Practical Issues

The time required to do a tally for a poll will be proportional to the number of pairwise candidate combinations, and to the number of ballots:

n Pairs Time/Ballot Time/Poll
33 6s0.5h
46 12s1.0h
510 20s1.7h
615 30s2.5h
721 42s3.5h
828 56s4.7h
1045 90s7.5h
20190 380s31.7h
  1. n candidates =
    n (n – 1) / 2 pairs.
  2. A poll in BC generally contains 400 voters, give or take; Assuming a 75% turnout: 300 ballots.
  3. Assuming it takes as long as 2 seconds per pair, per ballot to count.
  4. As seen, with many candidates the manual processing time can quickly get out of hand.

This is clearly more demanding than tallying an FPTP election, and is often discounted as a manual process for that reason.

As a manual method, with these assumptions, we hit the wall around, say, seven candidates, though this will depend on the actual time per pair per ballot and the actual number of voters.

Depending on how the tally sheet is organized this could be reduced. (The proposed tally-sheet column arrangement faciliates a manual tally, probably more on the order of 1s/pair/ballot than the perhaps over conservative 2s assumed above.) Specific benchmark testing is, of course, in-order.

A manual count would be reasonable up to this vicinity, which limitation would also be facilitated if we continue to constrain parties to run no more than one candidate per electoral district.

With an optical-scanner tallying process there would be no adverse processing consequence in allowing multiple candidates per party (though this could increase the campaign-cost per party), and since with Ranked Pairs, unlike FPTP (and to a lesser degree, unlike IRV), outcomes are not harmed by participation of “similar” candidates, there would be no electoral harm either.

An evaluation of recent BC Elections data using these manual-count assumptions shows how this might unfold in a real election:

/ District
Worst Case Scenario
Electoral District Largest Poll Time
2013 2 1 Kootenay – East 194 ballots 0.1h
3 15 Parksville – Qualicum 268 ballots 0.5h
4 35 Maple Ridge – Mission 260 ballots 0.9h
5 21 Fort Langley – Aldergrove 339 ballots 1.9h
6 8 Vancouver – False Creek 312 ballots 2.6 h
7 4 Abbotsford – Mission 235 ballots 2.7h
8 1 Vancouver – Point Grey 203 ballots 3.2h
2009 3 26 Saanich North
and the Islands
245 ballots 0.4h
4 36 North Island 264 ballots 0.9h
5 16 Shuswap 381 ballots 2.1h
6 7 Vancouver-False Creek 230 ballots 1.9h
2005 3 4 Burnaby-Edmonds 284 ballots 0.5h
4 19 West Vancouver
512 ballots 1.7h
5 27 Port Moody-Westwood 544 ballots 3.0h
6 19 Delta South 446 ballots 3.7h
7 6 Okanagan-Vernon 336 ballots 3.9h
8 3 Comox Valley 400 ballots 6.2h
9 1 Vancouver-
Mount Pleasant
267 ballots 5.3h

Though in each of these elections a large majority of electoral districts and polls within them could easily be manually counted, we nevertheless see some districts with large numbers of candidates, or coupled with ballot counts that would present challenges to a manual count.

These are “general” poll data, as well, and in many of these cases there are considerably higher ballot counts for the Advanced polls.

These cases could stand to be split, either into smaller polls, or to divide their ballots among multiple counting teams. There are too many cases where we’re pushing the envelope, however, for a practical manual count.

Prudence dictates that we must be able to accomodate electoral districts with large numbers of candidates, or if our timing expectations are off, which means that we must have available some manner of computerized counting tools that we can deploy where needed, even if not a full-scale roll-out.

A computerized solution could take any of many forms. The salient problem to be addressed is tallying the actual ballots. Aggregating tallies, and the final analysis to determine the final rankings are of minor difficulty.

Perhaps the most direct, straightforward, approach for counting is to use an optical reader, and to ensure the ballot conforms to this usage:

  • Moving from a paper-ballot voting system all the way to a computer-voting station, is probably a bridge too far for voter acceptance. I suggest the transition into a new voting method is a “big-enough” change for the first few elections. (After that, why not? — and then an optical-reader-compatible ballot would be of no more import.)
  • Counting-time requirements with an optical-reader approach would be a tiny fraction of a second per ballot. Such technology is reliable and robust, inexpensive, and has been tried and true for many decades.
  • This requires no Internet connection. The printed tally report would be signed and included with the ballots, as now, as well as called-in by phone or faxed, as now, to the district office. But it could also be extended with an Internet connection to update the data directly to the district office.

This is all straightforward in principle, but it might well also suggest additional process changes such as having a single reader station at each polling station and the ballot read and stored in a single location, much as is done in some BC local-government elections, vs whether to maintain a more per-poll approach as now.

Resolving such particulars, however, is premature at this point; in the event the Legislature determines to consider this matter the ensuing discussion requires full and active participation by Elections staff, upon whom it would fall to implement the system and to conduct elections with it.

(The Ranked-Pairs workbench application, which you can download here, is a Java program that can be used to process sets of ballot data, both from a number of scenarios bundled with the application, as well as allowing you to use your own ballot data. Among other things it demonstrates that the processing requirements for an end-to-end fully-computerized solution are not onerous and well within the capabilities of current consumer-grade notebook or desktop computers.)

Next: Condorcet Details

Share Button