Defeat strength: Difference between revisions

(Add some articles to Category:Voting theory metrics)
(→‎Standard: remove example methods—could easily confuse people into thinking Ranked Pairs has to use margins, or Schulze has to use winning votes)
 
(9 intermediate revisions by 3 users not shown)
Line 1: Line 1:
[[Condorcet methods]] require the resolution of [[cycles]]. Typically, this is done by ignoring "weak" pairwise defeats in favor of "strong" ones. The metric used to distinguish these is called '''defeat strength'''.
[[Condorcet methods]] require the resolution of [[Condorcet cycle|cycles]]. Typically, this is done by ignoring "weak" pairwise defeats in favor of "strong" ones. The metric used to distinguish these is called '''defeat strength'''.


Most election experts agree that, out of the standard ways to measure defeat strength, winning votes are the best, with margins in second.
Measures of defeat strength for a pairwise winner W over a pairwise loser L include:

== List of measures ==

=== Standard ===
* '''winning votes (wv)''' = number of votes for W>L if greater than the number of votes for L>W, otherwise zero.
* '''margins''' = (number of votes for W>L) - (number of votes for L>W)
** Gives more strategic incentive than wv but may be easier to understand
* '''pairwise (non)opposition''': number of votes for W≥L, or equivalently 1 - votes for L > W
** Strong intuitive appeal (pick the candidate opposed by the least voters)
** Example method: [[MMPO]]
*** Gives even less strategic incentive than wv (satisfies later-no-help and favorite betrayal).
*** Violates [[plurality criterion]]. A complete unknown can win with no real support, just because everyone forgot to rank them on their ballot.
* '''Relative margins''': Defeat strength = margin ÷ (votes for W≠L)

=== Cardinal ===

* '''Approval-based support''' = ''no. of voters approving of the winner but not of the loser of the defeat.'' Gives special influence to preferences which cross the approval cutoff and thus helps diminish certain strategies. Useful when one assumes that only these voters will support the corresponding "majority complaint"
* '''Cardinal rated''' strength = ''sum of difference in the candidates' cardinal ratings on all ballots which rate the winner over the loser of the defeat.'' Even more strategy-resistant than wv, but involves interpersonal comparisons of cardinal ratings.
* '''Winning approval''' = ''approval score of the winner of the defeat.'' Using this as defeat strength leaves only one immune candidate: the least approved of those who beat all more approved ones. Similar for other scores.

== Kinds of defeats ==

* '''(Pairwise)''' '''defeat''' = ''more voters prefer A over B than B over A''
* '''Majority-strength defeat''' = ''pairwise defeat which has a wv-strength of more than half the no. of voters.'' Using only such defeats can reduce incentive to truncate by reducing the likelihood that additional preferences will harm earlier ones. Voters adding a preference can create a majority-strength win, but they can't reverse the direction of one.

== References ==
<references />


* '''winning votes''': Defeat strength = votes for W>L
* '''margins''': Defeat strength = (votes for W>L) - (votes for L>W)
* '''relative margins''': Defeat strength = margin &divide; (votes for W&ne;L)
[[Category:Voting theory metrics]]
[[Category:Voting theory metrics]]
[[Category:Condorcet-related concepts]]

Latest revision as of 15:26, 26 June 2024

Condorcet methods require the resolution of cycles. Typically, this is done by ignoring "weak" pairwise defeats in favor of "strong" ones. The metric used to distinguish these is called defeat strength.

Most election experts agree that, out of the standard ways to measure defeat strength, winning votes are the best, with margins in second.

List of measures

Standard

  • winning votes (wv) = number of votes for W>L if greater than the number of votes for L>W, otherwise zero.
  • margins = (number of votes for W>L) - (number of votes for L>W)
    • Gives more strategic incentive than wv but may be easier to understand
  • pairwise (non)opposition: number of votes for W≥L, or equivalently 1 - votes for L > W
    • Strong intuitive appeal (pick the candidate opposed by the least voters)
    • Example method: MMPO
      • Gives even less strategic incentive than wv (satisfies later-no-help and favorite betrayal).
      • Violates plurality criterion. A complete unknown can win with no real support, just because everyone forgot to rank them on their ballot.
  • Relative margins: Defeat strength = margin ÷ (votes for W≠L)

Cardinal

  • Approval-based support = no. of voters approving of the winner but not of the loser of the defeat. Gives special influence to preferences which cross the approval cutoff and thus helps diminish certain strategies. Useful when one assumes that only these voters will support the corresponding "majority complaint"
  • Cardinal rated strength = sum of difference in the candidates' cardinal ratings on all ballots which rate the winner over the loser of the defeat. Even more strategy-resistant than wv, but involves interpersonal comparisons of cardinal ratings.
  • Winning approval = approval score of the winner of the defeat. Using this as defeat strength leaves only one immune candidate: the least approved of those who beat all more approved ones. Similar for other scores.

Kinds of defeats

  • (Pairwise) defeat = more voters prefer A over B than B over A
  • Majority-strength defeat = pairwise defeat which has a wv-strength of more than half the no. of voters. Using only such defeats can reduce incentive to truncate by reducing the likelihood that additional preferences will harm earlier ones. Voters adding a preference can create a majority-strength win, but they can't reverse the direction of one.

References