trie_rcv

https://crates.io/crates/trie_rcv
Ranked Choice Voting (RCV) implementation using Tries in Rust.

RCV differs from normal first past the post voting in that voters are allowed to rank candidates from most to least preferable. To determine the winner of an RCV election, the votes for the least popular candidate each round is transferred to the next candidate in the respective ranked votes until some candidate achieves an effective majority.

Example usage:

use trie_rcv;
use trie_rcv::RankedChoiceVoteTrie;
use trie_rcv::vote::RankedVote;

fn main() {
    let mut rcv = RankedChoiceVoteTrie::new();
    // remove all candidates with the lowest number of votes each round
    rcv.set_elimination_strategy(EliminationStrategies::EliminateAll);
    rcv.insert_vote(RankedVote::from_vector(&vec![1, 2, 3, 4]).unwrap());
    rcv.insert_vote(RankedVote::from_vector(&vec![1, 2, 3]).unwrap());
    rcv.insert_vote(RankedVote::from_vector(&vec![3]).unwrap());
    rcv.insert_vote(RankedVote::from_vector(&vec![3, 2, 4]).unwrap());
    rcv.insert_vote(RankedVote::from_vector(&vec![4, 1]).unwrap());
    let winner = rcv.determine_winner();
    println!("WINNER = {:?}", winner);
    assert_eq!(winner, Some(1));
    
    // alternatively:
    let votes = RankedVote::from_vectors(&vec![
        vec![1, 2, 3, 4],
        vec![1, 2, 3],
        vec![3],
        vec![3, 2, 4],
        vec![4, 1]
    ]).unwrap();

    let winner2 = rcv.run_election(votes);
    println!("WINNER = {:?}", winner2);
    assert_eq!(winner2, Some(1));
}

This implementation also supports ranked votes ending with SpecialVotes::WITHHOLD and SpecialVotes::ABSTAIN values:

1. SpecialVotes::WITHHOLD
   Allows the voter to explicitly declare for none of the candidates.
   Qualitatively this allows a voter to declare a vote of no confidence.
   Serializes to -1 via SpecialVotes::WITHHOLD.to_int()
2. SpecialVotes::ABSTAIN
   Allows the voter to explicitly declare for none of the candidates while also voluntarily removing himself from the poll.
   Qualitatively this allows a voter to indicate that he wants one of the candidates to win but isn't able to decide for himself and would thus want to delegate the decision to the rest of the electorate.
   Serializes to -2 via SpecialVotes::ABSTAIN.to_int()

use trie_rcv;
use trie_rcv::RankedChoiceVoteTrie;
use trie_rcv::vote::{RankedVote, SpecialVotes};

fn main() {
    let rcv = RankedChoiceVoteTrie::new();

    let votes_round1 = RankedVote::from_vectors(&vec![
        vec![1, SpecialVotes::WITHHOLD.to_int()],
        vec![2, 1],
        vec![3, 2],
        vec![3]
    ]).unwrap();

    let winner_round1 = rcv.run_election(votes_round1);
    println!("WINNER = {:?}", winner);
    assert_eq!(
        winner_round1, None, concat![
        "Candidate 1's vote should not count after round 1, ",
        "no one should have majority"
    ]);
    
    let votes_round2 = RankedVote::from_vectors(&vec![
        vec![1, SpecialVotes::ABSTAIN.to_int()],
        vec![2, 1],
        vec![3, 2],
        vec![3]
    ]).unwrap();

    let winner_round2 = rcv.run_election(votes_round2);
    println!("WINNER = {:?}", winner_round2);
    assert_eq!(
       winner_round2, Some(3), concat![
       "First vote is ignored in round 2, candidate 3 wins"
    ]);
}

Elimination Strategies

Technically the RCV algorithm specification doesn't state what to do in the situation that there are multiple candidates who all have the same, lowest number of votes in some round during RCV.

The elimination_strategy setting handles which candidates to eliminate each round.
Technically the RCV algorithm specification doesn't state what to do in the situation that there are multiple candidates who all have the same, lowest number of votes in some round during RCV - EliminationStrategies::ElimnateAll, EliminationStrategies::DowdallScoring, and EliminationStrategies::RankedPairs offer different ways to resolve that edge case.

1. EliminationStrategies::ElimnateAll
   Removes all candidates with the lowest number of votes each round.
2. EliminationStrategies::DowdallScoring (default)
   Among multiple candidates with the lowest number of votes each round, sort the candidates by their dowdall score and eliminate the candidate(s) with the lowest Dowdall score. The Dowdall score for each candidate is calculated by the sum of the inverse of the ranking (starting from 1) for each ranked vote. If a ranked vote does not contain a candidate, then it does not count towards the dowdall score.
3. EliminationStrategies::RankedPairs
   Among multiple candidates with the lowest number of votes each round, attempt to construct a directed acyclic graph establishing a pecking order between candidate preferences via ranked-pair comparisons, whereby candidate A is said to be better than candidate B if there are more votes that rank A higher than B and vice versa, and eliminate the candidate(s) that are at the bottom to the pecking order (i.e. there are no other candidates that it is "better" than the pecking order, and there is at least 1 candidate that can be said to be "better" in the pecking order)
   1. Note that if a ranked vote ranks A explicitly but not B, then that is counted as ranking A higher than B as well
   2. In the event that a directed acyclic preference graph cannot be established (such as when there are cyclic preferences between candidates), then the elimination behavior will default to eliminating all candidates with the same, lowest number of votes each round i.e. it will fall back to the behavior of EliminationStrategies::ElimnateAll
4. EliminationStrategies::CondorcetRankedPairs
   (Implementation of the majority rule according to this paper)
   Between the candidates with the lowest and second-lowest number of votes each round, attempt to construct a directed acyclic graph to establish a pecking order between candidate preferences via ranked-pair comparisons, and eliminate the candidate(s) that are at the bottom to the pecking order. This ensures that the winning candidate is a Condorcet winner if one exists in the poll results, and will revert to EliminationStrategies::ElimnateAll if the preference graph cannot be constructed.

Build instructions

Build crate using cargo build, run integration tests with cargo test