GOSH Community Call - Thursday February 25th 14:00-15:30 UTC

[briannaljohns]

Hey everybody!

Since August, The GOSH Governance Working Group has been working to create a governance structure for the GOSH community. You can see what’s been accomplished so far by browsing through our notes on the forum.

On February 25th the Governance Working Group will be hosting one more community call before finalizing the governance plan so that community council elections can begin!

The purpose of this call is for the working group to hear input from the GOSH community on setting up the election process and presenting the final proposal for the governance structure.

The agenda will be shared in advance of the call.

We’ll circulate the governance proposal document for comments one week before the call, and will also share an updated version after the call for any final comments.

How do I join?

You can join via this Zoom link.

We understand that using Zoom is not ideal, however, in circumstances such as community calls, it is best to ensure that we are using a platform that will allow for stable calls with larger groups, especially when important discussions are to be had.

Please let us know if you have any questions!

[hpy]

Thank you so much @briannaljohns for the announcement, happy to hear that there is another community call. I will do my best to attend.

You can see what’s been accomplished so far by browsing through our notes on the forum.

I read a couple of the most recent notes (very detailed, thanks!) including the one from 9 February (meeting #15), and just like to confirm a few things:

• Election models
  • To discuss on community call: simple vote count vs ranked choice?
    • Other methods are available, such as Schulze method

Have these options been set in stone, i.e. are we only choosing between “simple vote count” vs “ranked choice”? Or are they just examples and we can discuss other options?

I wrote a post in this forum analysing election mechanisms near the beginning of the governance process last year though it didn’t get much response at the time. I’ve been following the latest developments in election reform for several years, and in the post I argued against a simple majority vote for elections (aka first-past-the-post/presumably what “simple vote count” meant from the 9 February meeting notes). Such a voting method would be the opposite of the inclusivity that GOSH strives for.

At the time, I was hoping to spur a discussion and consequently didn’t want to push too strongly for any particular alternative method. However, since my original post didn’t gain much traction and/but the community call is coming up, I’d like to respectfully make a few points for your consideration.

“Ranked choice voting” (RCV) is actually a broad term that encompasses a variety of voting methods. I don’t know which one the meeting notes refer to, but the most common variety that people mean is instant runoff voting (IRV). I’m happy to go into details if there’s interest, but empirical evidence from real-life implementations of the instant runoff flavour of ranked choice voting shows that it rarely lives up to expectations. In a notable instance in the US city of Burlington, Vermont, the instant runoff voting for mayor - despite the best of intentions - led to the election of someone that most voters ranked last on their ballots. Burlington voted to repeal this voting system soon afterwards. There is a newly released study that examined 96 different implementations of ranked choice voting across the US which showed this kind of unintended consequence is sadly not the exception.

In case there is further interest, there are also peer-reviewed academic publications that demonstrate the unintended, but substantial pitfalls of ranked choice voting:

Burnett, C. M., & Kogan, V. (2015). Ballot (and voter) “exhaustion” under Instant Runoff Voting: An examination of four ranked-choice elections. Electoral Studies, 37, 41–49. Redirecting (PDF)

Ornstein, J. T., & Norman, R. Z. (2014). Frequency of monotonicity failure under Instant Runoff Voting: Estimates based on a spatial model of elections. Public Choice, 161(1), 1–9. Frequency of monotonicity failure under Instant Runoff Voting: estimates based on a spatial model of elections | Public Choice

The good news is that there are better methods avoiding the problems of instant runoff voting (the most common variety of ranked choice voting), a few of which I listed in my post. The even better news is that these replacement methods are easy to implement, online and offline.

One promising method is score-then-automatic-runoff (STAR) voting. With STAR voting:

• Voters are asked to assign a score (such as from 0 to 5, with 5 being the highest support; or even on a Likert Scale) to each choice just like book rankings on Amazon or the marks/grades you receive on homework.
• Voters are allowed to give the same score to more than one choice.
• The choice with the highest cumulative score wins. This is much easier to tabulate than ranked choice voting: Just add the numbers!
• In case more than one choice ends up with the same score, an automatic runoff round happens where the choice that got the most preferred votes wins.

In practice, the main difference for voters is that they score instead of rank the options. Turns out this is easier for many because the concept is easy to understand, and often it is hard to assign ranks outside of your most preferred couple of choices (“I’ve ranked my favourite candidate highest, but I don’t really care for the other two options, how do I pick which one to rank higher???”). Score-based voting such as STAR provides far more flexibility. Most importantly, it is more expressive and inclusive of different opinions.

Luckily, there is a web implementation of STAR voting that we can use immediately. From what I’ve read, there is a way to conduct an online STAR-voting election where each person is only allowed to vote once (this is a feature that not all voting platforms offer). The development team behind this platform is fairly responsive, and I would be happy to reach out to them with any questions/concerns that we may have.

On that note, the 9 February meeting notes also state:

Considering OpaVote and Helios options for voting

Again, are OpaVote and Helios the only two we are choosing from or are they just examples? My post from last year listed several options that are 100% open source (including Helios). On the other hand, as far as I can tell OpaVote is not open source (please correct me if I’m wrong). Considering the principles of GOSH and the sacred nature/sanctity of elections, I strongly urge against adopting any proprietary (i.e. closed source) solutions because they are, by definition, a black box controlled by others. It should also be noted that OpaVote is a paid service where it is only “free of charge for elections with up to 25 voters and 10 candidates.” I personally have no problem paying for a service, but OpaVote is likely closed source and does GOSH have a dedicated budget for running elections? If there is a budget, the money is much better spent supporting open source solutions.

Helios itself is fully open source, but “to create and administer an election [on their main instance], you will need to log in using Google, Facebook, Yahoo, or Twitter.” These log in options are not only proprietary, but also extremely invasive of privacy and do not respect our digital rights (I mentioned this in my post). To overcome this limitation, we would have to self-host an instance of Helios (maybe the tech people improving the GOSH website can help?). In addition, it is not clear from the Helios website which voting method(s) they implement. Can someone chime in on this?

Fortunately, there are other fully open source options. One is STAR voting, another is the Condorcet Internet Voting Service (CIVS) hosted by Cornell University. CIVS tries to improve on ranked choice voting by tabulating votes using pairwise comparisons between candidates (specifically, a Condorcet method using the Schulz algorithm mentioned in the meeting notes) to ensure that the winner individually wins against every other candidate. This is considered to be slightly better than instant runoff voting, but to my knowledge Condorcet voting does not alleviate all of the pitfalls of ranked choice voting. Lastly, my post also mentions other possible platforms such as Loomio.

Two frequently asked questions/comments I see are:

1. Q: Simple majority vote is terrible for democracy, surely anything else would be better? So we should just use ranked choice voting. Let’s not make perfect the enemy of the good.
   A: Indeed most other options are better than simple majority voting. But since we know there are options more inclusive than ranked choice voting and that are just as easy (and often easier) to implement, we should seize this rare opportunity to reach for something better.
2. Q: We can argue over the finer details of different voting methods all day/month/year, but we need to be practical! So let’s just go for simple majority voting.
   A: An exhaustive academic treatise analysing different voting methods (and I’ve only touched on a few prominent ones in this post!) would indeed take up multiple tomes but we don’t need to dig that deep for a good solution. GOSH is a prominent member of the open source community and the GOSH Manifesto lays out clear aspirations to improve accessibility, act ethically, democratise science, empower people, all with no black boxes. Adopting a better voting method ticks all of those boxes and would set a positive example for other open source communities.

With all of that said, I specifically suggest:

0. We not use a simple majority vote for the upcoming election. It is the worst of all worlds, goes against the GOSH Manifesto, and sends a negative message that we do not care enough about an inclusive voting method.
1. Instead, strongly consider STAR voting as an easy, more expressive and inclusive voting method compared to ranked choice voting.
2. If, for some reason, the GOSH Governance Working Group decides against 1., we should at least avoid proprietary solutions (and certainly not pay for one!). Again, other solutions such as CIVS exist.
3. If all of the above fails, GOSH commits to acknowleding this failure and seriously investigate and adopt a better solution for future elections.

To be clear, the above are obviously just my opinions with my own privileged biases and prejudice. Whatever GOSH does will of course be a collective decision, and like I said in my post most people here are far more experienced than I am in managing a community. But I just like to humbly submit these suggestions for your consideration.

I’m happy to discuss more, but given the limited time of the community call, hopefully this post provides a bit more detail. Looking forward to your feedback and/or critique.

[julianstirling]

While I am neither as passionate or as knowledgable as Pen about electoral systems I do 100% agree with the point that the most common voting systems leave people disenfranchised. This is worth giving some serious thought. I do think that making the perfect the enemy of the good can be a problem, but we should set a reasonably high bar on what we define as “good”. Thanks @hpy

[marinappdf]

Thank you for the explanation!

I’m at the governance working group and I have been studying election mechanisms, but I know nearly nothing about it. Thank you for your class.

At first I “gave up” for simple majority because I have read that ranked election could lead to some unintended results. Since that was the only option I knew, I decide that was better to use simple majority, which we know and understand its bads, than choosing an unknown method that could lead to unexpected results.
However, after your explanation, I understand how bad ranked election can be, and that score elections are an option. Now we can consider to implement it.

This definitely will be taken for our meeting. However, I’ll have to study more the other links you put to verify is possible to implement this method to the next election. I accept any help.

About STAR, I have a some questions:

Do you know what are the downsides of STAR?

Is there a way to avoid a tie?

PS.: I answered your other post, but I forgot to comeback to it recently, when we talked about it on the GWG, sorry!

[solstag]

@marinappdf

I have read that ranked election could lead to some unintended results

The thing is: ranked election may lead to somewhat unintended results under some circumstances (actually, a theorem shows this is going to be true for any given voting system), but majority voting leads to way less intended results much more often (as long as there are more than two options to choose from, of course).

[marinappdf]

Yes, I understand. But since I knew nothing about it, I thought was better to keep with something I could explain to people, than make a choice that I wouldn’t know how the explain how it works, neither the contras.
But I hope to know enough about it to be able to explain to people how it works.

[marinappdf]

Hi!

Star voting

I checked Star.voting and I think the “secure and professional” election they provide uses only google forms.

Check this:

STAR Voting has its own web app for running simple and functional polls. It’s very quick and easy to set up, so be sure to give it a try on star.vote. To run an election using paper ballots go to starvoting.us/elections
If you are interested in running a more secure and professional election for free, online, Google Forms is the way to go. Without needing to download anything, Google Forms can be customized and set up and run STAR Voting elections by copying and pasting in a simple script. This post is a step-by-step guide for how to do that.

Maybe I’m missing something, so please help me check this, but it seems is not an open option, not for us.

[marinappdf]

Star Voting

We can use the the simple poll here.
I didn’t find any information about the limitation of this poll.

• I was able to add up to 20 candidates, but it could be more;
• Can be set to be private;
• Can be set to only allowed on vote per IP (but not for person, or for login);
• We can set a custom URL;
• We can set a star and end time.

I didn’t find any indication that is open source.

Condorcet Internet Voting Service (CIVS)

@hpy

Condorcet Internet Voting Service (CIVS)

About this one, I think is a good option!

• Is open source ( here );
• No need of installation, or server, we can just use it ( here );
• No limits for voters (10.000. . . );
• Only authorized voters can participate (we have to add all the e-mails before lunch the election), but we can also create a public election;
• Allows to create the category “unacceptable”, this means, one can vote for people they don’t want to win.

The platform is good, we could use now. However, I’m still confused if the Condorcet method is a good option.

I’ve read here that this method is not good.
I’m also not sure, isn’t the condorcet method a ranked method?
Pen-Yuan give us plenty of reasons NOT to chose a ranked method . . . . .

I understand that ranked method still better than a simple majority. But how will I explain to the community that we choose a method that we know that showed unintended consequence, even thou I don’t know how to explain those consequences? Or prevent them?
I want to chose a more inclusive method, but I believe is better to first learn and understand the goods and bads, before announce anything as official. If we chose this method, I feel I have to understand which are those unintended consequences and etc.

I will follow up with Pen-Yuan reading suggestions, I’ll keep studying about it, and I ask @solstag and @hpy to help me find a good platform and method.
We could make some criterium for comparing platforms.
How about this?

Platform	Open source	Method	Budget	Disadvantage	Trustability	Praticality
X	not sure	Z	free	. . …	. . . . .	. . . .

PS: Also, there are other needs we have to take into account.
We did a quata for social minorities, and this can be tricky to work with some methods (as the ranked method).

Community Council seats will be awarded as follows:

• A minimum of 4 candidates who self-identify as social minorities will be seated, presuming there were 4 such candidates in the pool. If there are less than 4 such candidates receiving votes, all those candidates will be seated.
• Remaining seats will be awarded to the remaining unseated candidates who have received the highest number of votes.
  • An implication of this model is that other than the top 3 vote-getters could be at risk of being bypassed for a seat in favor of a social minority candidate who received less votes.

[solstag]

Ni!

CIVS actually provides modern methods superior to Schulze:

CIVS currently supports five rules for Condorcet completion: Minimax-PM (the default rule), Schulze (also known as Beatpath Winner or Cloneproof Schwartz Sequential Dropping), Maximize Affirmed Majorities (MAM), a deterministic variant of MAM called CIVS Ranked Pairs, and a runoff-based Condorcet algorithm called Condorcet-IRV. The Schulze and MAM rules are described elsewhere (follow the links); Minimax-PM, CIVS Ranked Pairs and Condorcet-IRV are described below.

The full description is here:

https://civs.cs.cornell.edu/rp.html

CIVS with Minmax-PM is pretty good, and we can justify the choice both from the documentation CISV provides and the trustworthiness of the project. Also, don’t forget that explaining a Condorcet Winner is quite simple, and that most complexity concerns only rules for tie-breaking in the absence of one. And the Minmax-PM rule is not that complicated.

Also note that all these issues with strategic voting are not likely to happen in GOSH. Debian has been using Schulze since forever and never had problems with that despite quite heated debates. In practice, healthy communities seem to do fine using any Condorcet method. And unhealthy communities won’t be saved by election methods.

More importantly, CIVS provides proportional representation elections, which is what we want for constituting boards and committees.

.~´

[solstag]

Oh, regarding source code, @marinappdf please note that @hpy provided links to the source code of STAR, CIVS and Helios in his other post.

[hpy]

Wow thanks for the feedback @julianstirling, @marinappdf, and @solstag! I’ll try to respond:

My previous post linked to this thread from the Snowdrift.coop forum (another open source community) that discusses community elections and governance. Just FYI, they have some good discussions there.

@julianstirling: we should set a reasonably high bar on what we define as “good”

I basically agree! I don’t think we need to commission a research project to choose a voting method , and in my opinion none of them are perfect.

@marinappdf:

Is there a way to avoid a tie?

Great question!! STAR is an improvement on pure scoring systems (for example, Loomio supports simple score polls but not STAR). One of the improvements is a way to break ties. The way STAR does it is that if there is a tie, then the option/candidate with the highest number of preferred votes wins. For example, if a voter scored option A higher than option B, then the preferred vote of that ballot goes to option A. This way ties are almost completely avoided. I know people have proposed other tie-breaking methods, but I’m not familiar with them and the https://star.vote website doesn’t implement them… yet.

Do you know what are the downsides of STAR?

Indeed I don’t think there is a perfect method, and there are possible downsides of STAR, too. Here is a sample of the possible cons:

1. STAR is a newer method and haven’t been used in many real life elections. I think the biggest one was a Democratic primary election in one of the American states (which purportedly went well). The concern is that because STAR hasn’t been used that much, we don’t know what we don’t know about what problems might arise. For example, for years I really liked ranked choice voting with instant runoff (and even tried it in a project I worked on) until studies showed that it has serious problems. As far as I know there hasn’t been similar academic studies on STAR yet (though STAR was developed by academics).
2. Apparently, in some cases such as when you know a three-way tie is likely, it is theoretically possible to vote strategically instead of for your true choice. Strategic voting is something all new election methods try to avoid.
3. STAR doesn’t have one benefit of Condorcet voting, which is a guarantee that the winner will definitely win against every other candidate in pairwise comparisons. I should note that while the Condorcet voting method has this guarantee, it has another problem which is the possibility of tied “circular” winners, i.e. there is no winner that wins pairwise against other options. There are many proposed methods for breaking Condorcet ties (see here), but I admit I am not familiar with them. Maybe @solstag knows more?

I checked Star.voting and I think the “secure and professional” election they provide uses only google forms.

Yeah I don’t like the Google option at all and suggest against using it! Fortunately there is https://star.vote which implements STAR voting.

[STAR] Can be set to only allowed on vote per IP (but not for person, or for login);

That’s what I saw, too. However, I think the developers behind the STAR voting website has a new beta version that allows you to do one vote per person. If there’s interest I’m happy to email them to see if the new version can be used and how.

However, I’m still confused if the Condorcet method is a good option.
I’ve read here that this method is not good. I’m also not sure, isn’t the condorcet method a ranked method?
Pen-Yuan give us plenty of reasons NOT to chose a ranked method…

To my knowledge, Condorcet voting is an attempt to improve the most common instant runoff “flavour” of ranked choice voting. In other words, Condorcet also has ranked ballots, but they are counted in a different way. I think the main difference between Condorcet and instant runoff is that Condorcet does the “winner wins in pairwise competitions with every other candidate” thing. Again, the challenge is how to break ties/the circular situation I mentioned earlier. I also have the impression (and this is just an impression) that Condorcet - because it is still a ranked method - is considered by some to only be slightly better than other ranked choice voting systems. Also as I mentioned, giving scores is more expressive and nuanced than rankings. This is why I was suggesting STAR more strongly over Condorcet, but I’m happy to be wrong!

In other words, I think generally speaking: score-based voting (such as STAR and others) > Condorcet > other ranked choice methods >>> simple majority vote

Since we’ve started a discussion about this, I want to note for your reference that approval voting is another system that’s considered much better than simple majority and ranked choice. Approval voting is where the voter chooses only yes and no for each candidate. The biggest difference with simple majority voting is that you are allowed to say yes and no to multiple choices, not just one (theoretically, a voter can say yes to all choices or no to all choices). The winner(s) is/are the option(s) that got the most yes votes. There are advocates for this system, but I didn’t mention it before because I am not aware of any online and open source platforms that support approval voting.

We could make some criterium for comparing platforms.
How about this?

I like this table! I think we can use an EtherCalc spreadsheet for this. What do you think?

PS: Also, there are other needs we have to take into account.
We did a quata for social minorities, and this can be tricky to work with some methods (as the ranked method).

Hmmm, good point. I am not 100% sure which one of the proposed platforms would allow us to implement the quota most easily. We might need to think creatively and maybe even create test polls on the different platforms?

@solstag:

CIVS actually provides modern methods superior to Schulze

The different Condorcet tie-breaking algorithms have always made me dizzy. Is there a good explanation for not just how they work, but what they mean? I’ve also seen that the Debian project has successfully used Condorcet voting for years without major problems (as far as I know), which is a nice show of confidence. @solstag are you familiar with how the tie-breaking methods work and if one or some of them are significantly better than instant runoff voting?

More importantly, CIVS provides proportional representation elections, which is what we want for constituting boards and committees.

Very true and I like that about CIVS. You piqued my curiosity and after some searching I found that STAR also supports multi-winner and proportional representation elections, but I need to check further if the web version allows that.

Oh, regarding source code, @marinappdf please note that @hpy provided links to the source code of STAR, CIVS and Helios in his other post.

Indeed, here are the source code links again to save a click :

CIVS
STAR
Helios (will need to self-host to avoid proprietary login and admin)

[solstag]

Ok, I’ll try to provide some clarification from what I know:

1. In modern terminology “Condorcet” is a criterion, not a method, and it can be asked of any kind of electoral system. As @hpy said, it means that if a candidate would win against every other candidate in a one-to-one election (this is called a Condorcet Winner), then he must win the overall election. Any electoral method that respect this criterion can be called a “Condorcet method”. I think everyone would agree this is an essential property for an election system. Therefore, modern methods that do not respect Condorcet are required to argue that the situations where they violate Condorcet are very unlikely to happen in practice.

2. When there is a candidate that would win a one-to-one election against all others, then all Condorcet methods agree on the result. It means it doesn’t matter what method you use in these situations. However, situations may arise where there is no candidate that would win a one-to-one election against all others. Each electoral method that wants to respect the Condorcet criterion must therefore provide a “tie-breaking” rule. That’s where you get the variety of “Condorcet methods”.

3. Ranked choice methods are methods where people rank candidates according to their preference, often with the possibility of saying placing candidates at the same rank. Among ranked choice methods, “instant run-off voting” does not respect the Condorcet criterion. This has always been known. People decided to use it anyway, because it’s simple. Turned out it was a bad idea.

4. Schulze, Minimax-PM and the other methods supported by CIVS all respect the Condorcet criterion. Therefore, they only differ when there is no Condorcet Winner. The way they differ will determine other properties of the election system, such as other criteria and how vulnerable they may be to different types of strategic voting.

5. Minmax-PM, the default method in CIVS, seems to do very well against strategic voting (from reading the paper CIVS refers to). Minmax simply means that in case there is no Condorcet Winner, then we pick the candidate who is closest to become a Condorcet Winner (i.e., whose biggest loss to another candidate is the smallest - in some precise mathematical sense). I think this is quite easy to grasp.

6. (here’s more opinion that knowledge) From what I’ve read, score systems may be more vulnerable to strategic voting than the best ranked systems. They also fail the Condorcet criterion, though their proponents argue this won’t happen in practice. I personally think score systems are needlessly complicated. Ranking is much easier: once we have considered the choices, we know what we prefer almost intuitively. Scores force you to put down to numbers something that has no actual numeric correspondence in your head (or in reality). They thus demand more investment from voters, may demotivate voters who have less time to vote, and may worsen the overall quality of votes.

7. In approval voting you can’t express preference of one choice over the other. So I feel its application is restricted to a few situations where this makes sense. And like for score voting, I feel these systems are less mature and not sufficiently understood.

8. I know it sounds great to say “oh, let’s leave it open and not fix on one method”, but I actually think picking a single method has very important advantages: People get used to it, they eventually come to familiarize themselves, understand it and trust it. This improves both their disposition to vote, the quality of their votes. It increases the proportion of people who actually understand and can check the results. And decreases the energy spent on discussions (see what is going on here) and mistrust from not understanding a method or not understanding why was it picked for a particular task. Picking one single method reduces the “black box” aspect of modern electoral methods. Honestly, they all do pretty much the same thing. I feel it is much wiser for an organization to choose one method and go with it. Again, see the case of Debian.

Well, hope this is useful =) cheers!

.~´

[marinappdf]

Thank you @solstag and @hpy for your comments. I won’t answer “thank you” and “agree” to all the comments because this will make this post unbearable. So here is my general Thank you.

First, I think is great we are having this debate on the forum. As you probably now, people like to question anything, so I can’t just make a decision based on trust on others. Since I’m a member of the working group, I’ll be asked questions, and if I don’t know how to answer, or if I say “I just picked one”, this will go bad. That’s why is good that you and others give your opinions publicly, and that we make this decision together.

Maybe this is easier for some people who are already used to those terms and etc. That’s my first time ever, and for many people too, so we have to take this into account.

Put this, we can continue the discussion.

From the reading, I prefer STAR method, however the CVIS platform seems better for now.
Even thou Ale says is easy to understand, I don’t agree. Is definitely harder than “sum all scores”, and the algorithms are much harder to follow. Also, I disagree that give scores is harder than preferences, mainly because the scores are the order of the preferences, with the advantage one can give the same score for more than one person.
But this decision is not only about the method, but also about the available platform. On this case, I prefer CVIS.

Summary

We have there options: CIVS, STAR and “approval voting” right?

CIVS

Platform: https://civs.cs.cornell.edu/

• Free and open source easy to set election platform;
• The same method have been used by a big and trustable community (Debian);
• Is known to be good;
• @solstag says the main issues, related to strategic voting, are not likely to be a problem to GOSH;
• The difficulty to an overall understanding is “medium” (my opinion);
• The difficulty to understand the algorithms is “hard” (my opinion).

STAR

Plataform: https://star.vote/

• Free and open source easy to set election platform;
  • HOWEVER, the option available control voting only by IP, not by e-mail (@hpy , can you check if they could implement this, please?)
• This method is new, there are less examples of usage than CIVS, but there haven’t been cases of unintended consequences, or “bad results”;
• Does not fill the Condorcet criterium, so it can’t guarantee that the winner will definitely win against every other candidate
• The difficulty to an overall understanding is “easy” (my opinion);
• The difficulty to understand the algorithms is “easy” (my opinion).

Approval voting

No platform

• Does not fill the Condorcet criterium, so it can’t guarantee that the winner will definitely win against every other candidate;
• The difficulty to an overall understanding is “easy” (my opinion);
• The difficulty to understand the algorithms is “easy” (my opinion).

You want to add something?

If you have to chose one, right now, which one would be?

.
.
.
.
.
PS: I want to keep reading and organizing this sort of information, lets make this ethercalc either way @hpy !

[solstag]

Ni!

@marinappdf

I don’t agree. Is definitely harder than “sum all scores”, and the algorithms are much harder to follow. Also, I disagree that give scores is harder than preferences, mainly because the scores are the order of the preferences, with the advantage one can give the same score for more than one person.

A bit more clarification:

Ranked voting allows you to give the same ranking to multiple choices. It also allows you to not rank some choices and they’l be considered tied in last place.

Again, the algorithms are not important, all one needs to agree with is (1) The Condorcet criterion, which is very much common sense and (2) That in case there is no Condorcet winner, then picking the candidate closest to being one is reasonable.

Also, if you use STAR with scores that are simply the order of preference then you’re just using STAR as a ranking method. And in that case there are better ranking methods.Scores gain power the more they’re not like ranking and let you express nuanced preferences, and that’s when it gets cognitively complicated and heterogeneous from person to person.

Anyway, my opinion between these choices is not strong in the sense that in my view any decent method will work in a healthy community. But it is strong in that we should stick with a single method, for the reasons I outlined, and that we put it in the rules of the community - like Debian put it in their constitution - so nobody should mess with it or create a fuzz unless there is a concrete and serious problem.

.~´

[hpy]

Many thanks @solstag for the fantastic explanation of the Condorcet criterion! After several years it finally makes more sense to me and helped me appreciate Condorcet much more. From the explanation, it sounds like:

1. At least some of the Condorcet tie-breaking algorithms choose the winner that is closest to being the Condorcet winner. So technically this means the final winner chosen by the algorithm is actually not the Condorcet winner, because it is just the closest one, right? I’m not saying this is good or bad, but technically does that mean such a result would not meet the Condorcet criterion?
2. Thanks for defining Condorcet as a criterion rather than a “method” per se. That makes sense. From what I can tell, since the ballots still asks voters to rank choices, can we reasonably call implementations such as CIVS ranked choice voting (just not using the problematic instant runoff method)?
3. Condorcet-based methods sound better than ranked choice with instant run-off. Do you know if there are issues with instant run-off that might still happen under a Condorcet system? Or would Condorcet solve most/all of the problems with instant run-off?

Again, thanks for the great explanation, but that actually leads me to @marinappdf’s point. It took me a looong time to understand Condorcet-based ranked voting methods compared to STAR. I remember when I first read about STAR a few years ago, it literally took me ~10 minutes to understand the whole thing including its algorithm. However, I’ve been reading about Condorcet for at least as many years, and I am only starting to feel like I understand the mechanism better. Even now, I think I can only explain the tie-breaking algorithms on a high conceptual level, but still have no idea how the math actually works. In comparison, I feel comfortable explaining the actual detailed algorithm of STAR (or approval or ranked choice with instant runoff).

Yes, the detailed implementation is abstracted away by the web platforms we are considering , but in a democratic system I think it is very important for voters to be able to understand what’s happening “inside the box” or “under the hood” with relative ease. This would greatly help build trust in the system.

I also agree with @marinappdf that scores are not harder than rankings to work with. Of course, this is just my opinion.

More importantly than that is a problem I’ve seen in real-life cases:

Let’s say there are three candidates Krusty, Homer, and Marge. And assume that I as a voter my favorite is Marge with Homer being a close second. But I really don’t like Krusty. So…

On a ranked ballot, I will have to put (1) Marge, (2) Homer, and (3) Krusty.

On a score ballot (e.g. 0-5 with 5 being highest), I can put Marge=5, Homer=4, and Krusty=0.

In my opinion, this example demonstrates that a score-based system is far more expressive and can capture nuances in opinion where a ranked ballot does not. And I think a good goal is to find a way for democratic participants to most fully express their opinions on the ballot.

Another practical issue that might make scoring easier than ranking is if, for example, I feel about the same for two (or more) candidates. In a score-based system, I can give those candidates either the same or very close scores. But in a ranked system, I am forced to put one option ahead of another, and sometimes that’s super difficult because how do I decide???

Quick note about strategic voting: First of all like @marinappdf suggested, I don’t think strategic voting will be a major problem for this community. That said, I think the single run-off stage of STAR disincentivises strategic voting because your vote will only make a difference if you give candidates different scores. This prevents “bullet voting” where you give your favorite candidate the top score and everyone else 0.

@solstag: I know it sounds great to say “oh, let’s leave it open and not fix on one method”, but I actually think picking a single method has very important advantages: People get used to it, they eventually come to familiarize themselves, understand it and trust it. This improves both their disposition to vote, the quality of their votes.

Great point! However, after thinking about it a bit more I am still inclined to leave this more open to change. A lot of what we’ve been discussing reflects developments in voting reform over the past several years. In another year or two we might have new knowledge and experience that suggest another option is better. I think it is a worthwhile trade-off to re-visit this topic regularly and consider improvements including possibly adopting another system. If we must commit to something (and I’m not saying we must), I would only suggest committing to not using a simple majority vote whenever there are more than two choices.

@marinappdf: I also thought more about your great point about ensuring minority representation and ensuring there are at least 4 on the Community Council. Both the STAR and CIVS websites show you the ranked order of winners after the polls close. So maybe it is easier than we think: Just go down the final list of results, starting from the 1st place winner, and seat candidates as appropriate. Does this make sense?

@marinappdf: If you have to chose one, right now, which one would be?

Thanks to @solstag I now feel much better about Condorcet-ranked (is that gramatically correct?) as implemented by CIVS. However, I am still more inclined towards STAR for the reasons I listed above.

@marinappdf: (@hpy , can you check if they could implement this, please?)

I have sent them an email with the main question: Can “one email one vote” can be used? I also raised concern about STAR possibly not meeting the Condorcet criteria, and if we can ensure the reserved minority seats can be elected with STAR. I really hope they will respond soon.

@marinappdf: PS: I want to keep reading and organizing this sort of information, lets make this ethercalc either way @hpy!

I’ve created an Ethercalc sheet here. Let me know if it works for you. For now I’ve used your assessment of easy vs hard.

Tell me if I missed something!

[hpy]

@solstag I just saw your most recent post right after I made mine. Good food for thought.

I’ll have to log off for today, but will study/reflect on your comment a bit more later.

[solstag]

@hpy Condorcet methods do not need to be complicated, no more than score methods do.

They may get complicated because people want to obtain methods that satisfy some weird criteria and so they have to come up with something equally weird. Score methods (for now) may seem simple, but some are already quite complicated in order to have better properties than others (for example the ‘usual judgement’ score method).

If the first time you saw an explanation of a Condorcet method you were unlucky to pick a complicated one, then it will all become spaghetti in your mind until you see a simpler explanation eheh.

For example, all the Minmax method does for a given candidate A is to calculate, for each other candidate B, on how many ballots A wins over B, and B wins over A, and subtract those two numbers. Then score A with the maximum of these values. The winner is the candidate with the lowest score.

Minmax-PM needs another line of explanation, but that’s it.

Answering specific questions:

At least some of the Condorcet tie-breaking algorithms choose the winner that is closest to being the Condorcet winner. So technically this means the final winner chosen by the algorithm is actually not the Condorcet winner

Not really. I suspect one day you saw some really bad explanation and it stuck in your head . Let’s get back to the basics: not all elections have a Condorcet winner. When that happens, you can’t pick the Condorcet winner because it does not exist! If it exists, all Condorcet methods will choose it, unequivocally. That’s the definition of a Condorcet method: a method that picks the Condorcet winner if it exists. Now, when the Condorcet winner doesn’t exist, then some methods (such as Minmax and its variations) will implement a choice that can be interpreted as “picking the closest to a Condorcet winner”. (in Minmax, by picking the candidate whose worse one-to-one loss is the smallest in terms of difference in votes, i.e., the one where the smallest number of votes need to be changed to make him a Condorcet winner).

can we reasonably call implementations such as CIVS ranked choice voting (just not using the problematic instant runoff method)?

Not only we can, we should. Instant run-off is just one method in the family of ranked choice methods, which includes all methods in CISV and many others. Some people use the term “ranked choice” to refer to IRV, but that is misleading, not to say wrong. IRV is a method, RC is a family of methods.

Condorcet-based methods sound better than ranked choice with instant run-off. Do you know if there are issues with instant run-off that might still happen under a Condorcet system? Or would Condorcet solve most/all of the problems with instant run-off?

All methods have “issues”. There is actually a theorem proving that no ranked voting system can satisfy all properties we would intuitively consider “desirable”, because some properties are mutually exclusive (it may be valid for score system depending on how you reinterpret the properties). On top of that there are issues with strategic voting, and again each voting method covers some risks and exposes others (there’s another theorem for that, which includes both rank and score systems eheh). People who study such systems then analyse and simulate different reality-like scenarios to try to figure out what are the methods with more serious risks of bad behaviour (I guess we’re all aware of that from browsing around). For example, CIVS uses Minmax-PM exactly because one such study compared it to tens of other methods, including score methods, and concluded it behaved better (link on the CISV page).

About the expressiveness of score, it makes some analysis of actual properties of elections a bit harder and approximative, but in my view the main point is that this expressiveness is overestimated. If rejection for an option is actually relevant, based on public understanding and not the whims of a few individuals, it will have an impact on both score and ranked choice. Votes like you suggest (5, 4, 0) actually expose one of the vulnerabilities of score to manipulation, as you can push your votes to the extremes to harm a candidate in favour of another. Also, note that ranked choice does allow one to give the same rank to different candidates.

In any case, I don’t mind if we choose STAR or Minmax-PM. I’m happy with both.

More importantly, I don’t see any reason to believe that a new revolutionary method will appear anytime soon. Election systems have been an endless debate where “best” is the real enemy of “good”. If we don’t fix a single method as the way to do things, we will, without a shadow of a doubt, waste immense and recurring amounts of energy on endless discussions whenever someone new comes up and doesn’t like what we’ve been using lately, or someone secretly feels method A is better for their proposition than method B. I’d rather spend all that energy discussing the substance of issues and not voting procedures.

.~´

EDIT: I’ve fixed some mistakes in this post after publishing it, sorry. Among them, I previously wrote score methods don’t satisfy many criteria, that’s not actually true, though most don’t strictly satisfy Condorcet.

[julianstirling]

@hpy, I think I agree with @solstag that Condorcet is not confusing if it is explained well. It was a term I had not heard until this thread, but I think that the CIVS website does a very good explanation.

I think the Condorcet winner is conceptually very easy. I think the criterion for winning and the the method of voting are so much more important than keeping the algorithm simple. Otherwise first past the post starts looking viable again .

[marinappdf]

Thanks again!

@solstag

They may get complicated because people want to obtain methods that satisfy some weird criteria and so they have to come up with something equally weird. Score methods (for now) may seem simple, but some are already quite complicated in order to have better properties than others (for example the ‘usual judgement’ score method).

I think that’s what probably happened to me. I understand it better now, thank you

@hpy

@marinappdf: I also thought more about your great point about ensuring minority representation and ensuring there are at least 4 on the Community Council. Both the STAR and CIVS websites show you the ranked order of winners after the polls close. So maybe it is easier than we think: Just go down the final list of results, starting from the 1st place winner, and seat candidates as appropriate. Does this make sense?

That’s actually super obvious and will definitely solve the question!! haha

@hpy

I have sent them an email with the main question: Can “one email one vote” can be used? I also raised concern about STAR possibly not meeting the Condorcet criteria, and if we can ensure the reserved minority seats can be elected with STAR. I really hope they will respond soon.

Ok, let’s wait their response!!!

@hpy

I’ve created an Ethercalc sheet here. Let me know if it works for you. For now I’ve used your assessment of easy vs hard.

Yassss!
I’ll check it later!
Thank you!

@solstag

More importantly, I don’t see any reason to believe that a new revolutionary method will appear anytime soon.

I agree with you, even so we should make it possible do change. I’ll propose that changes on the election method can’t be made easily, so we won’t have the problem of one election method per election.

Conclusion

In my understandig we could either chose for the CIVS, or Star. To chose STAR we need to first check if “on email one vote” will be a possibility. If it’s not, than the viable option is CIVS.

Either way, we’ll have to make a simple explanation page/wiki about it. I think I could use most of what you wrote here, if @hpy and @solstag are ok with it

[solstag]

@marinappdf

I agree with you, even so we should make it possible do change. I’ll propose that changes on the election method can’t be made easily, so we won’t have the problem of one election method per election.

Everything in an organization can always be changed Even bylaws can be completely replaced. All it takes is getting people to agree. The only thing we can do is make things harder to change. If we don’t want the election system to change without thorough discussion and the agreement of everybody in the organization, the right way to do that is what Debian did: put the rules in the bylaws (or constitution, or charter of principles, or whatever). That ensures the system can only be changed after everyone has been notified and had a chance to ponder on the reasons and the proposal(s) to change it.

Do note that, in terms of the energy required, even that is not such a big barrier. You only need to publish a page with the proposals, convene the people in an extraordinary assembly (or equivalent), vote on the proposals, and it’s done. Even if it’s an incorporated organization, the only extra work is for the secretary to collect signatures on a paper and submit it to some bureaucracy. That’s all there is.

@marinappdf

Either way, we’ll have to make a simple explanation page/wiki about it. I think I could use most of what you wrote here, if @hpy and @solstag are ok with it

Isn’t the forum CC-BY!? eheheh (But, jokes apart, if it isn’t then that’s something we could discuss at some point).