An election system

For the serious discussion of weighty matters and worldly issues. No off-topic posts allowed.

Moderators: Azrael, Moderators General, Prelates

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

An election system

Postby DrZiro » Sun Nov 06, 2011 1:26 pm UTC

I've been thinking about election systems. One problem with a plain proportional system is that it promotes extremism; a small party with only a few percent can get into the parliament, and if there are two main blocs, the small party might get undue inluence. On the other hand, any type of single-canditate election - whether it's a simple plurality vote, some kind of runoff system, or a condorcet system - favours people who are very close to the middle; this means that you get candidates expressing almost the same opinions, so the voters don't have much to choose between. In other words, proportional leads to too much diversity, and single-winner systems lead to too little diversity. The idea of this system is to find a balance by mixing the two.

In this system, there is a parliament and a president - we might think of them as legislative and executive power. The president is elected through a condorcet system. He gets to pick 15 people to form a cabinet and work for him. The president and the cabinet are also part of the parliament, but have only a small fraction of the votes there. The other 240 seats are filled through a proportional election.

Each party gets to make a list of their candidates for parliament, and a list of their candidates for president. When you vote, you pick the two lists from one party - the parliament list and the president list go together, and you are not allowed to choose your own candidates for either. This is to avoid campaigns focusing too much on persons rather than on ideas. You do however get to choose a secondary vote, sort of like in an STV election. That way, you don't have to be afraid of voting for a party that might be too small to get in. (There is also a system for local lists, but I won't go into that right now.)

So you get a parliament where most of the members are elected proportionally, but a few are decidedly more centrist.
The supposed benefits of this method include:

In many proportional systems, you get two blocs that are very close in size. If a small extremist party gets into parliament, they could get the deciding vote, even if they themselves only have a few percent. With this method, one of the blocs can ally with the cabinet instead, so the extremists don't get that much influence.

Even if there is no such extremist party, you might often get shifting majorities. One bloc gets 51% in one election, the other gets 51% in the next, and so the politics sway wildly back and forth. With the condorcet cabinet mediating, you get a mix of both instead of going back and forth, so things get more consistent.

The executive power is more likely to have the respect of both blocs, and of most of the people.

Each of the presidential candidates (and their corresponding parties, if they have any) will want to go as high as possible on the condorcet lists of their opponents. That means that it is not a good idea to slander your opponents. Hopefully this should lead to more civil political debates.

You will get decisions that are closer to what most people want, but you will still be able to hear the opinions of others.

What do you think, is this a useful idea, or would it not work in real life?

aoeu
Posts: 325
Joined: Fri Dec 31, 2010 4:58 pm UTC

Re: An election system

Postby aoeu » Sun Nov 06, 2011 2:50 pm UTC

DrZiro wrote:In many proportional systems, you get two blocs that are very close in size. If a small extremist party gets into parliament, they could get the deciding vote, even if they themselves only have a few percent. With this method, one of the blocs can ally with the cabinet instead, so the extremists don't get that much influence.

If you don't let them to be a deciding vote a more correct wording would be that they get no influence at all. Why not just make it a requirement that a party needs to get a minimum few percent of the votes to be given any seats?

User avatar
Yakk
Poster with most posts but no title.
Posts: 11129
Joined: Sat Jan 27, 2007 7:27 pm UTC
Location: E pur si muove

Re: An election system

Postby Yakk » Sun Nov 06, 2011 3:00 pm UTC

I dislike any system whereby individual politicians who have power in parliament owe more to their party than to the people who elected them. The point of having a bunch of parliamentarians isn't merely counting which party has more points, but rather those parliamentarians are supposed to do stuff, and each individually have power to keep the government from pulling a fast one.

Under a party list PR system, if you disagree with your parties position for whatever reason, you are basically screwed. You might as well be an entry on a spreadsheet.

Under an individually elected system, if you disagree with your parties position for whatever reason, and can get enough of the same people who voted for you last time to support you anyhow, you aren't screwed. You get to keep your job in parliament.

I want my parliamentarians to be have leverage against their parties.

---

The "even bloc" effect is actually a sorting effect. Suppose there is a central pile of political opinion somewhere. And suppose there is 75% agreement, all things considered.

Now some relatively small subset (say, 20% of the 75%) happens to also want X, while the rest are tepid. Why give them X if you have this much support? You tell them to to screw themselves.

And thus the monolithic agreement fractures -- because what is important isn't what is essentially important, but what is important enough to disagree on, which is a function of how much you agree on. If you politically all agree that we should have a modern, western democracy with freedom of speech, and agree on all economic topics -- but you disagree about the drinking age, then politics will be viciously about what the drinking age should be.

A X% concordant swing just means that the size of a "kernel" of political opinion (a party or coalition) is around 50%-X% instead of 50%. Which should lead to more, smaller political parties. And when you fall short of 50%-X% after winning the presidency, you still need to bargain with the smallest parties for a few extra votes.

So I don't like your party lists -- I want 240 individual politicians in parliament, not N parties. And pushing the executive down to parliament isn't all that good either: you could easily have an executive that lacks a parliamentary majority, which leads to perverse incentives for parties to make the system not work and blame the other side.
One of the painful things about our time is that those who feel certainty are stupid, and those with any imagination and understanding are filled with doubt and indecision - BR

Last edited by JHVH on Fri Oct 23, 4004 BCE 6:17 pm, edited 6 times in total.

Danny Uncanny7
Posts: 74
Joined: Wed Oct 12, 2011 12:53 pm UTC

Re: An election system

Postby Danny Uncanny7 » Sun Nov 06, 2011 3:54 pm UTC

I think that the ideal voting system is a preferential voting list, where each vote is a ranking of all possible candidates. So each person ranks the candidates in order of preference. In the first round, only the first picks are counted. Whichever candidate has the least votes is knocked off and all of his voters get their vote changed to their second preference. Then the second least popular candidate gets knocked off and all of his votes get swapped for the next preference and etc etc. This way you don't have to decide between voting for the party you want and voting strategically against who you don't want. If you want to vote Ralph Nader and he gets knocked off, your second pick and still go to the Democrats. I am sure that there is a name for this system but I can't find it right now.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Sun Nov 06, 2011 5:15 pm UTC

Aoeu:
There are a few problems with a minimum percentage system.
First, it leads to more voting tactics - people may feel forced to vote for a party they think will get in, rather than the party which they actually agree with.
Second, parties will need to use tactics too - smaller parties will be forced to join together to have a chance, even if they are not really compatible and don't make sense as a single party.
Third, the system will be unpredictable, as a small number of votes can change the result very much; just enough votes to get a party over the threshold can have a large impact, which wrecks the proportionality.
Fourth, even if we don't want minorities to set the policies, we do want to hear what they have to say. As Yakk points out, members of parliament are not just there to count the number of votes, they are also doing other things, and one of those things is to discuss things and voice different opinions (which is why it's called "parliament", after all).

Yakk:
The question of parties vs. individuals is of course an old one, which we won't be able to solve here. I'm quite firmly on the party side in that debate. Increased person voting leads to increased populism, partly for the obvious reason that you are voting for a person and therefore focus more on that person, but also because with that many people to pick from, you don't have time to care about their actual politics. You have a point that in a system where we have many more options, we are more likely to find an option that suits us, but that is probably too optimistic, since people won't have time to consider all the candidates. What's more, to advance in such a system, you would need to spend huge amounts on campaigns, whereas in a party system you can let not only the party but the ideology speak for you.

I've specifically designed the system to be as impersonal as possible, to make sure things focus on politics rather than personality. Still, like it or not, the president would be a single person - that means that you can certainy make a career in politics without a supporting party. In fact, I suspect independent candidates would be much more likely to win here than in for example the US presidential elections.

And as for those members of parliament who disagree with their party, one beneft of this system is that it is relatively easy for them to start a new party. I don't think it's a good idea that members of parliament be encouraged to go off on their own quests - they should generally stick to doing what they were elected to do.

I don't quite understand the first few paragraphs of the second half of your post, but it looks interesting, so perhaps you would like to explain again.

It is conceivable that there would be two almost equally sized blocs, and that the presidential election would be so strongly polarised that one of the sides would able to pick their choice. Then you would get the same polarisation as in ordinary proportional systems, and the same swaying majorities. But I don't think that's likely. I think it's more likely that a compromise candidate would emerge as president, and he would put forth suggestions that compromised between the two sides. Hopefully this should lead to a breakup of the blocs.

Suppose we have a political spectrum from A to E. Without the condorcet president, there are two coalitions, AB and DE. When we add the president, he will probably be from the C party, and he will have to mediate between the others. At first, it might lead to a strange mix of AB policies and DE policies, but eventually B and D will start coming together to make their own reasonable compromises, rather than stubbornly standing their ground. It will also be possible for smaller special interest parties to have some influence in the parliament without disrupting it.

As for an executive that lacks a parliamentary majority - yes, you would very likely have that, but that's not the same as one which has the parliamentary majority against him, as often happens in the US. Here we would get a president who has wide acceptance in the parliament, much wider than 50%, if all goes well.

Danny:
I think you're talking about the Instant Runoff Voting system, or IRV. That's for single candidates, so it doesn't really work for a whole parliament. For the president in this case, I chose condorcet rather than IRV. It produces a candidate which has a wide support, where as IRV produces one that has a stronger but more narrow support. Since in this system he is supposed to be mostly a mediator, that made sense, even though IRV candidates are better for some situations.

For a multiple candidate election like the parliament, the closest equivalent would be the Single Transferrable Vote system, STV, which I mentioned briefly. That can be good for smaller groups, but it becomes difficult to keep track of several hundred candidates. If you vote for a bigger party, the effect is basically the same as a proportional election (but, as noted above, the fact that you can't vote for individuals makes the system focus more on actual politics).

The voting tactics you mentioned, with the Nader example, is an important point. For the president election, the condorcet system pretty much takes care of that. In the parliament election, there could be a problem if you vote for a small party. That's why I've added the secondary vote, which does exactly what you say: You can vote for Nader first, and Gore second, for example. There could of course be a third vote too, and so on, but I don't think that would make much practical difference.

User avatar
Yakk
Poster with most posts but no title.
Posts: 11129
Joined: Sat Jan 27, 2007 7:27 pm UTC
Location: E pur si muove

Re: An election system

Postby Yakk » Sun Nov 06, 2011 5:28 pm UTC

I don't elect people to do specific policies. I elect people to make decisions about problems that I haven't even had a chance to consider yet. The millions of people in the nation really don't have time to do the in-depth research to have informed opinions on every issue that should come before parliament -- that is why I want to elect parliamentarians to do. (Imagine the cost if every person in a nation spent 40 hours a week researching every politically charged issue -- it would be dumb.)

I really don't see the point of having a parliament if the members are just voting point shares of their parties and have no autonomy. And that is what party list systems do -- they reduce parliamentarians to living point shares. I want each party to have to bargain with their members, who in turn are closer to the electorate that directly elected them, as a proxy for bargaining with the non-uniform desires of the electorate (even the electorate who voted for a particular party).

I want my parliamentarians to actually discuss issues, to have personal research budgets, and I want there to be 200+ opinions being put forward in parliament, not 4 opinions aped from a party play book.
Suppose we have a political spectrum from A to E. Without the condorcet president, there are two coalitions, AB and DE. When we add the president, he will probably be from the C party,

Why would the president be from the C party?
One of the painful things about our time is that those who feel certainty are stupid, and those with any imagination and understanding are filled with doubt and indecision - BR

Last edited by JHVH on Fri Oct 23, 4004 BCE 6:17 pm, edited 6 times in total.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Sun Nov 06, 2011 7:10 pm UTC

As I said, you're right that we can't expect people to consider every political detail. But we also can't expect them to consider the ideas of several hundred candidates in depth.
I'm happy to discuss the pros and cons of a more individual election system, but that discussion doesn't really apply specifically to this system, any more than to existing proportional systems.

If we think of the parties A-E as a political spectrum (which is a simplification, but in many cases a relatively accurate one) then the condorcet system will tend to elect the most central candidate. If we assume that the AB and DE blocs are roughly equally big, the central candidate will be from the C party. In a different situation, he might be from one of the less central parties, or he might be independent - this is just one example.

User avatar
Yakk
Poster with most posts but no title.
Posts: 11129
Joined: Sat Jan 27, 2007 7:27 pm UTC
Location: E pur si muove

Re: An election system

Postby Yakk » Sun Nov 06, 2011 7:22 pm UTC

That would be true of every PR system relied on party list based systems. Multiple seats per riding systems, for example, do not always require party list based politics.

Yours is compounded with the directly (concordant) elected president appointing yet another bunch of flunkies to the parliament who owe their political existence to one person's opinion of their acts. Yet more useless warm bodies cluttering up parliament, instead of independent actors with actual power.
One of the painful things about our time is that those who feel certainty are stupid, and those with any imagination and understanding are filled with doubt and indecision - BR

Last edited by JHVH on Fri Oct 23, 4004 BCE 6:17 pm, edited 6 times in total.

User avatar
Dream
WINNING
Posts: 4338
Joined: Tue Dec 04, 2007 7:20 pm UTC
Location: The Hollow Scene Epic

Re: An election system

Postby Dream » Sun Nov 06, 2011 7:37 pm UTC

Danny and Dr Ziro: The system Danny is thinking of is called PR-STV, Proportional Representation, Single Transferable Vote. It works very well for multiseat constituencies. Candidates are elected when they reach a threshold of the eligible electorate divided by the number of seats, plus one. Surpluses to that number of votes are transferred according to the preferences expressed in the ballot paper.
I knew a woman once, but she died soon after.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Sun Nov 06, 2011 8:28 pm UTC

Perhaps I should have clarified - I'm thinking of a system where the voters are not divided into constituencies, as they are in many countries.

STV is indeed sometimes referred to as PR-STV, to emphasise that it's the multi-seat version. It is used for multi-seat constituencies, but as far as I can see, it requires a division into constituencies. Apart from that, I suppose it would give reasonably similar results, ignoring the person-vs-party difference.

Yakk:
You are right that the cabinet, in their role in the parliament, would serve mostly as "vote markers" for the president. Another way would be to simply say that the president gets 16 votes himself - perhaps that would even be better. It is also possible that the mediator role and the executive role should not be given to the same person. Maybe the parliament should elect the executive, as they do in many proportional systems. Either way, I figure the mediator (and the executive, for that matter) could use a few assistants, and then it might be convenient to formally put them in the parliament too - after all, it's not just about voting, but also about talking. It's also possible to elect the whole cabinet as a group. Anyway, that's details. The point is that the mediator (person or group) gets a small fraction of the votes.

User avatar
Dream
WINNING
Posts: 4338
Joined: Tue Dec 04, 2007 7:20 pm UTC
Location: The Hollow Scene Epic

Re: An election system

Postby Dream » Sun Nov 06, 2011 10:04 pm UTC

DrZiro wrote:STV is indeed sometimes referred to as PR-STV, to emphasise that it's the multi-seat version. It is used for multi-seat constituencies, but as far as I can see, it requires a division into constituencies. Apart from that, I suppose it would give reasonably similar results, ignoring the person-vs-party difference.

The recent Irish presidential election was run with PR-STV in a single single-seat constituency.
I knew a woman once, but she died soon after.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Sun Nov 06, 2011 10:33 pm UTC

I think you may be confusing the terms. STV (sometimes called PR-STV) refers to the multi-winner system. The Irish presidential election used IRV (sometimes called "single-winner STV"), the similar system for single-winner elections. For single-winner systems, constituencies would not be necessary, or even make sense at all.

User avatar
Zamfir
I built a novelty castle, the irony was lost on some.
Posts: 7604
Joined: Wed Aug 27, 2008 2:43 pm UTC
Location: Nederland

Re: An election system

Postby Zamfir » Mon Nov 07, 2011 9:13 am UTC

DrZiro, your system seems designed against a rather theoretical problem, instead of real-life experience. PR systems rarely behave the way you fear. In particular, you're ignoring that such sytems can have a centrist coalition, a majority that excludes parties more to the edges of the field.

That means that small parties never get a perfect fulcrum position. A small party might have a position of strength when it needed to make turn a near-majority coalition into a full majority coalition. But if the party starts making overly radical demands, people will simply start looking for a more centrist coalition instead. When a small party can make strong demands in a coalition, it's usually because their demands have some resonance within the larger parties. So the parties like being in a coalition with them, they're not entirely forced to.

Your system looks designed to make sure that small parties are never relevant at all. You consider small parties to be "extremist", which I presume you mean as a bad thing. In which case I am wondering why you want to have some PR aspects to your system in the first place.

If you look at elections in district systems, they don't really feel like choices between equals, do they? Their campaign focus may be on the centre, but everyone knows that different parties will bring in very different people when they win. Putting on a centrist face from the right and from the left are in practice very different things. People near the centre tend to think that the differences are important enough. The people who consider both candidates cookie-cutter equals tend to be on the edges, the same extremists you don't want to give a shot at influence in your system.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Mon Nov 07, 2011 5:19 pm UTC

The problem of minor parties getting undue influence is of course not something that happens everywhere all the time, but I do think it's an existing problem. I don't know all that much about how it works in other parts of the world, but I know how it works here in Sweden. For a long time, we have had two blocs, with the power shifting back and forth between them, sometime leading to problems of inconsistency. Since the last election, neither side has a majority, and a small anti-immgrant party have a swing vote. In Denmark, they have a similar situation, and have had since 2001; a small party have been able to radically change the immigration policies despite very few votes, because the governing parties don't have a majority in the parliament.

So at least in this case, it is definitely not a question of anyone wanting to be in a coalition with them. Hopefully in my system they would have more of a choice. It's still possible for a coalition to include smaller parties, after all.

I don't necessarily mean that "extremists" are always bad, but sometimes they are. My goal is that they should be able to have their say, but not be able to effect politics that have little popular support. I also don't think that all small parties are extremists (but extremist parties do tend to be small). Certainly small centrist parties could have reasonable influence in this system.

Even in a plurality system, no, the candidates are not necessarily indistinguishable. But such a system clearly offers fewer options. Fewer opinions are heard, and the voter turnout is low; the US often ends up around 50%, whereas here it's usually around 85%. Some of the problems of plurality systems can be solved with more modern systems, such as IRV or condorcet, but I think a case can still be made that the diversity is too low.

User avatar
Dream
WINNING
Posts: 4338
Joined: Tue Dec 04, 2007 7:20 pm UTC
Location: The Hollow Scene Epic

Re: An election system

Postby Dream » Mon Nov 07, 2011 5:36 pm UTC

DrZiro wrote: For a long time, we have had two blocs, with the power shifting back and forth between them, sometime leading to problems of inconsistency. Since the last election, neither side has a majority, and a small anti-immgrant party have a swing vote. In Denmark, they have a similar situation, and have had since 2001; a small party have been able to radically change the immigration policies despite very few votes, because the governing parties don't have a majority in the parliament.

I see this as a problem with political parties, not with the electoral system. A coalition should naturally lean towards the most popular policies of its members, not the most extreme. The more popular a policy, the more likely it would be shared by, or compromised on by the coalition members. If the major partner in the coalition is spineless and given to caving in to threats from the minor partner, that can't be blamed on the voting system, or even the existence of the coalition, only on the political ineptitude of the party involved.
I knew a woman once, but she died soon after.

DrZiro
Posts: 132
Joined: Mon Feb 09, 2009 3:51 pm UTC

Re: An election system

Postby DrZiro » Sun Nov 13, 2011 12:09 am UTC

Just to clarify, the anti-immigrant parties in question are not part of any coalition. The ruling bloc does have the option of cooperating with the opposite bloc rather than with the extremist party, which I think they largely do in Sweden, but less so in Denmark.

You might say that it's not the election system that's the problem, but in this case the election system is what I'm concerned with; even if it is not the problem, perhaps it can be part of the solution.


EDIT:
I realised a flaw, or possible improvement.
One of the ideas is that the system should increase compromise, so that if one bloc gets 51% and the other 49%, the 51% should not have 100% of the power. But if there are only two blocs, no middle parties at all, then the system doesn't do that; the larger bloc is still free to pick the president. We could solve that by letting voters make their own condorcet lists, but I've specifically wanted to avoid that, partly because it leads to focus on persons above politics, and partly because people often misunderstand the condorcet system and vote "wrong". One solution is to keep the party president lists, but make them separate from the parliament lists. That would also solve the problem that people might want to vote for a particular president but no party supports him, which could lead to the rise of insincere parties just looking to promote a president candidate. Such a change would probably require that the condorcet elect would be separate from the executive office - that is, not automatically the same person, but perhaps usually the same person, other wise you could potentially elect a president who is not accepted by the parliament. Perhaps the actual president should be found through a condorcet election within the parliament.

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Thu Mar 15, 2012 7:08 pm UTC

Not really related to this thread, but we got kicked out of the Obama Impeachment thread for irrelevancy.

Edit: Qaanol's numbers appear to be correct. The more variety of voters you add, the better AV gets and the worse IRV gets. I've spoiled my old data for posterity sake.
Spoiler:
Edit: Numbers in this post are a little off. In particular, the number of non-condorcet winner elections is artifically high as there was a bug that disallowed certain candidates from being a condorcet winner. The IRV implementation was also slightly off, but the results didn't change much. See my next post for more detailed results.

We were discussing Plurality (First Past the Post) vs Approval Voting vs Instant Runoff for a single winner election. Examples in each case can be given where the result doesn't jive with various criteria, the two that came up was the most preferred and the condorcet winner.

So I ran 1000 elections with 5 candidates and 10 groups of voters, each group having a random size, and random preference for each candidate, and for approval voting, approving all people they preferred over a random amount between 30% and 70% of their preference range (preferences were random[0,10] and the cutoff was random[3,7]).

The two benchmarks I used was the candidate with the highest total preference and the condorcet winner.

When the Condorcet and Preference winner were the same (42% of elections), the results were (ie how often each system picked the said winner)
IRV 94%
FPTP 74%
AV 68%

When there was no Condorcet winner (42% of the time), the percentage of picking the preference winner was
IRV 55%
AV 64%
FPTP 46%

When the Condorcet winner was different then the Preference winner (16%), the results were
IRV: 16% preference 76% condorcet
AV: 51% preference, 38% condorcet
FPTP: 22% preference, 62% condorcet


So in the 42% of elections when Condorcet and Preference winners are the same, IRV is incredibly reliable, and AV is actually worse than FPTP. It does about the same when there's no condorcet winner, while IRV plummets in effectiveness. Overall, IRV is a tad better at picking the preference winner at 67% vs AVs 64% and FPTP lagging at 55%. When there's a disagreement with condorcet and preference, AV gives a slight edge to the preference winner while IRV and FPTP heavily prefer the condorcet winner, with IRV being the best at picking at least one of them at 92% vs AV at 89% and FPTP at 84%.

I ran it again with the AV cutoff being fixed at 5, and this improved it slightly to pass IRV in picking the preference winner, but still being the worst in the 'obvious' elections.

In terms of picking the Condorcet winner when there is one, IRV completely dominates, 86% vs 68% for FPTP and 58% for AV.

In terms of picking either winner, IRV wins again, but not by as much, at 75% vs 71% for AV and 68% for FPTP.
Last edited by mike-l on Mon Mar 19, 2012 4:46 pm UTC, edited 4 times in total.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

Роберт
Posts: 4285
Joined: Wed May 14, 2008 1:56 am UTC

Re: An election system

Postby Роберт » Thu Mar 15, 2012 7:59 pm UTC

While your experiment is not perfect, I think it shows pretty clearly that FPTP is terrible.
The Great Hippo wrote:[T]he way we treat suspected terrorists genuinely terrifies me.

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Thu Mar 15, 2012 8:02 pm UTC

Outdated
Spoiler:
Ok, fixed a few bugs that didn't alter much, and got more detailed results. Here are 1002 elections. Each table has rows that indicate what preference rank the condorcet winner was, and the columns indicate what rank the winner was in each election type

Code: Select all

FPTP
               
Cond    1       2       3       4       5       Total
1       407     81      46      29      6       569
2       36      97      12      7       1       153
3       15      5       21      3               44
4       3       1       1       3               8
5       1                                       1
#N/A    90      65      43      22      7       227
Total   552     249     123     64      14      1002

Match Condorcet: 528


Code: Select all

AV
            
Cond    1       2       3       4       5       Total
1       403     119     37      10              569
2       87      42      16      7       1       153
3       29      7       4       3       1       44
4       1       6       1                       8
5       1                                       1
#N/A    134     53      26      11      3       227
Total   655     227     84      31      5       1002

Match Condorcet: 449


Code: Select all

IRV
            
Cond    1       2       3       4       5       Total
1       487     32      25      9       16      569
2       16      126     5       2       4       153
3       6       3       32      3               44
4       2               2       4               8
5               1                               1
#N/A    74      63      51      27      12      227
Total   585     225     115     45      32      1002

Match Condorcet: 649


AV is the clear winner in picking the preference winner, or in picking one of the top 2, or top 3, etc. It's the clear loser in picking condorcet winners. FPTP has nothing going for it.

When the condorcet and preference winners agree (57% of the time), IRV dominates. When there is no Condorcet winner (22% of the time), it's the worst. AV is the opposite.

I'm of the mind now that the best system is probably a condorcet + approval system, where each person ranks the candidates and approves a subset. Then if there is a condorcet winner, it's selected, otherwise the approval winner is selected (or rather, the approval winner from the Smith set is selected).

Edit: Just tested a Condorcet method, the Nanson's method, which is like IRV but instead of counting just first place votes, it counts the Borda count (10 points for first, 9 for second, etc) and removes candidates with below average amounts. Since the condorcet winner never has below average borda count, this method elects a condorcet winner 100% of the time. This one is incredibly competitive with AV on the non condorcet case. Here's the last 1000 I ran:

Code: Select all

Non Condorcet winners

Pref     1    2    3    4    5
FPTP   133    53   28  10    3
AV     122    53   41  7     4
Nanson 114    54   43  12    4


Not sure why FPTP did so well here, repeating the runs puts it lower while AV and Nanson stay similar.

Similar to the Nanson method is the Baldwin which eliminates just the borda loser instead of the below average ones. For some reason this one performs terribly on non condorcet sets, dunno why.
Last edited by mike-l on Mon Mar 19, 2012 4:47 pm UTC, edited 1 time in total.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

Роберт
Posts: 4285
Joined: Wed May 14, 2008 1:56 am UTC

Re: An election system

Postby Роберт » Fri Mar 16, 2012 4:36 pm UTC

I approve of your voting system and pretty much any voting system mentioned except FPTP.

Should I also rank my order of preference? :wink:

(For people who don't get the joke, I'm wondering what voting system we should use to vote on which voting system we should use.)
The Great Hippo wrote:[T]he way we treat suspected terrorists genuinely terrifies me.

User avatar
Qaanol
The Cheshirest Catamount
Posts: 3069
Joined: Sat May 09, 2009 11:55 pm UTC

Re: An election system

Postby Qaanol » Fri Mar 16, 2012 6:00 pm UTC

I just finished my own simulation. I went with a somewhat realistic model, using 500 voters and 5 candidates chosen independently at random for 1000 simulated elections.

Spoiler’d for detailed description of the model
Spoiler:
Each person, whether voter or candidate, has an opinion on each of 3 issues. Opinions range from 0 to 1. Candidates and voters have their position chosen independently at random from identical distributions.

One issue is neutral, so any given person is equally likely to have any opinion (rand). One issue is unifying, so people are more likely to have opinions in the middle (rand + rand)/2. And the last issue is divisive, so people are less likely to have opinions in the middle (rand + rand + 1)/2 (mod 1).

Each voter has level of drop-off for each issue, indicating how quickly the person dislikes opinions a given distance away on that axis. This is a simple linear slope, ranging from 0 (no drop-off) to 1 (rapid drop-off) but more likely to be in the middle than toward either extreme (rand + rand)/2.

A person’s disagreement with another position is just the difference in positions multiplied by the slope. A person’s opinion of a candidate is 1 minus the sum of the of the disagreements on each issue (weighted manhattan distance) bounded below by 0 so all opinions are from 0 to 1. That opinion, or level of agreement, is taken as the voter’s true and honest amount of support for that candidate, and used directly for tallying the preference winner, denoted by Range voting below.

The ordering of the candidates in terms of this level of agreement is used directly as the voter’s ranking of the candidates for IRV, and for determining who this person would vote for in each head-to-head race when checking the Condorcet condition. The first round of IRV is used as the whole of FPTP.


Spoiler’d for approval voting threshold methodology
Spoiler:
In approval voting, each voter has a threshold and will only vote for candidates above that level of agreement. I used three different ways to choose the threshold, and assumed each voter is equally likely to use any one of them. In each election, every voter is randomly assigned to one type of threshold.

The first is an absolute threshold, meaning a fixed level of agreement that must be exceeded. Voters of this type choose their threshold independently at random between 0.25 and 0.75, weighted toward the middle (rand + rand + 1)/4.

The second type is a relative threshold, meaning a fraction of the difference between the most and least favorable candidate. The fraction is chosen independently at random for each voter, between 0.25 and 0.75, by taking k = (rand + rand + 1)/4 and using (1 - k)*minpref + (k)*maxpref. Here minpref is the amount this voter agrees with the candidate he or she agrees with least, and maxpref is the amount this voter agrees with the candidate he or she agrees with most.

And third is a dynamic threshold relating to the gaps between candidates. Among the candidates this voter rates between 0.75*minpref+0.25*maxpref and 0.25*minpref+0.75*maxpref, the voter finds the two adjacent candidates with the largest difference between them in terms of this voter’s level of agreement (or the gaps at the ends of the interval if they are larger than between any adjacent candidates) and puts the threshold in that gap. In other words, when looking at the sorted list of agreement with each candidate, this voter finds the largest gap among the central section of candidates and votes for those above that gap.


We have candidates and voters pulled independently at random from the same nonuniform 3D issues-space. Voters have independent strengths of conviction about their positions on each issue, and rate candidates on how closely they agree with them. We also have multiple types of dynamical behavior for voters deciding on their approval voting thresholds, which are independent and random. Here are my preliminary results:

Spoiler:
When there is a Preference winner (and there always is unless the random numbers exactly balance out, which they don’t) the different voting systems find it the following percent of the time:
100% Range (assuming perfect honest range voting, which is impossible, but this is the baseline here)
87.8% Condorcet (the Condorcet method is quite good at finding the most-preferred candidate)
86.9% Approval (right up there too)
71.2% IRV (not so great)
41.7% FPTP (The FPTP method is terrible at finding the most-preferred candidate)

When there is a Condorcet winner (which happened 97.4% of the time in this simulation, although on average I’m seeing low 90s) the different voting systems find it the following percent of the time:
100% Condorcet
90.1% Range
85.5% Approval
75.4% IRV
41.4% FPTP


From this, we see that Approval, IRV, and FPTP are clearly stratified in that order for both criteria.

I also tallied the number of times each voting system picked as its winner the various last-place finishers.

Spoiler:
Here is how many times each system picked as its winner the Condorcet loser:
53 FPTP (that’s 5.4% of the time)
0 IRV
0 Approval
0 Range
0 Condorcet

And the same for each method picking the preference loser as its winner:
63 FPTP
0 IRV
0 Approval
0 Condorcet
0 Range


I ran some more simulations, one with only 200 voters but 10,000 elections that showed the trends above continue rather evenly. I also did several with 200 voters and 500 elections to to test various parameter changes. Adding more candidates makes all the voting systems perform somewhat less successfully. Specifically, with 10 candidates, here are rates of success in finding the Condorcet and preference winners, when each exists.

Spoiler:
Condorcet:
100% Condorcet
83.5% Range
75.3% Approval
53.4% IRV
25.6% FPTP

Preference
100% Range
77.8% Approval
77.0% Condorcet
46.0% IRV
24.6% FPTP


Furthermore, altering the politics so all issues are divisive makes for a similar effect. Here’s that with 200 voters, 500 elections, and 5 candidates,:
Spoiler:
Rate of finding Condorcet winner when it exists:
100% Condorcet
83.0% Range
75.2% Approval
61.2% IRV
38.8% FPTP

Rate of finding preference winner:
100% Range
77.2% Approval
77.0% Condorcet
58.0% IRV
39.4% FPTP


Conversely, making all issues centralizing has the reverse effect:
Spoiler:
Rate of finding Condorcet winner when it exists:
100% Condorcet
92.5% Range
92.3% IRV
84.2% Approval
73.2% FPTP

Rate of finding preference winner:
100% Range
91.2% Condorcet
87.8% IRV
86.4% Approval
72.2% FPTP


And, to show the combined effect of divisive issues and many candidates, I ran another full 500 voter, 1000 election simulation, with 10 candidates and all issues divisive:
Spoiler:
Rate of finding Condorcet winner when it exists:
100% Condorcet
85.3% Range
77.4% Approval
33.2% IRV
12.2% FPTP

Rate of finding preference winner:
100% Range
78.8% Condorcet
78.2% Approval
32.4% IRV
12.7% FPTP


The biggest take-aways here, I think, are that Approval voting is robust to changes in the issues and size of candidate fields. It performs consistently well on both the Condorcet and preference metrics, regardless of other things. If I go back and make more changes, I’ll try adding in a “net preference” running total, so at the end we can see how much total agreement there has been between the voters and the candidates elected by each method.

Spoiler’d for long conclusion
Spoiler:
By contrast, both IRV and FPTP are highly sensitive to the details of how divisive issues are, and how many candidates are running. FPTP is as expected, quite terrible. IRV is mediocre most of the time, and quite poor sometimes. When circumstances align, such as having all the voters clumped together in one central ball, and only a few candidates, then IRV can shine, even slightly outperforming Approval voting in those rare cases.

Range voting, at least as implemented here, is impossible to use in real life. It is not rational to vote perfectly honestly. At the very least, your top candidate should get the maximum vote and your bottom candidate should get the lowest possible vote. I may go back and add in a more realistic range voting, that scales the individual voter’s range to the full [0, 1].

I also learned that tallying an IRV vote is tedious, and has no shortcuts. In juxtaposition, even finding a Condorcet winner is trivial. That’s just a matter of looking at the N×N grid of theoretical head-to-head match-ups, and tallying which of those bins get the vote of the current voter based on the voter’s internal preference ratings for the candidates.

At least the results of a local Condorcet ballot can be tallied, and then all the local results tallied later, just by adding bins. There are only NN-1)/2 bins, but for reasonable sized elections that’s fine. Even a 10-person race just need 45 tallies. However, there is not always a Condorcet winner. I suspect something about the symmetry or shape of my distributions for voters in issues space led to there being a Condorcet winner in my simulations with greater frequency than there would in a real election.

If one were going to take the trouble of having voters fill out ranked choice ballots, it makes the most sense to use those directly for determining the Condorcet winner, if there is one. That would have a high probability of being the preference winner as well, according to my simulations. But on that note, the Approval voting ballot is much simpler to implement, much simpler to tally, and much simpler to get people to use correctly. Approval voting also has nearly indistinguishable performance from the Condorcet criterion, as far as locating the preference winner goes.


I’m open to suggestions for improving my model simulation. I’ve written it in Matlab, but it runs rather slowly for large input values. I might transfer it to compiled code if I do anything with it.
Last edited by Qaanol on Fri Mar 16, 2012 6:51 pm UTC, edited 2 times in total.
wee free kings

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Fri Mar 16, 2012 6:07 pm UTC

I'm leaning to agree with Qaanal on running more numbers. If we're going to go through the trouble of ranking everybody, we might as well just use a condorcet method.

But there's something wrong with your simulation, IRV cannot pick condorcet losers, but you have that happening 3.5% of the time. And while I understand that our voting methodologies are different, I'm consistently getting every system picking preference losers about 1% of the time, so I'm surprised that 3 of your systems got 0.

I modified my sim to use your voter preferences, but still having my 10 blocks of voters with random sizes, since I'm a chump and am using excel, so it doesn't easily scale. I get almost 90% success rate on finding the Condorcet winner with IRV.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

User avatar
Qaanol
The Cheshirest Catamount
Posts: 3069
Joined: Sat May 09, 2009 11:55 pm UTC

Re: An election system

Postby Qaanol » Fri Mar 16, 2012 6:27 pm UTC

Hmm, you’re right. I just spotted a minor bug in the Condorcet checking that was letting some candidates who would have a perfect tie and no losses count as a Condorcet winner (and similar for having a tie but no wins count as a Condorcet loser).

And also my IRV algorithm has voters, in case of perfectly equal preference between two or more candidates, pick the one that comes first on the ballot. I wasn’t expecting perfect ties to come up, but for voters with strong opinions and candidates who are far away, there could be zero agreement, and that could make for ties.

If all a voter’s remaining options are ranked at 0, that might be the issue. But, that wouldn’t explain why I’m getting IRV and FPTP giving the same number of Condorcet losers—and preference losers—as each other every time.

I’ll go add a thing so people won’t IRV vote for candidates they have zero agreement with. But I suspect the problem might be simpler than that, just a crossed wire somewhere.
wee free kings

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Fri Mar 16, 2012 6:43 pm UTC

Ok, so my model only uses your first 2 approval cutoffs, and still has the 10 voter blocks, but is otherwise the same as yours. I get very different results.

I'm gonna swap into an actual programming language to get rid of my final restrictions.

Condorcet Winner (92% of the time)
Condorcet 100%
IRV 89%
Range 81%
FPTP 72%
AV 65%

Preference Winner
Range 100%
Condorcet 80%
IRV 74%
AV 69%
FPTP 67%
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

User avatar
Qaanol
The Cheshirest Catamount
Posts: 3069
Joined: Sat May 09, 2009 11:55 pm UTC

Re: An election system

Postby Qaanol » Fri Mar 16, 2012 6:53 pm UTC

Hmm, interesting. I fixed the little edge-case bugs I could find, but I’m still getting similar results to before.

As for the Condorcet loser stuff, that was human error. I read the output table by column, when I needed to read it by row. What I had originally posted was how many times the Condorcet winner was eliminated in the first round by IRV and FPTP (always together).

Perhaps the fact that your people are grouped so tightly makes the elimination rounds of IRV operate much more in tune with the Condorcet criterion than it does for independent voters? Try something like 100 independent voters for 100 elections and see how that goes (or, 100 groups of size 1 voter each, if you prefer.)

Also, how are your candidates and voters distributed? Are they being placed in an issues-space and having the preference level calculated from objective agreement as mine are, or are you just randomly assigning preference levels for each candidate to each voter?
wee free kings

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Fri Mar 16, 2012 7:07 pm UTC

I'm using your criteria for voter preferences. I'm switching to VBA right now so I'll be able to run with more than 10 voters soon.

Edit: Ok, 99 Voters (to avoid condorcet ties), over 1000 elections, I get pretty much identical Condorcet results on IRV and AV, 80% each. AV wins on preference, getting 81% vs IRV 68%. Condorcet gets 86% when there is a Condorcet winner, which is 88% of the time.

The only difference left between our models at this point is that I only use your first two methods of AV cutoffs. Can you try removing the third one (the largest gap method) and running for 99 voters?
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

User avatar
Griffin
Posts: 1363
Joined: Sun Apr 08, 2007 7:46 am UTC

Re: An election system

Postby Griffin » Fri Mar 16, 2012 9:22 pm UTC

I just have to say, I'm very impressed by the work you've both done so far and enjoyed following. I'm glad this moved here and ended up heading in such a productive direction.

Range (real, honest range) is, as expected, the best system, but I'm happy to see the simplified AV methodology doing so well.

And I think the ultimate conclusion (anything is better than FPTP) definitely seems to have come through.
Bdthemag: "I don't always GM, but when I do I prefer to put my player's in situations that include pain and torture. Stay creative my friends."

Bayobeasts - the Pokemon: Orthoclase project.

Роберт
Posts: 4285
Joined: Wed May 14, 2008 1:56 am UTC

Re: An election system

Postby Роберт » Fri Mar 16, 2012 9:36 pm UTC

Griffin wrote:I just have to say, I'm very impressed by the work you've both done so far and enjoyed following. I'm glad this moved here and ended up heading in such a productive direction.

Range (real, honest range) is, as expected, the best system, but I'm happy to see the simplified AV methodology doing so well.

And I think the ultimate conclusion (anything is better than FPTP) definitely seems to have come through.

Unfortunately, with FPTP voting, there's no way to vote for "anything but FPTP".
The Great Hippo wrote:[T]he way we treat suspected terrorists genuinely terrifies me.

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Sat Mar 17, 2012 1:12 am UTC

Griffin wrote:Range (real, honest range) is, as expected, the best system

That's because our metric is how well each system agrees with range voting. And this assumes that these 'preference' numbers are actually meaningful things and comparable between voters, and that optimizing them is what's best for the group. Which is a reasonable assumption for sure, but an assumption nonetheless, and one that a number of economists disagree with.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

User avatar
Griffin
Posts: 1363
Joined: Sun Apr 08, 2007 7:46 am UTC

Re: An election system

Postby Griffin » Sat Mar 17, 2012 7:24 am UTC

It seems to have done quite well finding your condorcet winners as well, though, no?

It's still impossible, of course. :P

Since this is a general election system though, I do wonder - aside from the voting method, what other changes do you support?

(I know in advance no one else is going to agree with most of mine, so other people's are far more interesting at this point
Bdthemag: "I don't always GM, but when I do I prefer to put my player's in situations that include pain and torture. Stay creative my friends."

Bayobeasts - the Pokemon: Orthoclase project.

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Sat Mar 17, 2012 2:40 pm UTC

Griffin wrote:Range (real, honest range) is, as expected, the best system


Griffin wrote:It seems to have done quite well finding your condorcet winners as well, though, no?
80%, while many systems are 100%. Hardly the best.

As for other changes, I like 2 house systems, I like some form of proportional representation in at least one of the houses. I'm against private funding of campaigns. Probably a host of others.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

Goplat
Posts: 490
Joined: Sun Mar 04, 2007 11:41 pm UTC

Re: An election system

Postby Goplat » Mon Mar 19, 2012 4:12 am UTC

mike-l wrote:
Griffin wrote:Range (real, honest range) is, as expected, the best system

That's because our metric is how well each system agrees with range voting. And this assumes that these 'preference' numbers are actually meaningful things and comparable between voters, and that optimizing them is what's best for the group. Which is a reasonable assumption for sure, but an assumption nonetheless, and one that a number of economists disagree with.
Assuming people put a number on how much they like something (such as an election result) may not be realistic, but is there any accurate model for how people should make decisions that isn't too complicated to deal with?

User avatar
Qaanol
The Cheshirest Catamount
Posts: 3069
Joined: Sat May 09, 2009 11:55 pm UTC

Re: An election system

Postby Qaanol » Mon Mar 19, 2012 10:07 am UTC

The following was just posted on the previous thread. I assume it is going to be removed for being off-topic on that thread, so I’m quoting it here. I haven’t read all the links, and I can’t vouch for any of it, but I wanted to make sure the post didn’t get lost from the discussion on account of being in the wrong place:

Spoiler:
broken lader wrote:Here's why Approval Voting is far superior to IRV.
http://www.electology.org/approval-voting-vs-irv

There's a lot of discussion of voting method "properties" (criteria) here, which is a common rookie mistake. It's like comparing the horsepower and aerodynamics of race cars, instead of just putting them on a track and doing timed trials. The timed trials for voting methods are "Bayesian Regret calculations". They measure average voter satisfaction, as a function of all failures of all "criteria", even ones that have never been invented — and using the actual frequency-times-severity of those property failures.
http://ScoreVoting.net/BayRegsFig.html

The criticism about Approval Voting being "un-majoritarian" are just incredibly primitive. This is the kind of argument people historically make when they just heard about voting methods the day prior, and are operating on intuition. I provide what I think is a pretty robust rebuttal to that argument, here.
http://www.electology.org/majority

The "it can elect a Condorcet loser" argument is similarly primitive. The key with these theoretical possibilities is to look at their FREQUENCY (probability) and the amount of DAMAGE (utility decrease) they cause when they happen. While it is possible that Approval Voting could elect a Condorcet loser, it appears that in practice, Approval Voting is more likely to elect Condorcet winners (when they exist) than real Condorcet methods.
http://ScoreVoting.net/AppCW.html

And when voters game Condorcet methods, you can get some pretty devastating outcomes.
http://ScoreVoting.net/DH3.html
http://ScoreVoting.net/CondBurial.html

Further, aside from the bullet-proof axiomatic proof that the "best candidate" is the one who maximizes the sum of individual voter utilities — NOT necessarily the Condorcet winner — there are some interesting thought experiments that can refute the Condorcet principle even from mere intuition. For instance, this argument is pretty brilliant and simple:
http://ScoreVoting.net/FishburnAntiC.html

Having studied this issue for almost six years, alongside Warren Smith (a Princeton math Ph.D. and voting theory luminary) I can only express great frustration at the regularity with which I encounter people speaking with great confidence on this complex and counterintuitive subject, only to find that they've barely scratched the surface of the modern science on the issue. There's a lot to read and ponder before you dismiss Approval Voting or Score Voting.

Clay Shentrup
The Center for Election Science


I’m working on rewriting my simulation as a native Mac application. It might take a few more days. Also, some of the pages linked in the quote here indicate that it is necessary to take into account dishonest voting, especially “burying” a rival on a ranked choice ballot, since people do so in real life. One of the pages in particular posits that in actual practice, approval voting elects the true Condorcet winner more often than real Condorcet methods. And of course another of the pages talks about how Condorcet is not the proper criterion for determining the will of the people.
wee free kings

User avatar
Diadem
Posts: 5654
Joined: Wed Jun 11, 2008 11:03 am UTC
Location: The Netherlands

Re: An election system

Postby Diadem » Mon Mar 19, 2012 12:33 pm UTC

mike-l wrote:Preference winner

You are using this criterion without defining it. I can't find any definition of this criterion anywhere in this thread, or even on google. I suspect you made it up, but since you're not saying how, that makes it kind of hard to judge what you are doing.



Also (this part is not aimed at anyone in particular), why do these type of discussions always devolve into discussions about artificial mathematical criteria, that have very little to do with real life situations, and of which it is often entirely unclear that they are even good criteria to begin with.

One example, but I could really take any other of these kind of criteria as well, is the Condorcet winner. Everybody always seems set on the idea that voting systems should elect the Condorcet winner. Why?

Imagine the following situation: 26% of voters have a slight preference of A over B over C. 25% of voters have a slight preference of A over C over B. The other 49% of voters have a slight preference of B over C, but really hate A's guts. In this situation A is the Condorcet winner. But should he really be elected? Maybe, I don't know, we can talk about that. But it's most certainly not obvious that he should be. Used as an absolute criterion Condorcet is not a very good one. Situations where a voting system fails to elect the condorcet winner may very well be a feature, not a bug.
It's one of those irregular verbs, isn't it? I have an independent mind, you are an eccentric, he is round the twist
- Bernard Woolley in Yes, Prime Minister

User avatar
Qaanol
The Cheshirest Catamount
Posts: 3069
Joined: Sat May 09, 2009 11:55 pm UTC

Re: An election system

Postby Qaanol » Mon Mar 19, 2012 1:26 pm UTC

Diadem wrote:
mike-l wrote:Preference winner

You are using this criterion without defining it. I can't find any definition of this criterion anywhere in this thread, or even on google. I suspect you made it up, but since you're not saying how, that makes it kind of hard to judge what you are doing.

It means the candidate who is most-preferred by the whole population. As in, take each voter’s true opinion of that candidate, with 1 being complete agreement and 0 being total disagreement. Add up those values for all voters. The result is the level of preference for that candidate among the voters. The preference winner is the candidate for which that value is greatest.

I described this in more mathematical detail here.

The problem is that there is no way to measure the true preference of each individual. The closest we can get is Range voting, but even there people will at the very least scale their preferences so their favorite candidate gets 1 and their least favorite gets 0. (That scaled preference method will be an option in my new simulator.)

The other thing I’m thinking of including, though I’m not sure how to model it, is some way for each voter to make a rational play at tactical voting. As in, give the voters some sort of “straw poll” results—maybe created from a small sample of randomly chosen voters—and then have each voter decide whether to “bury” one or more medium-ranking candidates that have strong straw-poll support. That means, to give dishonestly low opinions of the strongest rivals to one’s favorite candidate.

Diadem wrote:Also (this part is not aimed at anyone in particular), why do these type of discussions always devolve into discussions about artificial mathematical criteria, that have very little to do with real life situations, and of which it is often entirely unclear that they are even good criteria to begin with.

One example, but I could really take any other of these kind of criteria as well, is the Condorcet winner. Everybody always seems set on the idea that voting systems should elect the Condorcet winner. Why?

Imagine the following situation: 26% of voters have a slight preference of A over B over C. 25% of voters have a slight preference of A over C over B. The other 49% of voters have a slight preference of B over C, but really hate A's guts. In this situation A is the Condorcet winner. But should he really be elected? Maybe, I don't know, we can talk about that. But it's most certainly not obvious that he should be. Used as an absolute criterion Condorcet is not a very good one. Situations where a voting system fails to elect the condorcet winner may very well be a feature, not a bug.

Absolutely. Condorcet is not the right criterion to use. That’s why we’re going with the Preference Winner. The candidate that is most preferred by the voters is exactly the candidate that “should” win the election.
wee free kings

User avatar
Yakk
Poster with most posts but no title.
Posts: 11129
Joined: Sat Jan 27, 2007 7:27 pm UTC
Location: E pur si muove

Re: An election system

Postby Yakk » Mon Mar 19, 2012 2:24 pm UTC

Qaanol wrote:The problem is that there is no way to measure the true preference of each individual. The closest we can get is Range voting, but even there people will at the very least scale their preferences so their favorite candidate gets 1 and their least favorite gets 0. (That scaled preference method will be an option in my new simulator.)
It is worse than a measurement problem. Describing what you actually mean is next to impossible.

And even if you did manage to describe it, there is the "how do you get more political power" question -- what kind of behavior does your system encourage? In your case, if candidate X is the one closest to what you want, becoming a fanatic for candidate X increases your political power in the system (ie, self-normalizing your own opinions towards 1.0 for candidate X).

Even if we don't talk about that kind of thing, what you describe gives more power to fanatics, and less to people with nuanced opinions. Just because I disagree with some opinions of every political candidate, should I have less political power? Meanwhile, Bob over there has picked a candidate, and agrees with everything the candidate says, and thus has more political power?

The "Preference winner" concept, to me, fails both practically (defining, measuring, etc) and theoretically (fanatics shouldn't be given more political power). The fact that a transformation of a voters political position from a moderate to a fanatical one (even a fanatical one for a moderate candidate) increases the political power of the voter is a sign of failure to me.
One of the painful things about our time is that those who feel certainty are stupid, and those with any imagination and understanding are filled with doubt and indecision - BR

Last edited by JHVH on Fri Oct 23, 4004 BCE 6:17 pm, edited 6 times in total.

mike-l
Posts: 2758
Joined: Tue Sep 04, 2007 2:16 am UTC

Re: An election system

Postby mike-l » Mon Mar 19, 2012 2:41 pm UTC

I'm unclear how Bayesian Regret of sincere range voting could be anything other than 0, as it seems to be measuring the difference in preference outcome vs the preference winner, and sincere range voting should always elect the preference winner. Yet they list it as having a small positive value.

There are tons of 'proofs' that are of the form 'If all the voters vote this way then approval is better', and then assuring us that their assumptions are reasonable, without comparing them to reality, for example Australia, where, despite the claim that IRV leads to two party systems, there's currently a 4 party coalition running their house. The largest group ever to use AV on the other hand is IEEE, who found it quickly degenerated back into FPTP. I really would love to see how AV would do on a country scale though, and it's clearly at least as good as FPTP, and relatively trivial to implement. Here's hoping someone does it.

Edit:
Ok, It looks like there was no problem in either Qaanol or my code, it's just the number of voters that makes a difference. The more individual independently voting voters there are, the better AV does and the worse IRV does. It's still better than FPTP, but from a purely theoretical nonstrategic voting point of view, AV seems to mop the floor with IRV. AV is also surprisingly good at picking Condorcet winners, much better than IRV. Condorcet methods still top the charts though, and condorcet winners are more and more likely the more voters you add. (Ie the condorcet winner is the preference winner more often than the AV/IRV/FPTP winner is, though AV is close) For specific numbers, see Qaanal's first post, I can now confirm they are in line with mine.

I've completely shifted my support to Nanson's method, which is like IRV except when picking who to remove from the ballots, a Borda count is done and everyone below average is removed. This is Condorcet since a Condorcet winner cannot be below average Borda count so can never be eliminated. (Borda count = N-k points for each kth place ranking). From a 'moral' point of view, I prefer ranked systems as ultimately the choice is between the first and second candidates, and I'd like anyone who wants to have a say between them to be able to.

This still completely ignores voter behaviour with regards to strategy though, something that is much harder to model.
addams wrote:This forum has some very well educated people typing away in loops with Sourmilk. He is a lucky Sourmilk.

User avatar
Griffin
Posts: 1363
Joined: Sun Apr 08, 2007 7:46 am UTC

Re: An election system

Postby Griffin » Mon Mar 19, 2012 5:19 pm UTC

And the thing is, all of this? This is just to insure that the candidate people want to win actually wins.

But there's an important question left unasked - do we really want the candidate who professes to be closest to what the voters want? Or do we want the candidate who will do the best at giving the voters what they want?

For population level voting, I primarily support AV for many reasons, but I also don't particularly support population wide voting, because no matter the system used, I don't believe it's the best way to get the best candidate.

As I've mentioned before, I consider a representative sampling employed for a duration of time with an obligation to research and investigate for the sole purpose of electing a representative to be the best way to handle this system. Jury Pools of a needed size rather than elections.

But I'd like to hear other opinions and alternatives. Because the point of the whole democratic election system is to pick the "best" candidate, is it not? Are there ways to improve the chances of this happening that we can build into the election system?
Bdthemag: "I don't always GM, but when I do I prefer to put my player's in situations that include pain and torture. Stay creative my friends."

Bayobeasts - the Pokemon: Orthoclase project.

nitePhyyre
Posts: 1280
Joined: Mon Jul 27, 2009 10:31 am UTC

Re: An election system

Postby nitePhyyre » Tue Mar 20, 2012 7:07 pm UTC

Yakk wrote:Even if we don't talk about that kind of thing, what you describe gives more power to fanatics, and less to people with nuanced opinions. Just because I disagree with some opinions of every political candidate, should I have less political power? Meanwhile, Bob over there has picked a candidate, and agrees with everything the candidate says, and thus has more political power?

The "Preference winner" concept, to me, fails both practically (defining, measuring, etc) and theoretically (fanatics shouldn't be given more political power). The fact that a transformation of a voters political position from a moderate to a fanatical one (even a fanatical one for a moderate candidate) increases the political power of the voter is a sign of failure to me.
Yes, people will normalize their preference to whatever scale you give them. But, people aren't going to actually become more fanatical. What you are talking about is strategic voting. And yes, a system where someone who doesn't think to 'game' it is given less power than someone who cheats is flawed. But it isn't about fanaticism.

Griffin wrote:But there's an important question left unasked - do we really want the candidate who professes to be closest to what the voters want? Or do we want the candidate who will do the best at giving the voters what they want?

For population level voting, I primarily support AV for many reasons, but I also don't particularly support population wide voting, because no matter the system used, I don't believe it's the best way to get the best candidate.

As I've mentioned before, I consider a representative sampling employed for a duration of time with an obligation to research and investigate for the sole purpose of electing a representative to be the best way to handle this system. Jury Pools of a needed size rather than elections.
When the States was founded they had this. The franchise was limited to rich, white, males. At the time, the kind of people who had the time to read news. There was an extra layer of filter between the whole population and the government. While expanding the franchise to be more representative was definetly a good thing, I could see how reaquiring that filter could be a good thing also. Have the general population vote on a body of representatives that votes on the house, senate, and executive elections. Maybe use the people in state government?

Griffin wrote:Because the point of the whole democratic election system is to pick the "best" candidate, is it not?
No, not really. The point of democracy is to give people who are going to be governed a say in that governance. This tends to weed out the worst candidates, but it also weeds out the best.

I say that we should get rid of 'terms'. Just keep a running tally. People could change their vote at any time. If someone is doing a god job, why make them go out and campaign? If someone is doing a bad job, why wait 4 years to get rid of them?
sourmìlk wrote:Monopolies are not when a single company controls the market for a single product.

You don't become great by trying to be great. You become great by wanting to do something, and then doing it so hard you become great in the process.

User avatar
Yakk
Poster with most posts but no title.
Posts: 11129
Joined: Sat Jan 27, 2007 7:27 pm UTC
Location: E pur si muove

Re: An election system

Postby Yakk » Tue Mar 20, 2012 7:12 pm UTC

nitePhyyre wrote:
Yakk wrote:Even if we don't talk about that kind of thing, what you describe gives more power to fanatics, and less to people with nuanced opinions. Just because I disagree with some opinions of every political candidate, should I have less political power? Meanwhile, Bob over there has picked a candidate, and agrees with everything the candidate says, and thus has more political power?

The "Preference winner" concept, to me, fails both practically (defining, measuring, etc) and theoretically (fanatics shouldn't be given more political power). The fact that a transformation of a voters political position from a moderate to a fanatical one (even a fanatical one for a moderate candidate) increases the political power of the voter is a sign of failure to me.
Yes, people will normalize their preference to whatever scale you give them. But, people aren't going to actually become more fanatical. What you are talking about is strategic voting. And yes, a system where someone who doesn't think to 'game' it is given less power than someone who cheats is flawed. But it isn't about fanaticism.

I was responding to someone who had framed non-scaled preferences as some kind of "platonic ideal" against which voting systems should be measured. There is a measurement problem with non-scaled preferences (and your noted gaming), but even beyond the measurement problem, it wouldn't be a good system even if there was no way to game it (ie, you could read the minds of people and actually extract their non-scaled preferences).

If non-scaled preferences are not desirable, then using it (where we model people's behavior a as a function of their non-scaled preferences and their game strategy) as a metric (where you compare the results of the election to what would happen if you could read people's minds) to measure the performance of another (implementable) voting systems is a bad idea.
One of the painful things about our time is that those who feel certainty are stupid, and those with any imagination and understanding are filled with doubt and indecision - BR

Last edited by JHVH on Fri Oct 23, 4004 BCE 6:17 pm, edited 6 times in total.


Return to “Serious Business”

Who is online

Users browsing this forum: No registered users and 12 guests