistakes.
I've written software to implement both and in that sense they're pretty similar in complexity. There is high quality open source software implementing either that should be understandable and verifiable by 90% of the programmers out there. Practically there's no additional cost to implement one or the other. They use the same ballots, and take similar number of words to write down in legalese in a law. But, they get different outcomes, and it can matter.
Sometimes you vote and you don’t get what you want. That’s true now, it will always be true, but it could be better. If you don’t get your first choice, is your second choice still kinda good? Maybe you really wanted Bernie Sanders but Joe Biden will kinda do (2020 D), or maybe you really wanted Ron Paul but Mitt Romney is kinda good enough (2012 R).
"Aren't there lots of Condorcet cycle-resolution rules?"
Yeah, they're kinda all good-enough. Cycles are rare; times when cycle resolution rules get different answers are even rarer. I've done statistical simulations of millions of elections and simulated voters are happier with the outcome of any Condorcet method by a wide margin over the results of IRV elections.
Count all the first place votes. If a candidate has more than 50% of the votes, they win. Until there is a winner, disqualifying the lowest scoring candidate and move votes to each ballot’s highest currently active candidate.
The first algorithm was popularized by a French mathematician in the 1700s and is called “Condorcet’s Method” but I like to think of it as a “Virtual Round Robin Tournament” with a contest between every pair of candidates. The second algorithm has been called “Instant Runoff Voting” (IRV) because it holds a pick-one election, but then instantly rearranges the vote for a ‘runoff’ if no candidate got over 50% of the vote.
If more people want A than B, A should win. Let’s call that the ‘majoritarian principle of democracy’. Condorcet’s method is built around it and implements it as best as possible (with its built in tie-breaker details). IRV can fail this. It happened in 2009 in the election of Mayor of Burlington, Vermont; and then they repealed IRV there.
Currently IRV is getting more press than Condorcet. I think this is because it's slightly easier to explain (see the lengths of my paragraphs above), but this simplicity is not a good thing, it's a blind spot. Condorcet considers everyone's whole ballot at once; IRV only considers a tiny part of each ballot; and that leads IRV to make mistakes.
I've written software to implement both and in that sense they're pretty similar in complexity. There is high quality open source software implementing either that should be understandable and verifiable by 90% of the programmers out there. Practically there's no additional cost to implement one or the other. They use the same ballots, and take similar number of words to write down in legalese in a law. But, they get different outcomes, and it can matter.
FAQ
“But will my second choice detract from my first choice?”Sometimes you vote and you don’t get what you want. That’s true now, it will always be true, but it could be better. If you don’t get your first choice, is your second choice still kinda good? Maybe you really wanted Bernie Sanders but Joe Biden will kinda do (2020 D), or maybe you really wanted Ron Paul but Mitt Romney is kinda good enough (2012 R).
"Aren't there lots of Condorcet cycle-resolution rules?"
Yeah, they're kinda all good-enough. Cycles are rare; times when cycle resolution rules get different answers are even rarer. I've done statistical simulations of millions of elections and simulated voters are happier with the outcome of any Condorcet method by a wide margin over the results of IRV elections.
"Aren't IRV problems rare?"
Yes, but it did happen. And for no additional cost or complexity we could have something better, so we should, right?
"Isn't IRV better than what we have now?"
Yes, but for no additional cost or complexity we could have something better, so we should, right?
One risk of implementing IRV is that it elects the wrong person and then it gets repealed and we go back to pick-one voting. This happened in Burlington, VT. How long does it take for that to fade before we get to try again? Let's not pick the known-flawed system. Let's be smart, it's easy.
No comments:
Post a Comment