[EM] Legacy IRV limitations

[Greg Dennis]

Ballot real estate has been a factor in limiting the rankings to 5 in some
US elections, but never to 3, as far as I understand. Places that limited
the rankings to 3 did so only due to a limitation of older voting machine
models, specifically the old Optech IIIP-Eagle machine, which can only
support three columns of bubbles. That was the machine used in San
Francisco for their first RCV use, and now that they've upgraded their
machines, they allow up to 10 rankings (the limit of their new Dominion
ImageCast machines).

There have been jurisdictions, including New York City and some Bay Area
cities, that have limited the rankings to 5 despite the equipment
theoretically supporting more. I believe ballot real estate *does* factor
into there, to save on the (quite high) cost of paper and to make a simpler
experience for the voters and poll workers. My understanding of the data
out of Cambridge, MA (which has a 9-seat council elected by STV), that
rankings after 5 never matter in practice, despite the fact that they allow
up to 15.

Greg

On Tue, Jan 16, 2024 at 8:18 AM C.Benham <cbenham at adam.com.au> wrote:

> Richard,
>
> One reason for limiting ranking to just 3 "choice" levels is the issue
> of "ballot real estate."  Specifically, more choice levels take up more
> ballot space.  That's a big issue in U.S. elections where there are so
> many election contests.
>
>
> In  Australia elections for the House of Representatives and for the
> Senate are held at the same time.
>
> The voter is handed a white ballot paper for the House of Reps. and a
> green one for the Senate, and instructed
> to fill them in with a provided pencil (or the voters own pen or pencil)
> and then put them in different boxes.
>
> So why not in the US say have one ballot paper say for electing the
> President and VP and another for electing less
> important offices?
>
> As a reminder there is a simple way to correctly count such "overvotes."
>   Just pair up the ballot with equivalent similar ballots during that
> counting round.  Specifically, if two ballots rank candidates A and B as
> equally preferred, one of those ballots goes to support candidate A and
> the other ballot goes to support candidate B (during this counting round).
>
>
> Since Hare (aka IRV, aka RCV) meets both Clone-Independence and
> Later-no-Harm, limiting the number of candidates
> the voter is allowed to rank (or just limiting the number of preference
> levels) is especially eggregious.
>
> With limited preference levels (but allowing above-bottom equal ranking) I
> vastly prefer  Smith//Approval (ranking above bottom)
> or the even more simple Condorcet//Approval  or  Condorcet//FPP(whole).
>
> If voters are free to to strictly rank from the top however many
> candidates they wish, then IRV is acceptable to me (better than STAR,
> Approval, Score, Majority Judgement, Margins) and the Australian way of
> dealing with "overvotes" is simple and acceptable.
>
> But given that that IRV is being used on ballots with a limited number of
> preference levels, then it is important that above-bottom
> equal-ranking ("overvoting") be allowed and handled correctly.
>
> The way to do that is to initially order the (remaining) candidates
> according to what is produced by treating a top (among remaining candidates)
> preference for more than one candidate as equal fractions (summing to one)
> of a vote to each.
>
> Then count the  equal top preference ballots again, this time giving a
> whole single vote to whichever of the equally highest ranked is highest
> in the initial order.  Then eliminate the candidate with fewest (highest
> rank) votes.
>
> Without the latter step the method is quite a bit more vulnerable to
> Push-over strategy.  I think this is well in the spirit of the Single
> Transferable
> Vote, which is like a a sort of "Declared Strategy Voting" where the
> virtual voter has a single vote and is pursuing the not-very-sophisticated
> but simple
> and honest strategy of trying to minimise the chance that the voter's
> (current among remaining candidates) favourite will be eliminated.
>
>
> Chris Benham
>
>
>
> *Richard, the VoteFair guy* electionmethods at votefair.org
> <election-methods%40lists.electorama.com?Subject=Re%3A%20%5BEM%5D%20Legacy%20IRV%20limitations&In-Reply-To=%3C69d857e6-a6a7-43c9-89ad-20879503fdf8%40votefair.org%3E>
> *Sat Dec 16 17:35:17 PST 2023*
> ------------------------------
>
> On 12/16/2023 6:12 AM, Kristofer Munsterhjelm wrote:
>  > On 2023-12-16 13:44, C.Benham wrote:
>  >> Why do at least several US Americans here think there is something
>  >> problematic and/or weird about allowing both quite
>  >> a large number of candidates on the ballot and voters to strictly rank
>  >> exactly as many of them as they wish?
>  >
>  > I guess it's partly that some US locations do this de facto anyway (e.g.
>  > some places using IRV only lets the voters rank three candidates). And I
>  > *think* that's due to legacy hardware? Optical scan machines that can
>  > only read bubbles, and mechanical ones that can only read a certain
>  > number of holes.
>
> One reason for limiting ranking to just 3 "choice" levels is the issue
> of "ballot real estate."  Specifically, more choice levels take up more
> ballot space.  That's a big issue in U.S. elections where there are so
> many election contests.
>
> Otherwise, when there are more than 3 candidates, the number of choice
> columns interacts with the issue of "overvotes."
>
> It's the FairVote organization that promotes the myth that IRV cannot
> handle "overvotes."
>
> Apparently FairVote does this to allow using old data from Australian
> elections to certify new or revised IRV software.
>
> Australia previously, before machine counting of ballots became
> available, counted their ranked-choice paper ballots manually, by
> stacking ballots in piles.  (That's what I've read.)
>
> To speed up that manual counting, apparently Australia adopted the
> shortcut of stacking ballots according to which candidate is highest
> ranked after removing eliminated candidates.
>
> That shortcut means that during each counting round only a single stack
> of ballots needs to be looked at, and sorted, based on which candidate
> has become the newly highest-ranked candidate (after the latest
> elimination).
>
> An important part of this shortcut is to reject/dismiss/ignore any
> ballot when there is no longer just one highest-ranked candidate.
> That's probably when the term "overvote" appeared.
>
> In turn, this is why FairVote promotes the myth that when there are only
> three "choice" columns each choice column can have only one mark.
>
> If there are only three choice columns and a voter wants to indicate
> that one particular candidate is worse than all other candidates, and
> there are 5 or more candidates, all but the most-disliked candidate need
> to be ranked at choice levels "first," "second," and "third."
>
> Now that election officials in the United States and Australia count
> paper ballots using machines that read ballots, it's time to at least
> question this legacy limitation of not allowing "overvotes."  And
> hopefully we can soon abandon this legacy limitation.
>
> For clarification, in Australia a voter writes a number inside a box
> located next to each candidate's name.  Software can recognize those
> handwritten numbers as reliably as a person, yet much faster.  When
> there is uncertainty a photographic image of the ballot can be displayed
> on multiple computer screens for verification from several humans.
>
> This limitation of not ranking more than one candidate at the same
> choice level is due to a lack of ballot data (including results) against
> which new software can be verified.
>
> It's time to end this ridiculous limitation.
>
> Part of my frustration comes from the fact that Portland Oregon recently
> adopted counting rules that are even worse than just ignoring ballots
> with "overvotes."
>
> With "advice" from the FairVote-controlled "Ranked Choice Voting
> Resource Center" the Portland election officials chose to skip over
> overvotes instead of dismissing the remainder of the ballot.
>
> This means a voter who ranks candidates A and B as their "second choice"
> and candidate C as their "third choice" will get their ballot counted as
> support for candidate C even if candidates A and B have not been
> eliminated.  Yet ranking candidate C higher than A and B is exactly the
> opposite(!) of what the voter clearly intended!
>
> As a reminder there is a simple way to correctly count such "overvotes."
>   Just pair up the ballot with equivalent similar ballots during that
> counting round.  Specifically, if two ballots rank candidates A and B as
> equally preferred, one of those ballots goes to support candidate A and
> the other ballot goes to support candidate B (during this counting round).
>
> Now that we have machines and software to handle the correct counting of
> "overvotes," this extra "effort" does not impose any significant delay,
> or any significant increase in electricity to power the computer for a
> few extra milliseconds.  It does require extra effort from the
> programmer who writes the code, but that just involves extra effort from
> one person for a few hours.  (And if they don't know how to write that
> code they can copy from open-source software that correctly does this
> counting.)
>
> To repeat, the only reason for the legacy of dismissing "overvotes" is
> that we lack certified ballot data against which to certify upgraded
> software.
>
> Allowing overvotes will make it possible to meaningfully rank more than
> 6 candidates using only 5 or 6 choice columns.
>
> (A complication is whether an unranked candidate is ranked at the bottom
> printed choice level, or lower than all ranked candidates.  And this
> interacts with the complication of how to rank a candidate who is a
> write-in candidate on someone else's ballot.)
>
> Limiting ranked choice ballots to 6 choice columns is reasonable, even
> when the election contest has 10 or more candidates.  But doing so does
> require correctly counting 2 or more candidates at the same "choice" level.
>
> Richard Fobes
> The VoteFair guy
>
>
> On 12/16/2023 6:12 AM, Kristofer Munsterhjelm wrote:
> >* On 2023-12-16 13:44, C.Benham wrote:
> *>>>>* Why do at least several US Americans here think there is something
> *>>* problematic and/or weird about allowing both quite
> *>>* a large number of candidates on the ballot and voters to strictly rank
> *>>* exactly as many of them as they wish?
> *>>> >* I guess it's partly that some US locations do this de facto anyway (e.g.
> *>* some places using IRV only lets the voters rank three candidates). And I
> *>* *think* that's due to legacy hardware? Optical scan machines that can
> *>* only read bubbles, and mechanical ones that can only read a certain
> *>* number of holes.
> *> >* I'm not sure, though.
> *> >>* I prefer Smith//Approval, but accept that that is more complex to
> *>>* explain and sell and probably the most approved candidate
> *>>* will nearly always be in the voted Smith set.
> *> > > >* -km*
>
>
>

-- 
*Greg Dennis, Ph.D. :: Policy Director*
Voter Choice Massachusetts

e :: greg.dennis at voterchoicema.org
p :: 617.835.9161
w :: voterchoicema.org <https://www.voterchoicema.org/>

:: Follow us on Facebook <https://www.facebook.com/yeson2rcv> and Twitter
<https://twitter.com/yeson2rcv> ::
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20240116/27d9105b/attachment-0001.htm>