[EM] Simplest Condorcet method to hand count?

[robert bristow-johnson]

God, I hate this list server.  I have to always remember to hit "Reply all" for it to go to the list.

> On 05/22/2025 8:10 PM EDT robert bristow-johnson <rbj at audioimagination.com> wrote:
> 
>  
> > On 05/22/2025 7:16 PM EDT Etjon Basha <etjonbasha at gmail.com> wrote:
> > 
> > 
> > Hi Robert,
> > 
> > Alas, laborious but far from simple.
> > 
> 
> I think it's conceptually quite simple.  But laborious.
> 
> > Do you reckon pairwise matrices would be easily computed by hand by volunteers?
> 
> Yes.  But this work is distributed.  Even for a large city (like NYC) or an entire state (like Maine or Alaska), it's quite doable if each town can perform their own recount independently and the final results of the recount are published and are summable.
> 
> So, if there are N candidates (let's count "Combined Write-In" as one of those candidates), the ballot pile (and this is a pile of ballots for *only* the small town or for a single voting precinct, essentially for a single ballot bag that has the ballots that were inserted into a single voting tabulator) is processed N(N-1)/2 times.  For 2 candidates, that's once (like for FPTP).  For 3 candidates, that's 3 times.  For 4 candidates, it's 6 times.  These are finite numbers.  Totally doable.
> 
> For each pass, it's about a pair of candidates, A and B.  Each ballot has one of three states, which are tallied.  Either A is preferred to B or B is preferred to A or there is no apparent preference of either (like neither are marked or both are marked to the same ranking level).  Three tallies.
> 
> In Vermont, when we have a recount, when the bag is opened, the first thing is that the serial number on the bag seal is read and confirmed to be the same number that was recorded on election night when the bag was first sealed.  Then we dump the ballots out onto a big table in full visibility to everyone on scene.  Then we, in pairs of counters, count out individual mini-piles of exactly 50 ballots each and clip those ballots together with a unique mini-pile increment (like "Pile 1", "Pile 2", etc.) and a tally cover sheet with the pile number is clipped in with those 50 ballots.
> 
> Then, with teams of 4, and no more than two recounters can be from the same party in each team, we count the ballots of a mini-pile at least twice and make sure the tallies agree before we're done with that mini-pile.  In the recount, two persons (different party affiliation) are the Callers and two persons (different party affiliation) are the Counters or Talliers.
> 
> The Callers identify the state of the ballot.  For Condorcet (we don't do Condorcet yet in Vermont), they would say "A" if it's A>B and they would say "B" if it's B>A.  Or "Null" or "Tied" if it's neither a vote for A or B.  Or *rarely*, there is a dispute about that particular ballot.  The Counters will tally whatever the Callers say.  There would be 4 columns of tally marks, "A>B", "B>A", "Null or Tied", and "Disputed".  There would be tally marks under each column.  The number of those marks *must* add to 50 and the corresponding tallies between the two Counters *must* agree.  If they agree, the roles of the Callers and Counters are swapped and the process is repeated for the same 50 ballots.  If the two recounts of the bunch of 50 ballots *totally* agree, that mini-pile is done, the tallies are shown on the cover sheet, all are clipped together and go to the completed pile that the Clerk is monitoring and she puts the tallies for that bunch into a computer spreadsheet.
> 
> There are other issues in a recount and these include what to do with poorly-marked ballots and what to do if the callers disagree with what the ballot actually says.  This is the normal thing.  But, at the end of the day, every ballot is fully examined multiple times and a *consistent* discernment of what the ballot says is made and documented on the tally cover sheet for each mini-pile.
> 
> Now, for FPTP, this happens to the entire pile once.  For Condorcet, it would have to happen 6 times to the pile if there were 4 candidates.  So, instead of taking 1 hour, it might take 6.  But, because of summability, we can get bigger crews for a larger town or larger voting precinct (say, with 1000 or 2000 voters, that's the biggest we get in Vermont for a single voting precinct) and work, separately and simultaneously, on different portions of the big pile.  Because the end result will be separate summable tallies for bunches of 50 ballots, each numbered and clipped, so if there's a real legal fight, the District Judge will know which mini-pile and which funky ballot he/she has to look at to judge the ballot's value.  At the end of the day, each ballot is fully adjudicated and *consistent* *repeatable* tallies are published.  Everyone knows what those numbers are, media, competing campaigns, and of course, the office of the Secretary of State who will declare who the winner is.
> 
> > Perhaps I'm overly pessimistic on this one issue.
> 
> The recount process is nearly identical to what we do with FPTP now.  It would just have to be expanded to be making the very same decision for each ballot for the multiple pairings of candidates.  In a single pass you do each pairing for the whole pile, one pairing for that pass.  That's increased *quantity*, not increased complexity.
> 
> Another thing is, if the recount is concerned about only two candidates, like only the two leading candidates are disputing who won and the other minor candidates are not disputing their loss, then the recount of a Condorcet election only need to process the pass involving those two candidates.  Then it's no more work than a recount would be for FPTP.

--

r b-j . _ . _ . _ . _ rbj at audioimagination.com

"Imagination is more important than knowledge."

.
.
.