Talk:Strategic voting

IRV vulnerability

I don't see how instant runoff voting is vulnerable to the push-over strategy. I'm not so confident that I'll edit it out, but could someone doublecheck this and explain it here or in the article if I'm wrong? 66.41.154.228 (talk) 05:03, 16 December 2008 (UTC)

I've been researching these things for decades now, and I've never heard of a "push-over" strategy or tactic against runoff voting. In the past, we have had trouble with complaints being based on small sets of contrived, anecdotal examples. The peer-reviewed sources we have at Instant-runoff_voting#Tactical_voting seem entirely consistent with the likelihood that someone just made this up. GetLinkPrimitiveParams (talk) 04:12, 1 February 2009 (UTC)

IRV is definitely vulnerable to turkey-raising. A well-known example is the 2022 Alaska special election, where voters had an opportunity to support Palin, resulting in the elimination of Nick Begich (the Condorcet winner). –Maximum Limelihood Estimator 00:26, 18 April 2024 (UTC)

Move to Strategic voting

The ngram shows a 4x more frequent usage of strategic voting compared to tactical voting. Suggest changing lemma to strategic voting. HudecEmil (talk) 19:30, 16 October 2022 (UTC)

Changed HudecEmil (talk) 17:23, 19 October 2022 (UTC)

Wiki Education assignment: Political Parties

This article was the subject of a Wiki Education Foundation-supported course assignment, between 26 August 2024 and 5 December 2024. Further details are available on the course page. Student editor(s): Mikb26 (article contribs).

— Assignment last updated by PurdueGrad29 (talk) 18:45, 18 November 2024 (UTC)

Removal of sources

@Affinepplan: You say that the sources I added are essentially SPS. Could you give some references indicating that Constitutional Political Economy and Electoral Studies are unreliable journals with such low quality control that their articles can be considered SPS?

As for the other point, you may think that statistical simulation papers are inappropriate, but they seem to be accepted elsewhere on Wikipedia (e.g. Green-Armytage strategy incentive papers, and various papers showing probability of various failures under impartial culture). So I don't think they can be dismissed out of hand just by being Monte-Carlo simulations. Would you accept a rephrasing, something like "Statistical simulations suggest...", instead of "Thus, ..."? Wotwotwoot (talk) 12:57, 3 December 2024 (UTC)

I guess I'm trying to be tactful and avoid saying it outright, but both papers are just extremely amateurish and low-quality. There's a reason they have zero citations. Neither author has any professional or academic credentials in the field... they are just enthusiastic amateurs with some various technical background whipping up "simulations" and then spending a while to pretty up the result.

I could do this, you could probably do this, any data scientist with a few free weekends could do this.

Not all statistical simulation papers are created equal :( Affinepplan (talk) 15:31, 3 December 2024 (UTC)

First, I'm going to quote the prose and citations that were removed (adding links to the journals):

Thus, Condorcet methods incentivize candidates to position themselves closer to the median voter and appeal to a wider section of the electorate than instant-runoff voting does.

• Robinette, Robbie (2023). "Implications of strategic position choices by candidates". Constitutional Political Economy. 34 (3): 445–457. doi:10.1007/s10602-022-09378-6. ISSN 1043-4062.
• Ogren, Marcus (2024). "Candidate incentive distributions: How voting methods shape electoral incentives". Electoral Studies. 90. Elsevier BV: 102799. arXiv:2306.07147. doi:10.1016/j.electstud.2024.102799. ISSN 0261-3794.

Now I'm going to point out how comical that Affinepplan is referring to these as "essentially SPS". These are both recent papers, so it's not surprising that they haven't been cited yet (especially when there's folks out there like Affinepplan shittalking and burying their results, cloaking themselves in anonymity). The dismissiveness by Affinepplan seems libelous to me, and I'll eventually restore Wotwotwoot's change soon if someone doesn't beat me to the punch. -- RobLa (talk) 21:58, 3 December 2024 (UTC)

libel requires a statement to be false

anybody who has spent much time reading high-quality publications in this field recognizes these papers as... not that. Affinepplan (talk) 03:12, 4 December 2024 (UTC)

before restoring the change, maybe get some opinions of actual experts? I would say this is a big fat {{expert needed} } Affinepplan (talk) 03:14, 4 December 2024 (UTC)

pinging @Jannikp97, @DominikPeters whom I know to be domain experts. I would love your opinions on the inclusion of the edit in question (and opinions on the quality of the source are relevant by proxy). Affinepplan (talk) 13:02, 4 December 2024 (UTC)

I'm not particularly familiar with either Constitutional Political Economy or Electoral Studies. Both clearly are academic peer-reviewed journals and they have well-regarded editors (neither experts on voting), but I checked the editorial boards of both journals and they don't seem to contain social choice experts.

After a brief look at the papers, I tend to agree that they are ultimately unconvincing and in my view shouldn't have been published as they are (which is what I would have said if I had been a reviewer). However, I'm not so sure that we as wikipedia editors should be making such judgments.

A way to argue against these papers that I think is more robust is to say that they are very much primary sources (WP:PSTS): they set up some particular simulation with a thousand arbitrary choices and come to some conclusions, rather than providing an overview of several different studies/simulations that have been done about incentives that candidates face. (There is very substantial literature about this using the term "strategic candidacy", but it doesn't contain interesting conclusions about the IRV vs Condorcet fight. There is a wikipedia page about strategic nomination but it doesn't seem aware of this literature.)

If one wants to keep it, one could make the sentence under discussion more neutral by writing it like "In some simulations on random data, Condorcet methods incentivize ..."

By the way, a nice theoretical paper about a similar topic is this one: https://arxiv.org/abs/2303.09734, The Moderating Effect of Instant Runoff Voting, which is mostly about comparing IRV to plurality. DominikPeters (talk) 15:32, 4 December 2024 (UTC)

I mostly agree with @RobLa, it is peer-reviewed academic sourses and while at first glance the tone and some other aspects of it appear questionable, I wouldn't say such sources should be excluded. Their recency and topic make it unsurprising that there as not many citations.

I suggest they should be cited in the appropiate context, making it clear that it's based on simulations. Otherwise, they are mostly not about very revolutionary claims, are they? There are probably works with similar conclusions out there. Rankedchoicevoter (talk) 20:35, 4 December 2024 (UTC)

It sounds like rephrasing the statement would be acceptable to the people here. I'll wait a bit to see if anybody disagrees and then take a stab at it. Alternatively, we could try to get more people involved with an RFC if there's no agreement. Wotwotwoot (talk) 14:48, 5 December 2024 (UTC)

personally I do not think these sources should be included in the article at all.

if you insist, and are willing to phrase the inclusion in such a way that makes it very clear that these are fairly arbitrary "simulations" with many hand-chosen parameters and lack any analysis of empirical data justifying these parameter choices, then I will not die on the hill to block the edit.

but I would much much much rather you choose a better source based on outcome data and more robust analyses, like [2303.09734] linked above. Affinepplan (talk) 15:14, 5 December 2024 (UTC)

Rationale for separating sincere/insincere strategies

Hi @Wotwotwoot—first, I wanted to say thanks for lots of the edits you've made here, I see a lot of big improvements :)

First, I wanted to ask about the change to the discussion of Duverger's law—while I'd agree that the strength of the effect is going to depend on how often favorite-betrayal is incentivized, I'm not sure if the new formulation would really count as a "law" the same way Duverger's law (as generally stated) does, and I'm not sure this is discussed in the sources provided.

In the section on common strategies, I've tried to separate out discussion of sincere and insincere strategies, for a couple reasons. First, discussing the current "double-counts" the same . More important, though, I'm looking at a in preparation for an article on the Muller-Satterthwaite theorem, which establishes the conditions under which a mechanism (without payments) is strategyproof. Namely, a rule is strategyproof on a restricted domain of preferences if and only if it satisfies Maskin monotonicity, which is the combination of:

1. Independence of irrelevant alternatives, which prevents insincere burial/compromise (the goal is making candidates look "strong" by ranking them above many irrelevant alternatives).
2. Ordinality, which corresponds to exaggeration/leveling strategies (which exploit cardinal rules).
3. Positive responsiveness, which corresponds to turkey-raising strategies.

This is basically a "positive" version of Gibbard-Satterthwaite that lets you classify all the ways rules are manipulable, and also build strategyproof rules when the domain is restricted (e.g. the median mechanism for points on a line). – Closed Limelike Curves (talk) 18:10, 6 January 2025 (UTC)

About Duverger's law: to quote the Wikipedia article, Duverger argued that:

"a majority vote on two ballots is conducive to a multiparty system, inclined toward forming coalitions": that the two-round system encourages a multiparty system, but to a lesser degree than proportional representation does.

Since top-two has some compromise incentive, yet Duverger argues it supports a multiparty system, we can't say that compromise incentive alone leads to two-party rule. It must be "sufficiently strong" - i.e. stronger than top-two but not stronger than FPTP - to be consistent with Duverger's argument.

So I think qualifying the implication is consistent with Duverger's claim. At worst it's WP:SYNTH; but if it is, the reference to Duverger's law should be removed altogether since Duverger doesn't deal with immunity to compromising as a binary criterion. Strictly speaking, common party list methods like Sainte-Laguë also have some compromise incentive, but they also support multiparty rule.

About sincere/insincere strategies, I don't think it's necessary to distinguish them to avoid double-counting. Consider the straightforwardness notion of Gibbard's theorem: the method is susceptible to strategy if the voter's optimal ballot depends on the behavior of the other voters. This concept doesn't need to deal separately with the specific forms that strategy may take, and I don't think it double-counts any specific form of strategy.

To put it another way: we shouldn't need to distinguish between preference-sincere strategic ballots and preference-insincere strategic ballots for the same reason that we don't need to distinguish between strategic ballots where the voter ranks their honestly least liked candidate last, and strategic ballots where the voter doesn't. They're all ballots adjusted by the voter according to information about how the other voters are going to vote.

But I might be misunderstanding what you mean by double-counting.

About Muller-Satterthwaite, my impression is that his theorem deals with strictly ranked ballots. Quoting from the paper, "A ballot B is a strict ordering of the elements within S ... [i]ndifference is not allowed ... A voting procedure is a single-valued function v(B_1, ..., B_n) that evaluates the profile of ballots and selects one member of S as the group's chosen alternative". If it does pertain to strictly ranked ballots, then we wouldn't need to distinguish cardinal-sense sincere and insincere ballots. Wotwotwoot (talk) 18:23, 9 January 2025 (UTC)

but if it is, the reference to Duverger's law should be removed altogether since Duverger doesn't deal with immunity to compromising as a binary criterion.

I think we have a few sources describing favorite-betrayal as a binary criterion—Poundstone, Volić, and WDS's papers (WDS can be used with caution as an expert SPS, since he's a professor who's been cited by other RSes like Laslier 2018). – Closed Limelike Curves (talk) 20:26, 9 January 2025 (UTC)

I'm not referring to favorite betrayal being described as a binary criterion, but rather compromise immunity. However, the same argument holds for favorite betrayal. I'll first go through compromise immunity and then show that the same holds for favorite betrayal.

You said:

while I'd agree that the strength of the effect is going to depend on how often favorite-betrayal is incentivized, I'm not sure if the new formulation would really count as a "law" the same way Duverger's law (as generally stated) does, and I'm not sure this is discussed in the sources provided.

That is, as I read it, you said that my qualified statement about Duverger's law might not count as a law the way that the unqualified Duverger's law does.

My response in my last response was to say that Duverger already implicitly said that some methods with compromise incentive support multiparty rule. Therefore, to conclude, in an unqualified manner, that every method with compromise incentive leads to Duverger-style two-party rule is not something that can be concluded from Duverger's law itself. The version prior to mine said that compromising "typically (but not always)" leads to two-party rule, but I think even that is a too strong statement, as it doesn't follow from the sources. Thus, if we are to mention Duverger's law in the context of compromising, it should be qualified.

In other words: only methods with sufficiently strong compromise incentive leads to two-party rule, because Duverger explicitly mentioned some methods with compromise incentive as supporting multiparty rule.

For favorite betrayal, the argument is similar. Runoffs fail the FBC yet are stated by Duverger himself as allowing for multiparty rule. We thus cannot conclude that FBC failure leads to (or "typically" leads to) two-party rule.

Wotwotwoot (talk) 12:40, 10 January 2025 (UTC)

About Muller-Satterthwaite, my impression is that his theorem deals with strictly ranked ballots.
Yes, which is why I mentioned the third condition (must be ranked)—Muller-Satterthwaite shows an ordinal social choice mechanism, restricted to a given domain, is strategyproof if and only if it satisfies IIA and positive responsiveness. – Closed Limelike Curves (talk) 20:29, 9 January 2025 (UTC)

If Muller-Satthertwaite only deals with strict ranked ballots, then there should be little reason to bring up cardinal-style sincere vs insincere strategies, because they're outside the scope of the theorem. I would imagine it suffices to say "this theorem only pertains to ranked methods". Wotwotwoot (talk) 12:49, 10 January 2025 (UTC)

But I might be misunderstanding what you mean by double-counting.

What I mean by double-counting is that, as-is, we discuss the same strategy (exaggerating how good/bad a candidate is) three separate times—for compromise/burial/leveling—which leads to a lot of repetition. Whereas, if we separate out leveling into its own strategy, the classes of strategy are mutually exclusive and jointly exhaustive, which makes things much nicer. – Closed Limelike Curves (talk) 20:45, 9 January 2025 (UTC)

James Green-Armytage defines compromising as ranking or rating a candidate higher to get them elected, and burying as ranking or rating a candidate lower to keep them from being elected. (Section 2.2 as well as proposition 4 of "Strategic voting and nomination".) As the simulation results and summary table make use of this definition of compromise and burial, I think it could lead to confusion to exclude leveling.

For instance, readers of the SVN paper could look at the summary table, see that approval has a high compromising incentive, and go "but hold on, that doesn't make sense".

If we are to separate them out, then I think we should find a source that defines compromising, burial, and leveling separately, and then make a note that there's some disagreement about whether leveling is merely a subtype of compromising or burial; or we could just drop leveling as a distinct category. That wouldn't be as nice, either, but would be more true to the given sources. Wotwotwoot (talk) 13:02, 10 January 2025 (UTC)