xkcd

by **Griffin** » Tue Mar 13, 2012 7:13 pm UTC

Marzipan, you keep arguing against this sourmilk strawman. Stop it.

He said, over and over again, we have a right to remain silent, and thus, it would not be much of a stretch to argue that we could also have a right not to vote. He never said the two were the same thing, and there are ways to disagree with his argument, but you're not even acknowledging it.

mike-I:
I'm aware your proposal wasn't actually ranged voting. Similiar to how Range voting just becomes AV, your methodology is another word for FPTP. The only people who wouldn't give their lead candidate one and all the others zero are idiots, because all they've accomplished is making their vote worth less.
Any system that punishes people for using it correctly is probably not a good system.

Condorcet, yes, is the best. But its also more complicated, more expensive, harder to trust and (this is a big one) easier in some ways to tweak fraudulently with no one noticing.

AV is a distinct middle ground, with the huge benefit of our system already being 95% compatible.

by **Silknor** » Tue Mar 13, 2012 7:19 pm UTC

Yeah, that's not what I was thinking of by Range Voting. In that case, the rational thing to do is for everyone to vote exactly as they would in a FPTP election. The only difference is that some will vote non-rationally (eg. give .6 to good and .4 to lesser evil), thus giving more influence to people who vote rationally.

by **mike-l** » Tue Mar 13, 2012 7:20 pm UTC

I prefer IRV if we are staying in the non-condorcet round. I feel like it still lands on the 'understandable' and 'trustable' side of things. Your reductions have convinced me that range voting sucks

by **Griffin** » Tue Mar 13, 2012 7:29 pm UTC

IRV voting is basically... the worst. Statistically, it has no advantage of FPTP and significant flaws, from what I understand. I could be wrong - this is one of the systems I don't really have much experience with or knowledge on, but I've heard bad things about it. It's basically the rank system though, right?

Why are you so opposed to approval voting?

by **mike-l** » Tue Mar 13, 2012 7:33 pm UTC

Griffin wrote:IRV voting is basically... the worst. Statistically, it has no advantage of FPTP and significant flaws.

I haven't run any simulations myself, but my understanding is that IRV is significantly more likely to find a Condorcet winner than FPTP.

Why are you so opposed to approval voting?

The ability to elect a condorcet loser? The sensitivity to voters changing their 'threshold' for approval? The incentive to vote strategically? With arbitrary cutoffs, it doesn't even satisfy the majority critereon.

by **Griffin** » Tue Mar 13, 2012 7:35 pm UTC

IRV can also select a condorcet loser, so that's not much of an argument.
I don't understand your second objection.
And I don't understand the effective nature of your last one.

Edit:
Looking it up - by god this looks a lot more expensive and complicated than approval voting. How does this work in the field? Do people reliably manage not to screw this up? Is it actually affordable compared to an FPTP/Approval Voting approach? And hasn't Australia had problems with this approach? (name, its so complicated people just take party-distributed cheat sheets with them to the poll and copy them down verbatim)? Talk about strategic voting!

Also, let me get this straight - you support IRV because it isn't perfect, but is more likely to get a condorcet winner than FPTP.
You oppose AV, because although it's more likely to get a condorcet winner than FPTP, it isn't perfect.

Doesn't that seem rather strange to you?

by **mike-l** » Tue Mar 13, 2012 7:43 pm UTC

IRV cannot elect a condorcet loser, at best it can elect a condorcet second last place, because the winner ALWAYS beats the last remaining name left with it.

The example I gave before of
33% A>B>C
33% A>C>B
34% B>A>C
shows my second point, where A wins or loses depending on where voters choose to cut off their approval, and of course my last point, where A is preferred by an absolute majority but loses if group 1 approves A and B, but group 3 prefers only B.

I definitely prefer AV to FPTP. I just think IRV is better than either.

by **Dauric** » Tue Mar 13, 2012 7:52 pm UTC

Can someone clear up the alphabet soup? I'm finding it difficult to keep up with the acronyms without a scorecard.

by **mike-l** » Tue Mar 13, 2012 7:58 pm UTC

Dauric wrote:Can someone clear up the alphabet soup? I'm finding it difficult to keep up with the acronyms without a scorecard.

FPTP: First pass the post: Everyone votes once and most votes wins
AV: Approval voting: Like above, except you can vote for as many different candidates as you want
IRV: Instant Runoff Voting: Rank all candidates, count top choices, if anyone has a majority they win, otherwise, whoever has the least votes is removed from all ballots and all remaining choices move up. Repeat.

IRV is essentially an one round form of runoff voting, which is where you do a regular vote, and if noone has over 50% you drop the worst candidate and do another vote. It just does it all at once by assuming your preferences wouldn't change between votes.

The other terms we've used and/or are relevant are:

Condorcet winner: If one candidate would beat each other candidate in a 2 man runoff, they should win.
Condorcet loser: If one candidate would lose to every other candidate in a 2 man runoff, they should lose.
Absolute Majority Winner: If one candidate is preferred first by a majority of people, they should win.
Absolute Loser: If one candidate is least preferred by a majority of people, they should lose.

The first is actually surprisingly complicated to attain. You basically need a system that tries to find such a candidate. I don't know of any systems that don't have this as a goal that are able to produce it. All the systems we've talked about can fail to elect condorcet winners.

FPTP and AV can both elect condorcet losers, as well as absolute losers, which is my biggest pet peeve. As a left wing Canadian, I feel like it happens consistently here.
AV can fail to elect majority winners as well, while the other two cannot.

Before being accused of being disingenuous, there are lots of spots where IRV fails too.
It's not monotone, which means that voters can actually hurt candidates by ranking them higher, and it's not consistent, which means that it's possible A to beat B when looking at every district only, but for A to lose to B overall, and, similar to the first problem, it fails the participation criterion, meaning that some voters could actually hurt their candidates by showing up and voting in the first place.

Finally, reversing every ballot may not alter the winner, though I don't really see this as a bad thing, if half the people are A>B>C and half are C>B>A, B seems like a perfectly fine winner to me.

by **Griffin** » Tue Mar 13, 2012 7:59 pm UTC

You really think people should be forced to say "and yeah, we would, I suppose, if forced to choose, pick B over A, but neither is acceptable to us", giving the victory, through their own votes, to someone they DO NOT want to win? All for... what, exactly?

We clearly have different ideas of whats important in a voting system.

Acronyms:
FPTP - First Past the Post, standard USian election approach. Vote for preferred candidate, most votes wins.
IRV - Instant run off voting. Rank candidates. Depending on implementation, can get extremely complicated. I've read about it for a while now and still have no idea how it works. Most of the people who use it don't either.
AV - Approval voting. Vote for one or more candidates. Candidate with the most votes wins.
Condorcet - Look it up.

by **Dauric** » Tue Mar 13, 2012 8:04 pm UTC

Thanks. I can look up full words on Wikipedia easily enough (like condorcet), but the acronyms are a lot less specific when they're my only search terms.

by **Iulus Cofield** » Tue Mar 13, 2012 8:16 pm UTC

This discussion is what I meant earlier by the one thing I learnt in college is there's no good way to decide things by consensus. The only system of voting that everyone can agree is ideal is FPTP, but only if there is only two choices. But if there's only two choices, there's a good chance neither choice is ideal, much less wanted by voters. No one has devised a system in which everyone can be satisfied.

by **Yakk** » Tue Mar 13, 2012 8:17 pm UTC

I believe that voting systems are subject to strong corrupting forces. I want my voting systems to be robust.

1) A relatively uneducated person at a polling station can record, for themselves and independently, the information from each ballot. Call this the polling station "summary".
2) This summary can be transmitted really cheaply from each polling station. (Ie, you can say it verbally to someone, or broadcast it over the radio, or print it using very little paper).
3) Transmitting the summary shouldn't open a secrecy hole -- there should be no problem that the summary is publicly available.
4) By looking at each polling station's summary, you should be able to determine the results of the election.
5) The power of corrupting a summary (by capturing a poling station and all observers at it) should be limited, even if you corrupt it after everyone else has shown off their summary.
6) The information (ballots) required to build a summary can be kept, and the above repeated if there are concerns, for error and corruption checking purposes.

FPTP has these properties, at least with a sane number of people on each ballot. They can show off each ballot to scrutineers, who can record what each ballot voted for. Vote totals are a summary, which is easy to transmit, and from every polling station's summary you can know who wins the election. Corrupting a given polling station gives you the ability to swing up to (# of voters served) towards the candidate of your choice, and knowing other station's summary only gives you a significant advantage if the election was already close, which limits the power of corruption.

I really am not willing to give up this robustness.

Approval Voting/Range Voting gives you much of the above properties.

I have never seen an Instant Runoff system that has anywhere close to the above properties. Usually it becomes "feed results into computer". (With N candidates, there are more than N! possible ballots! And each such ordering must be kept intact in order to determine the winner usually!)

As noted, Range Voting reduces to FPTP while allowing people to "partially spoil" ballots in a sense.

by **mike-l** » Tue Mar 13, 2012 8:18 pm UTC

Griffin wrote:You really think people should be forced to say "and yeah, we would, I suppose, if forced to choose, pick B over A, but neither is acceptable to us", giving the victory, through their own votes, to someone they DO NOT want to win? All for... what, exactly?

We clearly have different ideas of whats important in a voting system.

You're allowed to leave spots blank and if all of your candidates are gone, you are no longer counted towards any of the remaining. Which is essentially what happens when I vote for the loser in any system anyway.

Yakk, I agree the biggest issue with IRV is that it's difficult to record and transmit by hand. But in an age of computers, I'm less concerned about that than electing absolute losers.

by **Griffin** » Tue Mar 13, 2012 8:24 pm UTC

I still don't understand how approval voting elects anything like "absolute losers", but IRV doesn't. But then, you've not exactly done anything to justify why thats a bad thing, even within whatever framework you're using it, or why electing the "second most loser" is such a huge improvement.

by **mike-l** » Tue Mar 13, 2012 8:29 pm UTC

Griffin wrote:I still don't understand how approval voting elects anything like "absolute losers", but IRV doesn't. But then, you've not exactly done anything to justify why thats a bad thing, even within whatever framework you're using it, or why electing the "second most loser" is such a huge improvement.

I don't want someone that the majority of people want least. I'm taking that as an axiom.

IRV cannot do that by construction, since they can never win a runoff.

AV includes FPTP, and that happens all the time in FPTP elections. In a 3 party system, it's exceedingly rare for anyone to get the majority vote. If the 2 losing parties both prefer the other loser to the winner, then the winner was ranked last by a majority of voters.

I live in Canada, where we use FPTP. our current government is pretty extreme, while the rest are relatively similar. It's not outrageous to believe that a good percentage of the people who did not vote for the current government actually would rank them last among all choices, and that comprises over 60% of the voters. It's a very real possibility that we're under an absolute loser right now.

by **Yakk** » Tue Mar 13, 2012 8:44 pm UTC

I find your faith in computers amusing. It isn't who casts the vote who has the power, it is who counts it. And you are asking people to trust a computer (which is a black box, even to a professional, in practice) to choose who wins.

To have anywhere near the robustness, you'd have to give each scrutineer an different hand-crafted piece of hardware and software to tabulate the results ("hand-crafted", to avoid single point of failure corruption problems. If everyone is using the same counting software, that counting software is a common single point of failure.) Then publish the results in a file somewhere. Then have each scrutineer compare their own file to the shared file. Then have multiple independent hand crafted software platforms that can check the results.

In practice, each party will have to hand out their own software platform to their scrutineers. Who will have to own their own portable device. You'll want multiple scrutineers at each location, ideally from different social circles and with different hardware and software on their device, because of the likelihood that said devices will be compromised without anyone noticing is higher (the monoculture problem), and isn't correlated to the local support for the parties political beliefs either (which mitigates this problem in a FPTP situation).

The published information allows for O(n) bits of data to be broadcast with each vote. I could, for example, say that I'll vote for some strange pattern, and you'll be able to detect it pretty reliably. With 20 candidates, there are 2432902008176640000 different voting patterns, to the level that I should be able to generate a pattern that isn't likely to occur unless I cast it (and the votes won't be uniform in that space, which means at a relatively few number of candidates, I can pretty much find a unique voting pattern that nobody else is likely to vote). This allows for the possibility of relatively reliable vote buying and vote extortion if you include every detail in the summary. So the summary of a polling location leaks information.

What is worse is that the implementations of IRV seem to embrace complexity, so the level that I doubt that the first few attempts by someone to write a program to determine the winner will be bug-free. Fractional votes being passed on, etc.

So no, IRV is not an acceptable voting system.

In my opinion, where there is a real binary decision (say, who is going to be the President), I'd go with FPTP. For a parliament, a scaled PR system, where the number of votes a person in parliament has is proportional to their support from the voters in their riding. (Ie, suppose each riding has 10 seats in it. If you get 50% of the vote, you personally win 5 seats. If you get 10%, you win 1 seat. The rounding problem to deal with fractional seats is not that hard, and not that interesting.) This generates a parliament that reflects the preferences of the population pretty well, with local representation that is directly responsive to the people who elected her (ie, if someone goes independent, and continues to have the exact same people voting for her, she doesn't lose her job).

Ie, AV or FPTP for a singular position: Localized PR for Parliament, who wields power in a FPTP fashion internally.

by **mike-l** » Tue Mar 13, 2012 8:51 pm UTC

All your concerns are addressed by a regular old runoff Yakk. But really, I rely on computers for my day to day safety, I'm ok with relying on them for my government too.

Incidentally, all 3 of our national parties use IRV for their leadership votes. (Though the conservatives doing something is probably an argument against it)

by **Yakk** » Tue Mar 13, 2012 9:03 pm UTC

With computers that work for my safety, my risk is incompetence (of the programmer). With computers picking my government, my risk is incompetence and corruption.

There is little profit in killing users or consumers (at best, you save money by being less careful than you could be). There is lots of profit in owning a government.

Political corruption should be assumed to happen, and systems that aren't robust against corruption are not good political systems.

by **mike-l** » Tue Mar 13, 2012 9:22 pm UTC

FPTP is also not a good political system. Things can be done to make other systems work better, FPTP is fundamentally flawed. Even a simple two ballot system is substantially better (more likely to elect the Condorcet winner, impossible to elect an absolute loser)

by **Yakk** » Tue Mar 13, 2012 9:42 pm UTC

A two vote system gives you n(n-1)/2 possible votes, which is low enough that it can be summarized well.

Describing the set of situations where the two vote preference system disagrees (in result) with the "infinite" IRV system might be interesting.

Note, however, that the Canadian system is an artifact of segmented double fptp. And given that we are doing double fptp, we could replace the first pass with a PR system (say, my local PR, where Ridings of 10 seats vote to allocate said seats between candidates, where a candidate can hold more than one seat) and get results that match what you would want. The system would be more continuous than any "winner take all" at the riding level, which reduces the impact of corruption, it generates the local people-driven representation that Canada currently has (as opposed to party-list PR systems).

The second FPTP system (in parliament itself) remains a problem by your metrics. But your approach doesn't attack that either.

by **mike-l** » Tue Mar 13, 2012 9:47 pm UTC

2 ballot is a 2 round runoff, you do a round, then remove all but the top 2 and vote again.

I'm not sure what you mean by the second FPTP system. If you are referring to votes on individual bills, there's no problem with... well, any system when there's only 2 options, yay or nay.

by **Griffin** » Tue Mar 13, 2012 10:18 pm UTC

Okay, please, please, explain to me how it could EVER be possible for a candidate the majority dislikes to get elected in an approval voting system, unless the majority dislikes EVERYONE. In which case even the condorcet winner is disliked by the majority, no?

Also, justify your assertations that

In a 3 party system, it's exceedingly rare for anyone to get the majority vote. If the 2 losing parties both prefer the other loser to the winner, then the winner was ranked last by a majority of voters.

for AV.

by **mike-l** » Tue Mar 13, 2012 10:21 pm UTC

I've done that, multiple times.

40% prefer A>B>C and I don't care if they vote for A only
30% prefer B>C>A and vote for B only
30% prefer C>B>A and vote for C only.

A is preferred least by 60%, but wins.

Any result in FPTP is possible in AV. This is what I mean when I say that AV is sensitive to where voters put their approval cutoff, in any situation where AV would give a different result than FPTP, by making the approval cutoff stricter you could change the result back to the FPTP result.

My assertion was for FPTP systems, sorry I didn't clarify that. I have no idea what a regular AV system looks like.

by **Griffin** » Tue Mar 13, 2012 10:24 pm UTC

Except, in that case, there is no candidate who isn't disliked by the majority. (Over 50% think A, B, and C all don't deserve to win)

There is literally no other option.

Which leads me to believe you completely ignored what I said.

If you've gotten to that point, it doesn't bloody MATTER what voting system you use - your fucked either way. IRV, Condorcet, whatever, you're going to get someone in office that the majority of people don't want to be there.

by **mike-l** » Tue Mar 13, 2012 10:25 pm UTC

But only A is ranked LAST by a majority.

But in this case, 60% prefer B over A, and 70% prefer B over C. So in any comparison with alternatives, the populace prefers B. FPTP and AV didn't find it, but IRV would. (Now, by changing where people set their cutoff, AV COULD find it, but there's no guarantee it would)

by **Griffin** » Tue Mar 13, 2012 10:28 pm UTC

What's your bloody point? SO?

They are ALL ranked "last" by the majority, because they all pulled a majority "PLEASE DONT WIN".

A had the most support, A should be the victor. Better someone gets what they want than no one does.

But I've begun to see the problem - your view. You see everything through the lens of your own system to such an extent that you think your rankings reflect some sort of objective reality instead of being a bullshit approximation. And there's nothing I can do to argue against idealogical conviction, so I'll leave it at this:

You've done nothing to prove AV is flawed, all you've done is prove that it would produce different winners than your favoured system.

Whoop-de-fucking-doo.

Not everyone agrees that the "least offensive candidate with no stances that commonly ranks in second or third because hes not objectionable or has a pretty face or is well known" is the one who should be winning elections.

You keep arguing AV is bad, but the extent of your argument is "because its not condorcet". Why don't you actually say something meaningfully negative about it if you're going to argue against it?

by **mike-l** » Tue Mar 13, 2012 10:34 pm UTC

Ok, look at my example

Say you are in the C>B>A group. Then if you think the layout is as I said, 40 A>B>C, 30 B>C>A, 30 C>B>A, then you should approve both C,B, so that your least favorite choice doesn't get in.

But if you are wrong, and the layout is actually 40 C>B>A, 30 B>A>C, 30 A>B>C, then if your group votes for both C and B, your favorite candidate loses, while voting C only, your group wins. So even if you are OK with B being in power, if you would still prefer C to be in power, it might be best not to 'approve' B.

I do see through a lens where the world is not so black and white that I'm either OK or not OK with someone. There are people I'm more OK with than others, and I'd like to be able to vote that way.

And yes, I'm taking it for granted that, when possible, we should pick the option that is preferred to all others. I apologize that I'm using established terminology and criteria. I'll try to be less of an asshole. My argument has never been that it's not condorcet, because neither is IRV. I've said that it can pick the condorcet loser, which I explained doesn't mean that it doesn't pick the condorcet winner, but that it picks someone who is unpreferred in any pairing. I've said it's sensitive to where voters set their cutoff, as you can see in this post, but I also said many posts ago. I've said it can fail to elect an absolute winner, and can elect an absolute loser. All of these are well established terms, and the last 2 have actually been called "Especially Intolerable Failures" by some authors.

by **Griffin** » Tue Mar 13, 2012 10:43 pm UTC

Your taking for granted that a ranking system accurately gauges preference as well.

Lets take your example, but instead of using a stupid, meaningless, context-less ranking, we'll us PREFERENCE POINTS!

And we'll assume that someone won't vote for a candidate unless they have 70 preference points out of a hundred in favour of them.

40% prefer A>B>C and vote for A only, preference rating 100! Other two have preference rating of 20 and 10.
30% prefer B>C>A and vote for B only, preference rating of 80! Other two have preference rating of 20 and 10
30% prefer C>B>A and vote for C only, preference rating of 100! Other two have a preference rating of 20 and 10.

Now lets add up our results:
4000+300+300 for A, giving a total preference of 4600
2400+800+600 for B, giving a total preference of 3800
3000+600+400 for C, giving a total preference of 4000

And yet in YOUR system, B, very clearly the least popular of the candidates, the least liked, the one people want to win THE LEAST, ends up winning. While in my system, AV, A (the most popular candidate by far) comes away with the victory, just as it should be.

Your system is JUST as good at picking the least popular candidate as mine, as long as you stop assuming that just because its ranked, its reality.

Again, your argument breaks down to "it's not a ranked system so its bad" without any support for why thats so other than continuing to repeat "because its not a ranked system". And I'm sorry, but I can't support IRV when it, as quite clearly demonstrated above, proves itself so capable of picking the least liked candidate.

by **mike-l** » Tue Mar 13, 2012 10:46 pm UTC

If we could accurately measure preference, that would be great.

But you were the one who said any rational voter would never report their preferences accurately. And may I remind you that you were the one who said 'IRV is basically the worst' but haven't given any arguments why.

by **Qaanol** » Tue Mar 13, 2012 10:51 pm UTC

mike-l wrote:Define 'reflect the will'.

Spoiler’d for long maths:

Spoiler:

Consider a multidimensional Euclidean “issues-space”. We can look at the finite-dimensional case, although in theory there could be infinitely many issues at play, and we will start by considering just one dimension.

Along any given “issues-axis”, each individual internally has some function representing his or her level of agreement with that position. For example, someone might have a Gaussian normal distribution centered at the position they prefer. An “open-minded” or “undecided” person might have a very broad distribution, and someone with strong opinions might have a sharply peaked, or perhaps mesa-like, distribution. We may think of these distributions as having their squares normalized to area one, although people with no opinion have a flatline 0 function, and in general people who don’t care much about a given issue may have less area under the square of their function.

Essentially, each of these functions represents the internal views of each person with regard to that issue. The only requirements are the function must be non-negative, square-integrable, and have square-integral no greater than 1. Having an area 1 under the curve means having a maximally-strong opinion on the issue, which we can assume many people do at least for some issues.

Since we have a finite number of people, each with a function bounding finite area, we can sum up all those functions to get the combined “view of the people” function for that issue. This could have many maxima and many minima, depending on the individual distributions. There could even be multiple points tied for the global maximum. In the extreme case, each person’s opinion might be almost a delta function of his or her own position. That is fine.

It might be tempting to repeat for every issue, meaning every dimension of our space, and sum them. But in reality a person’s opinion of a given position might not be found by linearly combining that person’s opinions of its coordinates.

Instead we will simply repeat the process for the whole space at once. Each individual has a function representing his or her level of agreement with each possible position. Again, we enforce a maximum square-integral of 1 over the whole space, although people who don’t care very much can scale that lower. Someone with no opinion whatsoever would have a flat 0 function.

We now have, for each person, a distribution over the whole issues-space. Summing these gives the combined “view of the people”. It could have multiple maxima, and so forth. It could have step jumps, odd behavior, whatever. The important bit is that its square-integral is well-defined on every region, and it bounds a finite volume.

Now for each candidate, consider his or her own internal function giving level of agreement with each possible position. Actually, we ought to use the function he or she campaigns on and advertises, since that’s what people will be voting on.

In any case, we have these two distributions. One for the candidate and one for the combined view of the people. Normalize them, then take their inner product. That is to say, after scaling them to each have a square-integral of one, multiply them pointwise and integrate over the whole space. We have to make sure discontinuous functions and so forth are integrated properly, but that’s simple enough to do.

The result of that calculation gives the level of agreement between that candidate’s views and the views of the whole population. Repeat for each candidate, and the one with the highest agreement is the best choice to represent the will of the people.

Spoiler’d for brief conclusion:

Spoiler:

mike-l wrote:Because there are situations where FPTP elects the condorcet winner and approval may not. Eg 33% of people each prefer A>B but approve both, 33% approve A>C, and 34% approve B only. Then A is the condorcet winner, and is elected in FPTP, but B is elected in approval. (In no way defending FPTP, I agree that approval is better, though still being able to elect condorcet losers makes me cry)

That example pretty well demonstrates how much better Approval voting is than FPTP. Under FPTP, candidate A would have gotten elected with 66% of the vote to 34% for B, when in fact 67% of the voters approve of B but only 66% approve of A.

This also goes to show that the Condorcet condition is not necessarily best for determining the will of the people. Indeed, perfect range voting where each person has a level of approval between 0 and 1 for each candidate, and those scores accurately reflect how much that voter agrees with that candidate, gives results that are optimal for determining the true will of the people.

mike-l wrote:If you're going to do Approval voting, why not go with Range Voting, instead of just yes/no on everybody, assign everybody a score. Then normalize everybody's votes so that each ballot adds up to 1 and see who got the most points.

Approval voting is range voting. It is the simplest form of range voting, where the range is {0, 1}. It is extremely easy to understand, implement, and tally. And it does not require any change to existing ballot procedures, just a yes-or-no vote for each candidate.

It is true that Approval voting does not achieve the same level of precision as when all real numbers [0, 1] are allowed in the range voting, but it is still pretty close when a large number of people are voting. And much simpler to explain, put in practice, and get people to use.

Silknor wrote:@Qaanol: I'm not defending FPTP, I think it's a fairly bad system. But approval voting is also a fairly bad system, even if it's not as bad as FPTP.

Approval voting is very nearly the best of all possible voting systems. The only one with theoretically superior results is another range voting method with more values in the range, but those have major practical disadvantages. Notably, complexity of explaining the process and getting people to understand and follow it, and difficulty in changing ballot-counting methods to account for them.

Approval voting has all the benefits of range voting, since it is a form of range voting, and all the simplicity of an up-or-down vote. It produces results that reflect the will of the people even more accurately than Condorcet, with much simpler voting and tallying procedures than any Condorcet method.

Silknor wrote:No doubt "would you be OK with them winning" is an easy choice the vast majority of the time (it's that tricky point at which 3 or more parties have roughly equal support where it's not so easy. But just about any voting system would pass this test, so it hardly seems important.

FPTP fails that test, because a voter may be okay with several candidate and unable to make a vote reflect that. IRV is just multiple rounds of FPTP, so it too fails. Even though on the face a voter seems to be indicating all acceptable candidate (and in order too!) in actuality the optimal will-of-the-people candidate could very well lose in the first round.

The best way I can explain it is “the candidate who is everybody’s second favorite”. Suppose everyone likes a certain moderate candidate a lot, maybe 80% agreement between each individual and that candidate. But there are a lot of other candidates, and every voter has some other candidate they like better, maybe one they’re in 85% agreement with. Those other candidates are only the top choice of a small fraction of the voters each, and near the bottom for more, whereas everyone likes the moderate.

FPTP would elect whichever of the extremists has the most followers. IRV would elect one of the extremists, since the moderate would lose in the first round (with zero votes!) Condorcet could pick any of the candidates depending on specifics. So far Condorcet is better because it at least has the possibility of electing the moderate. But range voting, including Approval voting which is a type of range voting, is even better, as it will elect the candidate that everyone likes.

Silknor wrote:Nor does approval voting always outperform FPTP. 80 percent prefer Candidate A to Candidate B, but not by much. They also really dislike Candidate C, so they vote for A and B. Despite the majority of voters favoring A, B can easily win if either B has supporters who slightly prefer B to A, but are smart enough to vote tactically instead of approving A and B, or if C's supporters slightly prefer B to A, and vote for C and B.

Another prime example of how accurately Approval voting results reflect the will of the people. You have 80% of people who like both A and B a lot, with only a minuscule edge to A in each of them. There are not enough remaining voters to get a different candidate elected, but if one of A or B is strongly preferred among the remaining voters then clearly that is the candidate with the most support among the voters. Approval voting triumphs.

Silknor wrote:You might say this is contrived, that it assumes people are casting ballots that don't represent their best interest (which is clearly to vote A and indicate that B is a close second somewhere besides the ballot box). But that's exactly the point. Approval voting encourages people not to vote according to their best interest, but rather by who they'd be ok with. This creates an opportunity for people who vote tactically instead of sincerely to produce results that would be avoided under other systems (sometimes even under FPTP!).

Not at all. In order to vote tactically, you need to already have at least a fairly strong preference for one of your acceptable candidates over another. If that is the case, then “voting tactically” by saying no to the less-preferred candidate is a valid simple way of indicating that strength of preference exists. If you are approximately equally in favor of either acceptable candidate, then you don’t really mind which of them gets elected.

Furthermore, the regret-minimizing strategy is to vote for each candidate you approve of, just as the Approval voting moniker suggests.

Silknor wrote:I think the example he gave should be:
33% prefer A>B>C, approve A and B.
33% prefer A>C>B, approve A and C.
34% Prefer B>A>C. Approve B only (the tactically correct choice, even if they'd be willing to approve A if B didn't stand a chance).

Then A gets 66% and B gets 67% and the win, despite that a majority prefer A to both B and C.

This once again demonstrates the superiority of Approval voting to not only FPTP and IRV, but also Condorcet. Since the people who like B best, like B much more than another candidate, whereas the people who like A best only do so by a small margin and also approve another candidate, it follows from the numbers you provided that the net amount of approval for B is greater than for A. Thus the will of the people supports B.

Griffin wrote:Condorcet, yes, is the best. But its also more complicated, more expensive, harder to trust and (this is a big one) easier in some ways to tweak fraudulently with no one noticing.

AV is a distinct middle ground, with the huge benefit of our system already being 95% compatible.

Condorcet is not best. Best is “perfect honest range voting” with all real values between 0 and 1 allowed. But that’s also infeasible on practical consideration. Approval voting is nearly as good as perfect range voting, and as you say already compatible without our current voting system.

Yakk wrote:As noted, Range Voting reduces to FPTP while allowing people to "partially spoil" ballots in a sense.

Only if you normalize it. But if you normalize it, then it’s not Range Voting. In reality, for rational voters and/or strategic voters, Range Voting reduces to Approval Voting.

mike-l wrote:2 ballot is a 2 round runoff, you do a round, then remove all but the top 2 and vote again.

This is in theory a strong idea, if the first round is Approval voting and the second round is heads-up. But in practice it requires either an IRV-style ballot so people indicate (by way of transitivity in IRV ranking) who they prefer from every possible heads-up runoff, or else running an actual separate runoff elections. Neither of those options is viable, since IRV ballots are a fiasco of complexity, both for voters to fill out and for poll workers to tally.

mike-l wrote:Ok, look at my example

Say you are in the C>B>A group. Then if you think the layout is as I said, 40 A>B>C, 30 B>C>A, 30 C>B>A, then you should approve both C,B, so that your least favorite choice doesn't get in.

But if you are wrong, and the layout is actually 40 C>B>A, 30 B>A>C, 30 A>B>C, then if your group votes for both C and B, your favorite candidate loses.

And yes, I'm taking it for granted that, when possible, we should pick the option that is preferred to all others.

In the Approval voting system, if a given voter approves of those other candidates and would be fine with one of them winning, he or she should vote for them. That way either a candidate they don’t approve of gets elected but there’s nothing they could have done about it, or else a candidate they do approve of gets elected, in which case they contributed to the victory.

If instead that person had voted only for a subset of the candidates they approve of, it’s possible that a candidate they disapprove of gets elected, but would not have been if this voter had voted for all the candidates he or she approves of.

That’s what I mean by the regret-minimizing strategy being best. It’s up to each individual to decide his or her own personal cutoff for “Do I approve of this candidate?” but once that question is answered honestly, the regret-minimizing strategy is to vote for exactly those candidate you approve of.

So in your example, if the B and C voters really prefer their favorite candidate to the exclusion of all others, then their relative rankings for the other candidates don’t matter very much because they disapprove of them all. Similarly for the A voters. But notice that if enough of the A voters also approve of another candidate, that other candidate gets elected, and the same for the B and C.

The point is, “Would I be okay with this candidate winning?” is exactly the right question to ask. If you’re okay with a candidate winning, then even if there’s another candidate you prefer more, you’re still okay if the first candidate wins, you won’t regret it. And if you’re not okay with a candidate winning, you definitely want some candidate you do approve of to win.

In Approval voting, the best strategy for an individual is also the best way to get the results to reflect the will of the people.

by **Griffin** » Tue Mar 13, 2012 10:54 pm UTC

Uh, yes, yes I did. Its needlessly complicated for the effect (see people have to use cheat sheets to get through it in countries that have it). It's easy to coordinate gaming of the system. It's much harder to track and keep honest. It would entail an overhaul of how votes are handled, processed, read, and tallied. It requires several iterations with an increased magnitude for error. It's likely to confuse people. It can, as I just demonstrated, lead to victory for the least liked candidate, because it assumes every step of every ranking is of equal value, which isn't enough to sink it alone but makes it functionally no better than AV, and its just plain more expensive.

You have pretty much entirely failed to argue for a single benefit of IRV that AV doesn't have. And now, here, my counter example, AV versus IRV. Votes in parens () have enough support to get an AV vote.

60% (A) > (B) > c
20% (C) > (B) > c
20$ (B) > a > c

Under an AV system, the most popular candidate (here, clearly B) runs away with the election. Under an IRV system, he is completely eliminated right from the get go.

It's absurd.

by **mike-l** » Tue Mar 13, 2012 10:56 pm UTC

IEEE used AV for a few years. They abandoned it because everyone just ended up voting for one candidate.

Also, in your example, there's no runoff. A wins outright.

I've had enough of your incredibly rude and asshatery tone. I admit I succumbed a bit to your level in my last post, for which I apologize. Anywho, I'm out, I may respond to Yakk, but I'm henceforth ignoring Griffin.

by **Griffin** » Tue Mar 13, 2012 10:59 pm UTC

That's a bit of strawmen, don't you think? Not really any more relevant than listing people who switched away from IRV.

Also, in your example, there's no runoff. A wins outright.

That was exactly my point! The candidate wins, even though more people want B to win. That is called "a flaw".

I've had enough of your incredibly rude and asshatery tone.

I'm sorry - can you clarify exactly what the problem was here? Via PM if you wish? Because I was trying to focus on your arguments and the topic at hand, and if there's some particular thing I'm doing that you find overly rude I can certainly make an effort to try and avoid it.

by **Qaanol** » Tue Mar 13, 2012 11:02 pm UTC

mike-l wrote:IEEE used AV for a few years. They abandoned it because everyone just ended up voting for one candidate.

They also didn’t have Ralph Nader, Ross Perot, or the Ronpaul.

The current US political system is being held back by the 2-party monopoly that cannot be broken in FPTP due to fear of spoiler effect. Under Approval voting, I expect we’d see the Libertarian party and Green party get a lot more support, and within a couple election cycles they’d be regularly winning some local prominent elections, and probably a few on the national stage too.

by **mike-l** » Tue Mar 13, 2012 11:07 pm UTC

Qaanal I 100% agree. FPTP is probably the worst system out there.

AV would address a number of issues with it, I just think IRV addresses more/better issues. The reason I pointed out the IEEE is because they are the most prominent group to have tried it. Other groups include the MAA and the ASA (mathematics and stats associations), but it's largely unused. IRV is used by numerous countries and many political parties within even more countries. It's been moved away from on occassion as well, but not on the scale of the IEEE.

Yakk, I guess we disagree that the choice that selected should be the one that is acceptable to most people vs the one that most people would prefer compared to any alternative. Given the former criterion, obviously AV is better at that if voters vote sincerely. But even if regret minimization is optimal, that does not mean people will do it (eg, the IEEE). Also, yes, by two ballot I meant 2 entirely separate voting days and processes.

by **Ghostbear** » Tue Mar 13, 2012 11:21 pm UTC

zmic wrote:This is an assumption for which you have no proof. There isn't even a proof that democracy is the best system to begin with, let alone that it works better by forcing non-willing people to vote.

What assumption? That democracy is about having a say in your government instead of choice? You provided just as much "proof" (none) as I did. I don't know why this is under debate, but we can take the very first sentence from the wikipedia article on democracy:

Wikipedia wrote:Democracy is an egalitarian form of government in which all the citizens of a nation together determine public policy, the laws and the actions of their state, requiring that all citizens (meeting certain qualifications) have an equal opportunity to express their opinion.

Which very much aligns with what I said, and doesn't align with what you said at all. By "all citizens of a nation together determine public policy" it indicates that you have a say in your government. Democracy has never been about choice, but about being able to influence the body that governs your life. You can easily have a democracy where the choice to not vote is stripped from you; or would you consider Australia, Brazil, France, or Belgium to not be democracies?

I have no idea where the comment about there being no proof that democracy is the best method of governance came from at all, or why it's relevant.

zmic wrote:It is not up to me to produce a good reasons for non-mandatory voting. It is up to you to produce a good reason for mandatory voting because it takes away a freedom from the citizen. If this is the best you can come up with, then I see no reason to make not voting illegal.

It causes the government to more accurately reflect the people, and not just the people who care enough or are well enough off to vote. It'd essentially make voter suppression no longer an issue. It'd cause frequently marginalized groups (due to low participation) less marginalized. It gives elected officials or those seeking to become elected officials more information on how the populace feels to adjust their policies.

If you're going to ignore the reasons provided, it's very easy to say that there's no good reason for mandatory voting, but that doesn't make your argument any better. It's not like this is the first time reasons have been provided; you can say they aren't good enough if you like, and provide reasoning for such, but willfully ignoring them doesn't mean they aren't there.

zmic wrote:I want the nation to respect my will not to vote.

And many people want their nation to respect their will not to pay taxes. Just because you don't want to do something doesn't mean you're entitled to not do it.

by **Diadem** » Tue Mar 13, 2012 11:38 pm UTC

I think people, in general, spend way too much time discussion formal mathematical properties of voting systems, without asking how important these defects are in the real world. Should I care about a voting system being able to elect the Concordet loser? Well if it happens in 50% of the cases, I'm gonna care. It it happens once in a billion, I don't. You need to look at realistic voting situations to judge a voting system, not formal mathematical properties.

Even better is reducing the number of elections where you can only elect 1 person. For a president it is necessary, of course. But for parliament the obvious solution is to just not use a district system. Then there's no FPTP, no RV, not IR, everybody wins.

by **Qaanol** » Tue Mar 13, 2012 11:43 pm UTC

Ghostbear wrote:It'd essentially make voter suppression no longer an issue. It'd cause frequently marginalized groups (due to low participation) less marginalized.

This is the first good argument in favor of mandatory voting that I’ve seen in this thread.

Diadem wrote:Even better is reducing the number of elections where you can only elect 1 person. For a president it is necessary, of course. But for parliament the obvious solution is to just not use a district system. Then there's no FPTP, no RV, not IR, everybody wins.

This is a good thought, but it has some drawbacks. Notably, if the population is actually split into two camps on some particular issue. Ideally the parliament/congress should have the same proportion of people from each camp as there are in the general population.

But if the split is, say, 60-40 in the population, and everyone votes for representatives with the top N vote-getters all being elected, then if this one issue is the decisive point, you’re going to end up with a parliament containing only people from the 60% camp, and none from the 40% camp.

by **yurell** » Wed Mar 14, 2012 12:53 am UTC

Griffin wrote: And hasn't Australia had problems with this approach? (name, its so complicated people just take party-distributed cheat sheets with them to the poll and copy them down verbatim)? Talk about strategic voting!

It's not really all that complicated. There are two houses that need to be voted for: the House of Representatives and the Senate. For the House of Representatives you just number the parties according to your preference (1-n), and you're done.
For the senate, there are two options — 'above the line' and 'below the line'. If you vote above the line, it works the same as the House of Representatives; namely you number the parties 1-n and you're done. If you vote below the line, you number all the candidates individually, instead of by party.

Often parties will hand out 'cheat sheets' that you can follow to ensure your runner-up votes go to the party they want to win if they don't themselves win, but that's not an 'it's complicated' stance so much, but 'if you want us to win, and we don't, you probably want these guys to win.'

In short, the system is as complicated as 'list these people in the order you want them to win', and the cheat sheets exist to help those who aren't really up-to-date with the politics decide who they want to win.

Who is online