Instant-runoff voting: avoiding the third-party "spoiler" problem

Great Debates
1

Since there’s a GD thread going on right now, “What do you think about proportional representation in the US House of Reps?” (http://boards.straightdope.com/sdmb/showthread.php?t=261571), I thought I would take this opportunity to open a debate on a related but very different electoral reform: Instant-runoff voting. It is much less controversial than PR, and also much simpler and easier to understand.

PR is used to elect multi-member policymaking bodies such as Congress or your
school board. IRV is for electing somebody to fill a single office such as president or governor or sheriff – or, under a single-member-district system, members of House, the Senate or a state legislature.

Most American public officials are elected by a “plurality” system, meaning that the candidate who gets the most votes wins – even if “most votes” means less than 50%. That works fine if there are only two candidates in the race, but what if there are more than two? Then its possible for a candidate whom most of the people don’t want, to win the office. E.g., if candidate A were running only against candidate B, then A might win by 55% to 45%. But it candidate C enters the race and “splits the opposition” to candidate B, then A wins only 35%, B wins 41%, and C wins 24%. In the latter scenario, B wins the office – even though the majority of the voters don’t want him! This is, to say the least, an undemocratic result. And it (rightly) detracts from the winner’s perceived mandate to govern – in his first term, Nixon suffered from the stigma of being a “minority president.” It also reduces voters effective choices – third-party candidates or nonpartisan independent candidates are discouraged from running in the first place, for fear of being labeled “spoilers.”

Some jurisdictions solve this problem by having a second-round “runoff” election if nobody wins a majority in the first election. But this has several disadvantages. First of all, it costs money. The public has to spend tax dollars on a second election, the surviving candidates have to raise funds for a second round of campaigning – and we have enough campaign-financing problems as it is! Furthermore, voter turnout for the runoff is usually lower than for the first election. By this time, everybody is sick of the whole thing – who wants to sit through two campaign seasons back-to-back?

A better way to solve the problem is instant-runoff voting. Under IRV, if more than two candidates are running for the post, you the voter don’t have to pick just one – you can rank-order them by preference. If no candidate gets a majority of first-choice votes, then the candidate who runs last is eliminated, the votes are retallied, and those voters who made first-choice votes for the eliminated candidate get to have their second-choice votes counted towards electing their second-choice candidate. The process repeats until a candidate with a clear majority emerges.

E.g., when Nader ran for president in 2000 he was reviled by liberals as a “spoiler” who was “splitting the opposition” to Bush, drawing away votes that otherwise might have gone to Gore. (The same problem arose when Teddy Roosevelt ran for president in 1912, Strom Thurmond and Henry Wallace in 1948, George Wallace in 1968, John Anderson in 1980, Ross Perot in 1992 and 1996.) Under an IRV system this would not have been a problem. A Nader supporter could have voted as follows:

1st choice: Nader (Green)
2nd choice: Gore (Democrat)
3rd choice: Browne (Libertarian)
4th choice: Bush (Republican)
5th choice: Buchanan (Reform)

(At least, that’s how I would have voted.)

If Nader did not get a majority of first-choice votes, a vote for him still could have counted to elect the voter’s second choice – in most cases, Gore, who probably would have won the election with a solid majority mandate. And if Gore still hadn’t won, well, at least nobody could have blamed Nader for that. Nader, in fact, would have added to Gore’s votes, by turning out voters who otherwise would not even have gone to the polls at all. IRV increases voter turnout.

IRV also discourages negative campaigning. If there are several candidates in the race, and each one knows that another candidates supporters are also, potentially, his own supporters – then the candidates are much less likely to “go negative.” Candidate A is more likely to say, “I respect candidate C’s views, but they’re just not as wise as they could be, and I have better ideas with respect to the following issues . . .” Much more civilized.

Note that IRV is not just a “third-party” reform but can also be used to open the field wider and ensure a more democratic results in nonpartisan elections. In the U.S., most municipal elections are nonpartisan. And, of course, independent candidates, running in partisan elections but not nominated by any party (where this is permitted), can also benefit from IRV.

The city of San Francisco approved a switch to IRV for city election in a 2002 referendum; it will be used for municipal elections this November. Pro-IRV groups have formed in New York (http://www.nysirv.org/), Utah (http://www.utah.fairvote.org/), Massachusets (http://www.massirv.org/), and Illinois (http://www.instantrunoff.com/). See also the Instant Runoff Project at http://www.instantrunoff.org/, and the Center for Voting and Democracy at www.fairvote.org.

So, what do you think? Is IRV a good idea or not?

2

The chief reason it would be a bad idea is that it makes voting more complex.

With some sort of web-based voting application, it should be doable.

If you have to use punch cards, it dramatically increases the chance of voting error. This is not worth the “fairness” benefits, because in todays political climate, the margin of preference between the two candidates is very close to the margin of error in recording the votes. We saw this in Florida – the margin of preference was actually smaller than the margin of error.

Anything that increases the margin of error is very bad, including a complex voting choice.

Again, with a web-based voting application, you could actually force the voter to make a valid choice before he could submit the vote, and in this case, the margin of error would be small.

3

By “web-based,” do you mean electronic touch screens? Those are what are supposed to be in place everywhere by November, and they could, as you say, force the voter to record a valid choice – but they aren’t “web-based,” they’re stand-alone units with no connection to the Internet. (In theory, anyway . . .)

4

IRV is the way we do it here in Australia, but we call it preferential voting. It certainly gives third parties and independents more of a chance.

All these punch-cards and electronic machines seem so complex and easily broken - you could do what we do and have hand-written ballot papers. The instructions are very simple - number the candidates in the order of your preference - put a 1 next to the one you like the most, 2 beside the next and so on. It means that you have to count them by hand, but you generally know the overall result a couple of hours after polls close, except in very close races. No hanging chads or dodgy Diebold-type companies and so on.

5

Its a variation on the single transferable ballot system isn’t it?

Personally I like it and I fail to see how it can be considered complicated. The ballot design should be simple enough.

6

No, it isn’t. It’s easy to confuse them. But single-transferable voting, aka “choice voting” or “preference voting,” is a method of PR, that is, a way of electing a multimember policymaking body so that it fairly represents the full range of viewpoints in the community. From the website of the Center for Voting & Democracy:

7

There’s also approval voting, which is another take on the typical American first-past-the-post plurality system. With approval voting, you just cast a vote for each candidate that you like. There’s no minimum or maximum of total votes you can cast, but you’re limited to one per candidate. So if I wouldn’t mind seeing Bush, Browne, or Buchanan in office, I could vote for all three of them and ignore Gore and Nader. Most total votes wins.

Then there’s the voting system I devised a while ago (it may exist in some other form), wherein you’re given multiple votes (2n-1, where n is the number of candidates) and you can use them on any or all of the people running, registering your preference intensity by giving some candidates more of your votes than others. You could choose to use as many or as few of your apportioned votes as you wish. Most total votes for one candidate wins.

More food for thought.

8

It must the hour, becasue I’m not sure of the difference.

9

In the first, IRV, called preferential voting here in Aus, the process is used for a single seat / electorate. If you have 50% + 1 of the first-preference votes, you win the seat. If not, the candidate with the smallest number of first-preference votes is eliminated, and their second preferences taken into account. No absolute winner? Repeat the process as many times as required.

For example:
Candidate A gets 45% of the vote
Candidate B - 45%
Candidate C - 10%

No clear winner, so C is eliminated. Perhaps nearly all C’s voters chose A as their second preference, so their votes go to A. A now has more than 50% of the vote and wins the election.

In the other example, “single transferable vote” or “preference voting”, which I’m not familiar with, it seems you have preferences as well, but you’re electing more than one seat, perhaps on a local council. I don’t know this one well enough to explain it.

10

Just to make sure we’re absolutely clear, no voting system is going to be without its problems (cf. Arrow’s Theorem), given some requirements as to what a “fair” outcome is. For instance, consider the following example for IRV:

There are three possible ballots (i.e. everyone who votes ranks the candidates in one of three ways), with the number in parentheses representing the percentage of voters who ranked the candidates in that manner:

A > C > B (45%)
B > A > C (40%)
C > B > A (15%)

In this situation, C, having received only 15% of the first place votes, gets dropped, transferring his votes to B (the second choice), resulting in B’s getting the win. However, A’s not winning causes two problems, the first of which is that the plurality choice did not win. A was first on more ballots than anyone else, and did not win. But, more importantly is the second problem: Accepting that the plurality choice might not win leads us to conclude that the next logical step is to allow the person who was most likely to be the first or second choice to win. However, that person is also A.

Rather than allowing the plurality choice to win, satisfying 45% of the voters completely, and 45% somewhat, we instead have candidate B winning, satisfying 40% of the voters completely and 15% somewhat. This clearly seems to be an undesirable result.

This is a bit of a contrived example, but in general, IRV is fine when the third parties are weak, and it is mainly used to show support for that party, rather than to actually put the party in power. Once the third party begins to take a significant amount of the vote away from one or the other party, it becomes less clear that the final result is actually representative of what the populace wants.

11

It is true that in some situations like that supplied by MilTan there are some situations in which no candidate is preferred by the voters to each of the other candidates. One problem with the example is that, if there was a first-past-the-post system with voters’ preferences as indicated, it would be be in C’s interest to drop out of the race, since C presumably prefers B to A as a candidate and C can only be elected if A drops out of the race. It would also be in the interests of C’s supporters, knowing from opinion polls that C is coming third, not to vote for their first choice but to vote for their second choice B to stop their third choice from being elected (the well-known “don’t waste your vote on a third-party candidate” argument).

However, in real life, I suspect that examples of the Arrow paradox are pretty rare. Typically, candidates fall along some political spectrum (e.g., from left to right), and voters’ preferences reflect that spectrum. So if A is left, B is centre and C is middle, you’ll find that most of A’s voters will prefer in the order A-B-C, most of C’s voters will prefer in the order C-B-A, and B’s voters will split between B-A-C and C-A-B.

What you can get there is a case where the two extremes are more popular, and the middle is less popular, e.g.

A-B-C 45%
B-A-C 6%
B-C-A 3%
C-B-A 46%

In a plurality system, C wins (unless B withdraws from the ballot). In a single transferrable vote system, A wins (with 51% of votes after B’s preferences are distributed). But in two-way contests, B would beat A (with 55% to 45%), and B would beat C (with 54% to 46%). So in a single transferrable vote system, it would be in the interests of C (the plurality winner) to withdraw from the ballot in order to get B elected.

I believe that something like this (a major party candidate getting out of a race in order for a centrist candidate to get elected rather than a candidate on the other end of the political spectrum) does happen in real life in some elections in Australia. It is reasonably common for independent candidates to get elected – much more common than you would expect given the fairly low level of support for independent candates generally – and paradoxicaly they are more likely to get elected in seats which appear “safe” for one of the major parties.

Typically what happens is this:
(1) There is some controversy within the major party holding the seat, e.g. over a preselection (the equivalent within Australian political parties of primaries in the US), and either a poor candidate is chosen in the preselection or a popular party member (who may already be an MP) resigns from the party.
(2) Either you have a former party member running as a centrist independent, or you have a locally popular person running as an independent against the major party which has held the electorate as a “safe” seat.
(3) The other major party runs dead: either it doesn’t nominate a candidate at all, or it nominates a poor candidate and doesn’t campaign for him/her. It’s public knowledge that they are doing this because there is a good independent running, who they would prefer over the candidate of the other major party.
(4) The “safe” major party loses a lot of support to the independent, and the other major party encourages its supporters to vote for the independent too. (and, of course, if the other major party is still running a candidate, it directs preferences to the independent on its “how-to-vote” cards).
(5) So you might get a result like this:
“Safe” major party 45%
Independent 30%
Other major party 25%
(There can be other candidates of course, but they don’t affect this big picture.)
On those figures, if the preferences from the other major party hold tight by more than 80% (i.e., more than 80% of the 25% gon on the second count to the independent), the independent wins.

12

In 69% of elections with three candidates, the outcome can be changed by changing to a different vote counting rule.

13

In principle, strategic voting in first-past-the-post systems is IRV with the first round being in the court of public opinion.

PS-- I like approval voting better

14

Well, from the point of view of your experience in the voting booth, they’re pretty much the same: You’re presented with a list of names. You ask yourself, “Which of these would I like to see in public office?” And then you rank-order them by preference. The difference is in how the votes are tallied. In IRV, there’s only one slot to be filled – only one of the people you vote for will be elected. If the first-choice votes produce no majority, the name with the least first-choice votes is eliminated, the votes are retallied, and the process is repeated until a majority winner emerges. But STV is used for filling multiple seats in a multimember policymaking body; it is possible that several of the people you vote for will wind up in the legislature, or the city council, or whatever.

15

I couldn’t see any mention of that 69% in the paper at the URL (though it is an interesting paper).

However, the fact that changing voting systems can change outcomes is not a proof of the Arrow paradox. That paradox is that no voting system for choosing between three candidates will always satisfy certain conditions (and all of those conditions look reasonable for elections).

16

There’s something I’ve noticed about this thread as compared to ready29003’s thread on electing the House of Represenatives by proportional representation: On that thread, some posters have raised objections to PR on principle – the principle, I guess, that losing is losing and if your party can’t muster enough votes to win representation in a single-member-district plurality system, it’s just a fringe group anyway. (I’ve already responded to that as best I can.) But in this thread, nobody has objected to IRV on principle. Jawdirk objected that IRV might be too confusing for the voters, and Giles brought up the “Arrow paradox” – i.e., political scientists have proven that no voting scheme, however sophisticated, can perfectly translate the voters’ preferences into reality. But nobody has called the idea unfair, or disruptive, or in any way a fundamentally bad idea. I attribute this difference in reactions to two factors:

  1. IRV is much simpler and easier to understand than PR.

  2. IRV is much less likely than PR to upset the status quo, and if it does upset it, the process will take a long time. If we had had IRV in 2000, I’m sure Nader still would not have won. Minor parties have a better chance under IRV, and a new incentive to organize, but they still would remain mostly shut out of government for some time to come.

My thinking is, if these movements go anywhere, we will have IRV (or AV) before we have PR. That is, if we adopt the far less controversial reform of IRV, and a few third-party candidates do get elected to a few offices, they will be in a position to call for further reforms, including PR. Which will at least spread the word about the idea, which is more important than anything else. As I’ve said before, the main obstacle to PR in the United States is not that the people are against it but that most people have never heard of it, and explaining takes more words than you can fit on a bumper sticker. Same problem with IRV – but explaining it is much easier.

Why? I can think of only one reason why AV (which is IRV without the rank-ordering – you just check the names you find acceptable and leave the rest blank) is preferable to IRV: It’s easier to tally the ballots under AV. But who cares about that? We have computers. IRV produces a more sophisticated result, in that with IRV, the losing candidates can look at their place in the preference-ranking to precisely gauge their level of public support – e.g., if we had had IRV in the 2000 presidential election, Nader could point to the fact that he got more first-choice votes than Browne or Buchanan (if, in fact, he did).

17

It is not open to strategic voting. Because it ignores preferences, the monotonicity criterion cannot be violated. Why is this good? Because you can never hurt your favorite candidate’s chance of winning by voting for them.

18

Errmm . . . sorry, could you please explain that in greater depth?

19

It would be awfully silly to reject instant runoff voting because of Arrow’s Impossibility Theorem. After all, what Arrow shows is that no voting system accurately translates voter’s preferences to outcomes in a way that conforms to the requisite (and quite reasonable) conditions.

This, of course, includes first past the post voting schemes. IRV is no worse off by Arrow’s lights than FPTP is. So the question becomes, which system will, under conditions commonly met, generally produce more desireable results.

I think IRV beats FPTP, hands down, in terms of choosing a single candidate to represent a constituency. It’s still subject to strategic voting (i.e., voting for someone other than your preferred candidate to prevent someone else from winning), but less so than FPTP. It mitigates to a large extent the effects of vote-splitting between candidates with similar policies, which can often allow minorities with other views to beat out the majority. Most importantly, it returns a lot more information than FPTP voting does. For example, nobody really knows how many US voters would pick Green/Libertarian/whatever as their first choice, because most of those who would are pragmatic enough to vote for whichever of the two main parties is least offensive to them. This results in a government which believes the voters want it to behave in manner X, when in fact the voters want it to behave in manner Y. And democracy, after all, is supposed to be (to a large extent) about the government behaving as the voters desire. So if it were to turn out that a significant number would rank so-called fringe parties ahead of the major parties, if they could do so without feeling they were wasting their votes, then IRV voting would reveal that support (or at least, reveal it to a greater extent than FPTP) and politicians could then be pressured to respond to those voting trends.

Approval voting has most of these advantages, although without the rank ordering it returns slightly less information. I disagree with eris that it isn’t open to strategic voting. It is, but strategic voting would manifest itself in a slightly different fashion. For example, suppose one is voting in a region where three candidates are neck and neck. Let’s use my riding in the upcoming Canadian federal election, although I suspect that the Liberal candidate is actually a fair bit behind the NDP and Conservative candidates. We’ll pretend she isn’t, though. Now, suppose my preferences are as follows. I prefer, for the sake of this example, NDP over Liberal by a substantial margin, but prefer that the Conservatives not win by an even greater margin. Now, obviously I’ll put an X on my ballot next to the NDP candidate, but whether or not I put an X next to the Liberal candidate depends on how likely I think the respective chances of the Liberal and Conservative candidates are. If I think the Conservative is likely to come in last of the three parties, I’ll only vote NDP. If I think she won’t, I’ll vote both NDP and Liberal. So you see, while I won’t be tempted by strategic reasons not to vote for someone I want to see win, I will be tempted by strategic reasons to vote for someone I don’t want to see win. IRV would eliminate this particular dilemma, although it quite admittedly results in strategic voting in other regards. I doubt there is any voting system that would avoid strategic voting.

20

Wikipedia’s entry explains it better than I probably could. Look under the “Potential for Tactical Voting” heading. Of particular interest:

Now, to what affect this can actually happen, that’s a debate. But anyone who thinks the spoiler effect is real should see the potential problem here.