Instant-runoff voting: Difference between revisions

<br />

=== Variants ===

See the [[Equal-ranking methods in IRV]] article. IRV can be done with equal ranking allowed. The two main ways of doing this are either fractional (split the voter's ballot equally between all of their highest-ranked candidates that are ranked equally (3 candidates ranked 1st each get 1/3rd of a vote)), or whole votes (give each highest-equally-ranked candidate one vote (3 candidates get 1 vote each and 3 votes total)).

With whole votes equal-ranking, there are two ways to find a winner (which give the same result in standard IRV but differ for whole votes): either eliminate candidates until only two remain, and declare the one with more votes the winner, or eliminate candidates until one or more candidates are supported by a majority of active ballots, and then elect the candidate with the largest majority. Some have argued<ref>[https://www.reddit.com/r/EndFPTP/comments/e6bt6s/proportionality_failure_in_stv_with_equalranks/f9s5yno/?context=3]</ref> that in order to limit opportunities for pushover strategy with whole votes, a ballot that equally ranks candidates should be allowed to help those candidates win, but not prevent those candidates from getting eliminated.

IRV is equivalent to [[runoff voting]] (supposing no change in preferences) when there are 3 or fewer candidates. This is used to argue both for and against it; advocates claim it is cheaper and easier for the voters to vote once, while opponents argue that a delayed runoff actually gives voters a second look into the candidates in the runoff, potentially improving the quality of their decision-making, and that because ranking candidates is harder than picking one candidate, that runoff voting is actually easier for voters. Note that though IRV is called instant runoff, this is more because it elects a candidate who could win or tie a runoff ([[pairwise beat]] or tie) against at least one other candidate, rather than because it is equivalent to runoff voting in all cases.

One simple way to modify IRV to address many of the issues IRV opponents have without changing IRV fundamentally is to allow voters to approve candidates (using an [[approval threshold]]). If there are any majority-approved candidates, elect the most-approved of them, otherwise run IRV. Even if voters [[Favorite Betrayal|Favorite Betray]], they can still approve their honest favorite, giving that candidate a chance to still win. In addition, this allows voters to better avert the [[center squeeze effect]]. The standard argument made by IRV advocates against [[Approval voting]], that it fails [[later-no-harm]], has little to no relevance to this modification, since voters seeking to avoid hurting their favorite candidates' chances of winning in the approval round can simply refrain from approving anyone, forcing the election to run under IRV rules.

Though IRV is often praised for passing [[later-no-harm]], which is claimed to encourage voters to rank all of their preferences, it doesn't tend to use as much of the information provided by the voters as other ranked methods, such as [[Condorcet methods]]. This is a less extreme analog to how [[First Past the Post electoral system|first past the post]] technically passes [[later-no-harm]] by ignoring later preferences altogether. So IRV's [[later-no-harm]] compliance has to be evaluated in context of the other criteria it fails due to using less information than other methods - that is, there may be ambiguity to how much IRV is truly protecting a voter's interests by not using their later-preference information at all.

From this perspective, the main criticism of IRV is essentially that it determines who the worst candidate is using 1st choices. In so doing, it uses [[first past the post]] to determine who to eliminate, and ignores most of the voter's ballot each round. This ignoring of most of the ballot is what gives IRV its [[later-no-help]] and [[later-no-harm]] properties, but also leads to its vulnerability to [[center squeeze]].

In contrast, [[Nanson's method]] is a method that ''does'' examine the whole ballot for each voter for each round , and whose logic is otherwise the same as IRV. It passes [[Condorcet criterion|Condorcet]] but fails both [[later-no-help]] and [[later-no-harm]]. Both Nanson and IRV are nonmonotone, so the lack of monotonicity can't be attributed to IRV not looking at the whole ballot.

Note that when the top candidate doesn't have a majority, but the top two candidates each have over 1/3rd of the active votes (i.e. they combinedly have over 2/3rds), they are guaranteed to be the two final remaining candidates in IRV, so all other candidates can be eliminated (or equivalently, the pairwise matchup between the two can be tallied) to find the result.