Friday, June 5, 2009

Presented

Recently, I gave a brief presentation to the local chapter of the Green Party. I wouldn't call the following a "transcript", but it does hit the same points.


Between 1996 and 2000, the Green party increased their share of the votes in the presidential election by over 400%. But between 2000 and 2004, despite flagging support for the incumbent Republican, a lackluster Democratic challenger, and an increase in public support for the Green stance on virtually every policy issue, the Green party share of votes dropped by 96%. How is that possible?

Because Nader, and the Green party, was branded as a spoiler. When you spoil an election, your support dries up as voters go back to voting for the lesser of two evils. This is a known problem with our current plurality election method, one that economists, mathematicians, and political scientists have known about for years, and they have also known for years of alternative election methods that don't suffer from this problem. And yet, it doesn't seem to come up as something worth fixing.

Thankfully, recently, there has been some movement towards trying different election methods. Unfortunately, one often-proposed method, instant-runoff voting, which is a plank in the Green party platform, does not fix this problem. I'm here to show you some methods, called approval voting and score voting, that do fix this problem. Allow me to show you some examples.

This is a simple example with just nine voters; four vote for option 'A', and five for option 'B'. B wins. But what if we add a new option, 'C', that appeals primarily to the B-voters.

C doesn't win, but now, neither does B, so C spoiled the election. This is basically what happened in Florida in 2000. Instant-runoff voting (IRV) claims that it can fix this problem. On an instant-runoff ballot, instead of listing only one choice, you can list several choices in order. The candidate with the fewest votes is eliminated, and votes for them are re-allocated to the voter's next choice in line, and you repeat this until one candidate has a majority of votes. So, in our example, C has the fewest votes, so it's eliminated, and the votes go to the voter's second choice, which we know from before is B. B now has five votes, and wins, as they should have all along.

This looks great; it looks like IRV fixed the problem. But you still haven't won any elections. Support for a new party grows slowly over time; you don't go from zero to victory. So let's give C one more of B's votes.

Okay, now, you're doing really well; you've actually got more votes than one of the major parties. But now, we have to ask a question that we haven't answered before: what's the second choice of these last two B voters? If one of them (let's call him "Al") has A as his second choice, we get this:

Now, A wins. Remember, without C involved in the vote, B would win. But because C is here, A wins. So C is, again, a spoiler. And here you'll still lose 96% of your support in the next election.

A better option is score voting or approval voting: I'll do an example with approval since it's a bit simpler. With approval voting, instead of indicating just one candidate, you can indicate as many or as few as you want. So, those first two votes for C can come without in any way diminishing B's five votes, as can the third; and as your support continues to grow, maybe you convince some of A's voters to approve you as well, until you win. Without ever being a spoiler.

Score voting is only slightly more complicated, in that it gives you the option to explicity state a willingness to compromise, by letting you give each candidate a score in some range, such as zero to ten. So you could say, for instance, I give C my full approval at 10, but I'd be pretty satisfied with B as well, so I give it an 8.

Either way—score or approval—no spoilers, no abandonment by the electorate, and so your support can continue to grow until you actually win.

Now, what I want from you is I'd like to promote this idea to the rest of your party; your national conference is in July, perhaps something there. Also, IRV has had some pilot programs conducted, in California, Vermont, North Carolina; I'd would like to ask for your support in getting some sort of pilot program using score voting here.


Then there was a brief question-and-answer period. I was asked to reiterate the example of how a spoiler happens under IRV (the important part is that not all B-first voters will vote C-second), to speak to IRV's "success" in San Francisco (before IRV: one Green elected to council, after IRV: one Green elected to council. Change: zero), and to discuss briefly proportional representation (also a good idea.)

Overall, I thought the experience was very positive, and I will be put in touch with some local Green party members who are also voting-reform advocates. It was also recommended that I contact Ralph Nader directly.

The presentation took just five minutes, with five minutes more for questions and discussion. If you would like to give a similar presentation in your area, feel free to use everything I've said; the images are quite easy to reproduce on a chalkboard or whiteboard.

17 comments:

  1. I've heard some reservations that the irv-spoiler example I used is a bit hard to swallow; that it's hard to believe that half of the B-first voters would side with A over C.

    I have to responses to that: one, this example is actually a pretty lop-sided result: 44% vs. 55%, as in the 4-to-5 initial graphic, would be considered a landslide in an election with thousands of voters, and the closer the election the smaller the percentage of B->A "defectors" are needed, down to an arbitrary small number of votes (one) in a close election. And remember, it's the CLOSE elections which make-or-break the system (or a party).

    Example: As before, all C-first voters have B as a second choice, but there are 999 voters. A:499 v. B:249 v. C:251; only one B-first voter (less than half a percent) needs to have A as their second choice to spoil the election.

    Second, a real world example: in Burlington Vermont's recent 3-way mayoral IRV race, approximately one-third of Democrat-first voters had the Republican as their second choice, with just two out of three preferring the Vermont Progressive candidate (of course, in the topsy-turvey world of Vermont, the Republican was a spoiler for the Democrat, rather than the Vermont Progressive being a spoiler for them; so actually a *majority* went for the other "major party" in this election.)

    The moral is, don't think of it as "half of the remaining B-first voters"; think of it as "JUST ONE of the remaining B-first voters."

    The moral for me: use a bigger number than 9, in order to make the point clear.

    ReplyDelete
  2. Your alternate example is a bit contrived as well, as the B+C votes tally to 500, vs 499 for A. If we're talking about a 1 vote difference (or 0.1% of the vote total, in this example), I don't think "spoiler" is the correct term to use. If that one voter really prefers candidate 'A', then 'A' should win the election, right?

    One of the problems with approval voting (or Condorcet voting, etc), is not the principal, but the complexity. People intuitively understand IRV because it maps directly to traditional runoff elections, and better yet, an untrained person could determine the winner of an IRV election by hand. Not so with the the more exotic methods.

    ReplyDelete
  3. I can re-jigger the example with any number you like; as long as each of the three candidates are at or above 25% and below 50%, the example can work, so the B+C vs. A difference can be as great as a few votes less than 2:1; i.e., 334 to 332+333. Then it takes half of B's voters (so, 166) to pick A second for the election to be spoiled.

    Point being, there's a wide range of possibilities, and "just one vote" is just one end of the scale, if that's all that's giving you problems, and is analogous to a plurality election that spoils 499 to 498 to 2, when those 2 could have made it 499 to 500.

    But it each of those cases it is CERTAINLY correct to call it a spoiler; B wins if C doesn't show up, but A wins if C does. That's the definition of a spoiler: doesn't win but changes the winner.

    Finally, approval is VERY simple.

    "Approve as many choices as you want; most-approved choice wins."

    IRV, meanwhile, takes a four-step iterative process to explain. So I find it very hard to justify IRV as "simpler".

    ReplyDelete
  4. But it each of those cases it is CERTAINLY correct to call it a spoiler; B wins if C doesn't show up, but A wins if C does. That's the definition of a spoiler: doesn't win but changes the winner.

    But if C shows up (which should be considered a positive in a democracy), they have voted their preference. Just because some low information voters don't understand that, for instance, Nader's politics are closer to Gore than Bush, doesn't negate the intent of their vote. That's the issue to solve - the intent of the voters, preferably the intent of a very large number of voters.

    WRT the complexity issue, it may be easy enough to vote (albeit slightly more involved than simply ranking choice), but calculating the results is not as straight forward, even if the actual math is simple. And I'll have to respectfully disagree that ranked-choice voting tabulated in an IRV fashion is difficult to explain - it's exactly the same as runoff elections without the issues of different voter turnout from one election to the next. It's difficult for me to imagine any form of "complex" voting reform passing in this country, whereas IRV, in my mind, is easier for most people to conceptualize.

    ReplyDelete
  5. I've always been a fan of IRV, but had not appreciated that it still allowed spoilers. Another way to parse your example is to classify the voters on a one-dimensional right-left axis, R:4 CR: 1 CL: 1 L: 3, where the C party is assumed to get CR, CL, and L in a two-way with R. This is not contrived per se, but this distribution with very few voters in the center is unusual. Compared to the usual plurality system, there are many fewer spoiler situations, but it does shake my support for IRV/

    ReplyDelete
  6. Granted, approval voting isn't that hard to understand... I'm stretching things there.

    ReplyDelete
  7. Calling the 2nd biggest party the spoiler for the 3rd biggest party seems silly. Your example should be compared to a FPTP between A and C, in which A would also win, so IRV spoils nothing. Is there an IRV example where the 3rd biggest party spoils it for the biggest?

    ReplyDelete
  8. "Just because some low information voters don't understand [...] doesn't negate the intent of their vote. That's the issue to solve - the intent of the voters."

    Yes, precisely! The goal is to determine the true intent of the aggregate of the voters. Systems which are prone to spoilers fail at this, because they force a terrible choice on the voters: be truthful, or take the lesser of two evils.

    The problem with spoilers isn't that they "negate" a voters intent, the problem is that they give incentives for voters to vote other than their true intent.

    ReplyDelete
  9. Anonymous: your one-dimensional axis idea is something I examine in greater detail in this post.

    Also, it's true that there are relatively few examples of election returns under our current plurality system that you could point to where there's the potential of a spoiler had the election been held under IRV rules.

    However, third-party vote-shares tend to grow over time... until they spoil an election. Since IRV raises the bar to be a spoiler (in a three-candidate race) from "just one vote" to "just one vote over 25%", we would expect to see third-party vote shares improve under IRV until the rate of spoilers matched the current rate as it stands under plurality.

    The plurality spoiler rate, however, is also rather low; the question is, do any election outcomes change for the better before we hit that rate? And since third-parties under IRV still have the opportunity to become spoilers before they reach the point where they actually win an election, we could expect IRV to result in the same degree of two-party political domination; and indeed, that's what we see in Australia (to the degree that one believes Lib/Nat to functionally be just one party.)

    It's more the _threat_ of spoilers, than actual spoiled elections, which keeps the number of political parties at 2. Eliminating spoilers--which approval and score voting do in the three-candidate case--would add some much-called-for opportunity for more choice in our elections; while IRV would not.

    ReplyDelete
  10. @Krommenaas

    "Silly"?

    We're trying to look at elections analytically; the definition of "spoiler" I'm using (and it's a well-accepted one) is: "A candidate who does not win, but whose presence changes the winner."

    And "C", even though he gets the 2nd-most 1st-preference votes, certainly fits the bill.

    (Note that "B" is not a spoiler, because in AvC, A wins, but after adding B, A still wins. Also "A" is not a spoiler (assuming all A-first voters would vote B second), because in BvC, B wins, but after adding A, A wins; the winner changes, but changes to the new candidate.)

    I challenge you to provide a rigorous definition (i.e., one that includes no subjective terms) of spoiler that somehow _doesn't_ include a C-like candidate.

    The situation we're trying to avoid here is the retrospective "If only I had voted for the lesser of two evils." That lamentation exists because of spoilers, and IRV has spoilers.

    ReplyDelete
  11. I'm going to have to respectfully disagree that spoiler is the correct term to use. In a two-way race, there is a dearth of choices for the voter. We should expect different outcomes with three viable candidates.

    In your example, while a head-to-head matchup between A and B would lead to B winning, a head-to-head matchup between A and C, the candidates with the most votes, would lead to A winning.

    ReplyDelete
  12. I challenge you to provide a definition of "runoff election with three candidates" where the candidate with the 2nd highest amount of votes would be excluded from the runoff in favour of the candidate with the least amount of votes.

    ReplyDelete
  13. @Winnowhead

    I'm not sure what your contention is. Yes, with three viable choices, things get more complicated. I'm arguing that under IRV, like under plurality, the incentive for voters is to unfairly discount choices outside the usual two, and that this discounting is done because both systems suffer from spoilers in the three-candidate setting.

    And you're correct about the implications of my example, but I don't see what your point is.

    AvB, B wins. AvC, A wins. This is because it is not the case that all of the voters who prefer B over A also prefer B over C.

    Many IRV supporters seem to have great trouble with this; it's like they want to put all candidates into two boxes, and no one is allowed to rank anyone from their chosen box below anyone from the other box. Sure, if you allow this "boxing", IRV works. But you haven't improved voter choice; you're assumptions have just re-created the two major parties. And as soon as even ONE voter steps outside of your assumptions, you get a spoiler.

    With approval, no spoiler in the 3 candidate case means there's room for another option.

    With IRV, you're still stuck at 2.

    ReplyDelete
  14. @Krommenass

    You have managed to miss the whole point. Just because a candidate came in 3rd in the runoff doesn't mean they're the worst choice. You're assuming they are; perhaps this is why your having difficulty wrapping your head around the idea.

    The example to prove it is right here in the original post! AvB, B wins; BvC, B wins. But AvBvC, B doesn't make it to the runoff; rather it's the two second-tier players, as the super-star sits on the sidelines.*

    This is precisely analogous to the famous "Lizard vs. Wizard" election that Louisianan saw while it was using a top-two-runoff election (look it up).

    The problem is, you're incorrectly correlating "number of votes in first round" with "quality of candidate"; ultimately, I don't care (and neither should you!) about number of votes in the first round (if that's all I cared about, I could stick with plurality!) What I really care about is who the best candidate is. And any time I see a spoiler, I know that the best candidate might have lost.

    With approval (and score), there's no spoiler in a three-candidate election; therefore the winner of the election is the true best candidate.

    But with IRV, we'll still see spoilers, which means we're not always electing the best candidate.

    *This reads like an endorsement of Condorcet methods, but it's not. Condorcet (although better than IRV) still has its own spoiler problems in 3-candidate cases (but they're even more esoteric than IRVs); but Condorcet's insight is useful for illuminating the absurdity in this case.

    ReplyDelete
  15. So, whatever happened with this local Green Party chapter, did they ever adopt Approval and/or Range/Score voting as part of their plank, or were they ultimately disinterested?

    ReplyDelete
  16. I moved about 3,6000 miles away and fell out of contact with them.

    And there doesn't seem to be an active Green party in my new locale, although another approval advocate I collaborated with has spoken with the local Libertarian party.

    It's slow going.

    ReplyDelete
  17. Shame, would have been interesting to see if you had left some effect on them.

    That's interesting though about the local Libertarian party, do keep us updated if anything there suffices.

    ReplyDelete