Correction and Update: Some have pointed out that instant runoff voting doesn’t mathematically eliminate the possibility of “vote splitting”. They are correct. While it provides a way around the problem, it doesn’t guarantee everyone will fill out all the rankings. I stand by my claim that, in general, it is a solution to the forced vote splitting inherent in single-choice elections.

Some in the comments thread have claimed that approval voting eliminates vote splitting. I have offered trivial examples to show it does not, and contacted the Center for Election Science, a group that advocates the use of approval voting, whether that claim is true.

Here is the response from Aaron Hamlin, CES’ director:

This is more of a game theory issue. Approval voting allows voters to simultaneously hedge their bets while also supporting their favorite candidates. IRV fails to do this because IRV can divide first-choice votes and eliminate good candidates prematurely. IRV is, however, more resistant to this effect than plurality voting because of the way it can deal with candidates who are genuinely weak.

See here for an example:

We can also tell this from polling data (blog links to polling data) that compares voting methods.

CES agrees generally that both ranked-choice voting systems and approval voting systems are better than single-choice voting systems.

In short, approval voting does not mathematically eliminate vote splitting. I stand by my claim that Rob Richie of the Center for Voting and Democracy (which advocates instant runoff voting) was libeled in the comment thread below.

/End Correction and Update


The open mayoral election has exposed how inadequate our first-past-the-post voting system is for a modern city.

We have three urbanists running for mayor (Jessyn Farrell, Cary Moon, and Mike McGinn), seemingly splitting the urbanist vote and cancelling each other out. We have two strong lefty candidates, who are threatening to knock each other out in favor of the most popular urbanist, whoever that ends up being. We have fifteen other candidates, most of them more fiscally conservative than the anointed six front-runners, essentially cancelling each other out.

The conventional wisdom is that Jenny Durkan will survive the primary election, not just because of her financial advantage, but also because she represents a constituency that has managed not to split its votes among multiple candidates, at least in this election.

It would be enormously depressing to end up with a more strongly-whipped party mechanism, in which urbanists would agree to align behind a single candidate, while the anti-development coalition does likewise. I happen to like the chaos of several candidates, and the ritual of playing them off against each other to get changes made in how government works. I wish elections like these would happen more often.

Many of you are already familiar with ranked choice voting, in which each voter gets to rank all the candidates, as many or as few as she or he cares to. It is also known as instant runoff voting, as the ballots are used to simulate a series of run-off elections as the last-place candidates are eliminated one by one until someone has a majority of votes among all the ballots that didn’t run out of expressed preferences.

Among the commonly-mentioned benefits of ranked-choice voting are:

* resolving the “vote splitting” problem;
* discouraging negative campaigning that tends to plague one-on-one elections;
* reducing the costs of elections by only having one;
* better enfranchising military and overseas voters, by allowing more time for ballots to get to them, and get back.

Why has it not already happened here? The simple answer is inertia. We have always used first-past-the-post, so we always will, until someone in power (e.g. the city council) resets the default.

But many in Seattle have gotten to use ranked choices in public elections, such as those for the Associated Students of the University of Washington. Indeed, ranked choice voting is used in the elections of various bodies on at least 60 college campuses all over the country, including five other Pac-12 schools (Oregon State, Arizona State, Cal, UCLA, and Stanford), as well as the Associated Students of Western Washington U, the Student Bar Association at Seattle U Law School, and the Associated Students of Portland State.

Ranked-choice voting could be implemented via a city charter amendment, ideally proposed by the city council, with an eye toward allowing enough time for any lingering technology issues to be dealt with. (The technology exists, but the county’s current supply of vote-counting hardware, especially for tallying paper ballots, may not be up to the task.)

The “vote splitting” phenomenon of the 2017 election is not a problem to be solved, but a teachable moment that will hopefully lead to the replacement of our antiquated voting system that doesn’t handle the free market of ideas very well. Indeed, holding an expensive election, just to get one piece of information out of each voter in each race, and then holding a second election (enabling insincere voting in the first election) is government inefficiency of the worst kind.

107 Replies to “Ending “Vote Splitting””

  1. Technically, Instant Runoff Voting is only one type of Ranked Choice Voting. The terms are often used interchangeably, though, because the other types of RCV are either easy to manipulate (Borda Count) or lead to some ridiculous outcomes (Condorcet methods).

    1. Can you elaborate on your comments about Condorcet methods? I think that they are indisputably better than IRV and I would like to understand the skepticism.

      1. Sure. First, any Condorcet method must have a way to determine the winner when there is no Condorcet winner, and there is no algorithm for that determination that follows directly from the principles behind the concept of Condorcet winner.

        The main thing I was refering to, however, is that a candidate can be a Condorcet winner without any first-choice votes. Thinking this is ridiculous could be an artifact of living with a first-past-the-post system, but for a position in charge of a sizeable bureaucracy, I think it is important that the chosen candidate be someone who can mobilize at least some support for themself as a first-choice candidate. In contrast, IRV will never allow the candidate with the fewest votes to be elected.

        Also, as discussed below, IRV can be implemented without requiring full preference rankings from every voter. Borda and Condorcet cannot.

      2. Thanks for the response, though I’m not quite sure I understand your first point. I also think the assertion that Condorcet requires full preference rankings is incorrect, perhaps you could elaborate?

        I think the situation you describe in p2 is a rare edge case, but even if it does happen, is there really a good solution for it? Do you DQ a candidate if they don’t get enough first-place votes? Do you hold a runoff when you already have every voter’s preference rankings? Do you just say ‘eff it’ and hold a new election altogether? IMO there is no good solution to the situation you describe, no matter what you end up w/ a less-than-ideal candidate winning. I agree that you don’t want a highly undesirable candidate winning, but look at this mayoral election: it’s very possible that one or both candidates will move on w/ <20%. That's unavoidable really.

      3. Consider the following election:

        3 candidates: A, B, and C
        30 voters rank them A B C
        35 voters rank them B C A
        33 voters rank them C A B

        Who wins with a Condorcet method?

        After thinking about the full-preference rankings, I realized that you are correct on that point.

        My solution to the situation of having a Condorcet winner with no (or few) first choice votes is to not use Condorcet methods.

        Arrow’s Theorem says that, for certain definitions of “fair” and “election system”, no election system is fair. Unless you can find fault with Arrow’s definition of fair, any search for a fair election system must step outside of his definition of “election system”. FPTP, Borda, IRV, and Condorcet all meet his definition. AV and Proportional Representation do not, and thus can be considered when searching for a fair election system.

      4. IRV lets you win with one first-place vote, so I don’t see why “Condorcet lets you win with no first-place votes” is such a big problem.

        The big problem with IRV (and two-round runoff, even) is that it creates really weird strategic incentives. You can make your chosen candidate win by voting against them, whereas in plurality vote you can only make a compromise lesser-of-two-evils win that way. Concrete example:

        In the polls, the right-wing candidate is unambiguously #1 in the first round, with 31%. The center-left and centrist candidates are neck-and-neck for #2, in the low 20s; the rest are minor candidates and undecideds. In the second round, the polls have the center-left and right-wing candidate neck-and-neck and the centrist candidate clobbering the right-winger 2-to-1. The marginal supporter of the right-wing candidate should then strategically vote for the center-left candidate, not in order to elect the center-left as a lesser of two evils, but to advance the weaker opponent into the second round. This is not a farfetched scenario – it describes the 1999 prime ministerial election in Israel a few months before the election. (Ultimately the centrist’s support eroded and he dropped out, as did the minor candidates, and the center-left candidate beat the right-wing one 56-44.)

        This is two-round voting rather than IRV, but you can construct the same scenario in IRV pretty easily: redistribute minor candidates to the main three, and then have the wingers voting R > C > L, the leftists voting L > C > R, and the centrists split between C > R > L and C > L > R such that in an R vs. L match the result is close.

      5. Alon illustrates my biggest concern w/ IRV, which is that it is too susceptible to strategic voting.

        As for the example you give (A B C), assuming I did everything correctly, there is no Condorcet winner, but both ranked pairs and Schulze gave B as the winner. IRV and FPTP would also give B as the winner (assuming everyone voted for their first choice in the latter case).

        When it comes down to it, when only one candidate can win and there is no clear favorite, either you end up w/ a winner who is both loved and hated by near-equal measures of the population, or a winner who nobody particularly likes or dislikes. I’m not sure either alternative is preferable, but when it comes down to it I think RP or Schulze makes the best out of a bad situation (I also reiterate that I think such situations would not be common enough to really warrant worrying about).

        In the end I think we will have to agree to disagree. If it came down to it I would vote for IRV to replace FPTP, but then I’d take just about anything over FPTP.

  2. I like the idea of ranked choice voting for a primary. Nirvana bassist Krist Novoselic has done a good job of championing it’s merits. Elections are expensive both for the candidates and government. It’s also hard to get decent turnout for just a single primary so I don’t like the idea of a series of run-off elections. The only “flaw” of a caucus system is it requires grass roots voter participation. That traditionally was the way to “weed out” candidates of a similar ilk. But I don’t know how you could address every candidate just declaring they are a new party and ending up on the ballot anyway. I guess the disincentive is the sway having the endorsement and funding of a major political party.

    1. I don’t like ranked choice voting of any sort, because I’ve found when talking to people that they mostly do NOT actually rank their choices. They do not have a rank preference. They usually say
      “Well these guys are good, these guys are OK, and these guys are bad”. Rarely more specific than that.

      Approval voting is by far the simplest system to count and solves most of the problems. There’s also a “the good, bad, and the ugly” system if you really want to be more complicated.

      If you have to do ranked choice voting, for the love of all that is holy, do not use IRV. It doesn’t work. It has nearly all the problems of first-past-the-post and is hard to count too. It entrenches a two-party system! It has a “spoiler” problem!

      1. Fairvote has been doing quite well at getting reforms passed in public elections. Approval voting advocates, much less so. We must be working with different definitions of “incompetent” and “liars”.

        As for Pierce County, that was an odd overlay on party nominations. I’ve studied election systems for years, and I still can’t explain what the heck the charter review commission coughed out, as it had little resemblance to classical IRV with no primary. Again, we must be working with a different definition of “liars”.

    1. Consider that the IRV advocates cited on your web page don’t even understand IRV. See this quote for example:

      “I want to be able to vote for my favorite candidate without worrying that doing so makes the election of my least favorite candidate more likely.”
      –Janice H., Lacey WA

      IRV doesn’t solve that. It fails the Favorite Betrayal Criterion. See this layman friendly explanation by a co-founder of the Center for Election Science, who did his math PhD thesis on voting methods.

      This happened in the 2009 IRV mayoral race in Burlington, VT, after which IRV was repealed there. Specifically some Republicans could have gotten the Democrat (their 2nd choice) instead of the Progressive (their 3rd choice) if they had insincerely ranked the Democrat in first place. I.e. they were “punished” by voting for the “spoiler” Republican. Like getting Bush instead of Gore because you chose to be honest and vote for Nader. (I’m dating myself here a bit.)

      1. What Clay said. IRV does not do what it is claimed to do.

        (I should specific that if you have a *multi-winner* election, like electing a city council, then *single transferrable vote*, which looks superficially similar to IRV, works well. But for a *single winner* election, IRV is a disaster.)

        Use approval voting. It actually works. And it’s dead simple.

        Everyone votes up or down on each candidate (“approve” or “disapprove”). The candidate who the most people *approve of* wins.

        It gives you a rock solid guarantee that the most popular candidate won. Unlike every other system.

      2. The problem in that example is political parties, the bane of electoral evolution, not IRV, since you show the problem crosses over among different systems.

        The example really doesn’t apply to larger sets of candidates in nonpartisan elections who defy putting voters in neat ideological buckets.

      3. The founders thought that the problem was political parties; that they were the bane of electoral systems. That’s why they completely ignored parties when drafting the Constitution, leading to our current problems. Parties exist whether or not the electoral system acknowledges them. When they are merely nebulous collections of voters who share similar values (as is the case in the mayoral election) rather than actual organizations, the problem still exists, just in a less prominent form.

        The extent to which voters ‘defy’ being put in “neat ideological buckets” is equivalent to the extent to which voters don’t know what they believe. Personally, I consider it a problem that people don’t know what they believe. If we engage in the increased level of political discussion necessary to help people figure out what they believe (whether or not they agree with us), then voters will sort into ‘ideological buckets’ and ‘nonpartisan’ elections will behave more like partisan elections, where ideological groupings are more overt.

      4. Brent said:

        > The problem in that example is political parties, the bane of electoral evolution, not IRV, since you show the problem crosses over among different systems.

        No, it has nothing to do with political parties. You could completely remove party names from the 2009 IRV mayoral race, and the effect still happened. Voters got a worse result by voting for the candidate they preferred. Period.

        Here again, you prove the point I just made. IRV is so complicated and counterintuitive that it’s own advocates don’t understand it. This is concerning.

  3. What the author is saying is “let’s make voting harder and more confusing”. Um, no. Turnout is pathetically low as it is. Let us not make it worse. Also, there would likely be a significant number of invalid ballots if a voter accidentally doubled up a row.

    Granted, ranked voting would largely be done by a well informed and educated minority (or maybe a lot of random choices). If we are going to do this, why not go back to the good old days-namely white male, landowners? Heck, I’m eligible.

    Nope, note a good idea at all.

    1. Um, Jay, you realize that you can use RCV to vote just like you do now, right? If you like a single candidate and don’t care about the rest, you pick them as your first choice and leave the rest blank.

      That’s as easy as voting is now.

      However, RCV gives those of us who like multiple candidates the option to rank them.

      1. It depends. Some systems allow you to choose your favorites and leave the rest blank, while others require you to rank all the candidates or the choice is invalid. I’d prefer the former, because after choosing my top four mayoral candidates, I really don’t care about the others (except perhaps to put Tsimerman last).

      2. I don’t think the standard IRV algorithm works if you “put Tsimerman last” without filling in all the spots ahead of him.

        For example, suppose you vote (picking at random) Mary Martin 1st, Tsimerman 21st, and nobody in the middle 19 positions. In each round of IRV your ballot counts once for its top remaining candidate. So in the first round your vote goes to Martin. But she finishes last and is eliminated. In the second round your ballot has 19 blank spaces and a 21st-place vote for Tsimerman.

        The system could cast your vote for Tsimerman because he’s the only candidate left that you voted for at all. That’s clearly not what you want — your 21st-place vote was intended as a negative vote, a vote for anyone but that guy. If it was the last round and only one candidate stood against Tsimerman it could count your ballot for the other candidate, but that’s not possible here because there are 19 candidates standing against Tsimerman. You have to pick one. But you’re not around to ask, you’ve already mailed in your ballot.

        If you want to guarantee you always have a vote against Tsimerman in every possible scenario you have to vote every position on the ballot, even if the system is designed to recognize “negative” votes. Fortunately Tsimerman isn’t really campaigning and has no chance to get on enough ballots to win; it’s sufficient to leave him off, along with the rest of the no-chancers. But if there was a candidate that was truly odious, and truly had a chance to win, you’d want to vote for every other candidate, even if you had to draw straws to rank them.

      3. Theoretically Andres, if that is allowed. But one vote per office is the current system so there is little reason to change. This idea only works if all candidates are ranked, but wait, that happens anyway when votes are tabulated. Nope, I stand by my opinion, let us not make voting confusing.

      4. If you think IRV is too complicated, how about approval voting? Voters just mark off all the candidates they like, and the candidate with the most votes wins.

      5. Yes, I didn’t expect that you could leave intervening candidates blank. But the point is that in some systems you can rank only your favorites or all of them, in others you have to rank all of them. I don’t see how it hurts the tabulation to treat blank as “the lowest possible vote”.

      6. I am an advocate of approval voting. I’ve instituted it in every club and organization where we’ve needed to vote on anything, and it works wonders.

        With approval voting, *know* you chose the most popular result, which you *do not know* with first-past-the-post or IRV.

        And it’s *dead simple to count*. It’s very difficult to spoil a ballot with approval voting.

      1. Thanks DJW for sharing the data. There are a *LOT* of misconceptions around RCV, mostly because many people think very lowly of average citizens. They think RCV is “too hard” or “too confusing,” but it really isn’t.

        The data shows this – but so does just thinking about it. People rank choices almost every day in their lives. “What kind of pizza should we order?” “I’d love pepperoni, but if our group demands vegetables then I could go for olives. Really, I’m fine with anything as long as there are no pineapples.” That’s ranked choice voting.

        Stop assuming voters are idiots – give them more credit. As long as you don’t explain things in an overly-complicated way or roll out your system in an inefficient and confusing manner, like Pierce County did a decade ago, RCV can work wonders very easily – and it already does in cities, counties, and organizations around the country.

      2. I agree. I find I usually take a dimmer view of the typical voters’ political acumen than most people seem to do, so I’m always surprised to hear people say RCV is too complicated. It’s incredibly simple and highly intuitive.

      3. It’s not that they think voters are dumb, it’s that many voters are used to only one system and are suspicious that any changes might be a scam to favor the proponents. When there’s active voter suppression and gerrymandering in some areas and the Powers That Be won’t put a stop to it, that’s a reasonable fear. Add to that American education which short-shrifts history and other countries, leading people to assume that no foreign experiences really exist or are applicable to the US.

      4. In any case, RCV will only be established if it passes a referendum or initiative. The public-minded politicians will realize they can’t change people’s rights so fundamentally without public confirmation, and the self-minded politicians will realize they can’t get an advantage with RCV so they won’t want to adopt it unilaterally.

    2. Well, Seattle is already a well educated local. This is the first time the idea RCV might be confusing has ever even entered my head. I suppose my middle class white upbringing and education shows. I do think we should have more appointed positions to reduce the load on voters. If most voters do not understand the position or know very little about the candidates, it is essentially a guessing game. That is part of why I’m apposed to making the ST board elected. I still think RCV is worth the gain, but I agree with you that it could end up confusing people. I think confusion could be mitigated with good ballot design and an education media blitz.

      1. IRV / RCV isn’t worth the gain because there is no gain.

        It just doesn’t work. What IRV does is to eliminate the “spoiler effect” *only for candidates who were going to lose anyway*. If there are two genuinely popular candidates, like, say, Jessyn Farrell and Cary Moon — under IRV they WILL STILL SPLIT THE VOTE and cause a spoiler effect.

        IRV also badly subject to dishonest tactical voting: you benefit from strategic, dishonest voting. IRV has been used in Australia’s lower house for so long that the major parties distribute tactical voting instructions to their members, telling them to vote differently from their true preferences to maximize the chance of the party member winning. People follow these “ballot instruction cards”. And there’s a two-party entrenched system.

        You want to fix the problem? Use approval voting. The ballot is dead simple: vote “yes” or “no” on each candidates. It’s also much simpler to count: whoever gets the most “yes” votes wins. Under approval voting you can never benefit from dishonest tactical voting (there’s always some tactical voting possible but it has to be honest).

      2. We must be working with different definitions of “vote splitting”. Yes, votes would have split between the urbanist candidates, but most of Farrell’s and McGinn’s voters would have ranked Moon, so the urbanist vote would have eventually accumulated behind one of these candidates (probably Moon).

        Under approval voting, a lot of Farrell supporters would have strategically not approved Moon, and a lot of Moon supporters would have strategically not approved Farrell. That sounds a lot more like “vote splitting” to me.

    3. I personally think that, while high turnout is a worthy ideal to look to, encouraging knowledgeable voters to vote is even more important than that. I think a whole lot of voters who don’t really know who they’re voting for “but at least they’re voting” is the opposite of what we want, and can in fact be disastrous. Just look at the presidential election. (the Republican primaries are an interesting case study on vote splitting, as Trump was helped out a ton by votes being split between Cruz and Rubio, and Trump mostly stood out on his own).

      If a lot of people think “hey, Donald Trump would make a cool president,” and we tell everyone to just get out and vote without any sort of emphasis on research and voter education, that’s what’s a bad idea.

      1. The trend is that American elections have a better outcome when there’s a higher turnout. So the voters that are hard to turn out are disproportionally moderate, pragmatic, and left-leaning. That’s the reason behind the voter suppression: while it presumably loses votes from both sides, the voters who still turn out are more right-leaning and tribal-minded than average.

  4. I’ve supported ranked-choice voting ever since I heard it was working successfully in Australia. That’s the key to getting it adopted: reassuring the public that it won’t have (un)intended skews worse than the current system. At the national level I think we need to proceed slowly and carefully to give time for everyone in the far corners to get comfortable with the idea. But city and state elections would be a good proving ground. Wikipedia says San Francisco, Oakland, Berkeley, Minneapolis, Burlington VT, North Carolina, and a few other cities and local party organizations use RCV for some offices. Actually, that should be more widely known, because I didn’t know it. The way to get it established in more places is to publicize critques of how it’s working in existing places. And given many Americans’ reluctance to consider foreign experiences as valid or applicable to the US, it will have to be US places.

    Wikipedia also says our own Pierce County had RCV in 2008-2009; it was lost in the general primary changes in that era, when the parties fought for a “top candidate from the top two parties” system which was established and then replaced, and then Pierce voters repealed RCV. Perhaps they would have kept it if the statewide convulsions in the primary systems (at least four systems in ten years) hadn’t occurred.

    Our winner-take-all system is also why there are only two main parties in the US, because voting for anyone else is ineffective and loses your opportunity to say which of the top two is better or less bad (a major issue when one of the top two parties is irresponsible and dangerous).

    “The technology exists, but the county’s current supply of vote-counting hardware, especially for tallying paper ballots, may not be up to the task.”

    Pshaw, when the 2000 election had all the voting-machine and butterfly-ballot irregularities, my Canadian friends said, “We have a very simple system that’s reliable. You mark an X on a paper ballot, or you can circle or underline or such. Actual people tally the ballots, the result is known right away, and there’s a paper trail in case there’s doubts about the counting.” Is our hardware really helping things, or is it just to put computers everywhere for its own sake? Likewise they said something nteresting about the role of the Queen in Canada: “It’s a very simple system. If Parliament were to declare a dictatorship, the Queen’s army would come and re-establish democracy.”

    1. But it’s NOT working successfully in Australia.

      IRV / RCV is used in Australia’s lower house.

      They have an entrenched two-party system exactly like ours. They have the “spoiler effect” in elections just like we do.

      It’s basically worthless.

      What we need is approval voting, which actually helps. And yes, do it with simple paper ballots like the Canadians do. It is *incredibly simple to count*.

      1. Again, approval voting:
        — vote yes or no on each candidate
        — candidate with the most “yes” votes wins

        Guaranteed to make the most popular candidate win. No possibility of “spoilers” or “vote splitting”. Third parties are on an equal footing with “major parties”. Trivial to count. Easy to explain.

      2. Plenty of Farrell supporters who preferred her to Cary Moon would have declined to “Approve” Moon, to improve Farrell’s chances of winning. Plenty of Moon supporters would have done likewise to Farrell. Your claim that Approval Voting ends vote splitting is trivially absurd.

        If Farrell hadn’t been in the races, many of those Farrell supporters would have instead “Approve”d Moon. Therefore, your claim about AV eliminating “spoilers” is also trivially absurd.

        RCV allows voters to offer up a lot more information than AV does. That’s why there is no serious political movement to install AV in public elections. It is a far weaker opening up of the process than RCV is.

        If systems that encourage more candidates have occasional unforeseen results, so be it. Mathematicians should be more concerned about how open the process is than in how many weird effects can occasionally happen.

      3. Also AV is much more difficult to vote. Not everyone would use the same algorithm you would. I’m still not sure how I would have cast my ballot under AV for the 21 mayoral candidates. All I know is I would have filled in the bubble for Jessyn Farrell, and definitely not filled in the bubble for Alex Tsimerman. Filling out the rest of the list would take a lot of thought and attempts at clairvoyance.

        It is one step better than plurality, but at least plurality is simple to vote and explain how the count works. That doesn’t make it good, from the standpoint of those of us who want to encourage more serious candidates.

      4. > Plenty of Farrell supporters who preferred her to Cary Moon would have declined to “Approve” Moon, to improve Farrell’s chances of winning.

        That just depends on their preferences and their assessment of the odds. In my case, my honest rating would have been Ferrell=10, Moon=8, and the closest competitor maybe a 4. So voting for Moon would make sense, because if it caused Moon to defeat Ferrell, I’m only a little bit less happy. But if it caused Moon to defeat, say, Durkan, then I’d be very happy. Also, if Ferrell wasn’t a frontrunner, then I’d absolutely want to vote for Moon.

        > Your claim that Approval Voting ends vote splitting is trivially absurd.

        Not “absurd”, but “mathematically proven”. Approval Voting satisfies the Favorite Betrayal Criterion (FBC) and the Independence of Irrelevant Alternatives (IoIA) Criterion, the two fundamental criteria used to define “spoiler-proof”. Take a stack of Approval Voting ballots and eliminate any non-winning candidate, and observe that the result can’t change. This isn’t true of any ranked voting method, such as IRV or Condorcet. Those systems also FAIL the FBC, so ranking your favorite candidate in first place can HURT YOU.

        > If Farrell hadn’t been in the races, many of those Farrell supporters would have instead “Approve”d Moon. Therefore, your claim about AV eliminating “spoilers” is also trivially absurd.

        Again, that depends on the factors I discussed at top. Either way, that would be their choice. NO voting system can be free of spoilers as a function of voters strategically changing their minds. But as I just noted, Score & Approval AT LEAST are spoiler free in the sense that you can’t get a different outcome by just removing a non-winning candidate without any change in voter behavior.

        Whereas if you remove the Republican from the 2009 IRV mayoral race in Burlington, Vermont, then the Democrat wins instead of the Progressive—even if voters are 100% honest and provide the exact same ranking otherwise. THAT is a true spoiler effect. Also, Republican voters in that race were “punished” for ranking the Republican in first place, because it caused them to get their 3rd favorite instead of their 2nd. I repeat: Score Voting and Approval Voting can NEVER punish you for giving the maximum absolute support to your sincere favorite candidate. You can never ever have the feeling of “I shouldn’t vote for the Green, because that could cause me to get the Republican.”

        > RCV allows voters to offer up a lot more information than AV does.

        This is a common misconception. The math is here.

        Imagine we’re comparing Approval to the top-3 ranking used in SF. With 3 candidates, both systems give you 6 ways to vote. With 5 candidates, Approval gives 32, and ranking gives 60. But then with 9 candidates, Approval gives 510 and ranked gives 504. And for any greater number of candidates, Approval is better. Score Voting of course dominates from square 1.

        But even in the cases where IRV allows more information to be expressed, it IGNORES most of that information.

        Which is WHY it performs worse.

        That’s why there is no serious political movement to install AV in public elections. It is a far weaker opening up of the process than RCV is.

        If systems that encourage more candidates have occasional unforeseen results, so be it. Mathematicians should be more concerned about how open the process is than in how many weird effects can occasionally happen.

      5. Our current tech-obsession seems to lead to the idea that “more data is always better,” but this is not necessarily true. The NSA is collecting Terabytes worth of data to analyze, but (aside from the civil liberties concerns) having this much data actually makes it more difficult to analyze. Similarly, voters providing more data on the ballot is not necessarily an improvement. This is what I meant by my remark that AV avoids the problems exhibited by ranked-choice systems because it avoids extraneous data.

        However, that was not quite accurate. After all, FPTP leaves out most of the data included in a ranked-choice list, and it is still subject to Arrow’s Theorem. The advantage to AV comes from the fact that an AV ballot cannot be definitively constructed from a preference ranking; one additional piece of information is required: which distinction on the list the voter considers to be the most important. AV adds this piece of data, which removes it from Arrow applicability, and then removes all other extraneous information.

        Score voting feels better because voters can provide more information, but all that extra information is pointless, since optimal strategy under score voting devolves to approval voting.

        To deal with the risk of the quasi-spoiler effect from a large number of bullet-voters, I like the idea someone on here has suggested of using AV for the primary and then a two-person run-off. However, the run-off may need some creativity. In a situation like the State Treasurer race last year, the two Republicans could have run a strategy of getting all their supporters to vote for the two of them, and they likely would have gotten between 40% and 45% each. If the 52.5% of voters who preferred a Democrat voted for 2 Democrats on average, then that would be a vote total amounting to 105% of the voters across the three Democratic candidates, for an average of 35% each, making it likely that the two Republicans would have advanced to the general election, just as happened with the current system. While more difficult to implement, this same strategy could be used in a non-partisan race.

        My proposed solution is as follows:

        1) Count the AV ballots under normal AV rules.
        2) The candidate with the most votes qualifies for the run-off. (In places where the first round is the main election, if the candidate with the most votes has a majority, they would win outright, but since our main election is the run-off, we want to still have a choice available for the election with higher turnout in November.)
        3) Remove all ballots that approved of the top vote-getter from consideration.
        4) Recount the remaining ballots under normal AV rules.
        5) The candidate with the most votes on the second count qualifies for the second position in the run-off.

        I’m sure there are potential problems with this system, but I do not see them. I look forward to people pointing them out.

      6. Clay went through a lot of math to try to prove something I just trivially disproved by real-world example.

        If Jessyn Farrell hadn’t run, it would have possibly changed my vote from not approving Cary Moon to approving Cary Moon. If a bunch of voters used the same thought process as me, it could have changed the outcome of the election.

        Vote splitting is trivially possible under AV, and trivially able to change the outcome of the election under AV.

      7. > Clay went through a lot of math to try to prove something I just trivially disproved by real-world example.

        No, you certainly didn’t prove that Approval Voting fails IoIA or FBC, or that IRV passes them. These are just facts.

        > If Jessyn Farrell hadn’t run, it would have possibly changed my vote from not approving Cary Moon to approving Cary Moon. If a bunch of voters used the same thought process as me, it could have changed the outcome of the election.

        You’re just saying that the Approval Voting outcome can change if voters change their behavior. Of course ANY voting system can change if voters change their strategy. In the 2007 *IRV* mayoral race in San Francisco, 53% of voters bullet voted for only one of the 12 candidates. Do you think they would have all abstained if their favorite candidate hadn’t run?

        But with IRV it’s even worse than that, because the result can change WITHOUT a change in voter behavior. E.g. in the 2009 IRV mayoral race in Burlington, VT, the Republican was a spoiler. If he dropped out after the votes had been cast, then the winner would have switched from Progressive to Democrat—an improvement in the eyes of most Republican voters, who saw the Progressive as the most liberal, and as such ranked the Democrat 2nd between the Republican and Progressive. Those Republican voters were punished by voting for their favorite candidate. If they had insincerely ranked the Democrat first, then they’d have at least gotten their 2nd choice instead of their 3rd.

        THAT is the issue most people are talking about when they worry about spoilers. They normally feel afraid to vote for their sincere favorite candidate if that candidate isn’t “electable”. But with Score Voting and Approval Voting, it’s always safe to top-rate or “approve” your sincere favorite, without regard for electability. In general, you want to approve everyone you favor to your favorite frontrunner.

        > Vote splitting is trivially possible under AV, and trivially able to change the outcome of the election under AV.

        Hopefully you can see how misguided this criticism is now. In any case, you want to focus on Bayesian Regret, not your intuitive sense of how much of an impact such an issue as this has.

  5. I think IRV is a good idea. Hell, we should use it to replace low-turnout primaries.

    Imagine if instead of a primary that narrowed your choices held in the heat of August with inexcusable low turnout and a certain transit hero having to leave the state legislature to fundraise; we could have one IRV in November. It would also save taxpayers money as running elections is expensive.

    Also cutting down on the cost of elections? The very real consequences of going negative in IRV – risking losing a 2nd or 3rd choice vote. Right now, I accept as legitimate the fear of many transit advocates – including one Snohomish County candidate in this election & friend of mine – of having to fundraise for transit board candidates. IRV might just make big money negative campaigns a thing of the past in this state.

    Oh and we’d also see libertarians and independents (I’m the latter) stand a real shot of winning these elections. Something else to think about.

  6. I looked at this and agreed that we need to do this now. I suppose that the real issue is that I really don’t have an order of preference beyond maybe the first 8 or 10 candidates. And most people wouldn’t find ordering them to be a mostly useless exercise.

    But then I realized that if my partner was faced with the ballot shown, their likely reaction would be to simply not vote. While many of us would be excited about the opportunity to vote this way, some people like them would be much more likely to set the ballot aside and never pick it up again before election day.

    1. How about approval voting? Voters just mark off all the candidates they like, and the candidate with the most votes wins. Simple!

      1. Approval voting sounds like it violates “1 person 1 vote”

        Also seems like it creates tactical voting that rcv is trying to reduce. If there’s a candidate I would tolerate but not love, do I vote for them? If yes, the candidate I hate is less likely to win, but so is the candidate I love. With rcv, I can distinguish between candidates I love, like, tolerate, and hate.

      2. Approval voting sounds like it violates “1 person 1 vote”

        Of course it doesn’t. Each person has the same ballot, weighted the same, with the same power to influence the election. What leads you to this bizarre conclusion?

      3. The public will never accept a system where they can’t distinguish their #1 choice. That’s where it feels like taking their real vote away from them. I wouldn’t call it a “one-person-one-vote issue” because the voters are still equal, but it seems to undermine democracy in a broader sense. (“This isn’t what our ancestors fought for.) People don’t want to just say who they approve of, but which one they want.

      4. In first past the post, everyone gets 1 vote. In RCV, everyone gets exactly 1 vote in each round, even if that vote changes based on candidates being eliminated.

        In contrast, with approval voting, a voter who is dead set on exactly 1 candidate gets fewer votes than someone who decides to approve of multiple candidates.

        Imagine you’re going to lunch with coworkers. You say “I really want pizza for lunch”. A coworker says “I want Chinese food or sushi or Mexican food”. It feels like they have a much bigger influence on lunch then you even though each of you should have an equal say.

        That problem could be fixed by alloting each person “100 points” and letting them split it however they want. 100 points for 1 candidate, 20 points each for 5 candidates. 50 points for 1 candidate, 10 points for 5 other candidates. But at that point, you basically have RCV voting.

        My main objections to approval voting are that it adds complexity to first past the post without a clear benefit. It doesn’t remove tactical voting, and it basically asks voters to say “meh, I don’t hate any of these 5 people, so I guess anyone of them would be cool”.

        I also think much of this could be solved at the party level without needing to change voting systems. If there are 3 candidates who are basically policy clones of each other, then someone should step in to say which 1 of those 3 gets to run in the general election. That’s what a “normal” primary or nominating convention would do in other states, and it feels like it’s missing in Washington. See last year’s state treasurer race. If the Democratic party had somehow enforced only 2 candidates competing in the primary, they wouldn’t have been shut out of the general election.

      5. While I agree with Mike’s point that AV won’t be accepted because it doesn’t allow voters to distinguish a first choice, Larry’s contention that it violates one person one vote is wrong. Each voter gets a yes-no vote on each candidate, thus giving everyone the same number of votes.

        All voting is strategic, but not all strategic voting is equally bad. AV elliminates all incentive for voters to rank candidates insincerely (in this case, vote for someone who they dislike more than a candidate for whom they do not vote). The optimal strategy with AV if the two front-runners are known is to vote for one of the front-runners and everyone you rank higher than them.

        The optimal strategy with the points system is to give all your votes to whichever of the frontrunners you prefer, effectively replicating FPTP.

      6. This is a completely false and inaccurate understanding of approval voting.

        Think about it this way. You mark every candidate “yes” or “no”. A person who supports only one candidate marks their favorite “yes” and everyone else “no”.

        Everyone gets exactly the same number of votes. You vote “yes” or “no” on each candidate.

        A subtle refinement of approval voting is to state that the winning candidate must get more “yes” votes than “no” votes — just like passing a referendum.

        If *multiple* candidates get more “yes” votes than “no” votes, the one with the *most* yes votes wins — this is the tiebreaker.

        What could you possibly dislike about that?

        ONE MAN ONE VOTE, on EACH CANDIDATE. With approval voting.

        Versus ONE MAN NO VOTES. With first past the post, when you vote for a third party candidate. (They don’t even count third party votes in some places because there’s no point.)

      7. This is one of the few ways in which Nathanael is right about approval voting: Everyone gets an equal opportunity to cast a vote on an identical ballot.

        Some have decried IRV as not being one-person-one-vote, but every election system that gives everyone one and only one chance to fill out the same ballot is, by definition, one person one vote.

      8. > Approval voting sounds like it violates “1 person 1 vote”

        False. See discussion of the relevant legal precedent here:

        > Also seems like it creates tactical voting that rcv is trying to reduce.

        This is backwards. Approval Voting satisfies the Favorite Betrayal Criterion, whereas IRV fails it. You can see how both systems fare in Bayesian Regret calculations with various mixtures of strategic and honest voters.

        There’s even a theorem that Approval Voting tends to elect Condorcet winners when voters are strategic.

        > If there’s a candidate I would tolerate but not love, do I vote for them?

        It depends. Threshold strategy is discussed here.

        > With rcv, I can distinguish between candidates I love, like, tolerate, and hate.

        This is a misleading illusion.

    2. This is an unusual election. Usually there’s an incumbant and the challengers are few. It’s also possible that RCV might encourage candidates to keep their numbers to a manageable level. Or there could be a signature threshold or past-record threshold to get on the ballot. Or for those perennial candidates who don’t really expect to win but just want to get their views publicized, there could be some alternative system to publicize their views. (I don’t know what. A “best extremist candidate” award?)

  7. Don’t forget what happened down in Pierce County about 5-10 years ago?. They had it and wind up electing a treasurer that apparently was not qualified. They wind up recalling him eventually and Pierce County voters got rid of the ranked voting after that experience.

    1. That was because of voters not knowing that he was bad, not IRV. In a first past the post system, I can easily imagine two qualified candidates going too negative on each other, thus allowing the third, unqualified but unknown candidate to win. Or in our current runoff system, the unknown candidate being helped by the neg bombs of the two other candidates against each other into the second round, where supporters of the losing candidate and voters turned off by the neg bombs vote en masse for the unqualified but unknown candidate.

      1. No, IRV has a tendency to elect crank candidates like this. The problem is that IRV is badly subject to strategic voting, where you rank the “serious opponents” behind the “joke opponents” in order to reduce their chances.

        It happened in the Australian Senate, which is actually Single Transferrable Vote, but it’s also happened in the Australian House. In Australia they’ve been using IRV long enough that the major parties know how to break it, and they mail tactical voting pamphlets to all their supporters.

        Approval voting is a solid system; IRV is a weak system. There are mathematicians who study this stuff and figure out how to break voting systems.

      2. Please define “joke opponents”, “solid system”, and “weak system”. And try to set aside the scourge of political parties (which bring some pretty good mass clairvoyance into the mix) and define them in nonpartisan terms.

  8. One thing that stands against that ballot is how daunting it is! A ranked-choice ballot filled out to be counted by hand could be a lot simpler: just a single column of blank lines, and you write a name in each line, as far down as you care to go. A similar interface could be designed for a computerized voting machine. The requirements that the ballot be non-interactive paper and machine-readable give us this intimidating grid. I can’t think of a better way.

    “Approval voting”, where you vote for as many candidates as you approve of and each vote is counted once, is surprisingly good at eliminating vote-splitting and incentives for “harmful” strategic voting. Every ballot is strategic in a sense, because each voter has to decide where to draw that critical line between “approve” and “disapprove”, but voters don’t have to consider whether they should strategically vote for a candidate they like less over one they like more. Of course, approval voting can be unsatisfying for highly opinionated people, because they don’t get to articulate who their real favorite is. But picking the top two finishers from an approval vote, then running them off against eachother, is also a reasonable system. In Seattle we already have a two-round system with top-two general elections, and approval voting doesn’t have any problem selecting multiple candidates from a field, so converting just the primary to approval voting would be perfectly viable. I also think approval voting could be a pretty good system for in-party primaries with large fields.

    Some people like approval voting because they think it reduces incentives to campaign negatively. I’m not sure whether that’s true or not. But it would have an interesting impact on how people think about their non-top-choice candidates, and how candidates position themselves to people that won’t put them first: in a big field, each candidate has to fight to get on the ballots of people that like other candidates more. So in this mayoral field, Durkan and Hasegawa have each said/not-said a bunch of things that turned off young people and urbanists… but it didn’t matter, they weren’t going to be the first choices of these voters anyway. Even in a ranked-choice scenario, when Durkan made a bad joke about Nikita Oliver and Hasegawa ranted against road diets, I looked over my imaginary ranked-choice ballot and thought, “Well, I really dislike that, but it somehow doesn’t move them on my ballot.” Either one would have to be a lot worse for me to rank them behind, say, Harley Lever, or that angry libertarian dude that knows less about how city government works than I do. But in an approval-voting scenario it would have mattered. Durkan and Hasegawa have each been on my imaginary approval ballot at some point, but have each worked their way off it.

    So there’s my pitch and proposal. Approval voting in the primary. Take the top two from that and run the general just like we do today.

    1. This is just one example of a ballot format, and it may be more to illustrate what RCV is than a plausable ballot. Other formats have people write numbers next to the names. And this is an unusual election: we don’t usually have twenty candidates. This size is not typical of most races.

    2. OCR to read hand-written digits with a team of people to check results that are read with low-confidence shouldn’t be that hard. Would want multiple humans to check each of those results to make sure nobody is “ballot-stuffing”.

      1. OCR is possible, and we already have a ballot-tracking system, which presumably could be used by people that really wanted to check that the machine read their votes correctly. I’m a little concerned with the possibility of snooping on or manipulating votes with such a system. I’m already a little worried about that today (stealing neighbors’ election mail, forcing family members or employees to show their ballots… some pandas are giant pandas, some pandas are red pandas, and some pandas are trash pandas…), and of course, the more computers are involved, the more malfeasance is possible at-scale (the kind of panda that I am is a computer panda, so that’s how I think about computers, in terms of everything that can go wrong with them).

      1. Thanks for the examples. This is what I feared, that other systems may have unknown skew factors that theoreticians haven’t discovered, the public isn’t aware of, or candidates in power are manipulating for their own benefit. Having said that, vote splitting is a terrible problem.

        I voted Libertarian in the 90s and in 2000 Gore was my second choice, but in 2000 with so many people voting L or Nader, we ended up with Bush. There were irregularities in Florida, but the point is that if all those had voted for Gore his margin would be more comfortably ahead and less assailable. An increasing part of the country, following longstanding trends in Washington state and a few others, doesn’t like how much control the two parties have and wants more real choices. That’s why the primary system changed 4+ times in the 2000s, as the public fought the parties for a system that would benefit them. After 2000 when the R party took several turns for the worse, I feel like I can’t vote for anything but D. That doesn’t mean I agree with everything the D’s do, but they are the only one large enough and powerful enough to stop the R’s. (And I lost my interest in libertarianism after seeing Bush implement half of it.)

    3. There might be some tricky mathematical catches with running approval voting in the first round and a “top two” runoff in the last round.

      I *think* it should work, but it’s worth having the mathematicians analyze it. Could you contact the Center for Election Sciences and ask them to look for “holes” (as in, tactical tricks which could be used to abuse it?)

      Approval voting in one round is immune to abusive tactics. IRV and first-past-the-post are heavily subject to abusive tactics. But I haven’t seen an analysis of approval voting followed by a top-two runoff. It’s worth getting one.

  9. I’m a co-founder of the Center for Election Science ( I’ve studied election methods since 2006.

    It’s incorrect that Instant Runoff Voting (the ranked system you’re describing) solves the vote splitting aka “spoiler” effect. Here’s a simple demonstration by Andrew Jennings, a fellow co-founder of CES who did his math PhD thesis on voting methods.

    IRV is also worse than the other four commonly discussed alternative voting methods in virtually every way, most importantly “Bayesian Regret”, a measure of how well the voting method produces results which satisfy the preferences of the average voter.

    Two better and vastly simpler systems are Score Voting (aka Range Voting) and Approval Voting. You can see why in these links:

    A newer interesting system is Score Runoff Voting, which uses Score Voting to produce a “top two” by highest total rating, and then performs an instant runoff based on the voters’ scores for them—an instant head-to-head majority winner matchup. The benefits are summarized here.

    While CES generally advocates rated systems (Score/Approval) over ranked systems, it should be noted that the above arguments against Condorcet systems are misguided. While a Condorcet system could theoretically elect someone with zero first-place rankings:

    1) IRV can elect someone with only TWO first-place rankings—hardly a significant difference.
    2) The likelihood of someone winning with zero first-place votes (given the candidate and his mom can vote for him) is effectively zero anyway.

    Note that Approval Voting does NOT violate “one person, one vote”, as is explained by an attorney and election methods expert here.

    Here’s a simple Approval Voting mock election for Seattle mayor.

    Clay Shentrup
    Berkeley, CA

  10. “The open mayoral election has exposed how inadequate our first-past-the-post voting system is for a modern city.”

    Brent, what’s your evidence there’s anything wrong with it at all- even leaving out “compared to what?”

    Any chance at all that Seattle will end up with a Mayor blatantly in the pay of a vested interest, be it a person or an organization, or industry? Or associated with people or forces capable of violence?

    Or lacking sound mind as the law understands it? Or substance-addicted to the point it renders them unfit for office? Or visibly stupid? Or any resemblance at all to our country’s present Chief of State?

    Except in his own estimation, Alex Tsimerman shares only one thing with Donald Trump: If he gets elected, it’ll be because the people of Seattle are voting out of sheer spite, and will deserve exactly what they get.

    The exact reason why, whoever wins the election nobody will get the job who any STB reader will move rather than have for Mayor. I didn’t leave because Ed Murray won. But because of the people of Seattle’s general agreement with forces and mentality I’d refuse to live with if I could afford it.

    Seattle didn’t lose the Waterfront Streetcar by tyranny or conspiracy, but because neither public officials nor their constituents could decide whether to keep or drop it. A decade’s oxidizing track and catenary- elegant witness.

    Until “Urbanisim” becomes a political ideology, let alone a political party, wide range of election arrangements will bring same general results.

    Mark Dublin

    1. Two candidates will advance to the general election after the two of them combined got just 45% of votes cast. That’s an ugly result, violating our basic notion of majority rule, that will make candidates who have some agreement on issues from running against each other in the future.

      If the election were held over, expect that Hasegawa, McGinn, and Farrell would decide not to run. That’s a loss to the public, who deserve competitive elections in order to make government accountable.

      If we even allowed three candidates to advance, when something like this happens, it would be an enormous boost to electoral openness.

      I didn’t vote for Nikkita Oliver, but a system that would allow us to rank Durkan, Moon, and Oliver in November seems a lot more fair than one that forces us to choose between the two candidates who collectively got just 45%.

      (And no, Nathanael, having to approve or disapprove the candidate I like second-best would be an absurd and laughable system.)

      1. >Two candidates will advance to the general election after the two of them combined got just 45% of votes cast. That’s an ugly result, violating our basic notion of majority rule

        Whether it’s “ugly” is subjective. But the Bayesian Regret figures show Approval Voting produces highly accurate outcomes.

        The “basic notion of majority rule” is a figure of your imagination. For instance, IRV can elect candidate X, even though Y was preferred to X by a huge majority of voters AND got twice as many first-place votes as X.

        Many people see it as a benefit that in a runoff situation, Approval Voting would pit two similar centrist candidates against one another, so that we can debate broad strokes in the first round, and debate nuances and qualifications in the second round. This inverts the current situation where e.g. Democrats debate what kind of health care program we should have, and then in the general their nominee goes up against an opponent to argue whether we should even *have* a health care program.

        > that will make candidates who have some agreement on issues from running against each other in the future.

        I don’t see what your evidence is for this. You can vote for multiple similar candidates, and in many case you want to.

        > If the election were held over, expect that Hasegawa, McGinn, and Farrell would decide not to run. That’s a loss to the public, who deserve competitive elections in order to make government accountable.

        This is LESS the case with Approval Voting, because the only kind of “spoiler” effect it can have is when voters actually change their behavior. Whereas ranked voting methods can fail IoIA even if voters DO NOT change their behavior.

        E.g. in the 2009 Burlington mayoral race, the Republican was a spoiler. Because he ran, the Progressive candidate won. If the Republican had stayed out, then the LESS liberal Democrat would have won.

        Your criticism is more applicable to IRV than to Approval Voting.

        > I didn’t vote for Nikkita Oliver, but a system that would allow us to rank Durkan, Moon, and Oliver in November seems a lot more fair than one that forces us to choose between the two candidates who collectively got just 45%.

        I’m not sure what you’re talking about. How do we make an election less “fair” by picking two people instead of one, and then having yet another round? Maybe it’s more expensive or wasteful of time. But I don’t see how it’s more or less fair.

        > (And no, Nathanael, having to approve or disapprove the candidate I like second-best would be an absurd and laughable system.)

        If you’re talking about Approval Voting with a top-two runoff, you’re very wrong. It gets good Bayesian Regret and absolutely creates an incentive for many voters to approve more than one candidate.

      2. Let’s take a more trivial example in which the candidates for Most Popular Mathematician are Rob Richie, Clay Shentrup, and Nathanael.

        40% of those polled on some random list approve Nathanael, for reasons the other 60% can’t understand, and disapprove the other two.

        31% approve Clay and disapprove the others, as they just don’t agree with the other two.

        29% approve Rob and disapprove the others, for similar reasons.

        Nathanael wins with 40% approval.

        Switch to IRV with the same set of voters.

        Nathanael’s 40% rank Nathanael first, and Clay second, knowing that it is to their advantage to rank all the candidates, even though they were hoping someone who could concern troll any voting system to keep the plurality status quo is what they really want.

        Clay’s voters all rank Rob second, as they want some sort of reform, and are willing to settle for IRV if they can’t get score runoff or whatever later invention comes forth. Rob’s voters all rank Clay second, as they are willing to try just about anything to get us out of the plurality system trap. Clay wins with 60% in the second round under IRV. Clay also wins Condorcet, 71-29 over Rob and 60-40 over Nathanael.

        The pro-reform voters played crabs-in-a-bucket to advantage their favorite reforms under approval voting and paid they price.

        Yes you can show an example of where they might not have done so. But the Nathanael in this article’s comments, not to be confused with the fictional Nathanael anti-reform concern troll in this sample election, is making a claim that “vote splitting” universally never happens under approval voting. I have to discount every claim Nathanael makes about any voting system unless he either talks back this claim or shows the math. Clay is at least providing links.

  11. Yo, so just out of curiosity, I’m running an experiment. Not a scientific poll, but just because the Seattle mayoral election is super weird and super interesting, and I’m too lazy to build a website for it…

    I’m curious how various voting/counting methods would affect the mayoral election this year. So I set up an email address, seattlesim2017 -AT- gmail (dot com) for you to send your ballots to. Send each of the following:

    1. A single vote for the primary as it exists today. Vote as strategically as you would in the primary :-).

    2. A ranked-choice list as far down as you care to go. I will use this for IRV, Condorcet, and a few variations on Borda Count (Borda Count doesn’t handle partial rankings very gracefully, especially in such a large field, so… I’ll probably have to do some cheesy nonsense unless everyone submits full lists). Voting non-earnestly in any of these systems requires knowledge of the electorate that none of us have, so I wouldn’t bother. If you have “gaps” in your numbering, any vote after the first gap is ignored.

    3. An “approval” list for approval voting. Approval voting inherently has a strategic element to it, but you’ll usually have multiple earnest choices, and it’s never a good strategy to vote non-earnestly.

    4. A “score” list with 0, 1, 2, or 3 points for each candidate you care to list. All non-listed candidates will receive 0. Strategic concerns in score voting are similar to approval voting.

    I’ll accept any shorthand I can understand — for example, in score voting I’ll accept stuff like, “Give 1 point to everyone else remaining except Tsimerman”, and in ranked-choice I’ll accept stuff like, “Take everyone remaining except Tsimerman, shuffle them randomly, then put Tsimerman last.” Sorry to pick on Tsimerman — I’d pick on some of the other candidates if I could remember their names/spiels.

    You’ll have to trust me to respect your privacy (if I’m too lazy to build a website I’m probably also too lazy to spam you), I guess, and I’ll trust you not to stuff the ballot too much (winning the election doesn’t matter, stuffing the ballot is a waste of time). I’ll throw out extremely contradictory ballots (i.e. contradictory beyond plausible range of strategic voting), messages that aren’t ballots, and excessive ballot-stuffing.

      1. Score voting is there, at #4, with an 0-3 range. I don’t really like super-high resolution score-voting, personally…

        I don’t really buy “Equal Vote”‘s critique of “bullet voting” or “tactical minimization” in normal approval/score voting — it’s a perfectly reasonable expression of some voters’ earnest beliefs, that the difference between their favorite and second-favorite candidate is bigger than all the other differences. Or the critique of “tactical maximization”. If a voter believes the distinction between one particularly loathsome candidate and the rest overwhelms the differences between the others, that voter should maximize that distinction on their ballot. The system shouldn’t encourage them to draw some scoring distinction between as many of the candidates as possible to maximize their chances of getting counted in the runoff, possibly understating the magnitude of the distinctions they care about most (I’d call this “tactical Borda-ization”). This is the beauty of a two-round system: let people express their big-picture preferences first (on a pure approval or score ballot), then the small distinctions later, after more focused campaigning has occurred. For some people the runoff is easy — it’s between candidates they clearly rank differently — and for others it’s hard and hinges on smaller distinctions they didn’t have to think about in the first round. But they don’t have to worry too much about those distinctions in the initial round of voting.

        I also don’t like “Equal Vote” and “Fair Vote” as brand-like names rather than more neutral, descriptive names, but that’s sorta down to taste…

      2. A strategic change from 3,4 to 0,5 is a bigger distortion than from 3,3 to 3,4. This may explain why Score Runoff Voting outperforms Score Voting in some measures.

        There’s also the practical matter. IRV advocates routinely raise the bogus but politically effective argument about “bullet voting”. People intuitively like the idea of “majority winners”. SRV helps counter this naive criticism.

      3. The question is whether the 0,5 ballot is really a “distortion”. Maybe you could call it a “distortion” if you model abstract elections where voters and candidates occupy fixed positions in some multi-dimensional “political space”, but that ain’t how politics works — campaigns are dynamic, voters are dynamic, candidates are dynamic. In an approval or pure score-voting system every ballot is tactical, but voters choose between various sincere options with real tradeoffs. The “bullet vote” is a choice by a voter to emphasize just how much they prefer their favorite candidate over the rest of the field. If it annoys another candidate, if that candidate thinks they’re entitled to secondary support from these people… it’s that candidate’s job to go out and win that support. That’s how politics works.

        So if some group of primary voters says, “My candidate is the only real choice, all the others are owned by the banksters!”, with their bullet-voted ballot, and the top two finishers are among those owned by the banksters, let them back to express which of the bankster-owned candidates they prefer. They can probably be bothered to have some kind of opinion!

      4. > The question is whether the 0,5 ballot is really a “distortion”.

        Anything other than exact utilities is a distortion. There’s already some distortion caused by normalization. But if you start with normalized utilities and then push them to the min/max, you add even more.

        > Maybe you could call it a “distortion” if you model abstract elections where voters and candidates occupy fixed positions in some multi-dimensional “political space”

        The choice of utility generator is irrelevant. The issue isn’t _how_ the utilities got to be where they are, it’s just _what_ they are.

        > In an approval or pure score-voting system every ballot is tactical

        The rules don’t formally state what “honest” means, so there’s some subjectivity.
        But you can just normalize your sincere utilities and that’s an unimpeachable definition of “honest” and NOT tactical.

      5. Thanks for the reminder, I’ll make sure to bring my crystal ball and utils-chart to the poll next time… I can never remember the conversion rate between 2019-Dollars and vague feelings of belonging and progress.

      6. You bring your utilities with you any time you make any decision. That’s why revealed preference experiments can produce changes in behavior by changing probabilities.

        I guess the crystal ball comment refers to the fact that you make decisions based on your _estimates_ of utilities, since candidates may not actually do what you think they’re going to do. This was taken into account in Smith’s Bayesian Regret calculations via “ignorance factors”.

      7. Al — I totally agree with you. Bullet voting for Bernie is a sincere expression of contempt for all the other candidates. Anyone-but-Trump voting is a sincere expression of… well, you get the picture.

        These are things which people sincerely want to express. Frequently. If they can’t, they might stay home.

        I’ve always liked the idea of approval voting with a majority requirement. If none of the candidates get more “yes” votes than “no” votes, the election is rerun. We can then say that we *guarantee* that the winner got majority support!

  12. P.S. Next time an election like this happens, it could be a great opportunity for local activists to put together a physical straw poll to demonstrate the importance of the voting process. Coordinate with the campaigns to get them to drive turnout. Make an event of it. Then the candidates might see, “Oh wow, Plurality Voting was really unfair, but these other systems did so much better.”

    We missed the boat this time. :(

    1. And we will keep on missing the boats because your group and FairVote are working against each other instead of against the common enemy. … behaving as if the national debate on election systems is an approval voting contest and, with our behavior, proving on a large scale that approval voting can, and is reducing down to plurality voting, with the advocates of the status quo laughing at the flame wars between advocates of different systems.

      And then there are those who believe advocates of the status quo may be behind some of the flame wars amongst reformers.

      Prove to those who think some of the newer purported advocates of different voting systems who are trashing IRV are actually being disingenuous that you are genuinely pro-reform. Sit down with FairVote and talk through your differences. Come to some sort of agreement to let a thousand alternative flowers bloom, and end the crabs-in-a-bucket that leaves us stuck with plurality everywhere.

      1. I wouldn’t say that we’re working against FairVote. I and others in my camp have been open to IRV as a significant improvement over Plurality, and generally viewed it as a modest improvement over top-two runoff.

        However, FairVote and virtually all IRV advocates repeatedly make false and misleading claims that make it practically impossible for the public to make an informed decision. In some cases it seems an honest mistake, though in Rob Richie’s case it can only be explained by deliberate intent. Either way, this is very problematic.

        Electology’s mission is centered around education, so it’s only reasonable that they/I debunk the false claims of IRV advocates. I don’t accept your charge that this fact checking helps to preserve the status quo. And if it did, then I would simply suggest IRV proponents some the problem by creating with the false and misleading taking points.

  13. Sorry, Mike, but with memory of watching Albert Gore’s own debating style lose him an easy election to a candidate who should never have seen his tail-lights, I won’t stand by and see Ralph Nader framed for Al’s silent plea of “For God’s Sake Kill Me!” Al would’ve lost if he’d run unopposed. Which except for Ralph Nader, he did.

    Still curious about exactly what ‘tarians need Liberation from. In the capital city of our English cousins, the valiant struggle for the right to rent buildings with siding made out of rocket fuel endures unchecked. But our own Democrats nationwide are barely-alive proof that history should be the only graduation requirement in Washington for the next 20 election cycles.

    Before fleeing in terror from the Right’s sneering way of saying”Liberal” and taking refuge in “Progressive”, proof was right there in actual books that smell like they haven’t been opened since Roosevelt (the first one).

    The Progressives were the Republicans who finished the 19th century by bringing in the civil service exams that robbed so many good working-guy donors of the jobs and paving contracts they’d earned by loyally supporting their party. Often the Democratic one. So it’s not only lately that the Democrats have walked into a self delivered Everlast glove.

    Returns from every year since the Democrats adopted the description make it obvious that some Opposition Research has discovered The Progressives. And is making the most of it.


    1. Here are the facts:

      – Gore lost Florida by 537 votes
      – Nader Drew 97,488 votes in Florida

      You can argue that if Nader hadn’t run, maybe a lot of those 100k Nader voters wouldn’t have voted at all. But it would have taken nearly all of them to abstain and still see Bush elected.

      But back to the real issue: with a better voting method, it is almost certain Gore would have won. And Nader would’ve gotten like 10 times as many votes—close to a million in Florida.

      1. Well, Gore also got more votes in Florida than Bush. This was suppressed by the “Brooks Brothers Riot” or “Preppy Riot” orchestrated by the Republican Party, to prevent the counting of votes. Then 5 traitors on the Supreme Court ordered that the vote counting be stopped permanently to *make sure* that the person who got the most votes did not win.

        Never forget the treasonous coup in 2000. The traitors still need to hang for their crime.

  14. I’m with you this far, Clay. We certainly do need a change in our voting system: Eliminate the Electoral College. How much of a fight did Albert Gore ever put up to do that in his whole career?

    I also seem to recall there were some serious irregularities in vote counting, some with racial overtones, that went to court. Recount was stopped by one Justice- am I right? What would Albert, or the Country, have had to lose if he’d refused to concede the election?

    Whether or not he got the Presidency, by Inauguration Day the Electoral College would’ve been taking shell-fire like Fort McHenry when the song got written. But mainly, some personal memories that lose all my patience with any idea that Gore had a right to those Nader votes.

    First of all, what does it say about someone twice Vice President of the United States, and a career in the Senate, who couldn’t raise 540 votes to beat somebody with, politically and personally, nothing? And more important…what reason whatever to doubt he could be a worse President?

    In 1964, I sat in the TV room at school watching Lyndon Johnson lie our country into the Vietnam War, from which it’s never recovered and could well die on account of. His thinking? Next election, he didn’t want anybody to call him Soft on Communism. How would President Gore have handled the way the Right would’ve treated him when the planes hit the buildings?

    One sentence: The only thing worse than a right-wing Republican is a Democrat of any self-description that’s afraid of one. Or lacks any indication he won’t stand up to one.

    Anyhow, anybody want to help put a dead Electoral College into the Democrat’s next platform?


    1. Yeah, I’ll happily help with a campaign to kill the electoral college. But the most *effective* way to do this is something called the “National Popular Vote Interstate Compact”. If successful it will render the electoral college irrelevant. Which will make it much easier to eliminate it by amendment.

      1. Washington State has already entered into the interstate compact to elect the president through popular vote, via legislative act.

        It is telling that those dreaded IRV proponents have been working hard on that, while proponents of other systems only seem to show up to tear down campaigns to install IRV.

        It reminds me of BRT advocates who show up to oppose light rail, and then disappear back into the woodworks not to be seen when BRT is on the table, or ever work to get BRT onto the table.

  15. A ballot like the one shown, a visual nightmare, would be a total turnoff to a large percentage of voters.

  16. I seem to remember that Pierce County’s RCV was undermined by their elections supervisor, who worked rather hard at implementing it poorly, and then blamed RCV. Any voting system can be poorly implemented. In jurisdictions where it’s been implemented well, it’s generally been well-received.

    A few years ago when the King County charter was being reviewed, a few of us worked hard to encourage RCV in KC. What we ran into was that elected officials have been elected under the current FPTP system, so have a vested interest in keeping it.

    RCV would make our democracy more democratic. FPTP is archaic.

    1. Sure, IRV is better than Plurality Voting. That’s not saying much. Still, it’s the worst of the five commonly discussed alternative voting methods. Among ranked systems, Condorcet systems and even Borda are generally superior.

      Score Voting and Approval Voting are radically simpler and even better in general. So it’s just bizarre that activists spend so much time talking about IRV, a system that most of its very own advocates don’t even understand.

      I go through some of IRV’s flaws, and compare to Score and Approval here.

      1. It’s amazing that IRV is worse than Borda — Borda is awful if anyone votes tactically. But it is nevertheless true that IRV is worse than Borda.

        (I once talked to the leadership of a club which used the Borda count specifically so that the leadership could control the results, because they knew how to vote tactically and the rest of the club members didn’t. IRV is actually *worse*.)

      2. Score voting and approval voting are not radically simpler, unless you have some counterintuitive definition of “simpler”. How would you have voted on your mayoral ballot this year?

        IRV and Condorcet are both quite simple: You rank the candidates in order of your preference.

        Sure, there are ways to game the system, if you can visit the future and see how everyone else voted. Sure, you can even guess in partisan races how many people will vote for each party because the similar elections have been run over and over. In a nonpartisan race, that advantage to voters who have looked at past results goes out the window.

        The only way I knew how to vote an effective IRV or Condorcet ballot in the mayoral primary was to rank the candidates honestly. I would have been totally lost with approval voting, and not at all sure what to do with score voting.

      3. People’s voting choices aren’t even the biggest weakness of Borda. If you want to game Borda, just run as many candidates from your coalition as possible.


    Don’t believe me? Look at Australia’s lower house. They use it. They have spoiler-candidate and vote-splitting problems.

    Want the mathematics behind it? It’s been published repeatedly. Phone up the Center for Election Sciences, they’ll explain it to you.

    For your mayoral election, you want APPROVAL VOTING, which is dead simple. Look it up in Wikipedia. IT WORKS.

    1. Where has approval voting worked?

      How the heck am I supposed to figure out how to cast an effective vote under it? I would have been totally lost trying to figure out what to do with it in this mayoral race. If there are more than two candidates, many voters won’t know what to do with it.

      No voting system is perfect, according to Arrow’s Theorem, but approval voting doesn’t even let us differentiate between the candidate we like most, the candidate we like second-most, and the candidate we like third-most. You can’t algebra your way around that simple basic Achilles’ Heel of approval voting.

      You can’t do it with our traditional first-past-the-post either, unless you throw in an expensive primary election. So at least we all seem to agree FPTP is a gawd-awful voting system.

      1. It is a common misconception that Arrow’s Theorem proves no voting system is perfect. Arrow’s Theorem proves that, for a given definition of fair, no full or complete ranked-choice voting system is fair.

        While there are surely problems with Approval Voting, the lack of rankings allows it to sidestep the hypotheses of Arrow’s Theorem, so Arrow cannot be used to prove that AV is unfair. This advantage to AV is precisely *because* it does not allow differentiation between each ranking (first, second, third, & c.); voters are free to choose whichever distinction is most important.

        With any voting system other than FPTP, many voters won’t know what to do with it. However, for Arrow-eligible voting systems, the optimal strategy is often dishonest (i.e. ranking a less-liked candidate above a more-liked candidate). Unlike all Arrow-eligible voting systems, the optimal strategy under AV is always honest: it is never strategically advantageous to vote yes on a candidate that one likes less than a different candidate on whom the same voter votes no.

      2. Approval Voting has been used in many large organizations, and in public initiatives, not to mention numerous large experiments. People seem to have little problem filling out their ballots. The general strategy is to vote for the same candidate you would with the current Plurality Voting system, _plus_ everyone you like better. You actually see that people asked to rank a long list of candidates often only rank a handful of them, so this isn’t altogether different.

        Having visited Kenneth Arrow in his Palo Alto residence, I can tell you Arrow’s Theorem most certainly doesn’t apply to Approval Voting.

        It’s noteworthy that the designers of Tezos, a new alternative to Bitcoin, chose Approval Voting as the voting mechanism to allow participants to decide on future design changes. This is a business critical and they want a system that’s highly resistant to strategy. Consider that Bitcoin reached a market cap of 120 BILLION dollars, and the Tezos initial fundraiser raised 160 million dollars. These brilliant mathematicians chose Approval Voting over a ranked system like IRV, with strategic voting and game theory on their minds.

        Moreover, IRV has a number of major flaws that I highlight in this presentation.

      3. As Eric points out, approval voting avoids some mathematical oddities by forcing voters to offer less information. Not a good selling point.

        The only claim about approval voting in all the comments that I’ve seen hold up is that it is easier for the administrators to count than RCV is.

Comments are closed.