[EM] The Frequency of Condorcet Winners in Real Non-Political Elections

[Joshua Boehme]

One of the interesting takeaways for me is something I haven't seen mentioned yet: IRV fails to pick a weak CW 6% of the time, conditional on one existing (which is almost always the case unless the number of voters is tiny).


On 3/31/24 20:07, Andrew Myers wrote:
> Here is a paper I recently presented at the 61st Public Choice Society Conference <https://publicchoicesociety.org/conference/2024>, containing results from analysis of tens of thousands of polls run on the CIVS system since 2003.
> 
> The Frequency of Condorcet Winners in Real Non-Political Elections <https://www.cs.cornell.edu/andru/papers/civs24/>
> 
> Andrew C. Myers
> 
> Condorcet-consistent voting is attractive because it follows the principle that the will of a majority of voters should not be denied. However, it is in general possible that there is no Condorcet winner: a cycle of candidates might exist in which each candidate is preferred to the next. The possibility that such a cycle occurs, and the uncertainty about how to handle it, have been an obstacle to the adoption of Condorcet methods. This paper reports on the experience from CIVS, a long-running open-source voting service that supports Condorcet-consistent voting methods. Over a period of about twenty years, users have conducted tens of thousands of polls using CIVS, including many with real-world consequences. During this time, CIVS has thus accumulated probably the largest existing corpus of data about how Condorcet voting functions in practice. CIVS supports multiple completion methods for handling cycles, but the data show that it usually does not matter which completion 
> method is used, because there is rarely a cycle in polls with a large enough number of voters.
> 
> Full text here: <https://www.cs.cornell.edu/andru/papers/civs24/> <https://www.cs.cornell.edu/andru/papers/civs24/>
> 
> Cheers,
> 
> -- Andrew
> 
> 
> 
> 
> 
> ----
> Election-Methods mailing list - see https://electorama.com/em for list info