Talk:Spoiler effect

X

Alternative voting systems which avoid the spoiler effect include instant runoff voting, also known as single transferable vote.

This is wrong on two accounts:

1. IRV and STV are not the same thing. The first is a single-winner system; the second is a multi-winner system. The second reduces to the first in the case of a single district.
2. IRV and STV can suffer from the spoiler effect. For example, if the "p-ist" vote was split between 3 p-ists, the most popular and "major party" p-ist could be eliminated early, even though they would handily beat the winning q-ist.

Perot?

Where is Perot? He won Clinton the White House in two consecutive terms then disappeared forever. Where is the discussion on candidates planting a spoiler to ensure their win (as Clinton obviously did with Perot)? 198.209.0.252 (talk) 13:42, 21 April 2011 (UTC)

I also wondered why there was no mention of Ross Perot when I read this article. To add him, you would need a reliable source describing him as a spoiler or claiming that he cost Bush the election (all the candidates from that election are still alive). here is an article by Dan Quayle in which he claims Perot gave Clinton the election. here is another article claiming that Perot voters would have split evenly between Clinton and Bush, but without out any reference to a poll to back up the claim, though this may have been a further result from exit polls by Voter Research and Surveys mentioned earlier in the article. (Both articles are referenced in the Wikipedia article on Ross Perot's 1992 campaign.) Not really enough evidence for me to add Perot to the list of spoilers, though better evidence may exist.--Wikimedes (talk) 20:03, 27 August 2011 (UTC)

Nader must be a Millennial thing because Perot was a much more significant force, in the elections of the last 59 years. However, the whole concept of a spoiler is an antidemocratic conspiracy that demands the electorate vote for candidates who identify within the binary party system which has evolved to today's Republicans and Democrats. To say that any spoiler exists is to say that the two party system is threatened by voters who insist on their right to vote for whom they choose. Oh the humanity! CredibleSources (talk) 12:55, 2 November 2019 (UTC)

The frequent "Perot spoiler" claims are always funny because in 1992, Perot was the only candidate who wasn't a spoiler. Perot was the majority-winner (as he would have beat either Clinton or HW in a one-on-one election), so HW was a spoiler. (If he'd dropped out, Perot would've defeated Clinton.) Closed Limelike Curves (talk) 18:07, 5 April 2024 (UTC)

LIIA equivalent to IWA

@Wotwotwoot The LIIA == IWA equivalence holds from the standard social choice ⇒ social ordering construction (which I believe I noted, although I might not have made it explicit enough).

The second-place finisher is defined as the winner if the first-place finisher is removed; third-place is defined by results if first+second are removed; etc. This construction implies the first part of LIIA (removing the first k places does not affect finishing order of the remaining candidates) by definition, so we only need the second part: removing the last-k candidates (according to this ordering) does not affect the first-place candidate. Closed Limelike Curves (talk) 18:00, 6 April 2024 (UTC)

That's a good point. I think most readers would intuitively think of an order constructed by the method itself, like Plurality's three-candidate ordering being in the order of "Most first preferences, second most, fewest", not "Most first preferences, pairwise winner of the two others, pairwise loser of the two others", and get the wrong impression from the section stating that LIIA is equivalent to IWA. I'll just note it for the social choice construction at the end of IWA. Wotwotwoot (talk) 18:31, 9 April 2024 (UTC)

I think you're right that's what most people would think; at the same time, "remove and repeat" is probably the most mathematically natural/"correct" way to define the ordering "constructed by the method itself". It's better as a measure of candidate strength (was Le Pen really the 2nd-strongest candidate in 2002?). It's applicable to all systems (What's the "placement order" for River?) and gives a single coherent definition that works for every system, rather than having a different definition for each single-winner system. Closed Limelike Curves (talk) 00:45, 11 April 2024 (UTC)

I think this is a very subtle, but important, point: "worst candidate" needs to be defined consistently across systems for IWA to make sense. Otherwise, you can "hack" IWA (make it trivial) by defining the worst-place candidate according to any set of ballots as the one that, when deleted, doesn't affect the result. (Even if that candidate was actually the runner-up.) Defining "worst" like this lets you claim any system independent of at least one candidate satisfies IWA.

LIIA works because it asks "What's the worst alternative, according to the system itself?" Elimination order ranks candidates from worst to best according to the base method, not according to the new method. The worst candidate, according to a sequential loser method, is the one who could only win if every candidate ahead of them dropped out. Closed Limelike Curves (talk) 04:22, 11 April 2024 (UTC)

@Wotwotwoot Is there some other objective way to consistently define last-place alternative for all systems, besides the last-candidate-to-win construction?

If not, I suppose you could define a criterion like "independence of some alternative", which requires that for any possible ballot profile, at least one candidate other than the winner is not a spoiler. (IRV would satisfy that, since the min-first-place-votes candidate can't be a spoiler; FPTP would fail, since the Condorcet loser can win.) –Maximum Limelihood Estimator 18:15, 17 April 2024 (UTC)

Most systems are already social welfare functions (social ordering functions) and would be familiar to readers as such. For instance, Plurality generates an order of finish from the winner (with the most first preferences) to the loser (with the least). And Arrow's theorem, to pick a well-known result, explicitly deals with methods that return a social ranking.

Treating every method as a social choice function and then leveraging that to create a social welfare function would be confusing, because the social welfare function you end up with differs from the common definition of the method as an SWF.

I am not aware of any analogous election criterion mentioned in literature and directly relating to social choice functions. But I'm not aware of any reference to "independence of worst alternatives" either, so perhaps the best way to deal with the confusion is to delete the IWA section and just refer directly to LIIA. Wotwotwoot (talk) 17:06, 18 April 2024 (UTC)

Most systems are already social welfare functions (social ordering functions) and would be familiar to readers as such. For instance, Plurality generates an order of finish from the winner (with the most first preferences) to the loser (with the least). And Arrow's theorem, to pick a well-known result, explicitly deals with methods that return a social ranking.

I actually had Arrow's theorem in mind when explaining the social ordering construction. The reason I added it is because of a misunderstanding on an old version of the Arrow theorem page, which tried to claim Arrow's theorem wasn't important (because it only dealt with rankings instead of choosing a winner):

In social decision making, to rank all alternatives is not usually a goal. It often suffices to find some alternative. The approach focusing on choosing an alternative investigates either social choice functions (functions that map each preference profile into an alternative) or social choice rules (functions that map each preference profile into a subset of alternatives).

The point of this SRF construction is it's much more relevant to a practical electoral context. Under this construction, the SRF ranks candidates from second-strongest (heir apparent) to weakest (could only win if every other candidate was hit by a bus).

If we use the obvious SWF as the SRF, Arrow's theorem becomes less obviously relevant to an electoral context. After all, what if the change in irrelevant preferences only affected which candidate came in 2nd or 3rd place? Then there'd be no practical importance to IIA if we were holding an election.

The ranking construction makes IIA directly relevant to who wins an election, and questions like "what happens if a candidate drops out"—the common interpretation of Arrow's theorem as describing what can happen when a candidate is added or removed depends on this construction.

The reason for the description in terms of "weakest alternatives" is that requiring the order-of-finish to stay the same if we remove the first-place winner feels quite arbitrary. Explaining the weakest-alternative definition sidesteps all that. –Maximum Limelihood Estimator 03:54, 19 April 2024 (UTC)

I would just say something along the lines of "voters may object that the SWF's change may not involve the winner and thus isn't important. However, a standard construction from social choice functions shows that it also affects winners, hence Arrow is robust in the sense that it can affect the relative ranking of anybody". Or find a source that explains why Arrow is important (clearly there must be one, since it is so widely considered to be important). This would then avoid the confusion of having to deal with SWF/SRFs that look the same but aren't, and would avoid IRV proponents saying "but obviously, if you remove the IRV loser then the order doesn't change, clearly this definition is useless". This would limit the scope where the standard construction is needed so it doesn't cause confusion outside of that scope. Wotwotwoot 13:42, 19 April 2024 (UTC)

@Wotwotwoot what was wrong with the previous edit—I thought you'd suggested removing the information on Independence of Worst Alternatives?

The previous edit tried to make it much more explicit that LIIA was defined according to the candidate-strength ranking. Are there any suggestions you have on improving it?

I'm trying to explain the motivation behind LIIA as intuitively as possible. It doesn't really make sense to me why I'd care if removing the top candidate caused someone other than the second-place finisher to win. In many scenarios it's outright desirable: removing Bill Clinton from the election should cause the more-moderate Ross Perot, not Bob Dole, to win.

OTOH, preventing very weak candidates from spoiling the election (i.e. candidates without any hope of winning) seems very intuitive. –Maximum Limelihood Estimator 01:19, 22 April 2024 (UTC)

Citations are from untechnical sources and do not support content

I plan to add an issue tag to this page denoting such in 2 days. I attempted to remove one of the offending statements but again am stymied by the editing bureaucracy here. Affinepplan (talk) 19:57, 23 June 2024 (UTC)

considering the multitude of citations from nontechnical sources and the lack of rebuttal, I've added the tag. please do NOT remove it without consulting this talk thread. Affinepplan (talk) 17:42, 25 June 2024 (UTC)

The material you removed is well-cited, including multiple citations to academic and technical sources.

Given your editing history, I would strongly suggest not starting another edit war on the exact same subject, or there's a pretty good chance you'll get a topic ban. I'd suggest that, if you're WP:HERE to help us build an encyclopedia, it would be a good idea to try and work on improving articles in different fields. Closed Limelike Curves (talk) 02:41, 26 June 2024 (UTC)

Center for Election Science is an advocacy agency with a political agenda and should not be used as a reference for technical claims

William Poundstone is a columnist, NOT a researcher or a scholar and should not be used as a reference for technical claims

I am trying to help improve Wikipedia's reliability for articles on this topic. I really wish you would stop sabotaging my effort.

Please tell me, what are your credentials to have such an iron fist over these pages? Do you hold any degrees, or have any published research, or any professional experience in the field? Affinepplan (talk) 11:01, 26 June 2024 (UTC)

The claim you attempted to remove has two separate citations to scholarly journals, one to a well-researched popular science book, and one (you allege) biased source.

If you think one or two of the citations are not reliable, you can remove them. That said, you can't look at a sentence and say "this claim has two mediocre citations and two strong ones, therefore it does not have strong citations and can be deleted." That's not how Wikipedia, or logic, works at all; this isn't a court or bureaucracy where you can get a well-cited claim thrown out on a technicality because you happened to find a problem in one of four separate citations. Closed Limelike Curves (talk) 16:12, 26 June 2024 (UTC)

I agree with @Affinepplan - Center for Election Science should not be used, William Poundstone is not an authoritative source and this article needs more fingerprints on it and definitely more diverse reliable citations Superb Owl (talk) 08:24, 28 August 2024 (UTC)

unfortunately it seems those with authority to edit this page are more interested in maintaining the (politically-motivated & unscientific) status quo rather than improving the content. I gave up trying to fix it. Affinepplan (talk) 14:10, 28 August 2024 (UTC)

No one editor has authority with this article, which could really use your help in improving it if you aren't too discouraged Superb Owl (talk) 17:36, 28 August 2024 (UTC)

I attempted. but Wikipedia bylaws seem to heavily heavily favor preservation of existing content (no matter the quality) vs removing misinformation or biased content. I struggled paddling upstream for a bit but the bureaucratic obstacles seem pretty insurmountable so I have no motivation to continue. Affinepplan (talk) 17:38, 28 August 2024 (UTC)

I would recommend learning about (if you are up for it) inline flags (like 'better source needed') to point out specific issues. If those issues are not addressed in a week or so it should be ok to remove the text or improve it. That should hopefully make things easier and force a discussion or a fix by another editor who thinks the text is supported by a more reliable source and wants to find it. Superb Owl (talk) 17:44, 28 August 2024 (UTC)

> force a discussion

the problem lies here: I have been instructed that "consensus" must be reached when an edit is contentious. but if @Closed Limelike Curves simply chooses to never consent to my changes, then the edit can never be made. Affinepplan (talk) 19:34, 28 August 2024 (UTC)

I have been in your position before and that can be really frustrating. If more editors agree on a specific change, then changes can be made. Do you have any outstanding issues that you want to discuss that we haven't already? Superb Owl (talk) 20:20, 28 August 2024 (UTC)

@Superb Owl the incident @Affinepplan is referring to is a ban he received for 1) WP:SOCKPUPPETRY and 2) Edit-warring against a consensus of 3-4 other editors on the later-no-harm page over a several-month period. – Closed Limelike Curves (talk) 20:24, 28 August 2024 (UTC)

describing that situation as having "consensus of 3-4 other editors" is so misleading that I'd go so far as to call it a lie.

but I'm not going to get into this again. Affinepplan (talk) 20:25, 28 August 2024 (UTC)

Thanks for letting me know and we don't need to rehash old disputes here - let's just move forward if we can Superb Owl (talk) 20:26, 28 August 2024 (UTC)

Ranked Choice Voting is 'highly' vulnerable to the spoiler effect?

This lead asserts that RCV is similarly highly vulnerable to the spoiler effect, but both sources cited indicate that it is less vulnerable than plurality.
"In terms of the performance of the different election systems, we confirm the results of Merrill (1984, 1985, 1988) that in multicandidate elections run-off and sequential elimination systems perform far better than plurality elections, in that they are more likely to pick the Condorcet winner, and have a lower variance in their outcomes" Superb Owl (talk) 01:49, 28 August 2024 (UTC)

@User:Closed Limelike Curves, where are you getting the 'highly sensitive' from and why are we equating the two when all the sources clearly say we should not? Superb Owl (talk) 02:26, 28 August 2024 (UTC)

you would be correct. the authors of this content are not experts, have no formal training in this topic, and are relying on hearsay from other amateurs. Affinepplan (talk) 14:11, 28 August 2024 (UTC)

Hi; the sources indicate it is less vulnerable than plurality, but that both are "highly sensitive", with the phrase being taken from McGann 2002. The quote is included at the end of the reference.

Whether RCV-IRV turns out to reduce rates of spoiler effects is a very complicated question, though. The empirical research on the topic usually finds small or unclear differences. There are two reasons for this:

1. Unlike in McGann or Merrill's models, primary elections in the US winnow the field down to 2 major candidates, meaning that the US uses something like a de facto two-round system.
2. RCV-IRV tends to have high rates of spoiled votes and exhausted votes, which can offset the gains from wasted votes.

I think this is described a bit in the body, but I'll try and pull up sources to add more discussion on this to the lead. – Closed Limelike Curves (talk) 16:17, 28 August 2024 (UTC)

We need clear quotations from sources, because all 3 I have seen show a clear difference between the two and as it is written, it is a clear false equivalence Superb Owl (talk) 16:19, 28 August 2024 (UTC)

McGann 2002 is a result based entirely on a single synthetic statistical model of voter behavior and does not even attempt to study empirical real-world outcomes.

And even within the context of using simulations to study election rules, it's 22 years out of date. Statistical models have evolved over time (to become more relevant to real-world outcomes) and compute power has grown quite a bit. McGann 2002 uses a 1-dimensional mixture model over some normal & uniform preferences and derives all its results from only 1000 trials.

I understand that technically it's an "academic source" but the actual content of the paper is so hopelessly irrelevant to the point trying to made, especially in light of the fact that far better research exists just a Google Scholar search away, that it should not be considered as a proper citation for the content in this wiki page. Affinepplan (talk) 17:14, 28 August 2024 (UTC)

100% agree - we should not be relying heavily on primary sources of novel analysis - we really need secondary sources (like meta-analyses) that look at more than one study and summarize those findings Superb Owl (talk) 17:34, 28 August 2024 (UTC)

Hi @Superb Owl, I'm still in the process of editing this to introduce more nuance. This question turns out to be very complex, and depends a lot on a country's institutional setup. Many major misunderstandings come out of people importing papers talking about the alternative vote in the UK, then assuming it will have the same results in the American context. In the case of the United Kingdom, which has a "pure" plurality system with many different parties, you'd expect a substantial reduction in spoiler effects if you switched to the alternative vote. If the system you're comparing to is the two-round system, then as the papers I cited discuss, the differences are typically very small.

What creates a lot of complexity is that first-past-the-post often evolves something like a two-round system naturally, by developing a two-party system which narrows the race down to two contenders. This is most obvious in the United States, where partisan primaries replace the first traditional "winnowing" round of a two-round system. Instead of running in the general, candidates here typically seek the nomination of one of the two major parties (think Bernie Sanders in 2016). Similarly, if voters engage in strategic voting, there's usually very strong agreement between the results under IRV and under plurality voting. This is because under FPP, many voters (in the United States) adopt a strategy of watching the polls, then abandoning their favorite candidate if they're in last place and moving to support the next-best choice. This procedure mimics IRV automatically.

IRV may have some minor residual effects thanks to eliminating the wasted votes soaked up by third-parties (which typically pull about 5% of the vote). However, this is generally offset by the disenfranchisement of many voters through ballot spoilage, which also tends to hover at around 5% in RCV-IRV elections, and also by the substantial problem of exhausted votes. (This also doesn't include the common problems of reversed or cured ballots, where it can be ambiguous how a voter actually meant for their ballot to be interpreted.)

I apologize for not being very clear in explaining this, and for not communicating this properly via the talk page. I totally understand why it might have felt like I was ignoring your concerns and trying to remove information about how RCV-IRV can have advantages over FPP in some contexts. I do plan to incorporate a broader discussion with all this information when I have the time. – Closed Limelike Curves (talk) 22:58, 30 August 2024 (UTC)

Inclusion of the primary system might be helpful, but unlike in a runoff, there are still more than two candidates by the last round.
Another concern is that the research does not sound like it is definitive on this issue and therefore we should also be very careful when making strong claims. If we can make weaker claims that would be helpful, in my opinion. And the language needs to be clearer and using more secondary sources (less one-off studies) Superb Owl (talk) 23:11, 30 August 2024 (UTC)

> if voters engage in strategic voting [IRV agrees with Plurality]

this is very plausibly / probably true of any single winner rule (under strategic behavior most will all agree, most of the time) so framing it as specifically a characteristic of IRV, solely because IRV is one of the only non-standard voting rules for which data exists on which to observe such behavior, is pretty disingenuous. Affinepplan (talk) 15:11, 31 August 2024 (UTC)

Authorship and Technical flag

Noting that @Closed Limelike Curves has 88% authorship of this article currently. I think we could use some fresh perspectives, especially less technical ones to try and make this article more easily understandable for those of us not well-versed in statistics. Superb Owl (talk) 08:22, 28 August 2024 (UTC)

Hi, I think part of this is related to my having moved lots of the material from an older page called vote splitting (which is now a redirect). I'm a bit confused as to what you think is technical here, or how it's related to statistics (which I don't think is related to this page's topic. I'd definitely welcome any improvements on this front, though, and there's also a need for more citations in some sections. Happy for any help :) – Closed Limelike Curves (talk) 20:22, 28 August 2024 (UTC)

ah ok yeah I just noticed that Vote splitting was merged here and into another article. Just wanted to note it in case we needed it looked at from more angles, which is always nice Superb Owl (talk) 20:25, 28 August 2024 (UTC)