Whilst pure Condorcet passes it, any of the variants that does somehow pick a winner if there's no condorcet winner no longer passes it (but does beat the always a winner criterion in that case). Considering that no non-dictatorial ranked system that always produces a winner is able to satisfy this criterion, I think it's sensible to look at weaker forms of the criterion to still get something out of it if we want a ranked system. The local Independence of Irrelevant Alternatives is one such weaker form that still isn't satisfied by many. It is only concerned with the last place dropping out not changing the winner, and the 1st place dropping out always causing the rumner up to win. IIA is for a large part related to the spoiler effect, where passing it would mean resistance to the spoiler effect. Whilst condorxet methods aren't perfectlt resistent to the spoiler effect, they do have more resistance than most methods; if there is a condorcet winner, they there is no way that another candidate dropping out would change that winner, and a new candidate joining could only chamge the winner by either becoming the winner themselves, or by becoming part of the Smith set (which the previous winner would also still be part of). A criterion that is then particularly useful if we also care about the condorcet criterion, is the Smith-IIA, which states that no candidate outside of the Smith set should affect the winner. This also pairs nicely with the Smith criterion, a generalization of the Condorcet criterion, which states that the winner should always be from the Smith set (if a condorcet winner exists, then the condorcet winner is the only candidate in the Smith set and thus a method that satisfies the Smith criterion must necessarily also satisfy the Condorcet criterion). There's also independence of clones*, which means that if a clone* candidate is added or removed, it should not affect the chances if winning/losinf for any candidates outside that set of clones*. The Ranked Pairs Condorcet method satisfies all of these weaker forms of the IIA (as well as satisfying all other criteria, except the pure IIA, discussed in this mini-series). And seeing as no ranked voting method is able to satify pure IIA, I'd argue that ranked pairs condorcet is the best ranked choice system out there. To make things even better, it's relatively simple to understand how a winner is chosen, and easy to verify, which can be considered valuable properties of a voting system in practice to have. *A clone in this context is a candidate who is ranked by all voters directly above or below the candidate they're a clone off, without any other candidates between (both are considered clones). Potentially a set of multiple candidates could all be considered clones if they are always ranked together without any others ever ranked between them. The main point is that voters see them as close enough that they always move as a group within the rankings, and are never split up, even if they're not exactly equal.

Will you apply these criterion to the US electoral college?

@CarneadesOfCyrene

7 жыл бұрын

I did not plan on it, there are many more complicated systems of voting that are actually used, at some point we will look at these other systems of voting, but as of now I don't plan to in this series.

I'm curious why you haven't included Range/Score voting, Single Transferable Vote and Mixed Member Proportional systems in this series. I suspect STV and MMP would prove well placed on a chart of Bayesian Regrets of Election Methods. (See the graph at the bottom of the homepage for rangevoting.org .)

@CarneadesOfCyrene

4 жыл бұрын

Interesting voting methods. I included only a subset of methods because there is only so much time. But perhaps in a future series.

@nienke7713

16 күн бұрын

STV is just IRV that selects multiple winners, if STV is reduced to a single winner it becomes IRV, and as such as the same issues as IRV. MMP is a system that is inherently for a representative body with multiple winners (whilst trying to also maintain some amount of local representation), which means it doesn't really fit in this series, as this series focuses on methods that can select a single winner (and may also be able to select multiple winners and/or a ranking, but that's not the focus of the series). Range/score voting has a very high chance of tactical voting, as people will rank their most preferred candidate(s) the highest possible score (even if they don't particularly like any candidates) and their least preferred candidate the lowest possible score (even if they don't particularly dislike any candidates); they also have incentive to inflate the scores of any front-runners in the polls who they prefer over other front-runners, whilst undervaluing any front-runners they like less. E.g. let's say my honest scores (from 0-10) would be A 6, B 4, C 8, D 2 First of all, I just like C the most so I'll push that up to 10, and I like D the least so I'll push that down to 0. In addition, the polls suggest the race is likely going to be close between A and B, with C and D not really having much of a chance to win, and since I prefer A over B, and my true preference and most despised candidate are both unlikely to win, I might as well give A 10 points as well and B 0 points as well, in order to maximally contribute to the chance of A winning over B. This then creates a prisoner's dilemma kind of situation where if you don't vote tactically like that but most others do, then you're effectively wasting your vote, whereas if you're the only one voting tactically like that, you're really increasing your chances to get a more preferable outcome. So everyone always has an incentive to vote tactically, even if nobody voting tactically might be better for the collective good. If everyone votes tactically, you've effectively gotten approval voting; if not, then it's just harming the people who didn't vote tactically. Approval voting is also very susceptible to tactical voting. Let's say my honest evaluation would be I approve of W and X, and not of Y and Z, but I prefer W to X, and Y to Z. If polls predict that things will mainly be between W and X, even though I honestly approve of both, I have incentive to only approve of W in order to increase the chance of them winning. Similarly, if the polls predict it's going to be between Y and Z, I have incentive to say I approve of Y, even though I don't really approve of them, simply because I prefer them over Z and want to increase the chance of them winning rather than Z. Scores are useful for judging competitions where you have honest ratings and no motive to vote tactically, and approval is useful for making quick decisions with a group of friends where you just want to find what works for most of you, but they don't work well for the complexity of elections where people have a big stake in the outcome of it.

Is this the same as neutrality?

In this particular example (7:00), 4 of 7 voters (A-D) have a slight preference of H over J; however, 3 of 7 voters (E-G) have a very strong preference of J over H. In short, for this example, I think most of voters may be "happy" with J as the winner (J is in top 2 of 5 of 7 voters), so, I don't see why Independence of Irrelevant Alternatives is a must have criteria.

@nienke7713

16 күн бұрын

You can't tell, it's a ranking, not a rating, they might have a strong preference if H over J but marginally dislike the ones below J even more. E.g. A's preference could result from something like the following score evaluation: H7 J6 I5 K4 (difference if 1 between H and J) But it could also be from something like this score evaluation: H10 J3 I2 K1 (difference of 9 betweheen H and J) Similarly B could be something like: I7 H6 J5 K4 (difference if 1 between H and J) Or something like: I10 H9 J2 K1 (difference of 9 between H and K) And similar for C and D Meanwhile, E/F (and somewhat similarly G) could be something like: J10 I6 K5 H1 (difference of 9 between J and H) Or it could be something like: J7 I6 K5 H4 (difference of 3 between J and H) And that doesn't even consider options where they'd be scored the same (which is a typical feature of score voting) but being forced to order one over another (required in some, but not all, ranked systems) Now I'm not at all a fan of score voting, but to try to gain any idea of how strong the preference of voters for one over another is merely by looking how many positions they are removed is simply arbitrarily and unrealistic, and one of the major problems of the Borda count which effectively does that by assigning scores to the ranks (and requires each option to be ranked differently)

Suppose we are picking two winners by a system that will makes the two highest ranked candidates winners. X is liked best by 30%, which is enough to be one of the winners, but Y's 18% is not enough, and that candidate is beaten by Z's 20%. Thus, X and Z are the two winners. In our second go around, the relative rankings between X and Y are unchanged: nobody who liked X more than Y has changed her mind, and nobody who liked Y more than X has changed hers. But now, nobody cares at all for Z or any of the other candidates. Thus, under our election rules, the new winners are X and Y. I.e., with no change to the ordinal relations between X and Y on any voter's list, Y, which had been a loser, is now a winner. This scenario seems to be violative of IIA as you have defined it in this video. Is that intentional, or should you have put the principle differently? Thanks.

@nienke7713

16 күн бұрын

I think it assumes single-winner elections where the only way to get more than one winner is a tie. I think (but idk for sure) that a more generalised version of this would be that: -if no voters change their preferences regarding all winners compared to a loser y, then any number of voters changing their preferences regarding any other losers should not make y a winner AND -if no voters change their preferences for all losers compared to a winner x, and any number of voters change their preferences regarding other winners, then x should not become a loser. The principle is closely tied to the spoiler effect, where the introduction or removal of a similar candidate can affect the outcome. It is strongest in plurality/FPTP voting, but can to a lesser extent still happen in other systems, e.g. in instant runoff, whilst votes can be given to the next preference, the importance of 1st/higher preferences still means that additional candidates can change the outcome. Your example with picking the two candidates with the highest percentage also very much suffers from the spoiler effect, being effectively plurality/FPTP but with two winners instead of 1