How can we get instant runoff voting in the United States

Politics & Elections
21

I previously responded to this post in great detail but I don’t see my response. So to summarize, your intuition is wrong here. IRV ignores much of the information you indicate on your ballot and is seriously harmed by tactical voting so it actually produces worse outcomes than Approval Voting.

This is the expressiveness fallacy. What matters is how happy you are with the election outcome.

Like this.

Here’s an example where it would have.

Plus with Approval Voting we’d have many more options. Sanders and Bloomberg (and many others) chose not to run because they don’t want to be spoilers. Clinton and Trump wouldn’t have a chance with Approval Voting. (Though, of course, the Electoral College is a whole other can of worms at the Presidential level.)

Not true. Score Voting, Approval Voting, and Borda are all simpler. And there are several Condorcet methods that are simpler than IRV too.

Clay Shentrup
Co-founder, The Center for Election Science

22

Although interesting, the links only partly answer the question–in particular, the experimental results are a reasonable counterargument.

To be clear, though, “range expansion” is neither dishonest nor strategic in the normal sense. It doesn’t require anyone to lie about their preferences.

The range is arbitrary to start with–what would a “true” 99-point candidate even look like? An imaginative person might realize that a truly ideal candidate is easily 10 or 100 times better than what we normally think of as a solid candidate. That doesn’t mean we should always score near the bottom, just because there is some theoretical candidate that could be much better.

So it’s just common sense that voters should, at the very least, score the worst candidate at 0 and the best at 99. They can–depending on their preference for honesty vs. strategy–alter their positioning of the middle choices, but the endpoints should be at the extremes.

And I think this leads to bad outcomes. Consider 10 people voting on candidates A and B. 4 people prefer A, and vote A=99 and B=0. The remaining 6 vote A=0 and B=99. The B voters win, both with this scheme and a normal first-past-the-post scheme.

Now C comes into play. A voters hate him, and give him 0 along with B. But the B voters are split: half think C’s a naive idealist, and the other half think B is just maintaining the status quo. We end up with 3 people giving B=99 and C=10, and another 3 giving C=99 and B=10 (A still gets 0). These are honest votes: both groups really like their preferred candidate, while also indicating their preferred order for the less-favored candidates.

A then wins with 396 points vs. 327 for the others. It’s just the bad outcome that we’d like to avoid with a new voting scheme. I don’t think my setup is too artificial, either.

IRV at least would give the election to B or C. Perhaps counter-intuitively, it’s the A voters that will likely decide which of B or C gets elected, but maybe that’s not so bad.

As I mentioned, the experimental results are somewhat convincing, but not completely so. I think there must be a learning process involved where people learn to maximize the utility of their vote while still being honest. And that necessarily means range expansion for the endpoints, but I wouldn’t necessarily expect everyone to figure this out right away.

With enough truly strategic voters–that is, ones that realize giving their less-preferred candidate a “token” score instead of the maximal 99–one can avoid the bad outcome. But again, isn’t that what we’re trying to avoid? We want a system where honesty is rewarded.

23

Interesting. Much obliged.
Was thinking more along the lines of strategic voting having swayed a secret ballot election result rather than just a quirk of how the numbers fell.

I looked further at the other example posted on rangevoting.org, the by-election in Frome, South Australia in 2009 because I have more data and to hand on the result.
Firstly, was clearly was a case of non-monotonicity.

The hyperbole and outrage about the Frome 2009 result expressed on rangevoting.org way way exceeded the local commentary about the result, which in a nutshell was “Meh”

There’s no conspiracy here as suggested. The AEC two party preferred results for all elections are available on a polling booth-by-booth, but on subscription service. Also both parties scrutineers will keep that tally non any election when a recount is necessary.
*

Boylan lost because he didn’t get enough first preference votes. Due to asymetric preference flows he theoretically could have won on a lower primary vote but he didn’t seek that result pre-poll with tactical voting and had no influence over the conduct of the count post-poll.
*

Given that voting is compulsory here, that argument’s a non-starter.

Antony Green is considered Australian’s leading election analyst (he’s not a pollster). His take is on Frome 2009 can be reviewed here:
An Example of Non-Monotonicity and Opportunites for Tactical Voting at an Australian Election

I am aware of instances here where tactical voting has been applied to maximise election results. The NSW ALP Right v Left factional battles attempt all manner of such shenanigans but they are in “show & tell” ballots.

24

Thank you, clayshentrup. It seems obvious in retrospect, but it never even occurred to me to set my approval/disapproval threshold at the expected value.

What do you mean by “simpler”? The instructions on how to fill out the ballots might be simple, but the procedure for evaluating the ballots can end up fairly complex, and that’s the part that the public needs to be able to trust (“It’s not your vote that counts, but who counts the votes”).

Oh, and a general note to everyone: It’s true that IRV sometimes leads to some counterintuitive results, but it’s not enough to just point out those situations. It’s actually been mathematically proven that every voting system with three or more candidates has “paradoces of voting” of some sort or another, such as situations where a voter’s expected outcome is better from voting “dishonestly” than from voting “honestly”. So the question isn’t whether there are such situations; it’s just a matter of how likely it is for a given voting system for that system’s particular flaws to show up, and how bad the flaws are, compared to some other system.

In fact, you can even get a paradox of voting with only two candidates. Suppose, for instance, that 51% of the electorate thinks that A and B are almost indistinguishable, but that A is very slightly better, while the other 49% thinks that A is terrible while B is great. The optimum result in such a situation (the one that maximizes happiness of all voters) is for B to win, since A’s voters won’t mind that result very much and B’s will be ecstatic, but it’s tough to come up with a voting system that wouldn’t result in A winning (which will leave A’s voters feeling only slightly better, but B’s feeling much worse).

25

That’s where Approval Voting wins. 51% vote for A and B, and 49% vote for B. So B wins with 100% of the vote.

26

Why don’t we just let people vote as early and often as they want? Serious question. Eliminate any question of fraud, those who care more get more of a say, you can vote for any number of candidates from any number of parties in any proportions you want.

Or maybe we can just make official the method most people use: you don’t get a vote FOR anyone, you just get to vote AGAINST someone.

27

Is Center the “most overall preferred candidate,” or is it really a case of:
35% Left > “I don’t want Right to win”
33% Right > “I don’t want Left to win”
32% Center > others
and when Center ends up winning, Center discovers that the voters treat the result as:
32% Center
68% “I didn’t vote for Center!”

Hmmm…that does make a case for approval voting, doesn’t it?

28

Except that people vote strategically, and while strategic voting can have some subtle nuances that take some intelligence to unravel, almost everyone grasps enough to know that it’s pointless to approve everyone or disapprove everyone. So with only two candidates, anyone who even slightly prefers A over B will approve A and disapprove B.

You’re assuming that people are voting based on what they want, not what they need. Some people can afford to spend all day or all week or all month just going in and out of voting booths, but most can’t. Would it really be best for the country for everything to be decided by those people who have the most free time? I see no reason to expect it would be.

29

That’s not correct. Because with Approval Voting the question changes fundamentally. Instead of the ballot paper saying, “Put a cross against the candidate you think is the best person for the job.”, it’s saying, “Put a cross against all the candidates you think can do the job.”

30

That may be the principle, but why would anyone do that? A voters get their preferred candidate by disapproving B. They don’t care about maximizing overall happiness; they care about maximizing their own happiness.

This isn’t even strategic voting by the normal definition, which requires voters to lie about their preferences. This is just about voters maximizing their voting power.

31

Responding to: “3. If 200 new Rohde>…>Boylan voters had magically appeared… that would have made Boylan win. So those voters were very wise not to vote.”

This doesn’t actually address the point. They may have been compelled to vote, but they didn’t have to check the box for the specific candidate. They could have left that box empty.

If it would have served their purposes, they could have voted for some very unpopular third candidate.

Preferential voting of this sort is susceptible to “strategic voting” where votes are cast contrary to their ostensible purpose.

In very highly polarized elections, it can also be susceptible to a violation of the principle of “independence of irrelevant alternatives.” The classical example of this is: “If I vote for Jones, Smith wins. But if I don’t vote for Jones, Brown wins.” Jones can’t win either way, but my vote for or against him swings the vote one way or the other for the major candidates.

(“We have chocolate and vanilla.”
“Okay, I’ll have the chocolate.”
“Ah, we also have strawberry.”
“Well, in that case, I’ll have the vanilla.”
“Oh, I’m so sorry, we ran out of strawberry a little while ago.”
“Gosh, then I guess I’ll have the chocolate after all.”)

32

Yeah, well, IRV favors first choices. So yes, if Republican voters favored the Democrat to the Progressive but their first choice was the Republican, the position of the Republican in the count undercut the value of their preference for the Democrat over the Progressive. That’s consistent with the design theory.

Well, you kind of do. Not even STV is really PR. Proportionality has to be measured by something to be in proportion to.

33

Because it’s a fundamental shift in the way you vote and the meaning of your vote. Think of it like a group of friends trying to decide whether to go for BK, McDonalds, or KFC.

34

I have an active interest in the well-being of my friends. I’m willing to make small sacrifices so that no one is greatly put out. Happens all the time when there’s a vegetarian in the group.

Most of the voting public are not my friends. A huge number have interests actively opposed to my own. When possible, I’m going to vote in a way that maximizes my voting power.

35

No. It is the same as approval, only for single-candidate “strategic” voters. Those people can vote that way, sure; it doesn’t lend itself to paradox the way strategic voting does in other systems.

In elections with more than two candidates, single-candidate voters are giving up a share of their power, in that they are declining to have any influence on the sorting of candidates other than their favorite and least-favorite.

36

I didn’t say anything about single-candidate voters; just that whatever the selection is, the voter should min-max their choices. In other words, range voting should always be equivalent to approval voting.

clayshentrup’s link above confirms my suspicions, and says the threshold should be based on the expected value of the winner. Candidates you like more than the expected value of the winner get 99, and the rest 0.

Even a completely honest voter should at the least rescale their range so that the worst candidate is at 0 and the best 99. They give up nothing by doing this.

37

In a range-voting system, any voter who does not vote at the extremes is giving up a share of their power compared to one who does, no matter how many candidates there are.

Though the link clayshentrup posted suggests that it’s only a fairly small share of their power that they’re giving up, in practice.

Quartz, “can do the job” has different meanings depending on who’s running. If the general election were between Kasich and Trump, I’d interpret “can do the job” as “avoiding starting WWIII”, and I trust that Kasich can do that job, and so I would approve of him. If, however, the general election were between Clinton and Sanders, I might expand my definition of “doing the job” to include, say, reining in the excesses of Wall Street, and I’m not convinced that Clinton can (or rather will) do that, so I would disapprove of her. But this does not mean that I’d prefer Kasich over Clinton.

38

Definitely not.

This part is right. What I’m saying is, this creates no problems in range voting. And, having done that, honest/smart voters should also give intermediate grades to other candidates between their best and worst. Because thereby they wield additional influence, and thereby the overall result is more satisfying to more people.

Some voters may enjoy the less “powerful,” but more “expressive,” options, like giving most candidates Fs (zeros), and the barely-acceptable Ds (one point). Or whatever. The great thing is, this still works; that small, grudging distinction is still usefully compiled. Whereas writing in the dog’s name, or staying home, as such a voter may be tempted to do in a plurality (or approval) election among a disliked field, never helps.

I think the five-point letter-grade system is most practicable, for giving citizens quick fluency. I’m open to arguments for why a more granular scale would be better, but I haven’t heard a convincing one.

39

Read the link (and some of the sublinks, if you wish to see some proofs). A voter only ever loses voting power by not min-maxing their scores.

Essentially, they are just averaging between a “don’t care” vote vs. the min-max vote. They are writing in their dog’s name, but only fractionally. If a voter really wishes to express a degree of “don’t care”, that’s fine, but they’re clearly giving up power.

40

In an election where 31 votes cast differently one way would have changed the result, yes were 200 votes (from the same political grouping) cast the other way it also would have changed the outcome. In a ballot like Frome 2009 where the numbers & preferences were balanced that finely all three results are feasible, but the one with the biggest margin of error was the result declared.

Strategic voting has no material impact when the margin is clear and in extremely close ballots has such a narrow margin of error that a political operation would only seek to employ it if they knew how to calculate the precise number of votes required before polling. In which case there’s a bigger elephant in the ballot box than strategic voting.

This on focus on the marginal, unpredictable and oft-time counterproductive effects strategic voting is curious to myself when it would seem obviously the biggest distortions in place are winner takes all, gerrymander and low turnout.