Anyone have a rebuttal to the "stories" in fairvote's critique of Approval Voting?

You are here

19 posts / 0 new
Last post
Anyone have a rebuttal to the "stories" in fairvote's critique of Approval Voting?

Not sure if this is the best forum for this, but none of the others seemed quite right either... (Note: somehow this wouldn't format in normal paragraphs here, so I've made a bullet point for each paragraph to make it read better.)

  • Skip pages 1-6 and look at the "stories" told from the last paragraph of p. 7 to the middle of p. 11 of the "long" essay. They seem pretty damning to me, regarding the viability of score/approval voting in contested real-world elections. (The subsequent part on pp. 11-14 also seems worth addressing, though less directly damning of AV.)
  • Does anyone have a way to counter these stories, particularly in comparison to IRV/RCV? For me, this could not be accomplished merely by pointing out theoretical scenarios where IRV would have similar flaws.
  • Rather, it would have to either (1) demonstrate that these stories are somehow flawed (i.e., the problems they describe wouldn't/didn't actually occur for some reason), or (2) tell similarly convincing stories showing that IRV has had/would have similarly fatal flaws most of the time in contested real-world elections (which would convince me that our current system, annoying as it is, may be the best we can do short of a larger reform such as proportional representation).
  • In particular, I think probably the most important scenario for a voting system to handle well is one in which 45% prefer A (e.g., Obama or Hillary), 45% prefer B (e.g., Romney or, let's guess, Cruz), A's supporters strongly oppose B and vice versa, and 10% prefer C (e.g., Perot or Bloomberg), but significant numbers of both A's and B's supporters would be content with C as a tepid 2nd choice. This seems to be pretty typical of most recent U.S. presidential general elections.
  • The story on p. 8 makes it pretty obvious that using AV in such an election would result in a strategic nightmare in which pretty much all voters would be totally unable to figure out how they should vote to support their favored candidates.
  • Now, I realize it's easy to make up scenarios in which IRV can result in similar strategic dilemmas for voters, but in this particular scenario (45%/45%/10%), which seems to recur frequently in the most important elections, is there any reason to think IRV would fail as miserably as the story on p. 8 shows that AV would?

 

  • The other main issue is real-world experience: the essay claims that "in every single instance where such a [meaningfully contested] election has gotten competitive, approval voting has suffered a major breakdown," and later adds "especially if used multiple times."
  • The stories on pp. 9-11 (about use of AV in early U.S. elections, Dartmouth, and IEEE) describe major failures in all 3 instances--which the essay implies are the only historical cases where AV has been used in competitive, meaningfully contested elections that generated "much heat and energy."
  • Are the fairvote.org people somehow misinformed about (or spinning) these cases?
  • Or, have there been other such highly competitive, contested elections where AV has been used without serious problems? If so, that would at least be interesting to know about, although unless there have been quite a large number of them, AV would still seem fatally flawed as an election method.
  • Or, is it also true of IRV that in every single instance in which a meaningfully contested IRV election has gotten competitive (and used multiple times?), IRV suffered a major breakdown? (I've casually followed several recent IRV elections in Alameda county and SF, and it seems there have been no major problems there, and while many of these elections probably weren't as strongly competitive as the three AV examples on pp. 9-11 of the essay, I think at least a few of them have been pretty competitive.)
  • Perhaps someone has already written about this elsewhere, but I failed to find it. (Besides the thread linked above, I did find http://rangevoting.org/RVcrit.html, but that also fails to convincingly counter the essay in any of the ways I've described -- it mostly argues that strategic voting would still result in good winners rather than contesting that the various options for how to vote strategically would result in widespread voter confusion and electoral chaos.)

 

The first thing to understand

The first thing to understand is that FairVote's executive director, Rob Richie, lacks expertise in the science of voting theory, and has a tendency to mislead and even outright lie. Some examples:

http://sites.google.com/a/electology.org/www/fact-check
http://scorevoting.net/RichieRV.html
http://scorevoting.net/RichieOnApproval.html
http://scorevoting.net/LAgov2015.html

Regarding that essay, it goes astray right off the bat.

Approval voting magnifies the failure of Borda count due to tactical voting

This is false. Borda (as with virtually all ranked voting methods, including Richie's favored IRV) fails the Favorite Betrayal Criterion. E.g. if you prefer Green>Democrat>Republican>Libertarian, you want to tactically "polarize" the presumed frontrunners to the front and back of the list: Democrat>Green>Libertarian>Republican. That means the Green is doomed. Even if voters favor Green, if they don't think Green can win, strategic behavior will make that a self-fulfilling prophecy.

Whereas with Approval Voting, your strategy is to approve both Democrat and Green. This is obviously less of a distortion of honest preferences. It says you like Democrat and Green equally rather than that you prefer Democrat to Green. And it doesn't turn lack of perceived electability into a self-fulfilling prophecy.

They then produce arguments..that show approval voting working wonderfully well, based on the false supposition that voters will act like Borda’s “honest men” – e.g., they will vote like rational computers who all have read and understood recommendations by approval voting advocates on how to vote.

This sentence is self-contradictory. A "rational" behavior means "strategically optimal", not honest. It boggles the mind that Rob Richie cannot understand this incredibly basic distinction.

Real-world voting in fact is much more about psychology than math.

To the extent he's talking about the psychology of tactical voting, this claim is gravely mistaken. For example, in some models, Score/Approval Voting is better with 100% tactical voters than IRV is with 100% honest voters. In other words, the mathematical properties of these voting systems actually play a far larger role than the tactical effects of voter psychology. But as usual, Rob Richie makes these kinds of handwavy pronouncements without any scientific basis.

Violation of later-no-harm becomes utterly devastating – not just as a theoretical problem, but as a practical problem that makes the system unworkable.
un·work·a·ble (adjective) not able to function or be carried out successfully; impractical.
 
Consider that Plurality Voting, which for all intents and purposes violates LNH, is the most common voting system on Earth. And is massively more vulnerable to tactical behavior, since it often prevents people from voting for their preferred candidate. Is Plurality Voting "unworkable"? Let's get real.
 
And what does LNH really even mean? It roughly means "it's safe to vote for your second favorite candidate". Whereas Score Voting and Approval Voting satisfy the Favorite Betrayal Criterion, meaning it's safe to vote for your favorite candidate. Which do you think is more important, intuitively?
 
And the thing is, that's actually not even an accurate definition of later-no-harm. LNH says it's safe for candidate X if a voter ranks Y behind X. It does not say that it's safe for the voter. Voter tactics are about what's in the best interest of the voters.
 
These strategic actors would realize that “bullet voting” for only one candidate would be the best tactic. There might be certain elections where one major wing would know that the only way to defeat the other wing would be to approve of the centrist candidate, but any time they thought their candidate had a real chance to win, they would overwhelmingly cast a bullet vote. The result would quite likely be a plurality winner who potentially is unrepresentative and lots of finger-pointing about “spoilers” and tactical voting.
When would voters think their candidate had a "real chance to win"? Generally that would be when polling indicates that candidate is one of the frontrunners. So people will typically bullet vote when their favorite candidate is one of the legitimately most popular candidates. That is not "quite likely to be a plurality winner who is unrepresentative". It's likely to be the Condorcet winner.
 
Let's actually envision scenarios. Suppose initial polls show something like this:
 
Left = 51%
Center = 21%
Right = 48%
 
Then the best tactic for Center voters is to vote for Center and also vote for their 2nd choice. The best tactic for Left and Right voters is to bullet vote. Which means we'll get the majority winner between Left and Right. Exactly what we want!
 
Suppose instead it's like this:
 
Left = 51%
Center = 55%
Right = 48%
 
Then Center voters will want to bullet vote. Right and Left voters will want to make up their minds based on their utility differences. E.g. if you think Right=10, Center=3, Left=0, then you might want to just bullet vote for Right. If voting Center changes the winner from Right to Center, you lose "7 points of happiness". Whereas if it changes the winner from Left to Center you gain 3 points of happiness. Whereas the opposite holds if you think Right=10, Center=7, Left=0.
 
What if you don't have polling information? Then you can just assume all candidates are equally likely to win, and make up your mind purely on those utility differences.
 
Without that information, it’s all the more likely that voters would be uncertain what to do – and be that much more likely to listen to strategic actors who would be feverishly spreading the word that you should cast “bullet vote” for your favorite candidate and hope for the best.
Again Richie asserts that strategic actors will advocate bullet voting as the best strategy. But for many voters it's not the best strategy. E.g. a voter who favors Left and also likes Center, but hates Right, may want to approve Left and Center. Or say there are five candidates and you like four of them and hate one of them; then you might want to approve all four acceptable candidates. Strategic voting does not mean bullet voting.
 
Dartmouth College: A recent example of approval voting’s flaws comes from alumni seats on the powerful Board of Trustees of Dartmouth College, for which approval voting was used for several elections until 2009. In these elections with approval voting the Dartmouth Association of Alumni would nominate several candidates.
All evidence indicates that Approval Voting worked well at Dartmouth. The justifications for its repeal are nonsensical and apparently a smokescreen. It actually may have been repealed because it worked. And Dartmouth's student assembly elections were subsequently switched from IRV to Approval Voting.
 
Institute of Electrical and Electronics Engineers
 

http://scorevoting.net/FeerstTheory.html

The story on p. 8 makes it pretty obvious that using AV in such an election would result in a strategic nightmare in which pretty much all voters would be totally unable to figure out how they should vote to support their favored candidates.

People already have to figure out how to vote with Plurality Voting in situations like this. Approval Voting strategy is just, "Vote for the person you would have chosen as a tactical Plurality Voting user and then vote for any additional candidates you prefer to that tactical choice." So it's definitely easier in almost any circumstance.

We actually conducted an Approval Voting exit poll in Maine in somewhat similar circumstances. The independent (roughly the centrist) went from third place to first place. We saw no evidence that voters were paralyzed with indecision. We polled almost 800 voters. They were able to handle the process just fine.

Approval Voting was also used recently in an extremely gamed Republican straw poll, and apparently worked very well, with a significant fraction of voters supporting more than one choice.

I could go on and on, but the gist of it is that I think FairVote's arguments are mostly nonsense, made sellable by the complex and counterintuitive nature of the subject. No amount of empirical evidence can shake their faith, and their grasp of the underlying math/theory is extremely poor.

 

 

 

...not me.

they're spot on.   failure of later-no-harm is devastating, especially for a multi-winner election.

having said that, for single-winner elections, borda count is arguably better than irv.

and in multi-winner elections, stv is usually done wrong.  they should use the hare quota, and restore eliminated canddiates each election round, and order all combinations of candidates to eliminate, instead of just cumulatively eliminating.

 

failure of later-no-harm is

failure of later-no-harm is devastating,

No, failure of the Favorite Betrayal Criterion is devastating. Failure of LNH is more of a feature than a flaw, since satisfying LNH means ignoring preference information.

satisfying LNH does not imply

satisfying LNH does not imply ignoring preference information.

 

failure of FBC does not lead to bullet-voting, which is devastating.

btw, range voting also fails

btw, range voting also fails FBC.

lets say totals for candidates are a:100,b:201,c:200

favored candidate is a. lesser or two evils is c.

my ballot is a:10,b:0,c:5.

if i betray my favorite and swap a and c:

a:5,b:0,c:10.

totals become: a:95, b:201,c:205, causing the lesser of two evils to win.

=FBC failure.

You don't need to drop

You don't need to drop candidate a to 5 to raise candidate c to 10. Your ballot could be a: 10, b:0, c:10. Range/score voting does not fail favourite betrayal.

range voting fails favorite betrayal

you don't understand favorite betrayal.  it doesn't matter wether you "have to" or not, what matters is that you_ can_.

This holds for all criteria.  failing a criteria doesn't require a real person to actually vote that way.  That would be absurd.

"Your ballot could be a: 10, b:0, c:10.", yes, yes, it could.  then you'd be betraying your favorite AND bullet voting.

Thank you for demonstrating another problem with range voting

 

so now we have:

* later-no-harm

* favorite betrayal

* bullet voting

 

any more you'd like to add?

I do understand favourite

I do understand favourite betrayal. It's not about whether you have to or whether you can. It's neither of these. It's whether it would ever be beneficial to you to do so. If it was about whether you can, then obviously all methods ever would fail it. It's quite easy to not vote for your favourite! But with score voting, there's no benefit to you in dropping the score for your favourite, so it doesn't fail.

i think you missed my

i think you missed my implication that 10-10 is still favorite betrayal.

if u are expressing another candidate is tied then ipso facto you are no longer expressing that the 1st candidate is your favorite.

also unlike stv, you now have to worry about the failure of later-no-harm.  depending on the vote count, that dishonest 10 for the other candidate may be harming your favorite.  this would be true even if it wasn't a "betrayal" - even if it was e.g. a 9.  it would still be favorite harming.

OK. Failure of Later No Harm

OK. Failure of Later No Harm means that you can end up harming your favourite, but the favourite betrayal criterion is defined in a different way. So score voting doesn't fail the criterion as defined, even if you could argue that it is in some way a betrayal of your favourite.

i'm pretty sure the correct

i'm pretty sure the correct definition of FBC is how i defined it.

 

range voting and approval voting are listed erroneuosly as complying methods, even though they don't comply.

Voting two candidates 10 in a

Voting two candidates 10 in a Ranged vote isn't favourite betrayal as you have not been incentivized to lower your evaluation of your favourite. (every definition I can find doesn't consider ranking another candidate as equal to be FB) You have instead been incentivized to exaggerate your opinion of your backup candidate.

You're turning Favourite Betrayal into a hybrid of itself and later-no-harm. Range voting deliberately fails later-no-harm because it's an expressive system which doesn't regard consensus candidates as a bad thing.

Generally the most strategic way to use range voting is to bullet-vote, with occasional exceptions based on limited information voting if a race hasn't had any polling done for it. (basically where your preferences are A>B>C, you know there will be a significant votes for both B & C and don't want C to win, but you also want to vote low enough for B that you won't guarantee A will lose)

Later-no-harm is a criterion built of strict majoritarian thinking where your vote "belongs" to the top-listed candidate. I would contend that it's okay for an honest vote to cause your most preferred candidate to lose to one of your less-preferred candidates if they are a genuine compromise candidate for the electorate as a whole, and you go in understanding that voting that way puts you at a strategic disadvantage but makes the election as a whole more honest. (and really, this sort of strategic voting is only bad if certain groups of voters categorically exaggerate their choices more than others, which, pivoting back to the OP, can't be done with Approval voting, as it forces you to exaggerate strategically) To be honest I generally just view later-no-harm as an invented criterion to promote STV and IRV, which aren't terrible electoral systems, but do suffer from significant problems, some of which come from it being a multi-winner system, which are okay because they can't be avoided if you want the inherent advantages of multi-winner elections, and some of which come from it being a runoff system, which are not okay because there's no categorical reason to have a runoff system other than "it produces superior results."

Most of the criticisms for range voting boil down to incorrect criticisms of approval voting, which range voting simplifies down to in the perfect-information, perfect-strategy scenario. Yes, there is a risk in approving less-than-ideal candidates that they may be elected. Deciding whether that risk is acceptable is part of deciding whether to approve them.

Oh, and actual results of range voting and STV/IRV votes suggest that bullet-voting actually occurs more in STV/IRV elections than it does in Range ones. It's common enough in Australia that they also nicknamed it, "plumping." So if you're opposed to people only listing a single candidate in a vote, you probably don't want STV...

Is there a voting system that

Is there a voting system that doesn't suffer from any form of favourite harming?

If there was it'd be no good.

If there was it'd be no good.

If the candidate can express a>b, they should be able to express b>a, (or on a non-ordinal ballot, also a=b) and that expression should be meaningful in that it should be able to change the outcome.

if there's no case where a voter changing from a>b to b>a (or b=a) can make b win over c instead of c winning over b, then i question whether that's a fair system.

should be careful to

should be careful to distinguish between favorite harming and later-no-harm. (not harming you first choice by picking a second)

there are methods that comply with later-no-harm.   

You're most likely either

You're most likely either looking for something that violates Arrow's Impossibility Theorem, (which basically boils down to proving that voting systems can't be perfect) or referring to voting systems that obey monotonicity.

https://en.wikipedia.org/wiki/Monotonicity_criterion

Wikipedia says that Range and Approval are out-of-scope of the monotonicity criteria, but it just needs some language that's agnostic between ranking and cardinality for both systems to pass the criteria, so whoever wrote that bit is probably being needlessly picky about mathematical proofs.

STV and IRV both fail monotonicity in all of their variations used in major elections. Monotonicity is pretty highly regarded among voting system criteria.

Later-no-harm is a different kind of favourite-harming, where ranking a less-preferred candidate can cause your most-preferred candidate to lose. STV and IRV both pass later-no-harm, wheras Range and Approval both fail. Whether this is a bad thing or not is to a degree debatable, (as allowing later-no-harm also enhances a system's ability to elect compromise candidates who don't have more first-choices but do have higher average popularity) it's not a criterion that's universally desired by everyone who evaluates voting systems. Basically your vote has to be some form of Condorcet or runoff system to pass later-no-harm, and it's a criterion designed to catch systems where certain types of strategic voting ("bullet voting") has an influence on the result.

If there was it'd be no good.

-please delete, wrong part of the thread-

FairVote overstates the worst-case scenario

FairVote evaluates the worst-case scenario of approval voting in the paper you linked, and overstates their case.

Approval "bakes in" strategic voting to the system by removing the parts of a Range vote you could be dishonest about. Fairvote confuses the optimal strategy (approve every candidate you regard as average or better) with a sub-optimal strategy. (approve candidates only if they're not likely to win against your most-preferred candidate) In reality the viability of "bullet voting", and its reverse counterpart, compromising or "over-voting," depends on the available field, and often to arrive at the nightmare scenarios you have to assume not strategic voting but rather moronic voting, where two large groups of voters all simultaneously apply a sub-optimal strategy to conspire to pick a bad candidate. ;)

Many of their problems with Approval's disadvantages come from viewing Approval through an STV lens. They're looking at voter preferences as the only valid way of voting, and they don't want to risk their most-preferred candidate losing. This is a flawed way to evaluate any election system, you have to embrace it's limits and advantages together and form them into a mindset. In the case of Approval, that's one of "which candidates would I be okay with if they won?" Instead this rebuttal is looking at the big criteria that basically only STV wins and repeatedly trying to bash approval over the head for not meeting it.

Their point that systems should be evaluated by the likely level of strategic voting is well-taken, that's why I looked at actual Approval election data that suggests that in elections performed to date, voters were likely to approve an average of between 2.1 and 3.5 candidates, depending on which election you look at. Given that strategic voting in approval involves varying how many candidates you're willing to vote for, I'd say that's a pretty acceptable variance between the two extremes of strategic voting. (Unfortunately, like any system, the only way to know if someone voted strategically is to ask them to vote in a second election honestly, and such a result is still likely to be unconsciously biased) The STV equivalent of "bullet voting", where you express only a first preference in variants where that's a valid vote, is actually more common, at about 25% in real-world elections iirc. (of course, in STV, "bullet-voting" is categorically a sub-optimal strategy, rather than sometimes being good, sometimes being bad)

In real elections, you're likely to get a mix of naive voters, who make errors, aren't well informed, or just get confused about how to vote strategically, and mess things up, and some honest voters, who won't vote strategically even when it would benefit them because they want to "send a message," and finally you will of course get some strategic voters. There are likely to be some of each type in each group of voter, assuming your sample size reaches into the thousands, so while strategy or naivity will definitely have an effect, many of the naive and strategic voters will cancel each other out, too, mitigating the overall effect to something smaller than the total number of Nietschian strategists or bumbling low-information voters that are generally assumed in theoretical worst-case scenarios.

They refer to the Approval as not complying with "majority rule," which if they're referring to the Majoritarian criteria, Approval is a little out-of-scope and you have to tweak the definition to get it to apply, and what tweak you use effects whether it passes or not. (also, it's much more likely to meet the Majoritarian criteria when you assume all Approval voters are strategic to some degree, that is, willing to vary how much they'll compromise based on the rest of the electorate) Basically, Approval allows voters to choose possible consensus-candidates at an equal weight to their most-preferred candidate if they want to, and whether that's good or bad depends on your viewpoint. I consider consensus-seeking systems good, but not necessary, so I can take or leave this particular (arguable) violation of Majoritarianism.

Also, if you pay careful attention, you'll notice that in fact some of the criticisms that say certain voters are "uninformed" about Approval voting are actually demostrating strategic voting in approval, they're just voting strategically with less-than-perfect information. (that is, lowering or raising the bar of who qualifies as an approved candidate based on their perceptions of the race) Fairvote is providing an extreme example here, as this problem is not as dramatic in the elections that have been performed to date as the example paints it as being on average. Bullet voting is a relatively rare strategy with Approval, as I said above voters tend to vote for 2 or 3 candidates on average.

Fairvote's criticism of the very early US presidential elector system as Approval voting is incorrect. It's technically Bloc Voting, which makes it a dramatically different method that's strictly inferior to Approval and more resembles a multi-winner Plurality contest. You should disregard this entire example as none of it actually applies.

Misidentifying this as Approval and criticising Approval by association would be the same as if an Approval advocate used examples of Limited Preference voting (where you can only rank your top 3 candidates) to criticise STV as not being immune to clones and thus vulnerable to vote-splitting, which is objectively not the case. (it is, however, vulnerable to irrelevant alternatives, which is a problem at least as bad as violating later-no-harm)

Basically, the best criticisms the STV advocates over at FairVote have is the Dartsmouth college hypothetical, which as above, may not actually be a real example.

However, there are similar worst-case scenarios with STV systems violating monotonicity and even causing the no-show paradox, where some people could have voted most strategically by not voting at all, and these examples have actually happened in national or regional elections. The best way to evaluate these worst-case scenarios is to compare how bad each one is and how often it occurs, and that's too much work for me to do to rebutt a single claim, but from what I've seen so far I'm not concerned that Approval is anywhere near as bad as they think it is as a single-winner system. I'd consider it comparable, maybe slightly worse than, IRV, which is the single-winner flavour of STV.

It's typical of a one-system reform advocacy like FairVote to bag on competing systems, but if they're going to do it they should do it carefully and objectively, which the piece you linked to isn't.

Log in or register to post comments

Follow The Center for Election Science on: