Page MenuHomePhabricator

Rank "eliminated/not elected" candidates too in Single Transferable Vote
Open, Needs TriagePublic

Description

Currently, SecurePoll only ranks elected candidates in STV elections. See the screenshot below

Screen Shot 2021-08-04 at 4.53.30 PM.png (1×2 px, 311 KB)

It's useful to rank all candidates, because best losers may be considered as reserves.

And it should be very easy to do so. The sooner one candidate gets elected, the higher their rank; and the sooner one candidate gets eliminated, the lower their rank. So you should generate the final ranking both from top to down and from bottom to up.

fwiw the eliminated list is also ranked. It's only the unelected (seats all filled w/candidates remaining) candidates that get tossed into the back of that list.

All these "unelected" candidates also get eliminated if you continue the count. In each round of counting, you should either elect or eliminate at least one candidate. Since all the available seats have been filled, the remaining candidates can only get eliminated one by one, if you continue the counting process.

The edge case (which if you're referring to here sorry for retreading the ground) would be when multiple candidates are elected/eliminated at the same step. We don't distinguish those atm.

You can simply add an option to the software, so that a random selection gets possible. The election admin (the one who does the configurations when creating the poll) may choose to turn it off or on. At fawiki, we use another ranked voting method called Schulze, and we do the potential random selection manually (using page_random field from page table in database, to be more specific), but it would be nice if the software undertakes this task too.

Event Timeline

4nn1l2 renamed this task from Rank "eliminated/not elected" candidates too in Single Tranferable Vote to Rank "eliminated/not elected" candidates too in Single Transferable Vote.Aug 5 2021, 3:26 AM

Given we are pressed for time on this project right now, we probably won't be making that change at the moment.

I just wanted to make it clear that determining the best loser is critical even for the board elections. Suppose you have created a poll for selecting 3 seats among 20 candidates. It takes almost three weeks to cast ballots, scrutinize votes, tally them, and finally announce the results (18 August – 7 September). What if a candidate leaves the projects, gets disqualified, or even passes away in this period, and that candidate gets elected? Does any plan B exist in this scenario? You can't remove them in the middle of the elections. But the elections committee may choose to give the available seat to the best loser (i.e., the candidate ranked 4th).

Given we are pressed for time on this project right now, we probably won't be making that change at the moment.

I just wanted to make it clear that determining the best loser is critical even for the board elections. Suppose you have created a poll for selecting 3 seats among 20 candidates. It takes almost three weeks to cast ballots, scrutinize votes, tally them, and finally announce the results (18 August – 7 September). What if a candidate leaves the projects, gets disqualified, or even passes away in this period, and that candidate gets elected? Does any plan B exist in this scenario? You can't remove them in the middle of the elections. But the elections committee may choose to give the available seat to the best loser (i.e., the candidate ranked 4th).

@4nn1l2 I understand the concern. Like Tran said in T287806#7257828 this is already kind of happening because the Eliminated list is also ranked. Yes, ideally we would be ranking them all but to make a major change to the tallying algorithm while the Board election is in progress seems like a very risky proposition. We would like to prioritize this when there are no major elections in flights and it can be safely tested.

I am not sure this is actually something that works with Meek STV as I understand it. If someone withdraws after the election starts, the ballots of those who voted for that person need to be redistributed, as I understand it. See my comment in T309416#8773333 for more info.

@Risker this task might solve one of your concerns raised on that task at least

@jrbs is right. Our goal of proportional representation is proper, and the math of how you achieve that means that there is no single well-defined ranking of individual candidates. What matters is whether the set of winners does a good job of representing the voters. You can't just replace any member with the results of calculations which already assume they were elected.

See my comments on https://phabricator.wikimedia.org/T309416 also.