[EM] Rank Codes

classic Classic list List threaded Threaded
11 messages Options
Reply | Threaded
Open this post in threaded view
|

[EM] Rank Codes

Forest Simmons
It pains me to see all of the ranked ballot proposals that unnecessarily limit preferences to three or four alternatives because of ignorance of simple higher resolution ballots that can be easily marked and read (by hand or by machine) without ambiguity or confusion from poorly formed characters, stray marks, etc.

A method that allows only three or four candidates to be ranked cannot satisfy clone independence ... the only indispensable justification for scrapping First Past the Post Plurality. And (beyond that) it exacerbates the biggest IRV/STV/RCV defect, the high likelihood that one's choices will be completely exhausted before the final rounds unless you rank lesser evils at the expense of alternatives you like better, because of ranking limitations that highlight the effect of premature eliminations.

It is alleged that because of ambiguous handwriting and lack of room for more than a few "bubbles," only a handful of distinct ranks can be allowed.

But what if each bubble has a different value?:

[8]   [4]   [2]   [1]

The rank of a candidate is the sum of its darkened bubble values ... a number between zero and fifteen.

Suppose that there are to be 26 candidates, then instead of indicating their relative ranks with mere numbers, you can order them with standard alpha numeric code words ... Alpha1, Bravo2, Charlie3, Delta4, Echo5, Foxtrot6, ... Victor22, Whiskey23, Xray24, Yankee25, Zulu26. So the military already solved the ambiguity/ "noisy channel" commuunication problem in the early days of Morse code.

These 26 code words cannot be confused with each other no matter how illegible the hand writing.

If 625 alphabetically ordered code words are needed, there are that many easily distinguished five-letter words that satisfy the following rules:

The 1st letter of each word must be a member of the set {A, F, L, Q}.
The 2nd letter must be from {B, G, M, T}.
The 3rd from {C, H, N, W}, the 4th from {D, J, P, Y}, and the last from {E, K, S, Z}.

The important thing is that each of the five sets consists of four letters that cannot be confused among themselves. Futhermore letters in the same position come from different quarters of the alphabet, making alphabetical order easier to discern.

The ballots are to be accompanied by an easily accessible table of code words numbered in alphabetical order. However, the voters can skip numbers that they don't need when there are more code words than candidates in the race. Similarly, it goes without saying that the same code word can be applied to more than one alternative when equal rankings are allowed. And of course, these code words can be adapted for high resolution ratings if needed in the Range/Score/Cardinal Ratings context.

These suggestions are intended for absentee and other mail-in ballots ... electronic voting machines should allow in person voters to drag the names into a list in any order, and then print out paper copies for voter and precinct receipts.

I am sure there are better ways of doing this, but then why do we still keep seeing proposals with unnecessarily crippling limitations on the number of distinct ranks?

Comments? Suggestions?


----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Kristofer Munsterhjelm-3
On 13/02/2021 03.29, Forest Simmons wrote:
> It pains me to see all of the ranked ballot proposals that unnecessarily
> limit preferences to three or four alternatives because of ignorance of
> simple higher resolution ballots that can be easily marked and read (by
> hand or by machine) without ambiguity or confusion from poorly formed
> characters, stray marks, etc.

I have an impression that the problem is not real, but imagined: that
it's possible to do unlimited ranked ballots in practice without much
difficulty. Otherwise, the jurisdictions that currently use STV would
have encountered the problem and dealt with it already.

So the problem is more one of perception: it seems obvious that unclear
ballots are going to be hard to read, regardless of whether they
actually will. And so, as a precautionary measure, the method gets
limited to a few ranks.

(There may also be technology-specific limitations, e.g. the
jurisdiction in question uses mechanical voting machines that can't be
adapted to more than this many ranks.)

> A method that allows only three or four candidates to be ranked cannot
> satisfy clone independence ... the only indispensable justification for
> scrapping First Past the Post Plurality. And (beyond that) it
> exacerbates the biggest IRV/STV/RCV defect, the high likelihood that
> one's choices will be completely exhausted before the final rounds
> unless you rank lesser evils at the expense of alternatives you like
> better, because of ranking limitations that highlight the effect of
> premature eliminations.
>
> It is alleged that because of ambiguous handwriting and lack of room for
> more than a few "bubbles," only a handful of distinct ranks can be allowed.
>
> But what if each bubble has a different value?:
>
> [8]   [4]   [2]   [1]
>
> The rank of a candidate is the sum of its darkened bubble values ... a
> number between zero and fifteen.

I think these would confuse quite a few voters.

I'd probably just go with ordinary numbers and be fairly confident it's
going to work out. But if the problem is indeed one of perception, then
just saying "don't sweat it" isn't going to convince anyone who's sure
there will be problems.

Perhaps a study on ballot rejection rates would help provide evidence
that it works well most of the time? I seem to recall reading on
Reddit's EndFPTP forum that ballot spoilage rates are about the same for
FPTP and STV.

> Suppose that there are to be 26 candidates, then instead of indicating
> their relative ranks with mere numbers, you can order them with standard
> alpha numeric code words ... Alpha1, Bravo2, Charlie3, Delta4, Echo5,
> Foxtrot6, ... Victor22, Whiskey23, Xray24, Yankee25, Zulu26. So the
> military already solved the ambiguity/ "noisy channel" commuunication
> problem in the early days of Morse code.
>
> These 26 code words cannot be confused with each other no matter how
> illegible the hand writing.

Of the two suggestions, I think I prefer this one. You could make this a
minimal change by saying that a voter may use either ordinary numbers or
codewords, so that voters who want to be extra sure that their ballots
will be counted properly can use the codewords, while others may opt out
if they think it's not worth the hassle.

> These suggestions are intended for absentee and other mail-in ballots
> ... electronic voting machines should allow in person voters to drag the
> names into a list in any order, and then print out paper copies for
> voter and precinct receipts.

I'd prefer voting machines to be "Expensive Pencils" where the voter can
input preferences and have a paper ballot printed out, and where that
paper ballot is what gets counted. A voting machine is opaque; a
printout is not.

To mitigate chain voting, the machine could show the printout behind
glass and deposit it directly into either the trash or the ballot box
depending on the voter's choice. In addition, such a scheme would keep
fingerprints and DNA off the ballot paper.
----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Forest Simmons
Kristofer, thanks for your comments and suggestions.

Here's one more ... spell out the number names in old fashioned check writing style:

Zero, One, Two, Three, Four, ... , Ninety-eight, Ninety-nine, One-bundred.

Make this long form code optional as you suggested.


On Saturday, February 13, 2021, Kristofer Munsterhjelm <[hidden email]> wrote:
On 13/02/2021 03.29, Forest Simmons wrote:
> It pains me to see all of the ranked ballot proposals that unnecessarily
> limit preferences to three or four alternatives because of ignorance of
> simple higher resolution ballots that can be easily marked and read (by
> hand or by machine) without ambiguity or confusion from poorly formed
> characters, stray marks, etc.

I have an impression that the problem is not real, but imagined: that
it's possible to do unlimited ranked ballots in practice without much
difficulty. Otherwise, the jurisdictions that currently use STV would
have encountered the problem and dealt with it already.

So the problem is more one of perception: it seems obvious that unclear
ballots are going to be hard to read, regardless of whether they
actually will. And so, as a precautionary measure, the method gets
limited to a few ranks.

(There may also be technology-specific limitations, e.g. the
jurisdiction in question uses mechanical voting machines that can't be
adapted to more than this many ranks.)

> A method that allows only three or four candidates to be ranked cannot
> satisfy clone independence ... the only indispensable justification for
> scrapping First Past the Post Plurality. And (beyond that) it
> exacerbates the biggest IRV/STV/RCV defect, the high likelihood that
> one's choices will be completely exhausted before the final rounds
> unless you rank lesser evils at the expense of alternatives you like
> better, because of ranking limitations that highlight the effect of
> premature eliminations.
>
> It is alleged that because of ambiguous handwriting and lack of room for
> more than a few "bubbles," only a handful of distinct ranks can be allowed.
>
> But what if each bubble has a different value?:
>
> [8]   [4]   [2]   [1]
>
> The rank of a candidate is the sum of its darkened bubble values ... a
> number between zero and fifteen.

I think these would confuse quite a few voters.

I'd probably just go with ordinary numbers and be fairly confident it's
going to work out. But if the problem is indeed one of perception, then
just saying "don't sweat it" isn't going to convince anyone who's sure
there will be problems.

Perhaps a study on ballot rejection rates would help provide evidence
that it works well most of the time? I seem to recall reading on
Reddit's EndFPTP forum that ballot spoilage rates are about the same for
FPTP and STV.

> Suppose that there are to be 26 candidates, then instead of indicating
> their relative ranks with mere numbers, you can order them with standard
> alpha numeric code words ... Alpha1, Bravo2, Charlie3, Delta4, Echo5,
> Foxtrot6, ... Victor22, Whiskey23, Xray24, Yankee25, Zulu26. So the
> military already solved the ambiguity/ "noisy channel" commuunication
> problem in the early days of Morse code.
>
> These 26 code words cannot be confused with each other no matter how
> illegible the hand writing.

Of the two suggestions, I think I prefer this one. You could make this a
minimal change by saying that a voter may use either ordinary numbers or
codewords, so that voters who want to be extra sure that their ballots
will be counted properly can use the codewords, while others may opt out
if they think it's not worth the hassle.

> These suggestions are intended for absentee and other mail-in ballots
> ... electronic voting machines should allow in person voters to drag the
> names into a list in any order, and then print out paper copies for
> voter and precinct receipts.

I'd prefer voting machines to be "Expensive Pencils" where the voter can
input preferences and have a paper ballot printed out, and where that
paper ballot is what gets counted. A voting machine is opaque; a
printout is not.

To mitigate chain voting, the machine could show the printout behind
glass and deposit it directly into either the trash or the ballot box
depending on the voter's choice. In addition, such a scheme would keep
fingerprints and DNA off the ballot paper.

----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Juho Laatu-4
In reply to this post by Forest Simmons
Just a thought.

One simple technique that allows voters to use almost infinite number of different ranks would be to use one long box next to the name of each candidate. Voters would draw a mark in the box (or leave it unmarked). Marks towards the left side of the long box would be considered "good", and marks towards the right side of the box "less good". The ballots would be read by computers that would seek for a mark in each box, and assign a numeric preference value depending on the position of the mark in the box. There would be some additional perpendicular light lines across the boxes to help making the order of the candidates accurate.

BR, Juho


> On 13. Feb 2021, at 4.29, Forest Simmons <[hidden email]> wrote:
>
> It pains me to see all of the ranked ballot proposals that unnecessarily limit preferences to three or four alternatives because of ignorance of simple higher resolution ballots that can be easily marked and read (by hand or by machine) without ambiguity or confusion from poorly formed characters, stray marks, etc.
>
> A method that allows only three or four candidates to be ranked cannot satisfy clone independence ... the only indispensable justification for scrapping First Past the Post Plurality. And (beyond that) it exacerbates the biggest IRV/STV/RCV defect, the high likelihood that one's choices will be completely exhausted before the final rounds unless you rank lesser evils at the expense of alternatives you like better, because of ranking limitations that highlight the effect of premature eliminations.
>
> It is alleged that because of ambiguous handwriting and lack of room for more than a few "bubbles," only a handful of distinct ranks can be allowed.
>
> But what if each bubble has a different value?:
>
> [8]   [4]   [2]   [1]
>
> The rank of a candidate is the sum of its darkened bubble values ... a number between zero and fifteen.
>
> Suppose that there are to be 26 candidates, then instead of indicating their relative ranks with mere numbers, you can order them with standard alpha numeric code words ... Alpha1, Bravo2, Charlie3, Delta4, Echo5, Foxtrot6, ... Victor22, Whiskey23, Xray24, Yankee25, Zulu26. So the military already solved the ambiguity/ "noisy channel" commuunication problem in the early days of Morse code.
>
> These 26 code words cannot be confused with each other no matter how illegible the hand writing.
>
> If 625 alphabetically ordered code words are needed, there are that many easily distinguished five-letter words that satisfy the following rules:
>
> The 1st letter of each word must be a member of the set {A, F, L, Q}.
> The 2nd letter must be from {B, G, M, T}.
> The 3rd from {C, H, N, W}, the 4th from {D, J, P, Y}, and the last from {E, K, S, Z}.
>
> The important thing is that each of the five sets consists of four letters that cannot be confused among themselves. Futhermore letters in the same position come from different quarters of the alphabet, making alphabetical order easier to discern.
>
> The ballots are to be accompanied by an easily accessible table of code words numbered in alphabetical order. However, the voters can skip numbers that they don't need when there are more code words than candidates in the race. Similarly, it goes without saying that the same code word can be applied to more than one alternative when equal rankings are allowed. And of course, these code words can be adapted for high resolution ratings if needed in the Range/Score/Cardinal Ratings context.
>
> These suggestions are intended for absentee and other mail-in ballots ... electronic voting machines should allow in person voters to drag the names into a list in any order, and then print out paper copies for voter and precinct receipts.
>
> I am sure there are better ways of doing this, but then why do we still keep seeing proposals with unnecessarily crippling limitations on the number of distinct ranks?
>
> Comments? Suggestions?
>
> ----
> Election-Methods mailing list - see https://electorama.com/em for list info

----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Kristofer Munsterhjelm-3
On 14/02/2021 10.20, Juho Laatu wrote:

> Just a thought.
>
> One simple technique that allows voters to use almost infinite number
> of different ranks would be to use one long box next to the name of
> each candidate. Voters would draw a mark in the box (or leave it
> unmarked). Marks towards the left side of the long box would be
> considered "good", and marks towards the right side of the box "less
> good". The ballots would be read by computers that would seek for a
> mark in each box, and assign a numeric preference value depending on
> the position of the mark in the box. There would be some additional
> perpendicular light lines across the boxes to help making the order
> of the candidates accurate.
I've been thinking about such an interface for an Expensive Pencil too;
it would make it very easy to gather rated information. It would be more
noisy than a deliberate ratings system because it's hard to get a
pixel-perfect line going, but if there are enough voters, then perhaps
there's some rated analog of the Condorcet Jury theorem that would imply
that the effect of that noise goes to zero as the number of voters
approaches infinity.

However, there's one problem: the voters can't easily equal-rank under
such a scheme. Therefore, no method passes FBC unless it also passes
Strong FBC, which is an extremely hard criterion to pass. Pragmatists
may not care (e.g. Schulze has very low favorite-betrayal rates under
the impartial culture distribution anyway), but it's annoying if you'd
like your method to unambiguously pass.

-km
----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Richard Lung
In reply to this post by Kristofer Munsterhjelm-3

In 2007, STV was introduced to Scottish local elections, on the same day
as AMS for the Scottish Parliament, which had already been used in a
previous election. Despite these disadvantages to STV, its spoilage was
2%, whereas for AMS something like twice that spoilage. There was an
enquiry into the latter.

Enid Lakeman, in How Democracies Vote, goes into this issue of STV
spoilage, in considerable detail, to anyone concerned about it.

I think it was Mr Woodall, who investigated Meek STV to ensure it was
convergent. And this consideration applies to FAB STV. This is ensured
by the fact that it is a statistical count. It uses four averages, FA,
as the title says, and they all center the result to the most
representative candidates. I would like to point out, tho, that Binomial
STV is a different kind of count to the rest of the world of elections,
uninomial elections. With FAB STV, the last preferences count as much
against candidates, as first preferences count for them. This would
induce different voting behavior in an informed electorate of this
greater power to their vote.

Richard Lung.


On 13/02/2021 16:15, Kristofer Munsterhjelm wrote:

> On 13/02/2021 03.29, Forest Simmons wrote:
>> It pains me to see all of the ranked ballot proposals that unnecessarily
>> limit preferences to three or four alternatives because of ignorance of
>> simple higher resolution ballots that can be easily marked and read (by
>> hand or by machine) without ambiguity or confusion from poorly formed
>> characters, stray marks, etc.
> I have an impression that the problem is not real, but imagined: that
> it's possible to do unlimited ranked ballots in practice without much
> difficulty. Otherwise, the jurisdictions that currently use STV would
> have encountered the problem and dealt with it already.
>
> So the problem is more one of perception: it seems obvious that unclear
> ballots are going to be hard to read, regardless of whether they
> actually will. And so, as a precautionary measure, the method gets
> limited to a few ranks.
>
> (There may also be technology-specific limitations, e.g. the
> jurisdiction in question uses mechanical voting machines that can't be
> adapted to more than this many ranks.)
>
>> A method that allows only three or four candidates to be ranked cannot
>> satisfy clone independence ... the only indispensable justification for
>> scrapping First Past the Post Plurality. And (beyond that) it
>> exacerbates the biggest IRV/STV/RCV defect, the high likelihood that
>> one's choices will be completely exhausted before the final rounds
>> unless you rank lesser evils at the expense of alternatives you like
>> better, because of ranking limitations that highlight the effect of
>> premature eliminations.
>>
>> It is alleged that because of ambiguous handwriting and lack of room for
>> more than a few "bubbles," only a handful of distinct ranks can be allowed.
>>
>> But what if each bubble has a different value?:
>>
>> [8]   [4]   [2]   [1]
>>
>> The rank of a candidate is the sum of its darkened bubble values ... a
>> number between zero and fifteen.
> I think these would confuse quite a few voters.
>
> I'd probably just go with ordinary numbers and be fairly confident it's
> going to work out. But if the problem is indeed one of perception, then
> just saying "don't sweat it" isn't going to convince anyone who's sure
> there will be problems.
>
> Perhaps a study on ballot rejection rates would help provide evidence
> that it works well most of the time? I seem to recall reading on
> Reddit's EndFPTP forum that ballot spoilage rates are about the same for
> FPTP and STV.
>
>> Suppose that there are to be 26 candidates, then instead of indicating
>> their relative ranks with mere numbers, you can order them with standard
>> alpha numeric code words ... Alpha1, Bravo2, Charlie3, Delta4, Echo5,
>> Foxtrot6, ... Victor22, Whiskey23, Xray24, Yankee25, Zulu26. So the
>> military already solved the ambiguity/ "noisy channel" commuunication
>> problem in the early days of Morse code.
>>
>> These 26 code words cannot be confused with each other no matter how
>> illegible the hand writing.
> Of the two suggestions, I think I prefer this one. You could make this a
> minimal change by saying that a voter may use either ordinary numbers or
> codewords, so that voters who want to be extra sure that their ballots
> will be counted properly can use the codewords, while others may opt out
> if they think it's not worth the hassle.
>
>> These suggestions are intended for absentee and other mail-in ballots
>> ... electronic voting machines should allow in person voters to drag the
>> names into a list in any order, and then print out paper copies for
>> voter and precinct receipts.
> I'd prefer voting machines to be "Expensive Pencils" where the voter can
> input preferences and have a paper ballot printed out, and where that
> paper ballot is what gets counted. A voting machine is opaque; a
> printout is not.
>
> To mitigate chain voting, the machine could show the printout behind
> glass and deposit it directly into either the trash or the ballot box
> depending on the voter's choice. In addition, such a scheme would keep
> fingerprints and DNA off the ballot paper.
> ----
> Election-Methods mailing list - see https://electorama.com/em for list info
----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Juho Laatu-4
In reply to this post by Kristofer Munsterhjelm-3
I think I'm pretty much a pragmatist in your terms. In typical large elections it doesn't really matter if instead of voting A=B there would be a 50% chance that the vote will be read by computer as A>B, and 50% chance for A<B. In practical terms that is the same as voting A=B (comparable to flipping a coin). However from a pragmatic point of view it could be a problem if people would get irritated because of their inability to vote exactly A=B.

There might be ways to go around this "irritation", like drawing an X at one of the "smaller boxes" separetad by the light lines that I mentioned. That X would be interpreted as an exact value (unlike other marks). But this sounds already too complicated to be worth it. If people want to use exact ties, it might be better to split the long boxes into 100 small narrow (2 mm) boxes that would then correspond to some exact rankings/ratings. With these ballots you might at some point run out of ability to rank A>B, and be forced to rank A=B. So (from pragmatic point of view) there will be tradeoffs, of one or another kind.

My pragmatism gives also leeway in handling the various criteria (e.g. strategy related). Since I have learned that no method is strategy free, it doesn't make sense to me to meet some important criteria 100%, and then ignore some others that are not met 100%. I see all these criteria as pragmatic criteria in the sense that it is enough to eliminate each major risk in practice, but not necessarily 100% in theory. The best method might be one that intentionally does not meet any of the important criteria 100%. That could make it possible to meet as many important criteria as possible well enough, and thereby weaken the worst remaining vulnerability as much as possible.

BR, Juho


> On 14. Feb 2021, at 11.31, Kristofer Munsterhjelm <[hidden email]> wrote:
>
> On 14/02/2021 10.20, Juho Laatu wrote:
>> Just a thought.
>>
>> One simple technique that allows voters to use almost infinite number
>> of different ranks would be to use one long box next to the name of
>> each candidate. Voters would draw a mark in the box (or leave it
>> unmarked). Marks towards the left side of the long box would be
>> considered "good", and marks towards the right side of the box "less
>> good". The ballots would be read by computers that would seek for a
>> mark in each box, and assign a numeric preference value depending on
>> the position of the mark in the box. There would be some additional
>> perpendicular light lines across the boxes to help making the order
>> of the candidates accurate.
> I've been thinking about such an interface for an Expensive Pencil too;
> it would make it very easy to gather rated information. It would be more
> noisy than a deliberate ratings system because it's hard to get a
> pixel-perfect line going, but if there are enough voters, then perhaps
> there's some rated analog of the Condorcet Jury theorem that would imply
> that the effect of that noise goes to zero as the number of voters
> approaches infinity.
>
> However, there's one problem: the voters can't easily equal-rank under
> such a scheme. Therefore, no method passes FBC unless it also passes
> Strong FBC, which is an extremely hard criterion to pass. Pragmatists
> may not care (e.g. Schulze has very low favorite-betrayal rates under
> the impartial culture distribution anyway), but it's annoying if you'd
> like your method to unambiguously pass.
>
> -km

----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Forest Simmons
In reply to this post by Forest Simmons
IMHO Juho's idea of a kind of slider sketch is the best ever on this topic ... truly worthy of a Toby award!

If horizontal dashes are allowed, then the midpoint determines the rating, and two dashes with more than 80 percent overlap should count as equal ... the shorter the width of a dash, the more obvious lack of intended overlap, etc.

----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Forest Simmons
In reply to this post by Forest Simmons
A little straight edge with appropriately spaced numbered tick marks would be a perfect tool ...
----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Hahn, Paul
In reply to this post by Forest Simmons

I was about to suggest pretty much this very thing, though I might have used diagonal marks, and ½-2/3 overlap instead of 80%.

 

--pH

 

From: Election-Methods <[hidden email]> On Behalf Of Forest Simmons
Sent: Sunday, February 14, 2021 7:00 PM
To: EM <[hidden email]>
Subject: Re: [EM] Rank Codes

 

IMHO Juho's idea of a kind of slider sketch is the best ever on this topic ... truly worthy of a Toby award!

 

If horizontal dashes are allowed, then the midpoint determines the rating, and two dashes with more than 80 percent overlap should count as equal ... the shorter the width of a dash, the more obvious lack of intended overlap, etc.


----
Election-Methods mailing list - see https://electorama.com/em for list info
Reply | Threaded
Open this post in threaded view
|

Re: [EM] Rank Codes

Juho Laatu-4
I guess this idea overlapping marks interpretes as equal rankings would work with any kind of marks (dashes, diagonals, ...). The number of candidates a voter can rank would depend on how thin marks (probably vertical lines) the voter can draw (and how thin lines the computer can read). When the computer notes x% overlap, those rankings would be read as equal.

The computer program should probably be smart in the sense that if there are two separate markings on one row, maybe the first one would count, and the second one would be ignored.

Also, if the centre of the mark is considerably darker or thicker than the edges to the left and right, only the centre part could be interpreted as the intended mark.

The required overlap (x) could be different for different size markings (more overlap required for large markings). This would make it easier to mark ties when the marks are already getting thin. And it would be ok to make relatively large (slightly overlapping) marks without accidentally marking candidates equal.

If there is a set of candidates that are all connected to each others based on the x% overlap rule, but the leftmost mark and the rightmost mark are already far apart, it would make sense to break that chain of equals at some point. That point could be e.g. where the overlap is weakest. That splitting process would continue until the leftmost and rightmost mark would be close enough to each other, so that they can be considered "meant to be equal".

If there is a mixture of wide and narrow markings, even some more rules may be needed. The final algorithm could become quite complex and heuristic by nature, but that is not really a problem, since all the rules are anyway meant to understand the intended preferences of the voter better. It is not a problem if some of the candidates will be interpreted as equal although they were not intended to be equal (but just almost equal), and the other way around. I guess this is the main point of this mail - use of heuristics is quite ok here.

The pens that are used in making the marks should leave a clear mark, i.e. they should maybe not be hard pencils whose marks could sometimes be ignored by the computer (if the marks are weak enough). It is better to err in tie vs. not a tie than in ranked vs. not ranked at all.

A recount, made by the same computer and algorithm that made the first count, could give somewhat different results at the second time since the mechanics in the optical reading device could give slightly different images each time. This is not a big problem. One just has to be able to agree which reading process of the ballots is the final one.

(Also an electronic voting machine could have a similar user interface, maybe with ability to push the already marked candidates to the left or to the right. But that's already another story, with other challenges. That machine could print paper ballots with (more compact and exact) number based rankings.)

BR, Juho


On 15. Feb 2021, at 18.21, Hahn, Paul <[hidden email]> wrote:

I was about to suggest pretty much this very thing, though I might have used diagonal marks, and ½-2/3 overlap instead of 80%.

 

--pH

 

From: Election-Methods <[hidden email]> On Behalf Of Forest Simmons
Sent: Sunday, February 14, 2021 7:00 PM
To: EM <[hidden email]>
Subject: Re: [EM] Rank Codes

 

IMHO Juho's idea of a kind of slider sketch is the best ever on this topic ... truly worthy of a Toby award!

 

If horizontal dashes are allowed, then the midpoint determines the rating, and two dashes with more than 80 percent overlap should count as equal ... the shorter the width of a dash, the more obvious lack of intended overlap, etc.

----
Election-Methods mailing list - see https://electorama.com/em for list info


----
Election-Methods mailing list - see https://electorama.com/em for list info