User talk:Adamw/Draft/Board Election analysis

Criteria

The page is in parts rather speculative. It would be interesting to check whether the S/O/N thingy (as implemented by SecurePoll, which enforces some unwritten rules) satisfies some of the basic criteria, such as monotonicity, plurality, majority and Condorcet (it's easiest to exhibit counterexamples to disprove that it does). Nemo 11:39, 10 January 2016 (UTC)

Yes, thank you! I rewrote the essay just now, making it much more rigourous thanks to encouragement from @Ewulczyn (WMF): and @EpochFail:. Please give it another read if you find the time.

The majority criterion is weaker than Condorcet, so I'll skip it, but plurality should go on the list--it seems we failed. Adamw (talk) 12:44, 10 January 2016 (UTC)

You misunderstood the Condorcet criterion, that's not what it says. As for the other two you mentioned, your text is an attempt at counter-proof rather than a counterexample. [1] seems to be a good summary in 7 pages. If it doesn't help, I can produce an example counterexample later this week. Nemo 08:37, 11 January 2016 (UTC)

I'm disappointingly rusty at math, so please bear with me. I attempted to give counterexamples, a proof wouldn't have made sense in this case because the voting method we used might sometimes satisfy the criteria. As you suggest, my task here is to show that given our specific results, the criteria were not satisfied. Any help refining the examples would be appreciated, of course.

I see what you mean about the Condorcet criterion, now. We would have to use the same flawed S/(S+O) ranking to evaluate each two-candidate race, so we would have actually satisfied the criterion using this skewed version of "preferred". I've removed the section as irrelevant. Adamw (talk) 02:00, 12 January 2016 (UTC)

Controversial candidates

The results show the system was very effective at weeding out the controversial candidates, but it was the candidates who were controversial to the voters who were weeded out, and that included two candidates who had very high support levels as well as a significant number of opposes. Candidates who did well under this system were those with a lot of support and very few opposes. Imagine how this system would work if applied to US presidential elections...... For the community on this occasion there was clearly a significant anti incumbency sentiment and both incumbents were voted off by this system. This formula gives extra weight to oppose votes, support minus oppose would have reduced the anti incumbency factor and given all votes the same value. WereSpielChequers (talk) 05:01, 30 January 2016 (UTC)

@WereSpielChequers: I'm curious why you say that the support percentage formula was effective at weeding out the controversial candidates. A plain approval tally (total support votes) would have also weeded out the controversial candidates, because the controversial candidates all had low approval scores as well. Meanwhile, there's no clear reason to call Sefidari and Ayers controversial--or maybe I'm misunderstanding who you're referring to? Sefidari's and Ayers's support percentage scores were within 2% of Vrandečić's, but Sefidari had an approval score more than 10% higher (42% vs 32% for Vrandečić) which seems to imply the opposite of what you say. There was no significant difference in any of the top 5 candidates' support percentages, so our formula arbitrarily selected two people with significantly fewer support votes. Adamw (talk) 05:57, 30 January 2016 (UTC)

I'm counting candidates as controversial if there are votes against them. The current system enabled a relatively small proportion of voters to vote off two sitting members. If it had just been down to support votes then we'd have known how popular/unpopular candidates were, but not how controversial they were. WereSpielChequers (talk) 19:51, 3 February 2016 (UTC)

Obscure candidates

The report raised the theoretical possibility of an obscure candidate who almost everyone was neutral about getting a high percentage support. This is theoretically possible, but only with a non human electorate. If everyone was voting on the basis of only voting oppose where they actually saw a problem with a candidate then it would theoretically be possible for someone to be elected on 100% support from one support, no opposes and almost total indifference, however with a human electorate we can expect, as we saw in this election, that the minority who oppose all but their favoured candidate will be sufficient to defeat any candidate who lacked significant support. We can safely leave that theoretical anomaly to some future date where we haven't only discovered intelligent life on other planets, but we are integrating wikipedia with extrasolar equivalents. WereSpielChequers (talk) 05:26, 30 January 2016 (UTC)

The degenerate case was just an attempt to demonstrate math using simple numbers. The 2015 Board of Trustees elections give the concrete example. Adamw (talk) 06:00, 30 January 2016 (UTC)

All the candidates who didn't withdraw got over 500 opposes, this supports my point that any candidate however obscure will get a significant oppose. An obscure inoffensive candidate with few or no votes would still get hundreds of opposes. The Math that shows a candidate with 4 supports no opposes and thousands of neutrals is showing a result that would only be possible with a different electorate. WereSpielChequers (talk)

celebrity privilege

I think that it is important not to give unfair advantage to people widely known (often the incumbents) and sometimes relatively disliked. Known people get a lot of random support votes. Thus, specifically targeted "no" votes are also important, and leaving it just to counting support does not make much sense. I believe that a simple, plain vanilla system of S/N/O, with a result of net outcome (S minus O) is really good for that: it does not give unnecessary unequal weight to negative votes, but still allows to oppose a candidate. It is, in principle, giving everyone votes with the weights of -1, 0, and 1. I really don't like complicated formulas for voting, as I believe that voters should very easily understand the mechanics. This is important for democracy, to know what you're going :) Pundit (talk) 14:01, 30 January 2016 (UTC)

Net support (S - O) is still too complicated, I believe. It allows for tactical voting, and note that -1, 0, and 1 is equivalent to giving the options 0, 1, or 2 points to each candidate. It seems like that will penalize the "nice" voters: people who use an oppose vote for nonpreferred candidates will have twice the voting power relative to people who use neutral votes instead. I don't see a problem with offering an Oppose option, but it should not factor into the ranking. Instead, we could have a second criteria like "Candidates with more than 50% O/S are disqualified for being controversial."

It's simplest to use straight approval voting (tally of support votes), and do analysis of each election to keep track of how our outcomes compare to theoretically more ideal social choice algorithms.

I'm not sure what you mean by "random" votes, but I do understand the incumbent advantage you're describing. Having more seats up for election would help... Maybe you'd be willing to test your theory against the raw data, to see if you find a democratic way to counteract the incumbent advantage? I'd also like to play with the ballot dump, to figure out how many voted tactically, and come up with a few different measures for error in representation, given our outcomes. Feel free to clone the Jupyter notebook I've posted on Github. Adamw (talk) 02:06, 31 January 2016 (UTC)

Well, leaving an option to "oppose" and making it effectively useless (I don't recall any elections where your system would bring different results than just summing up supports) is not reasonable, too. I really don't think that summing up support and ignoring the opposition, unless it is gargantuan, is the way to go. How is simply deducting the opposing votes from the total support "too complicated"? I would certainly hope our voters are able to comprehend mathematics at this level. Also, keep in mind that while random support votes will happen (voting in support just because a name sounds familiar), random oppose votes will be rare. Thus, oppose votes will carry an informed choice. Pundit (talk) 22:00, 31 January 2016 (UTC)