Add support for Multi-Winner IRV Tallying

dsernst commented 1 month ago

Goal is to extend the existing Ranked Choice ballot design for voters, but to adapt the tallying algorithm to support declaring multiple winners, not just a single one.

Specific use-case:

There are 5 open nomination spots on the voting org's Executive Committee. 6 people have signed up as candidates.

The org's bylaws require all elections to use Ranked Choice Voting. Goal is to have voters fill out a single RCV ballot for the 6 candidates, but then be able to declare multiple winners.

Current Level of SIV Support

SIV does support ranked choice voting, with a nice front-end UI for voters.

But currently, the only RCV tallying SIV supports is single-winner IRV (Hare algorithm — eliminate lowest until someone crosses 50%.).

This PR would extend that to multiple winners.

Proportionality:

One unclear question is how important it is for the algorithm to respect Proportionality?

Without Proportionality, you get 5 clones of someone 51% of people like, rather than 3 of them and 2 for the 49%.

Must Do's before merging

[x] unit tests for tallying algorithm
[x] support to add number_of_winners to ballot schema in election-admin editor
- [ ] nice-to-have: Document the new feature below the editor under "Advanced Features".
[x] working tallying algorithm to calculate multiple winners
[x] front-end support on Election Status page to display multiple winners

vercel[bot] commented 1 month ago

@joshuaherr is attempting to deploy a commit to the SIV Team on Vercel.

A member of the Team first needs to authorize it.

CLAassistant commented 1 month ago

All committers have signed the CLA.

vercel[bot] commented 1 month ago

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
siv	✅ Ready (Inspect)	Visit Preview	Sep 27, 2024 4:28am

dsernst commented 1 month ago

Confirmed the unit tests (both old and new) all pass:

dsernst commented 1 month ago

How the election status page currently displays this (number_of_winners set to 5):

dsernst commented 1 month ago

[ ] It would be nice if we included even just a comment or something with a link to some reference documentation spec for the specific tallying algorithm this is implementing.

dsernst commented 1 month ago

Going to merge & deploy this for now, as all the Must Do tasks are complete.

But also noting down these additional nice-to-haves:

[ ] Lack of Proportionality support: The algorithm as implemented, I'm afraid sadly may not be proportional either. This page https://fairvote.org/archives/multi_winner_rcv_example/ goes over the Single-Transferable Vote tallying algorithm for RCV... The biggest difference is STV's "fractional weighting" feature, once a ballot helps to elect a winner, those particular contributed-to-the-winner-ballots get significantly downweighted (e.g. to only 8%, in Fairvote's example). Whereas this PR's algorithm currently keeps them all at 100%.
[ ] This algorithm grabs as many winners as possible in a single round, rather than limiting it to just one per round (as STV does, for example). Unclear exactly how much of a difference this makes. But again, wanted to highlight, and thought this could be another useful thing to cover with more unit tests.

UI Improvements:

Somewhat related to the above point, I found it a tad hard in the front-end UI to follow what's happening exactly. See this example (screenshot above), for example: https://siv-git-multiwinner-rcv-sivteam.vercel.app/election/1727427598586.

Some improvements I thought that might make it a lot easier to follow:

[ ] a. Show the question's number_of_winners parameter on the results page.
[ ] b. Show threshold % needed in the UI for votes needed to qualify as a winner: 1 / (num_winners + 1) (wiki: Droop Quota)
[ ] c. Highlight which candidates cross the threshold as winners specifically in the round where they do so
[ ] The "Tie among bottom candidate" note in the UI doesn't make any sense in the example link above, because all 4 of the candidates listed in Round 1 are actually ascending as winners, not getting eliminated as losers.

Handling ties among the winners:

Another slight issue to be mindful of— the same way single-winner IRV is a little ambiguous for who to eliminate in the case of bottom place ties, this new one is also a bit ambiguous in who to promote in the case of top-place ties.

E.g. if there are only two winner spots available, but three candidates all tie for the same top score.

(My guess is that as implemented, it will currently tie-break via alphabetical ordering, but haven't tested to be sure.)

[ ] We could modify the <CheckForTies /> warning to also highlight if/whenever this happens.

Lastly, the two remaining unchecked nice-to-haves mentioned in previous comments:

documenting this feature in the advanced ballot editor UI
linking to further documentation about the specific algorithm.

siv-org / siv