 Welcome everybody. My name is John Roberts. I'm a board member with the Center for Election Science and welcome to hear this discussion about how to judge voting methods which is a very critical topic. And to have this discussion about how to judge voting methods we're going to have with us Michael Rovinsky who is a board member of the Center for Election Science as well and answering these questions he'll be interviewing our esteemed executive director Aaron Hamlin and so with that I'll just leave it to you Michael take it over. Great thank you so much John. So yeah we're going to have a talk here this will be mostly Aaron probably doing most of the talking and I will be chiming in with things like wait what and we I'll try and keep an eye on the questions and on the chat for questions that come up related to what we're talking about at the moment I may not see that. John will try and keep track of all the questions so that we can have time for questions at the end if we don't get to things as we're going through. We usually have really knowledgeable people on these calls people with who are really interested in voting systems and a lot about the technical aspects of voting systems and then we I think also have people who are not nerds whose interests are more you know maybe more less technical and more social and interested in sort of how do I how do I make things happen in my community so I think we can I don't think we need to shy away from getting really technical in this but if it gets too in the weeds I might Aaron stop you and say something very polite like you know give it to me an English nerd and with that let's let's jump right in I I will start by I'll say that when I talk about voting methods with people I will often say just flat out say you know plurality voting is one of the worst voting methods and approval voting is one of the best voting methods so why like how would you how would how can I get away with saying something like that what are the criteria we would we would start to to be able to judge a voting method on whether it's good or bad sure so I think one starting point that we have on our site already that is nice just to kind of reference in general and I'll put it in the chat window which is an article that we have a lot back called what makes a voting method good and there's just really the kind of the introduction to the concept of okay you have this thing this this tool that measure support takes out takes in information and then provides a result in terms of selecting a winner given some number of candidates so given that's what it does it has all these things that are involved into it and so what that lays out is these factors for how we evaluate a particular voting method and when you take a peek at that you'll see some of them like for instance simplicity ease of administration whether it elects a good winner whether it actually captures candidates support in an accurate way whether the method itself is tractable and then kind of looking at it from a global perspective so those are the kind of kind of the main the main factors that we're looking at here and we can kind of dive into each of those factors individually and perhaps apply them a bit yeah so let's start with with let's start with what's a good winner how do you know if something's if someone's a good winner so that's probably one of the more interesting and challenging ones of the of the bunch so there are a number of ways of looking at that so and one of the challenges is not every voting method has had its day in the sun in terms of being able to conduct elections and so when we're looking at this we can use things like computer simulations we can look at game theory we can look at polling data and to the extent that we can we can look at real-world data so for example like with computer simulations there have been a couple approaches that have been done in the past one was by a mathematician Warren Smith in 2000 he had done a computer simulation comparing different voting methods and then more recently a few years back a former board member independently conducted a similar analysis comparing different voting methods and and here like so you may have like heard the term computer simulation and well like that's that's a bit to kind of unpack there so right how do you simulate what people want and how do you know that that's that is that is reflective of the complexities of real of the real people absolutely and doing computer simulation is is challenging so you have so the way that these are traditionally done in this context is you you create this these electorates so like we call them like agents within a model and you make certain assumptions about them such as like the distribution of their ideology their preferences and then you put you create candidates in the space this political space and they make decisions based on optimizing their own personal utility given whatever voting method that they're using and you can kind of move certain dials some of these styles can be how tactical the electorate is how many candidates there are in the race and other factors and the nice thing about this is you can take a method that hasn't been really used anywhere and you can create elections for it and run these and run these elections millions of times under different scenarios so that's one of the perks of this of this approach what's the advantage what what sort of measurement can you get if a voting method has been used in different places like what what further what what metrics metrics can you get from that so if a voting method has been used in different places so when we're talking about real world data it can still be a bit challenging so for example like say we have just like a traditional choose one voting method election a normal priority election and it's a three candidate race and we suspect something like there's some vote splitting for maybe a third party or independent candidate but like unless we collect actual data it's very it can be challenging like even in something where it's kind of suspect like but we don't really know it can be kind of challenging to look at that election say well what a different voting method have created a better result in this particular scenario because in order to be able to answer those types of questions you need to be collecting more more data so so the mere fact that and another thing to keep in mind that even if we're using another voting method the mere fact that that voting method selects a winner isn't necessarily cause for for celebration we need to be taking this extra information in in order to assess whether a voting method did a good job within a particular scenario and there are some ways that we can do that so for example CS in the past has done polling alongside major elections so we did this in 2016 and we had done this also leading up to the during the democratic primary in this 2020 election and we can run these polls alongside actual elections and and we can do what's called like a within-subject design which is when we ask the same people that we are that are respondents within this poll and we ask them well how would you vote with this method how would you vote with this method how would you vote with this method and in addition to getting that information of like a direct comparison between these respondents on how they would vote between each of these methods on top of that we can also ask them okay for for this question I want you just give me an honest answer regardless of viability how do you feel about this candidate like how much do you want them to win say on a scale of zero to five and we can also ask them to provide that information in a rank order as well using this kind of honest assessment control measure and that's not something that's really common that we haven't really seen that in the past but it's really these tools that are necessary to be able to bring in more information and get a better assessment of these methods so basically it's complicated I imagine that I imagine we could also look at things like what percentage if you have a voting method that's been used for a long time and look at what percentage of people actually vote for example what's a high percentage it might indicate that people are happy you know about the process and going to the polls or something like that yeah voter turnout is interesting but I think it's this is kind of a tricky one because like kind of it gives you this feel good feeling when everybody's coming out to vote at the at the same time however if we think about it in kind of a nerdy statistical point of view so like all the time when scientists go and they do sampling from some larger population they don't need to take a measurement from every individual within that population to be able to get a good prediction of what that population thinks about thinks thinks of as a whole so that the number of people turning out doesn't necessarily indicate a good a good result because you could have say like a terrible voting method where 100% of people turn out but you still get a terrible result there what I think so I'm just looking at the chat and there's questions about sort of like what is a better result what's a terrible result so that's sort of the that's sort of what we're trying to get to here yes about how do you define whether it's a good result or a bad result so when we're talking about this factor of winner selection one of the big factors and like what makes a voting method out of other factors what we're talking about is what voting method maximizes voter happiness to the greatest extent and so like going back to those computer simulations what they can do is they can look at the utility that a particular candidate brings them and then if that particular candidate at win how far off they got and on given all that information which candidate winning brings the maximum utility to that electorate and if a particular voting method didn't select that candidate how far off was it and how much utility was lost so when we're using these computer simulations that's what we're looking at and if we're looking at actual elections and say doing polling alongside it we can do that same kind of thing using these control measures and saying okay well this was the best candidate in terms of having a high utility winner which which candidate actually won what was what was the electorate's utility for that candidate and or happiness for that candidate and how far off was it in terms of who actually won and the the actual winner who would have made uh voters the happiest so uh great so moving on let's maybe move on from sort of voter happiness and what makes a what what what what makes the voters the happiest what what other there are other conditions that there are other factors of elements of a voting system that could you know arguably make it a good voting system that are that are separate from that that's right um such as uh um how well it measures support for example for not just the winner so obviously every voting method produces a winner and you can argue about whether or not that was a good winner or not but you know uh there's also a question of it's very useful to know how much support a particular candidate had how much support you know someone not just someone won but but some sort of measure of like how much do people actually like this person and additionally how much support people have for the other candidates as well could be very useful information for the next election people considering running stuff like that yeah absolutely and that's something I think is kind of overlooked a lot um so so when you're looking at at a race and we could see it like for instance in the 2020 primaries recently um there were a number of candidates uh under this choose one voting method that we're using where it looked like they hardly got any support and it's a reasonable question to look at that and say like really really like did they were were they really uh that unpopular or was it the voting method that just wasn't capturing their support uh and when we look at this uh through kind of uh more sophisticated data and actually taking like that data to begin with uh we can answer that question we can take this control measure of uh that asks people uh how much they would like to see a particular candidate be elected regardless of the viability factor we can take that measure and then superimpose that on top of the actual reflection of support that a voting method got and we can see what that discrepancy is so we see what the voting method measured and we can see what this control measure looks at and says well this is the actual uh support and look at that discrepancy and when we do that with different voting methods we can see um like varying degrees of discrepancy so like approval voting uh for instance not only does it do a good job in terms of maximizing better happiness uh it also does a good job gauging that support as well uh the same thing with a lot of uh rating systems like range voting uh or virtually any voting method that uses a type of range voting instrument when you're scoring candidates on the scale of say zero to five there's also tend to do a good job in terms of capturing candidate support in addition to getting a good winner on top of that um interestingly there are some methods some ranking methods in particular um can find this challenging so for instance if uh we look at uh ranked choice voting or instant runoff voting for instance it does uh an okay job uh at electing a good one to kind of uh maybe leaning negative to the so-so um in terms of uh of getting a winner that maximizes happiness on average uh but it does actually does much worse than that on uh capturing candidate support for instance uh and uh one you know who won you know who won but you don't necessarily know why and you don't know if that means that everyone really likes them or if just they're just the best of a bad lot so so the and in this uh scope the the struggle is being able to figure out the measure of support or the amount of support that other candidates got who didn't win for instance um so uh say like we're looking down the line uh with the democratic uh primary and we're looking at at at folks like um uh uh like say Andrew Yang or other candidates that had like their own hand uh show of support uh but they didn't show up very well uh when we look at these uh same polls like including other organizations that have done done these polls and we look at how these candidates do under same ranked choice voting even if we look at it in the um best possible light for that particular voting method uh which would be looking at the candidate right before they were eliminated so break trace voting it simulates a sequential runoff of rounds and so we could look at how much a candidate support how much support a candidate got right before they were eliminated which is like their maximum support for that voting method even if we look at it with that best possible light we see a really large discrepancy between uh what that candidate got who lost but still needs sort of support reflected how much that that candidate got in support compared to say a control measure and we see a pretty wide discrepancy there which tells us that this voting method isn't doing a very good job of capturing and using the information in an intelligent way to be able to tell us how much uh voters support these candidates it's very difficult to look at ranked choice voting results and have a sense of what happened i mean it's quite complicated to just sort of look at all those results there's not there's not sort of a clear metric you can point to for a candidate to say this is this this this is an easy an easy representation of something one of the things that we had done with the democratic primary results was and this was also a way to try to be able to make direct comparisons which it could be a little challenging because you have these different voting methods they take different types of information in and but you want to be able to make a direct comparison so you try to make the the metric that you're using as parallel as possible for these comparisons and when you look at the figures that we had done there comparisons you're talking about comparisons of support between different voting methods so so like and with interest voting you have ranking information approval voting you have these approvals you have other type of cardinal information with scores with other voting methods and you're trying to take these and put them together in a way that you can make comparisons between these so you so you can see okay well is this how much support this candidate got under this method compared to this method over here if you're comparing apples and oranges it makes it a little bit more challenging so we want to try to create the same kind of playing area or metric as possible and the way that we did that with the democratic primary was we looked at how much support that they had with these like the first choice rankings for them before they were eliminated but at the same time after they got all the transfers as possible before they were limited to like their maximum support when you look at a tabulation table for ranked choice voting the number that we would show for a particular candidate like that would be their max support and so that's the way that we were able to compare these different voting methods with support and even like with the max support like these candidates were really missing out a lot of support compared to these control measures with that particular voting method and it wasn't necessarily the same with other voting methods like approval voting did a great job in terms of measuring that support as well as other scoring methods of course our choose one voting method did its normal terrible job in terms of capturing that support right um encouraging people to participate like encouraging encouraging candidates to jump in the race if you don't like how things are going and you want to you want to fix things and you've got an idea it seems like that's a good that's a good measure how how how easy it is for someone to to jump in and and participate without being a fear for example afraid of being a spoiler yeah yeah so I think there are a couple factors that go into play here what is that that's that spoiler dilemma so it's not fun for anybody to really fulfill their democratic duty and then get thanked for it by being yelled at for for being a spoiler and causing a worse candidate overall to win nobody wants to go through that as a candidate and also as a voter like no one wants to be shamed by their their inner circle for voting honestly and then being told that they were a schmuck or being called names for trying to vote for the person that they felt aligned with them best and so um having a voting method that doesn't create that kind of dynamic is helpful for encouraging new candidates to run because then they don't have to deal with all that uh the second component is in terms of encouraging candidates candidates to run it's getting that reflection of support because if if you run even if you're not getting shamed uh if you run and you're getting like two percent three percent it's uh it's a bit demoralizing and a candidate as a candidate and isn't going to encourage you very much to uh to run again or to keep your campaign going and it's not going to encourage other candidates later to take up your cause that's exactly right so so it may be like uh every candidate that that runs isn't going to win even candidates do well like they don't always win every time but if you come into the to the playing field you've got good ideas and you have a voting method which is able to capture that support or that enthusiasm from the electorate and getting behind your ideas uh then your ideas can get some traction they can get co-opted in a way that just wouldn't be able to happen uh without that reflection of support it seems to me like uh scalability may also be an important factor like is there a can we reach a point where there's too many people in the race or do we want a system that that is flexible enough to allow anybody to jump in and sort of handle a large number of candidates in terms of the number of candidates in terms of the uh the number of candidates one of the things to think about is the interaction with ballot access laws um so in general we went with that uh the the barrier to entry to be uh kind of low uh one of the things that's nice with approval voting is that it works well with the number with a large number of candidates uh in a race say there are uh we're just using extremes here there are like 25 candidates within a particular race uh which isn't totally outlandish given the the primary so we've had in in the past in california in our primary at our last senate senate election in the in the open primary we had 37 37 people running yeah one yeah uh the the nice thing with approval voting is that if you have a long list of of uh candidates there it's not terribly uh more terribly like uh demanding mentally in order to go through that that that list with approval voting compared to say the way we're normally used to going about uh evaluating candidates um but like when we're thinking about the kind of and this gets into a little bit of the uh simplicity or complexity factor which is looking at a candidate list and how challenging is it for the voter to be able to use this voting method and assess a long candidate list so it's it's it's uh I I agree one of the things I love about approval voting is that it is scalable to any number of candidates like if they do is just go down the list once and say yes yes no no no yes no no as opposed to okay what's my first choice okay now what's my second choice okay now what's my and if you have like 37 people running um it's not so simple that's right yeah particularly so uh when we're thinking about different ways of doing it um there there are scoring uh methods that have you score each candidate um that's uh it does provide more information but there is more cognitive effort there uh as you're looking at each candidate in terms of uh figuring out like how to evaluate each individual one and say like you have a list of like 20 plus candidates and it can become even more challenging with a ranking for instance so one thing to keep in mind is you don't necessarily have to provide an assessment of each candidate with these other voting methods so some some places even truncate their ballot so that you can only provide a certain number of rankings even because by allowing you to rank all of them it would make their their ballot a mess but still like even if we're doing that what we're by say voters just saying like oh that's too much I'm not going to do that or we're truncating the ballot so we're not allowing them to provide the information uh we're losing out on the information that would be able to help us optimize the outcome as well as be able to uh capture that support for these uh for these candidates so it's kind of like well like do do we take a method that is as a complicated element in terms of how it allows voters to express the information do we go ahead and let them express all that information anyway or do we try to simplify it but at the cost of simplifying it lose out all this information possibly get a worse result or um or and uh not be able to capture the support for all these candidates even the ones who don't win so that we're not able to say see whether they have good ideas or help them get the information that they need as they're deciding whether to run again in the future or to try to build their party or or something like that so you want to use a voting system that's scalable that allows for for whatever it's going to encounter so that you can you can you can you can continue using it without making it worse by truncating the information just to make it simpler yeah that's right we've talked about uh a little bit about sort of being easy to understand I think is a is a is a quality you've talked about um what else uh I'm thinking ease of implementation perhaps well with the kind of the ease of of understanding and we could talk about the uh implementation but with with the ease like the the simplicity component as well as the um the complexity component it really has two parts there one part is that expression element and then the other part is understanding what's going in going on in that black box in terms of how that calculation is taking place so we can talk about like that part within the complexity or we could talk about the implementation I'll let you take the lead in terms of where you like to steer us um uh cover cover cover before implementation the implementation sorry I you said the implementation no the ease of understanding okay um so with the ease of understanding um this is keeping in mind like the other part of the voting method so we have that expression element all kinds of ways of expressing information with the voting method choosing one choosing as many as you want ranking and then using that ranking to simulate pairwise comparisons simulate sequential runoffs all kinds of things you can do that and then with the scoring information you could like take an average take a median you can mix and match and use certain some of the scoring elements and then use those scoring elements to imply rankings and do different types of pairwise comparisons there all kinds so what I'm getting at there is all kinds of complicated things that you can do with the information that you're providing so um when we think about the component of what you're doing with that information so with the normal way that we think about uh voting uh with just choosing one one candidate there we're just talking about simple addition we're just adding things up and the same thing with approval voting so you're not uh adding any complexity to the calculation element with approval voting you're just adding things up with say range voting when you're scoring each candidate on a scale say zero to five same thing you're just adding things up you can also take the average one tiny bit of complexity there is that if someone doesn't score a particular candidate what do you do with that the simple approach there there are different approaches that you can take but one simple approach is just considering that candidate as being scored is zero that's kind of the simplest approach that um with uh uh ranking methods say like condorsay methods uh with that one you're taking these rankings and then you're saying okay well this candidate was ranked higher than this candidate and this head-to-head pairing on this ballot uh this candidate wins you do this aggregation with all the ballots to be able to simulate these head-to-head elections or head-to-head um yeah elections between each candidate and seeing which candidate is able to win uh more head-to-head elections than any other candidate uh and that's and the candidate who's able to beat everyone head-to-head is called your condorsay winner uh and so condorsay methods uh address it that way and then you've got like these tiebreaker elements within condorsay methods which adds this other layer of complexity you get into things like uh uh Schultz smith sets all these other complicated things not even talking about the English there so so so the what I mean when as you as you as you talk about all these things so what I what I what I'm sort of thinking is well so there's a question of how detailed you can get and how much information you can grab and how specific and detailed you can be about measuring what the voters actually want but there's probably also some sort of uh important elements of how how happy how how comfortable is an average voter with what's happening and how well do they understand it and and you know because it's one thing to just you don't really want a system where where the measures will be going really well and ask them really really complex questions and even if they're willing to willing to sort of say that they might still come away feeling like well you're telling me this is this is who the winner is and you're telling me this is what I wanted but you know do I we we want a system where you don't have to trust the people who are doing it necessarily to to to do it right you know you don't want a system that just says yeah trust us it's too complicated for you to understand that that's right particularly like when we're thinking about like and kind of this stage and history in terms of where alternative voting methods have have come uh and what people are used to in terms of the of the of the complexity there's also an idea of diminishing returns with the added complexity that you're adding so like we've talked about some of these other factors like how good of a job of voting method does an electing winner and how like how much that satisfies the electorate and then there's also that capturing of support with with different voting methods and sometimes sometimes but not always adding a little bit of complexity to a voting method will be able to help on those other components sometimes however by adding additional complexity it may technically improve those other factors but it does so at a high cost and so you have to balance okay well how much like people not participating people people giving up people not showing up people not filling up whole ballot a high cost of complexity so so if say you want to improve these other the other measures but like even with say like approval voting and something like range voting you're already doing really well overall in terms of how these are performing and selecting winner and measuring candidate support but if you want to try to add to make that even better like you start to add extra layers of complexity and you have to start to think back and think okay well am i really getting my return on investment for the complexity that i'm adding or is what i'm doing here uh already doing good enough in terms of the goals and also thinking like these things like if we're talking about it in the real world we want to get these things implemented and we want to be able to do so with the high success rate and be able to have it to be able to gain the attraction necessary to spread over to other places so this isn't just like a one-trick pony for a particular community we want something that can that can move and it's easy to adapt but when you start adding these extra complexity measures you start to run into this kind of questions like well like am i really getting the return on investment for this complexity that i'm adding here compared to using this really simple approach that already does really well overall i imagine it's complicated by the fact that people are different in the amount of complexity they're comfortable with that's right the amount of complexity they think is worth it yeah that's that's right um we've got so it's uh it's almost 440 we probably should go to questions maybe in like five or ten minutes uh sure there's some is not not quite yet though john is erin is there's is there's other stuff that you wanted you thought we should cover about um things you want to voting method to do or possibly things you don't want to voting method to do um so maybe some other components just to kind of just kind of to touch on uh one is like ease of administration with the particular voting method so um and this kind of gets in with uh tractability a little bit as well and here we're thinking about things like does it need special software um uh does it require complex dot design um can you does it need extra components like do you need to have all the ballot data centralized and then is there a good method for doing uh what's called risk limiting auditing which is being able to take a sample of ballot data and being able to uh look at that sample and see whether um it was off for the official tally and you can um you do that statistically figuring out okay well what are the odds that we would get this result uh from this particular sample overall so so there are there are ways to you're talking about that's a way of uh checking that the results are probably uh weren't tampered with that's right uh and the voting method different voting methods have different protocols for uh risk limiting auditing and so one of the things we when we're looking at different voting methods it's helpful to see whether there's a protocol in place that's workable how much of like how difficult it is to actually implement it when we're thinking about the voting method because the voting method interacts with all these elements and somebody's got to run these things at the end of the day as well so we just have to keep in mind can you give an example for a particular voting system of an audit system whether it's a simple one or a not simple one sure so the current way that we uh though it's pretty simple um so we're just choosing as many as you want um there uh you would take uh in the current you mean choosing one correct yeah choosing one candidate uh there you would take a sample of ballots say you take a thousand of them uh and you look at them and you say okay um uh this is a breakdown of candidates that we got uh we can do some statistical analysis and look at uh confidence intervals for where we would expect the uh the true number to be in given the sample and if the actual tally is within that range we can look at and say yeah it seems like it's within reason if it's outside of that that range uh we can say well like I don't know we need to keep and you just keep sampling until possibly you sample and go through the entire uh uh ballot data okay that's a that's a simple way of auditing that's right and that would be the same for approval voting that would also be the same for approval voting uh you do have some more sophisticated approaches that you would have to do with ranking methods like rank choice voting although there is a protocol in place for risk limiting auditing for rank choice voting although it's much more challenging okay is it more challenging for range voting as well uh I'm I'm not aware of there being an explicit uh methodology in place for risk limiting auditing for range voting uh but given the type of uh tally that takes place there uh I would imagine it's somewhere in complexity between like approval voting and uh say um rank choice voting I I can't imagine it being as complicated or more complicated than rank choice voting giving the type of tallying that you have involved do you want to touch a little bit on um and we don't have to but about uh ways of ensuring that a governing body is the diversity of the voting population is reflected in the diversity of the governing body uh sure so which I don't know if that's necessarily a subject for a single winner which we're talking about single winner elections here yeah so so it it wouldn't be it would be uh so uh if you want to make sure that a particular uh uh electorate is represented within a governing body uh you're pretty much forced to use a multi-winner election uh that uses a proportional voting method the it's really without it's really outside of the abilities of a single winner method to be able to accomplish that when you have a single winner method um your best result is going to be getting a particular candidate that maximizes the happiness of everyone if we imagine say like a normal distribution within the electorate uh it would be that candidate like right in in the middle uh but it wouldn't be someone to say even if we imagine different types of distributions it wouldn't be um the candidate would be unlikely to appease people more out on the ranges you're only getting one person you can't you can't get you can't get someone who represents everybody when you only have to one minute maybe you get someone with like a split personality disorder dissociative identity disorder i i don't know but yeah you're pretty much stuck if you're looking at single winner is there anything i've missed that you wanted to touch on um let me see here uh so perhaps one other component would be tractability uh so there we might look at um what is what is tractability so tractability looking at the likelihood of a voting method to be able to uh to catch on uh given an attempt within a particular governmental election so there we might look at things like credibility so whether it has like an academic underpinning uh to it to so we could see whether it's uh certain aspects of it are predictable um what its track record has been in terms of passing with uh with ballot initiatives uh whether there are practicality components with feasibility which uh interacts to some extent with administration um so in terms of practicality but that that's one component as well in terms of thinking about um whether it makes sense to go with the particular voting method is whether it's tractable uh and practice so whether or not something will catch on but uh is that actually is that actually a quality is that actually a useful quality for a good voting method is whether or not it would catch on because i suppose a bad voting method could also catch on easy to catch on that's right yeah uh so uh for example uh an example of an excellently a voting method is excellently in tractability is our current choose one voting method uh it's done a great job uh to use everywhere and it's caught on real well um and so uh but that that also goes to say that uh just because of voting method does really well in any one particular factor um or even multiple factors uh we have to look at these factors as a whole uh in order to be able to assess uh like particularly from an organizational standpoint of uh where do we put resources behind because ultimately at the end of the day resources are limited uh and uh we may do ourselves a disservice by saying like splitting the vote with uh allocating in kind of multiple polls um good question i see one question let's just end with this and then we go we'll go to more questions from john um common weaver says is there is there literally anything good about first pass the post other than that we are used to it uh yeah yeah so we can always yeah there you go it's uh very very easy uh easy to fill out even with a long candidate list um it's uh easy to calculate use add those numbers up uh easy to administer like they they do it all the time uh although there are some unforced errors in terms of ballot design uh but that's not the voting method fault uh so uh very tractable as we uh mentioned uh mentioned earlier uh it uh kind of bungles the job completely in terms of selecting good winners uh as well as measuring candidate support uh and really uh and that's interesting too because you can see like here's an example of a voting method it was really well on certain factors uh super easy uh it's shown to be tractable very easy to administer uh but like boy is it terrible at choosing a good winner uh and particularly in any kind of like moderate to a heavily complicated election and also jeez is it really doing a bad job in terms of capturing candidate support unless unless you can make sure that there are only ever two candidates in which case it works perfectly that's right and uh there there are some uh uh what we'll call them smart legislators who have uh uh figured that out by making dot access laws really challenging uh so that it in some jurisdictions uh you have a de facto uh two two candidate race if a candidate doesn't run um how do you know that uh i'm trying to think of some sort of like a tree falls in the forest and no one's there to hear it if you if you if it's outside of the voting method if it's outside of the system if you're presented with two choices and you vote for one and one of the wins how do you know there's anything wrong if there's if you don't you don't know who didn't run that and that's uh although we're talking about that kind of uh you're bringing that kind of rhetorically like that does create a real history because someone may may look at their elections and say like you know what you keep talking about these complicated elections when you have all these candidates run we never have other candidates run uh and then never thinking about the causality of that like do you ever think maybe why you don't have other candidates run maybe it's because you have a voting method that disincentivizes candidates run in the first place cool um i think we could go to questions john do you do you want to take it sure yeah here's here's a question about metrics um aran uh someone asked what about the distribution of voter happiness in other words uh what if aggregate voter happiness by some metric like voter satisfaction or whatever let's say that a single number that voter satisfaction could be quite high but we wind up with one segment of the population that's really really much less happy with the result would you say that that's a good result and how does that impact voting methods yeah so it's uh so you can do modeling with different types of distribution so a lot of models use a normal distribution you can assume a lot of polarity and say uh do these models over a bipolar distribution uh with like kind of two humps uh but even even with that like at the end of the day you got to pick somebody uh and in terms of uh getting a good winner um like if you have uh two uh like a bimodal distribution there um they're like technically the candidate who is best for everyone would be someone in in the middle uh so like you're kind of forcing everyone to take a middle ground uh so like even with different types of distribution like that candidate in the middle uh you can if it's kind of a skewed distribution like you'd have to take that into account but like normally that's what we're talking about um so even in those cases uh you have to think just at the end of the day uh a single winner uh method is just not capable of appeasing everyone and so as a consequence uh you uh the maximizing voter happiness is really the the best that you can do so that's an argument for proportional representation for having yeah combining districts into multiple districts so that the sort of the smaller passionate group has a better chance of having at least one person in their representation yeah that that's right uh so there are some um circumstances when you have you're just kind of stuck you you inherently have a single winner election like mayoral governor races that kind of thing but when you can have a multi-member election and you have a multi-person body that's when you can use a proportional method and there you can acknowledge that okay uh we have a diversity of people within our electorate lots of different ideas if we want to be able to acknowledge them um then we have to use a multi-member system uh that uh uh elects people in a proportionate way in order for uh multiple views within the electorate to be acknowledged okay um here's another question about rank choice uh uh easier maybe a different question but uh why does rank choice do a poor job of capturing accurate support for all the candidates how common are rank choice voting weaknesses so do they occur once every hundred times or once in a thousand times so there are kind of two questions mixed in here like one is like the anomalies that you see within my choice voting in terms of selecting winner and then another part is that capturing support how common is it that it misses up in terms of capturing support uh with the first part it depends a bit on uh the complexity of the race um so one of the things that and here I try to I try to throw methods abound like as much as I can uh so there was a race in in Maine where a republican in belief his name is polkman um got really upset uh that rank choice voting didn't elect him even though he had the plurality of votes in the first round you had an independent in that election uh and that independence votes her votes transferred over to the the democratic candidate in that race um and when you look at it from that perspective you have kind of a small time spoiler there didn't get much support and you have a pretty simple transfer over there in those cases it really does make sense like it looks like right first voting given the information that we have it looks like it did a good job there and so when you have these kind of small time spoilers it seems to do an adequate job there the challenge there becomes when you have a more complicated race you have a bunch of people in the election and it's kind of close uh and those um it can not necessarily always but you can see these issues where um you uh you get votes putting in such a way where a candidate has broader support but because they're not getting as many first choice votes within a particular round it can unfairly knock them out and as a consequence you're eliminating really solid candidates that are ultimately better than the candidate who wins um and so uh and then there's the other part which is the um the measuring of support and that's something that actually ranked choice voting disconsistently poorly uh and we can look at this using uh research that we've done first hand and you can also look at uh rank choice voting advocates so when you look at their polling um you can see the reflection support for uh a longer candidate list particularly in the democratic primaries when you look at those you can you can see that if you look past like to say like the first uh three uh candidates or so you see these other candidates they get this like hardly any amount of support and you it doesn't look terribly far away oftentimes from our current choose one voting method of of uh of voting in terms of the support that they capture and there are some reasons for this one is that uh well there are two components one is something inherent with the uh measurement instrument which is ranking so ranking and statistics they call this an ordinal measurement and what that is is that when you have uh say first second third the distance in uh an amount between first and second isn't necessarily the same distance between second and third and and these other uh these other intervals and as a consequence it's uh you're kind of getting some sloppiness within that measurement and you also don't have a clear threshold in terms of where that order of rankings is where you're seeing satisfaction so like are you satisfied third fourth like maybe uh but you just don't know that that measurement instrument doesn't capture that utility component okay and it's like one one more part one more part in terms of uh the uh reflection of support under rank choice voting aside from like the the raw measurement is the calculation component and that's where uh and the reason this uh rank choice voting loses so much data within the calculation component is that um it's only looking at first choice preferences at any one point in time so even though as a voter you could have ranked all the candidates all the voters could have ranked all of the candidates so you have all this data there um but if you're looking at uh candidates who make it later uh in the rounds uh and you can see that they've ranked other candidates uh well if those other candidates that they've ranked have already been eliminated even though the voting method captures that information it's never been reflected in the tally at all and as a consequence you're never seeing that show up so that's the more complete answer to your to your question sorry about that john that's okay um here's another quote you mentioned the spoiler effect uh with respect to rank choice voting but more generally about metrics for measuring the quality of a voting method there was a question which method best eliminates the spoiler effect can you imagine what what's a metric that would measure uh to what degree there is a spoiler or the risk of a spoiler in an election yeah so uh the so the spoiler effect is one just for clarity is when a candidate runs uh they don't win and yet they cause someone else to win who uh wouldn't have won otherwise if not for their presence in the race that's kind of a a complicated but uh explicit definition it also is kind of has has an overlap with another criterion called independence of irrelevant alternatives which is an irrelevant person coming in changing the outcome of their race so there's a lot of overlap between those two so uh there like cardinal methods tend to do a good job there overall but in terms of like a particular voting method doing well there like cardinal methods do well and there like if say like uh you're getting like 98% of these taken care of the spoiler effect taken care of it's really thinking of like okay well for this extra little bit at the end how much complexity do I want to introduce there uh to to deal with that little sliver at the end so that's one of those things where you think okay well like how much complexity do I want to be adding here for this little a little bit of of utility and then also thinking like how much do we want to get hung up on individual components because like there are all kinds of different criteria to use and say okay well here's a particular criterion that I think is important and then thinking okay well how does that play into it overall whereas like we really want to be thinking about maximizing voter happiness for instance in terms of selecting a winner if at the end of the day for that particular component that's what we're thinking about yeah there's talking about maximizing uh overall hat voter happiness there are some very technical mathematical questions here regarding what the best measure of aggregate happiness should be should it be an arithmetic average some arithmetic mean utility or should there be a geometric average utility like using a geometric mean uh to introduce possibly more stability uh into uh the the choice of a voting method do you have any thoughts on this uh well I mean when you're doing a modeling when you're doing a model and you're trying to think okay well like what dirty method does the best like generally you're thinking about what the utility is for a particular voting method on average um and if you're using something like a geometric mean like you it deals a bit better with like more extremes within a data set that you may have but like these are like there are different ways of doing this so that we have two models that we look at but it's not to say that these are the only ways of doing these these models didn't overlap completely there were differences in terms of what they they got the same kind of general result overall but they weren't exactly the same and what that says is that within a model there are assumptions within that model there are different ways of calculating these metrics so uh uh it's unlikely to get too much of a different result even if you're saying using a geometric mean compared to an arithmetic uh with the particular mean um but like there are different ways of doing it I don't know that you get a big difference in terms of results okay uh another quick question someone asked is a two-party system evidence of bad results now uh related to that the spoiler effect so we have Duverger's law this two-party uh effect um is susceptibility to Duverger's law having only two-party system is that the same thing as measuring just the spoiler effect so to what degree is a two-party system evidence of a bad outcome of an election there's overlap there between Duverger's law and the concept of the spoiler effect um so Duverger's law um he's he's he's just saying that the voting method that you use can be a predictor for the number of parties that that carry out and there's the same predictor so it's not it's not an absolute rule despite the misnomer of his law being called a law but the the factors that he's looking at are twofold one is the threshold needed to get a particular candidate elected and within single winner methods like it's whoever has the most support within a particular pool of candidates and so you can't change that like single winner methods like you need to go to multi-winner methods where using proportional methods in order to lower the threshold for victory for that factor and then the second factor for predicting the number of parties is whether you can honestly support your honest favorite and so you're not worrying about throwing away your vote candidates so new ideas can come into play and there's candidates who are bringing new ideas to the table can get their support reflected and so some single winner methods do are able to address that particularly approval voting the one that we advocate for does a does a good job at addressing that component and letting you always support your honest favorite candidate and so what that says is like for example with approval voting it's not able to address one factor at all just by nature of being a single winner method you have to go to multi-winner methods using proportional methods to do with that for that second factor however it is able to address that which is always letting you be able to support your your favorite candidate so what that says at the end of the day is that for a single winner approval voting method you'll probably get more parties if the electorate wanted them but probably not as many as you would with say a proportional method okay maybe the last question here because getting late but do you think by the way if anyone has more questions I guess you can contact Aaron and Aaron's email is Aaron at electionscience.org is that correct Aaron that's right right so I'll just put that in the chat uh Aaron's our executive director and um so a last question would be do you think that moving forward with better voting methods is that prime how do we get there is it primarily a matter of educating the public or do you expect significant pushback from powerful people against a more dynamic democracy yeah I think that lays out the the challenge of of where we are absolutely like that that means educating folks because right now most people when you tell them use the word voting method they're going to assume you're you're talking about this choose one method that we're using now most people it's not even on their radar that the voting method is a thing and if they are familiar with the term they are likely not to be aware of very many of their other options in terms of pushback with what when we look at say like kind of thinking a bit back of the tractability component so far with approval voting we've done well although with a sample of one in terms of ballot initiatives we've had we had minimal pushback over there and some other elections there's been a little bit more resistance often circling around complexity as a rhetorical device for pushing back against a particular voting method after it's already been initiated and then pushing for its repeal and so as we're talking about kind of pushback one of the ways that we're thinking about is trying to not provide as much ammunition or being able to kind of disarm folks who would push away from approving different voting methods and one of the ways that we can practically address this is one by selecting a method that not only does a good job in these important metrics but it's also so simple that it makes it easier in terms of tractability so that it's difficult to use these kind of simple rhetorical devices and say like oh like nobody understands that so that's that's one way that we do with that okay well um I just want to thank everyone for a very productive discussion in chat the center for election science we've got a discord we have a chat going on at our we have a discord server and that's been posted in this chat you can contact Aaron I've posted in the chat also Aaron's email address my email address is johnjon at electionscience.org but in general I just want to thank everyone for participating please keep the chat going keep in on our at discord and thank you for promoting and better election methods please stay safe and healthy and we look forward to staying in touch in touch with you thank you very much thanks Aaron do you want to before we sign off yeah sure go ahead Aaron so there may be a couple things one is we've got an article series coming out it wasn't originally going to be a series but then it turned into 27 pages and then now it is a series um and so there's a kind of an assessment article that we have going on that compares different voting methods and using factors and applying them to different voting methods so if you've got particular questions you may want to hold it onto those questions until after reading that that series that's coming out um and then the other component is and I'll let maybe Michael take that one which is about giving to the center for election science to make sure that we can continue our work give it to the center for election science so that we can continue doing our work um we've we've made great strides in the last uh in the last 10 years um 10 years especially it's going on 10 yeah especially in the last couple years and we we had a great success in Fargo and we're very excited about st louis um it was we'll be voting on whether or not to adopt approval voting whether or not they're going to become the second city in the country to uh use approval voting in probably november yes uh so st louis is going to be um voting on whether to implement approval voting this november and then we've got Fargo is doing a two person commission race using a block problem of approval voting next month that'll be uh next month um and it it looks like someone's also asking when the series is coming up and how they can find it uh so uh we have a newsletter that you should totally sign up for uh you can sign up for that right over our site um and then if you look on the uh part of our website that uh says news uh you'll be able to find out all of our latest articles too okay um well again thank you all and uh keep keep up the good work and uh stay stay happy and healthy thank you very much for everyone's attention here take care