Why not instant run-off voting?

What are the major arguments against instant runoff voting? After learning about it (http://www.instantrunoff.com/), it seems to be the fairest way to elect officials.

I’m all for IRV but I have, in the course of my researches, encountered some contrary arguments.

From the Wikipedia (http://en.wikipedia.org/wiki/Instant-runoff_voting):

Regarding the “monotonicity criterion” (http://en.wikipedia.org/wiki/Monotonicity_criterion):

From “The Problem With Instant Runoff Voting” (http://electionmethods.org/IRVproblems.htm):

My own thinking is, the above arguments might have some weight, but the institutional barriers to third parties becoming major players in American politics are so all-pervasive that any reform that opens the process to them has got to help their chances more than it hurts them.

See also the following GD threads I have started, or to which I have contributed:

“A multiparty system is better than a two-party system!”

“What do you think about proportional representation in the US House of Reps?”

“Instant-runoff voting: avoiding the third-party “spoiler” problem”

“Yet another electoral-system reform: “ballot fusion,” or “cross-endorsement””

“Should the U.S. adopt alternative, pro-multipartisan voting systems?”

BTW: The City of San Francisco adopted IRV for municipal elections by referendum two years ago, and this November the voters will use IRV for the first time to elect their Board of Supervisors. Afterwards, we will have a chance to study that election and see how the pros and cons of IRV shape out in practice, under modern American political conditions.

I cannot think of an example of instant run-off voting which breaks monotonicity. In addition, I find it hard to think of a case where it will work: if you reduce the votes for your preferred candidate, then on each count your preferred candidate is more likely to be eliminated and less likely to be elected. Can anyone give an example where monotonicity is broken?

I can, however think of examples of strategic voting, where it makes sense not to vote for your number 1 preferred candidate so that that candidate 2 will defeat candidate 3. In addition, I have see this kind of strategic voting happening successfully in real life elections in Australia.

Paradoxically, in real life this seems to happen in most cases where candidate 3 is a relatively poor candidate in a seat which ought to be very safe for his party, so that candidate 1 from the opposite party runs dead, and lets his party’s votes mostly go to a middle-of-the road independent candidate 2. (And I’ve seen it working in both directions in Australia, with Labor candidates losing very “safe” Labor seats and Liberal or National candidates losing very “safe” Liberal or National seats).

I think it’s a great system, partly because (in cases like these) it makes the major parties have to work hard not to take safe seats for granted, and because it means that voters can vote honestly for independents and minor parties, without being told that “they are wasting their vote”.

IRV usually doesn’t “work” in the sense of making it more likely a third-party candidate will be elected to the office, Giles. It works by making that a realistic possibility, while encouraging the third party to organize and run its own candidates. If we had had IRV for the 2000 presidential election, Nader would not have been elected president – but his candidacy could have sparked a nationwide organizing surge for the Green Party, plus winning them permanent ballot access in a lot of states, and getting a high enough (first-choice) vote share to qualify for federal matching funds in future elections. In other words, IRV can help a minor, marginal third party grow into a real, competitive party – one that sometimes will get its candidates into office. And it also makes things simpler for the major-party candidates by eliminating the “spoiler” problem. E.g., in 2000 Gore, instead of sweating bullets over the problem of Nader’s presence in the race, could have pitched his message to Nader supporters, to persuade them to make him their second choice.

Sure, but as long as you’re reforming the system, why choose a voting method with all these known problems, when there are other systems that will serve the same purpose? Condorcet voting and approval voting both help third parties, without the flaws of IRV.

Let me repost the example I gave in a previous thread:

Sure, in certain scenarios preferential voting (what we call IRV here in Australia) can lead to an apparently aberrant result if many people vote strategically.

However, most people don’t vote strategically. They just go into the cubicle and choose the party they like the most and work down from there. Meanwhile, a few clever people from each party are plotting to manipulate the result, but their votes would largely cancel each other out. Therefore, strategic voting is not really a factor here.

But that’s not the real problem. IRV has the same flaws whether you vote strategically or not. The problem is that sometimes, ranking a candidate higher will make him more likely to lose (or vice versa), and it’s hard to know when.

Exactly: most people don’t vote strategically, so instead of using IRV’s flaws to their advantage, the flaws will work against them. Those people who go into the booth, rank their favorite party #1, and work down from there may end up hurting their favorite party instead of helping it.

How could a vote for anyone at #1 end up hurting a party, given a largely two-party system?

To clarify, Mr 2001, do you have some real life examples of an illlogical outcome in a place that employs preferential voting / IRV, perhaps in Australia?

Well, the point is to transition to a multiparty system, right? Even if we assume that IRV’s flaws won’t show themselves until third parties gain more support, what’s the point of adopting a voting method that can only be trusted during the transition period?

Real life examples? Nope. Frankly, I’m not sure how one would tell whether a real life outcome is “illogical” without interviewing each voter to find out his motivations, or at least recounting all the ballots and evaluating them under another voting method.

Still, I’ve shown how IRV can lead to counterintuitive results in theory, and I haven’t seen any reason to believe those problems wouldn’t occur in real life. The Condorcet method (among others) is easier to tally, produces more intuitive results, and works exactly the same in the voting booth; I have no idea why anyone would prefer IRV when there are alternatives.

Yes, whoever loses should instantly run off. :stuck_out_tongue:

Well, the point is to transition to a multiparty system, right? Even if we assume that IRV’s flaws won’t show themselves until third parties gain more support, what’s the point of adopting a voting method that can only be trusted during the transition period?


Obviously, because there has to be a transition period! The relative merits of IRV, the Approval Voting system and the Condorcet Method can be discussed after we have thoroughly and permanently broken the back of the two-party system!

My understanding of the Condorect method is that with it you have to compare each pair of candidates, and the candidate wins who beats all the other candidates.

My criticisms of it are:

(1) Sometimes no candidate would win against all the other candidates. If you have three candidates, and three voters, with these preferences:
A 1 3 2
B 2 1 3
C 3 2 1
then A beats B, B beats C but C beats A. That means you have to find some other method to brealk the tie. (With IRV as used in Australia, the method of breaking the tie in this case would be to eliminate one of the candidate with the lowest number of votes, choosing one at random – with just three voters like this, that makes the vote completely random, but in practice, this isn’t a problem when you have thousands of votes).

(2) With a lot of candidates (10 or 12 is not an unusual number for a single-member electorate in Australia), you would have a lot of counts to make, so it would conbsiderably increase the work.

(3) It would also increase the work required of the voter, unless a voter wanted to just follow a party list. It is common in Australia to use “optional preferential” voting: that is, you are just required to vote “1” for your preferred candidate, but may vote 2,3,4, etc., for other candidates. With IRV, if I know my candidate is one of the two likely to be in the final count, I can safely vote just “1”. With Condorect, I might worry about all the other possible pairwise combinations, and so need to fill up my ballot paper (making sure that I don’t make an error in the numbering), or I might have an option to vote for a party list (as is done in the Australian Senate or the NSW Legislative Council, where there are PR systems, and where there can be hundreds of candidates to vote for).

Note also that you also need special rules where a voter makes a mistake in the numbering. If a voter numbers the 10 candidates as 1,2,3,4,5,7,8,9,9,10, with IRV you can either declare the whole vote informal, or you can say the vote works validly up to the 5th preference, and is invalid thereafter, or you can say that the vote works up to the 8th preference, and is invalid thereafter. What would the rules be in Condorcet? And how do you explain them so that all those involved understand them?

Look, no voting system, in terms of translating the individual voters’ wishes into collective decision-making, is going to be perfect. No cite, but I once read an article in Scientific American where mathematicians demonstrated that it is logically impossible to devise a flawless voting system. Sure, PR might have its flaws. The important thing is that it’s a better system than we’ve got now. And I think a system where the voter gets a chance to rank-order candidates by preference – to actually express his or her choices in a finely nuanced way – will be easier to sell to the public than, say, an approval-voting system where the voter only gets to check yes-or-no on each candidate in the race.

Sorry, I meant to write, “Sure, IRV might have its flaws.”

But why use a voting method that we’ll have to trash as soon as the back is broken, when we could instead use a method that’ll work before, during, and after the transition?

Thing is, there are many systems better than the one we have now, and although none are perfect, some of them are better than others. “Hey, it’s better than what we have now” isn’t a reason to pick IRV over Condorcet; it’s a reason to pick one of several systems over plurality voting.

True. Two methods for breaking the tie are described here.

Well, it’d be more work than our current plurality voting, but less than IRV. With plurality voting, you need N counts for N candidates. With Condorcet, you need N*(N-1) counts, one for each matchup. With IRV, you need N factorial counts, one for each possible ordering of candidates.

Suppose you’re running a polling station in an election with 10 candidates. Under our current system, at the end of the day, you send the counts for each candidate to election headquarters - up to 10 numbers. If we used the Condorcet method, you’d need to send in up to 90 numbers. But if we used IRV, you’d need to send in the number of votes for each unique ballot ordering (which could be over 3.6 million if your precinct is large enough) - if you had 1000 ballots all marked differently, you’d need to send in 1000 numbers.

Once the ballots are counted, the counts can be entered into a computer and evaluated quite easily.

Hmm, my understanding was that Australia required a rank for all ballot positions. I heard something about Australia randomizing the candidates’ order on their ballots, because voters with no preference would just go down the list and rank everyone in order.

In any case, I believe you can do the same thing in a Condorcet election. If you only rank candidates A, B, and C, but leave D and E unmarked, then you’re saying A>D, B>D, C>D, A>E, B>E, and C>E, and expressing no preference about the matchup between D and E. You can still enter the known preferences in the matrix.

The simplest rule would be to declare the ballot invalid.

Or, I suppose you could say it works up to the 7th preference (or the 5th, if you want to be totally unforgiving), and partially works thereafter.

Let’s say the candidates are named A through J and ranked in the ambiguous order you described. Even though we don’t know whether the voter prefers H to I or vice versa, since they’re both a 9, we do know that he prefers A-G over both H and I, and he prefers anyone over J. We can still enter those preferences in the matrix, even though we don’t have anything to enter for the HI matchup.

Generally, in elections in Australia, the preferences are only counted after the number 1 votes in an electorate have been tallied. There’s a provisional count of those first preferences at the polling station, then the votes are packed up and sent in to a central tallying place for the whole electorate. So you don’t have to count n! different possibilities. On the other hand, often the result is not known for one or two weeks – though that possible delay is built into the timetable.

Again, generally there used to be a requirement that ballot papers be numbered completely. However, to reduce informal voting, most elections in Australia use an optional preferential system now. So a vote can be formal, but can become exhausted in a preference count when all the candidates that have been given a number have been eliminated from the count.

And, yes, candidates used to be listed in alphabetical order, but now are listed on ballot papers in a random order. In elections using PR (e.g. the Australian Senate and the NSW Legislative Council), candidates are grouped by party, and the randomisation is for the party groups. Within each party group, the candidates are listed in the order decided on by the party. And a more recent innovation has been allowing voters to vote for a party list, by putting a “1” into just one square, rather than having to list al their preferences. If they vote for the party list, hen their ballot paper will be counted in accordance with the list of preferences given in advance to the Electoral Office. In practice, 80-90% of ballot papers in these elections are marked with a “1” for the party list, rather than with preferences for all the candidates, and this is because pretty well all the parties recommend such a vote to their supporters (to minimise the informal vote).