SP Voting: Explanatory Video
rob Banned last edited by rob
one of your incorrect assumptions about the system is that it does not reduce to majority judgment in the case of two candidates.
I simply noted that Andy said that, if it isn't true, sorry, but that was a very minor side point. Your response certainly implied that it was true ("I’ve also given pretty comprehensible reasons, arguments and examples for my opinions about majority rule even in cases of two candidates") But the real point was that you don't seem to value game theoretic stability, or maybe you don't understand it in the way I do. Regardless I spent far more time on the example of the temperature voting thing to illustrate a similar point but with a bit more generality and because it applies more, given my hope (which I think you share), for there being more than two options so there actually is a middle ground/median. This was to ensure that we were on the same page on whether that was important (given that I still don't know what problem you are really trying to solve). Lijphart certainly seems to favor things that go for middle ground. I spent a bit of time reading his stuff (and about his stuff) and I'm totally on board that sort of approach.
Going back to the example of a team lunch with fruit cups and PB&J sandwiches from a long while ago, you mentioned that voters should not be allowed to declare being allergic to a candidate. And certainly, that would be absurd.
I'm not sure what you are saying would be absurd. (ignoring the kid with special requirements, or trying to factor that kind of thing into a way of voting on lunch)
I think it would be rather absurd, in the context of real-world voting, where someone can say "I simply can't accept this candidate, so you must accommodate me because my needs are more important than your simple preferences." In a small group, especially where there is continued interaction, it makes sense that we accommodate people with special requirements such as allergies or disabilities. With larger groups, such as political elections, it should be obvious that such a thing would be abused, so that isn't the kind of thing that voting is intended to handle. It also doesn't really fit with "one person one vote."
To me, voting is about preferences, not absolute requirements.
( The way to accommodate those, actually, is representative democracy. You vote for a person, who you trust to sort out special cases. But you don't get to declare yourself "allergic" to a candidate so you can get your way )
You also, in the context of my example of voting for a thermostat temperature, mentioned a voter who has OCD and doesn't like the number 73. In that case, it just seemed like an attempt to miss the point of a example that is intended to be simple (compared to voting for human candidates) for the purpose of illustration of a concept. Regardless, it doesn't seem like a sincere thing that you think is a likelihood in that scenario. (I have my doubts any real world office would feel the need to accommodate such a person, and also have my doubts that it has ever happened)
Notice that the "capital of Tennessee" example, used widely, elaborates in some detail about such assumptions to keep the example simple (in that case, that everyone would prefer a city closer to themselves, and it actually says that everyone lives in one of the four cities).
As I mentioned, "consensus" is in the sense of Lijphart, being the principle that representative governments--including their voting systems (how else?)--should be responsive to broad supermajorities rather than primarily to slim majority rule, which can be highly divisive.
I don't disagree with that, other than questioning what is meant by majority or supermajority in this context. I am very anti-party, especially when it devolves into two dominant parties (where the concept of majority becomes a real thing because you are left with a binary choice) This is why I keep returning to the concept of median, which is much easier to visualize for me when the thing in question isn't a binary choice but something that essentially lies on a spectrum. Maybe it isn't the best term either because it is hard to define what it means with discrete human candidates, but the temperature voting thing is the best way I can explain what I mean. (since in that case there really is a median, and it is also particularly easy to see the meaninglessness of the term "majority" as the number of potential "candidates" increases and you have options in the middle that have wide appeal even if only a small percentage considers them their first choice)
I don't think I ever got an answer on whether or not you consider this an issue with Condorcet-compliant systems, such as whether you consider Condorcet systems majoritarian. There are internal majorities, yes (between candidates pairwise), but overall they tend to settle on candidates that are near the center and therefore have wide appeal and tend not to be divisive. (as do various other systems like approval and STAR, which just use different approaches to generally elect what would be the Condorcet winner) To me any of those should accomplish your goal, to the best I can understand what your goal actually is.
@andy-dienes that just isn’t true. A very slim majority cannot guarantee their victory in SP Voting, it does not satisfy the majority criterion. Especially if the rank order variant is used. If you tried to actually explain how a majority might accomplish that, you would run into issues.
What I imagine you think would work is bullet voting. That’s impossible with a rank order variant. In any case, let’s say there are at least three candidates, and a slim majority aligns behind their first choice and last choice in an attempt to veto a candidate. Then all of the other candidates already have at least 50% of the electorate giving them no minimal scores. Then there are two cases:
First, if the majority first-choice is highly divisive, a large fraction of the non-majority is likely to score them minimally. This guarantees that there is some candidate with a broad supermajority of the electorate not giving them any minimal scores, and that candidate will almost surely prevent the highly divisive majoritarian candidate from winning the election.
Otherwise, if the majoritarian first choice is not highly divisive, then the system will be responding to the preferences of a supermajority, which is exactly what was desired in the first place.
In other words, the only way a majority can guarantee their first choice to win is either if they are also supported by a significant fraction of the minority, meaning that the majority top choice is not highly divisive.
51% of voters rate candidate Alice 5/5, every other candidate 0/5
Unless I am misunderstanding the method, please explain how any other candidate can win in this scenario.
That’s impossible with a rank order variant
Are you suggesting to disallow equal ratings?
@andy-dienes yes, the possibility of disallowing equal ratings was one of the first points addressed in the video. That turns SP Voting into a rank order system.
@cfrank Even in the context of truncation? So you will require voters to rank every single candidate? This opens up a whole new array of issues
@andy-dienes Of course, but it can be addressed in the same ways it would be addressed in any other system where that is a potential issue, such as Condorcet methods.
Another variant is to simply use the ballots themselves to define a discrete distribution.
where that is a potential issue, such as Condorcet methods
but it is not a potential issue in the same way. you are imposing forced strict rankings as a band-aid for an issue with the method. Condorcet methods typically function just fine with equal rankings allowed.
@andy-dienes no, it isn't a "band-aid," it's how the method can be defined to operate. SP Voting is a framework, it isn't a single method. By that logic any aspect of any voting system can be called a "band-aid." Disqualifying invalid ballots is always a necessary aspect of a voting system, and I'm sure there are reasonable defaults that can be defined to accommodate voters who don't want to rank all of the candidates, such as distributing their vote for unmarked candidates uniformly across all of the (unmarked) scores.
SP Voting is a framework, it isn't a single method.
My perception is that when I give a criticism of one possible instantiation of SP you are pointing to a different instantiation and claiming it does not have that issue.
Even if the 'framework' includes enough different variants such that you can always choose one of the variants without some particular issue, you cannot just assert that thus SP automatically gets the best of all possible worlds and is free of issues. This is moving the goalposts. By that logic, why not just propose that your "framework" is all possible voting rules?
If SP allows equal rankings, it has (in my view) major issues. If it does not allow equal rankings, then it has different major issues. Can you pick a particular instantiation and we can continue the discussion from there?
@andy-dienes SP Voting definitely does not automatically get the best of all possible worlds, nor is it free of issues. The issues you are raising are valid, and I'm glad you are raising them. I want to see if the parameters of the framework can be adjusted to accommodate them not just individually but all together.
The SP Voting methods I'm proposing are mostly for theoretical comparison purposes. I think the ranked order variant would give very good results, but as you mention, it has the issue of forcing complete rankings. What I'm curious about is, as we address these issues, whether the SP Voting framework converges in any sense to some kind of more fully specified voting system. And then, in that case, maybe the system can be simplified to capture the desired behavior. If it ends up being something like a Condorcet method, that would be interesting to me! That would pretty much convince me that Condorcet methods, which I am personally on the fence about, are the way to go. I still think that Condorcet methods could be relaxed a bit to stray away from majoritarianism toward a more consensualistic mechanism, but who knows.
For right now, if you're interested in considering this system, I'm thinking of an intermediary variant that does not require full ranking and allows equal scores: Let's say for now that voters can only indicate a single candidate in the minimum score position, and the default score for an unmarked candidate is S1 (unless perhaps they are unmarked by a supermajority (say 2/3) of the electorate, whereby they will be disqualified). I am still considering the geometric weighting with a common ratio of 1/2.
@andy-dienes yes, that makes sense. Alternatively, one might require that for every minimum score on a ballot, there is at least one corresponding non-minimum score. That would lessen the effect of teaming but also give a majoritarian advantage.
@cfrank Then whichever party has more candidates than the other is advantaged. This makes the clone failures even worse