Favorite betrayal criterion: Difference between revisions
Correct citation (copy-pasted the wrong one, sorry!)
(→Notes) |
(Correct citation (copy-pasted the wrong one, sorry!)) |
||
(8 intermediate revisions by 4 users not shown) | |||
Line 1:
{{wikipedia|Draft:Favorite betrayal criterion}}
The '''favorite betrayal criterion'''
It is passed by [[Approval voting]], [[Range voting]], and [[Majority Judgment]]. All these are examples of [[cardinal voting systems]], though some voters may be reluctant to have a compromise candidate tied with their favorite candidate on their respective ballots.
On the other hand, most [[Ranked voting systems|ordinal voting systems]] do not pass this criterion. For instance, [[Borda Count]], [[Copeland's method]], [[Instant runoff voting]] (IRV, known in the UK as the Alternative Vote), [[Kemeny-Young method|Kemeny-Young]], [[Minimax Condorcet]], [[Ranked Pairs]], and [[Schulze method|Schulze]] all fail this criterion. A few ordinal methods, like [[weighted positional method|Antiplurality]], pass it. Some [[Condorcet methods]] pass it when combined with the [[tied at the top]] rule, though this means they may not be Condorcet-efficient when some voters equally rank multiple candidates.
It is also failed by [[
== Definition ==
It is defined as follows:
:A [[voting system]] satisfies the Favorite Betrayal Criterion (FBC) if there do not exist situations where a voter is only able to obtain a more preferred outcome (i.e. the election of a candidate that he or she prefers to the current winner) by insincerely listing another candidate ahead of his or her sincere favorite.<ref name="small">Alex Small, “Geometric construction of voting methods that protect voters’ first choices,” arXiv:1008.4331 (August 22, 2010), http://arxiv.org/abs/1008.4331.</ref>
The criterion permits the strategy of insincerely ranking another candidate equal to one's favorite. A related but stronger criterion, the ''strong favorite betrayal criterion'' disallows this.<ref name="small" />
==Complying methods==▼
[[Approval voting]], [[range voting]], [[Majority Judgment]], [[MMPO|MinMax(pairwise opposition)]], [[MCA]] (except MCA-A and some versions of MCA-R), [[MAMPO]], [[Majority Defeat Disqualification Approval]], and [[Improved Condorcet Approval]] comply with the favorite betrayal criterion, as
▲==Complying methods==
[[Borda count]], [[plurality voting]], [[Condorcet criterion|Condorcet methods]] (except for Improved Condorcet methods, such as Kevin Venzke's [[ICA]], and Chris Benham's ICT) and [[instant-runoff voting]] do not comply.<ref>{{cite web|url=http://lists.electorama.com/pipermail/election-methods-electorama.com/2005-May/081273.html|title=WV methods fail FBC with 3 candidates|website=Election-methods mailing list archives|date=2005-05-13|last=Venzke|first=K.}}</ref><ref>{{cite web|url=https://munsterhjelm.no/km/yahoo_lists_archive/RangeVoting/web/2005-October/msg00059.html|title=I told why CC is incompatible with FBC. Why the continuing debate?|date=2005-10-18|last=Ossipoff|first=M.|website=RangeVoting Yahoo list mirror}}</ref>
▲[[Approval voting]], [[range voting]], [[Majority Judgment]], [[MMPO|MinMax(pairwise opposition)]], [[MCA]] (except MCA-A and some versions of MCA-R), [[MAMPO]], and [[Improved Condorcet Approval]] comply with the favorite betrayal criterion, as do ICT and [[Symmetrical ICT]].
== Examples ==
Line 383 ⟶ 364:
== Stronger forms of the criterion ==
FBC simply requires that for a given election, a voter always has some kind of [[strategy]] they can use to vote in such a way that they most support their favorite candidate. However, this means that some voting methods that fail FB can allow a voter to benefit by doing FB, even though they didn't actually have to. For example, several voting methods which pass FBC because they allow a voter to protect themselves by equally ranking multiple candidates 1st (implying that the voter has a simple way to always avoid FB i.e. equal-ranking, as opposed to some FBC-compliant voting methods where the non-FB strategy may be opaque or difficult to figure out and thus less useful for avoiding FB) are like this. [[Score voting]] passes a stronger form of FBC, which says that voters can never benefit by doing FB i.e. there is no possible strategy involving FB that can benefit a voter.<ref>{{Cite web|url=https://rangevoting.org/FBCexecSumm.html|title=RangeVoting.org - Favorite betrayal (executive summary)|last=|first=|date=|website=rangevoting.org|url-status=live|archive-url=|archive-date=|access-date=2020-05-14|quote=We've come a long way since the days when range and approval voting were the only known methods in which betraying your favorite is strategically avoidable. Now many other methods also are known with that "FBC property." [...] However, it appears Range and Approval satisfy FBC in a stronger and more obvious sense than these other methods. Specifically, with Range and Approval, betraying your favorite simply never is useful. With the other methods it can be strategically useful (cause X to win instead of Y, where the betrayers prefer X) but if so there is always a way to get the same effect (i.e. make X win) by some other dishonest vote not involving favorite betrayal.}}</ref>
== Notes ==
=== Least favorite betrayal criterion ===
A criterion related to FB is whether or not a voter can be hurt by giving no support to their least favorite candidates. Approval and Score voting pass this criterion, since if never benefits you to help your least favorite beat other candidates on approvals/points.
=== Criterion failure rates ===
Though many voting methods fail FB, they tend to decrease the incentive to do FB relative to [[FPTP]]. Example for [[Condorcet]]: <blockquote>25 A>B
26 B
49 C </blockquote>In FPTP, C would win with 49 votes to B's 26 and A's 25, so the A-top voters would have to vote B>A instead (i.e. put B as their 1st choice) to ensure B wins with 51 votes, rather than C with 49. In Condorcet methods, this isn't necessary, since the pairwise table is: ▼
▲In FPTP, C would win with 49 votes to B's 26 and A's 25, so the A-top voters would have to vote B>A instead (i.e. put B as their 1st choice) to ensure B wins with 51 votes, rather than C with 49. In Condorcet methods, this isn't necessary, since the pairwise table is:
{| class="wikitable"
|+Wins are bolded
Line 423 ⟶ 401:
and B is the [[Condorcet winner]]. See also the [[chicken dilemma]].
<blockquote>26 A>B
25 B
49 C</blockquote>A [[pairwise]] beats B beats C beats A, so there is a [[cycle]]. C would win in several methods, such as some [[Condorcet methods]], and [[IRV]]. To prevent A-top voters from having to vote B>A to make B the [[CW]]/[[majority criterion]] winner, it can essentially be observed that they have the incentive to do so, and if they do this, nobody else has incentive/ability to vote differently to get a different outcome, so B would automatically win.
A beats B beats C beats A. To prevent A-top voters from voting B>A to make B the CW, it can essentially be observed that they have the incentive to do so, and if they do this, nobody else has incentive to vote differently, so B would automatically win. However, if the 49 C voters instead had voted C>A, then they make it so that now the voting method recognizes the C-top voters have an incentive to do FB to elect A rather than B. So a cycle would be formed all over again. This means that this trick doesn't always work. See [[Algorithmic Asset Voting]]. ▼
▲
== Further reading ==
|