Accuracy and Goals for Proportional Representation and Multi Winner Elections.
Multi-Winner Elections are a way to elect a council or team from a group of Candidates. If each seat was elected one-by-one the majority would always win and the council wouldn't do a good job of representing the diverse electorate.
Proportional Representation (PR) was envisioned as an ideal that the governing body should match the demographics of the electorate as closely as possible. There are a lot of proportional representation based voting systems that have been and are used around the world but there doesn't seem to be a lot of science around how to look at which system is best and how to measure if a given election is accurate.
We here at RCV Oregon are committed to working for accurate election reforms that better represent the people. A lot of us are here to solve the spoiler effect and allow honest voting. We've agreed on the 6 criteria: Honesty, Equality, Accuracy, Simplicity, Expressiveness, and Viability but for PR systems there hasn't been any clarity around Accuracy in particular. Who should win?
If we are looking at a 1-Person-1-Vote system then it's pretty simple to say that for example if there are 4 seats you need a 1/4 of the votes to get a seat. If the numbers don't work out that perfectly we can choose to round up and down so the needed threshold would be half way between: 13-37%=1 seat. 38-62%=2 seats. 63-87%=3seats. 88+%=4seats. The other way to do it is to say that you need the full 25% to get one seat. (I think VSE would prefer the former.) It's critical to decide what a needed quota is to be able to win each seat so that when we are actually looking at elections we can tell if a candidate won fair and square or if they are a non-representative winner who shouldn't have won.
Once we've decided what is a fair quota for a 1-Person-1-Vote system we can use that same quota for more expressive ranked and score ballot systems.
Currently we have been looking at Ranked Systems like IRV-PR aka STV, Re-weighted Range Voting, SRV-PR, and also more conventional district based Multi-Winner systems or some combination of PR and district based elections.
Note: I edited the word "spoiler" to other vocab that means basically when the wrong winner wins. Apparently a "spoiler" is specifically just vote splitting.
Aaron Wolf Wed 29 Mar 2017 4:37PM
My goal itself is not about average or median distance. It's to minimize the number of voters whose distance is so great that they feel excluded from the process. But there's ways that a greater distance could still be felt as having some inclusion (or not) based on the subjective perspective of those stakeholders.
Trying to pin it to some measurable thing, it's something like a ratio between the farthest distance and the median or mean difference for that representative.
In other words, if I am more on the edge, I will still feel included more if the representative close to me (who isn't that close) is also pretty far from their other constituents. Thus, even though they aren't just like me, I'm not so far relative to the other constituents that the representative ignores me. So, a centrist representative in a polarized world where few constituents are centrists is one where the polarized constituents can feel that they all have influence and consideration from that centrist representative. I suppose it also relates to the distribution because a centrist with a much larger set of constituents on one side will veer toward that side, leaving the other side feeling excluded.
So, it's something like the sense that whatever distance I have to my nearest representative, I have adequate influence on them. So, the goal is to minimize the number of constituents whose closest representatives are too strongly pulled away from them in terms of the set of constituents around that representative.
A few extreme outliers may necessarily end up excluded, but the goal is to minimize that number. And maybe we can pin down the issue of exclusion as this quality of some measure by which the closest representative has too much closeness to other constituents or lopsided pressure in a direction away from me… Maybe some measure of influence on representatives that considers the relative distance and direction of all constituents who have that representative as their closest one.
Mark Frohnmayer Wed 29 Mar 2017 6:03PM
Seems like your metric should be quantifiable. The point of quantifiability is that you could then run VSE-like simulations on a variety of PR algorithms and see which ones actually perform well. The overall measure would include representative accuracy of the electorate as a whole as well as the percentage of how many voters have at least one "close" representative.
Aaron Wolf Wed 29 Mar 2017 7:22PM
Yeah, I think the point is to measure relative closeness as well as absolute. The point being that, in addition to the value of being actually close to a representative, there's the issue of being as close as other constituents.
I'm not sure what the ideal actually is, but among other things I've mentioned, if there's a 10% of the population that is really far from any representative but all really close to one another, then the ideal representation gives them some voice so they are included in the conversation. On the other hand, 10% being equally distant from representatives but all spread out from one another don't call for representation as much. And yet we don't want just representatives of consolidated groups. So, the ideal is that substantially-sized unified interest groups should be represented and diverse consortia of mixed voters should be represented, and especially in the latter case, ideally the representative is reasonably equidistant among constituents so not siding with some over others too much.
We don't want to empower only consolidated interest groups but neither do we want any such groups to be excluded. I'll feel included if there's a representative who is particularly in touch with my notable minority group and our concerns; and I'll feel included if I know that a representative listens to a wide coalition of supporters and I have, within reason, similar sway with that representative as other members of the coalition.
Clay Shentrup Sat 8 Apr 2017 4:16AM
Mark is right about making things quantifiable.
Also, I contend you cannot use binary metrics like "close". Would a voter rather have one representative who's "close", or 100 who are "just a little less than 'close'"?
Aaron Wolf Sat 8 Apr 2017 7:49AM
I think it needs to consider relative closeness. In some sense representing the sway a voter has on a representative or their representation relative to other voters.
If I agree with a representative 80% of the time, I will feel more adequately represented in a context where all other voters also agree about 80% of the time (though not necessarily the same 80%) than if most of other voters agree 95% of the time with the representative.
This is not the only factor to consider, but it's what I'm trying to get at. For quantifying closeness, it's about whether or not you feel your representation is closer to others than to you or is representing you as well (or at least in proportion to your (you and those who share your positions consistently) relative population in the group.
So, if there's 5 positions and 2 representatives: a representative could be equally close to positions A and B and a bit further from C. Another rep could be equally close to D and E and a bit further from C. In this scenario, all voters will feel equally represented if the C group feels they have a little representation from each of the reps to the extent that their total representation is still similar to that of the other groups.
But if a rep is much closer to A than to B and yet B is the same population as A and the same rep is the closest one and no other rep is anywhere near B, that situation will lead to B feeling underrepresented.
Sara Wolf Thu 30 Mar 2017 6:21AM
I like Aaron's idea there, that the most important thing is that people feel fairly represented. That perception is as or more important than actual perfection in representation because when people feel like things are fair and they have a voice then they work much better together. Proportional Representation is ultimately about the indirect goal of getting representation in practice, in legislation, and where it counts in governance and the key for that is teamwork.
Unfortunately we can't accurately measure representitive quality of legislation, so I am totally getting at trying to design or create a system of VSE for Proportional Representation here. What do you guys think about the idea of a quota (in a simplified system for extrapolations sake). If there are 4 seats should you need a full 25% to get a seat or should your groups percentage be rounded up or down to the nearest seat? The second question will be how to determine how to convert complex scores and rankings into overall percentages but one thing at a time...
Adam Zielinski Sat 8 Apr 2017 12:21PM
It's pretty much impossible to try to forecast how people are are going to feel about the field of candidates who have chosen to run, and if you feel like any of them represent you at all or to some degree. It depends on the individuals who decide to run.
Aaron Wolf Sat 8 Apr 2017 7:20PM
Yeah, but the point of quantification is to compare how people will feel about the outcome of an election, not how they feel about the candidates beforehand.
I'm concerned about scenarios where you and I both support candidate A over B but I feel poorly represented either way and you feel perfectly represented by A. In such a case, if most voters are like you, then A is fine and I'm just too small a minority to get stronger representation. But if enough voters are like me, then a good PR system would elect a candidate C closer to me if such a candidate runs… basically the goal being to minimize the number of voters who feel relatively less represented.
I could imagine quantifying by modeling a wide range of candidate positions. I suggest we're looking for which PR approach does the best job at minimizing the percent of voters who feel that "my closest representative is a stronger representative of other voters than of me"
@nardopolo and @clayshentrup does that sound like we're approaching something quantifiable and study-able?
Clay Shentrup Sat 8 Apr 2017 7:51PM
the point of quantification is to compare how people will feel about the outcome of an election
I don't think that's very important. For instance, a PR advocate could think it was just terrible if Congress was 100% full of centrists. But maybe it would pass laws that would improve net human welfare more than a superficially "better" PR system. Their immediate naive gut reaction would be, "Oh this outcome is just dreadful." But that would be wrong.
An analogy might be that you're randomly given either stock from Apple or Google, and you're disappointed that you got the Google stock. Then five years later, it turns out Google has performed much better, and you actually would have been worse off if you got the Apple stock.
A good system gives people what they want (long-term), not what they think they want (short-term).
Mark Frohnmayer · Wed 29 Mar 2017 4:10AM
@wolftune writes (from another thread): "... the criterion is about minimizing non-consent by minorities by doing all we can to require that they have a voice even though minorities should still not be able to overrule the majority ... I would love to see if anyone can figure out how to make this idea more measurable."
It seems to me that you're getting at a measure of average (or median?) distance of voter in issue space to the nearest elected candidate -- i.e. we want to maximize the number of voters who have a closely representative candidate. Is that accurate?