Loomio
Sat 24 Mar 2018 2:10AM

Critique of STAR (single-winner)

WW William WAUGH Public Seen by 300

Warren B. Smith's mathematical critique of STAR compared to plain Range Voting (don't know whether this has been treated in this forum already).

MF

Mark Frohnmayer Sat 24 Mar 2018 2:16AM

Rather than discuss this on this Loomio, I'd encourage folks to chime in on the related Election Science mailing list thread here: https://groups.google.com/forum/m/#!topic/electionscience/Gj1UPgmL5B0 - Jameson Quinn, Clay, Warren and others are discussing the claims made.

ED

Emily Dempsey Sat 24 Mar 2018 6:15AM

I'm not familiar with Google Groups and am not immediately sure how to reply.

That aside, I also confess to being nervous about stepping into the fray with multiple people I don't know as a totally unknown female in a mathematical conversation. (I don't anticipate being taken very seriously, especially given Warren's tone in the articles of his I've read - and the conversation above thus far.)

Anyway, after reading through the article, I do have an interesting cognitive dissonance that I do want to examine. That is: for each of the scenarios he presents given the proposed voter preferences, I feel simultaneously satisfied with the results of each election given the actual votes that were cast, but unhappy with the fact that some voters in those elections lied (in the Strongly Insincere sense!) in order to cause better results from their point of view.

That said, my instinct is that this dissonance can only happen when the electorate has cyclical preferences, and I'm partly wondering how likely cyclical preferences in an electorate is in the real world. And, curious how other voting systems perform when the electorate has cyclical preferences.

ED

Emily Dempsey Sat 24 Mar 2018 6:31AM

Also: He doesn't provide much support for Range+Runoff being better than STAR; he mostly just asserts it as true, and I don't buy it.

Range+Runoff would DEFINITELY be more susceptible to Weak Insincerity than STAR is. Both Range+Runoff and STAR have some incentive for a voter to maybe bury a candidate they kinda like under a candidate they think their favorite is more likely to beat in the runoff. It's somewhat risky in both systems because you could risk the candidate you gave an artifically high score to beating your favorite candidate into the runoff.
The difference is that in STAR, burying is also disincentivized by the fact that the voter's vote in the runoff is determined by the same ballot, so a voter who buried a candidate risks voting for the "wrong" candidate if the runoff ends up between the buried candidate and the candidate they buried with, whereas Range+Runoff has no additional disincentive for it; the voter can just cast a new ballot that totally swaps their stated preference between the two candidates in the runoff if their burying plan didn't pan out.

My instinct is that Range+Runoff would be about equally susceptible to Strong Insincerity? My gut feels very sure about this claim, but I'm having trouble thinking it through right now. (brain needs sleep...)

I feel super not sure sure about Restrictive and Expansive Sincerity but I will admit I don't feel terribly concerned about either of those in general.

ED

Emily Dempsey Sat 24 Mar 2018 6:57AM

(I really, really should be getting to bed, but just to share what I've googled so far:

There might be some academic papers that are relevant but I am not prepared to spring $40 for them in my current state.

The abstract for this one acknowledges that cyclical preferences in real-world electorates has not been found by studies that have gone looking for it, but sounds like this paper found it in one Danish election. https://link.springer.com/article/10.1023/A:1010304729545

The abstract for this article claims that in the 4 elections they looked at, there were clear Condorcet winners, and "collective preferences were ideologically constrained" (which.. I feel like I will need to read the article to get a good handle on what they mean by that second part, but there being a clear Condorcet winner means no cyclical preferences, yeah? Or at least, no cyclical majority?) https://www.sciencedirect.com/science/article/pii/0261379494900078

Anyway. Actually off to bed now. But the next thing I wanna investigate is how Range and Range+Runoff perform under the circumstances Warren presents cuz my gut anticipates they do wonky things when there's a cyclical majority too.)