RB
robert bristow-johnson
Sat, Jan 1, 2022 6:41 PM
On 01/01/2022 10:59 AM Richard Lung voting@ukscientists.com wrote:
As far as I know (which isn't far) Condorcet methods are means of cross-checking given methods, rather than a method in their own right.
Uhm, Richard, that's something that, at least some of us are trying to change. I am trying to make Condorcet a method in Vermont.
In my vocabulary, a "Condorcet method" in it's own right, is any method that is Condorcet-consistent.
And some methods that requires a "Condorcet completion" method (in the vocabulary of https://condorcet.ca/ ) begins with a straight Condorcet method where every pairing of candidates is examined and losers are marked.
By the way, the point is that an election method should make use of an exclusion count, as well as an election count.
What is "an exclusion count"? Or "an election count?"
(My pedantry showing again.)
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
> On 01/01/2022 10:59 AM Richard Lung <voting@ukscientists.com> wrote:
>
>
> As far as I know (which isn't far) Condorcet methods are means of cross-checking given methods, rather than a method in their own right.
>
Uhm, Richard, that's something that, at least some of us are trying to change. I am trying to make Condorcet a method in Vermont.
In my vocabulary, a "Condorcet method" in it's own right, is any method that is Condorcet-consistent.
And some methods that requires a "Condorcet completion" method (in the vocabulary of https://condorcet.ca/ ) begins with a straight Condorcet method where every pairing of candidates is examined and losers are marked.
> By the way, the point is that an election method should make use of an exclusion count, as well as an election count.
What is "an exclusion count"? Or "an election count?"
(My pedantry showing again.)
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
RL
Richard Lung
Sun, Jan 2, 2022 5:32 PM
Yes, Robert, Condorcet methods are not my specialty.
But 'What is "an exclusion count"? Or "an election count?"' That is a good question, not at all pedantic.
Roughly speaking, existing voting methods are election counts helped out by ad hoc exclusion rules. All traditional stv, including Meek, works this way, by getting rid of the candidate least in the way, when the transfers of surplus votes run out. -- "Premature exclusion."
It is also true of party list systems or their hybrids. So, the German MMP is designed to give full party proportional representation, with half the seats going to list candidates. But inconsistently has a 5% threshold defeating that purpose. As a result, the Liberal Democrats, dipping insignificantly in their per centage vote, lost all 40 seats, from one election to the next.
The point is that existing elections are essentially election counts, helped out by ad hoc exclusion rules, missing an exclusion count, in its own right.
The election count, I use (in Binomial STV) is Meek method use of keep values. Meek only uses keep values for candidates elected to a quota or more (because the virtue of Meek method is that it continues to count the later preferences to already elected candidates, requiring the services of a computer program).
Binomial stv also counts the keep values of candidates in deficit of a quota. Why? Because the binomial stv result depends on an exclusion count, as well as an election count.
The difference between election count and exclusion count is simply that an exclusion count is of the voters orders of preference in reverse order. Their respective counts are symmetrical, exactly the same process. Indeed, there is no reason why one persons preference order is not another persons unpreference order. So, there is no logical reason for not having an exclusion count as well as an election count.
The bottom line is that an exclusion count, as well as an election count, makes greater use of the preference information. My old statistics teacher taught us over half a century ago, that the best statistical test is the one that makes most use of the information from the data.
Regards,
Richard Lung.
On 1 Jan 2022, at 6:41 pm, robert bristow-johnson rbj@audioimagination.com wrote:
On 01/01/2022 10:59 AM Richard Lung voting@ukscientists.com wrote:
As far as I know (which isn't far) Condorcet methods are means of cross-checking given methods, rather than a method in their own right.
Uhm, Richard, that's something that, at least some of us are trying to change. I am trying to make Condorcet a method in Vermont.
In my vocabulary, a "Condorcet method" in it's own right, is any method that is Condorcet-consistent.
And some methods that requires a "Condorcet completion" method (in the vocabulary of https://condorcet.ca/ ) begins with a straight Condorcet method where every pairing of candidates is examined and losers are marked.
By the way, the point is that an election method should make use of an exclusion count, as well as an election count.
What is "an exclusion count"? Or "an election count?"
(My pedantry showing again.)
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
Election-Methods mailing list - see https://electorama.com/em for list info
Yes, Robert, Condorcet methods are not my specialty.
But 'What is "an exclusion count"? Or "an election count?"' That is a good question, not at all pedantic.
Roughly speaking, existing voting methods are election counts helped out by ad hoc exclusion rules. All traditional stv, including Meek, works this way, by getting rid of the candidate least in the way, when the transfers of surplus votes run out. -- "Premature exclusion."
It is also true of party list systems or their hybrids. So, the German MMP is designed to give full party proportional representation, with half the seats going to list candidates. But inconsistently has a 5% threshold defeating that purpose. As a result, the Liberal Democrats, dipping insignificantly in their per centage vote, lost all 40 seats, from one election to the next.
The point is that existing elections are essentially election counts, helped out by ad hoc exclusion rules, missing an exclusion count, in its own right.
The election count, I use (in Binomial STV) is Meek method use of keep values. Meek only uses keep values for candidates elected to a quota or more (because the virtue of Meek method is that it continues to count the later preferences to already elected candidates, requiring the services of a computer program).
Binomial stv also counts the keep values of candidates in deficit of a quota. Why? Because the binomial stv result depends on an exclusion count, as well as an election count.
The difference between election count and exclusion count is simply that an exclusion count is of the voters orders of preference in reverse order. Their respective counts are symmetrical, exactly the same process. Indeed, there is no reason why one persons preference order is not another persons unpreference order. So, there is no logical reason for not having an exclusion count as well as an election count.
The bottom line is that an exclusion count, as well as an election count, makes greater use of the preference information. My old statistics teacher taught us over half a century ago, that the best statistical test is the one that makes most use of the information from the data.
Regards,
Richard Lung.
On 1 Jan 2022, at 6:41 pm, robert bristow-johnson <rbj@audioimagination.com> wrote:
> On 01/01/2022 10:59 AM Richard Lung <voting@ukscientists.com> wrote:
>
>
> As far as I know (which isn't far) Condorcet methods are means of cross-checking given methods, rather than a method in their own right.
>
Uhm, Richard, that's something that, at least some of us are trying to change. I am trying to make Condorcet a method in Vermont.
In my vocabulary, a "Condorcet method" in it's own right, is any method that is Condorcet-consistent.
And some methods that requires a "Condorcet completion" method (in the vocabulary of https://condorcet.ca/ ) begins with a straight Condorcet method where every pairing of candidates is examined and losers are marked.
> By the way, the point is that an election method should make use of an exclusion count, as well as an election count.
What is "an exclusion count"? Or "an election count?"
(My pedantry showing again.)
--
r b-j . _ . _ . _ . _ rbj@audioimagination.com
"Imagination is more important than knowledge."
.
.
.
----
Election-Methods mailing list - see https://electorama.com/em for list info
KM
Kristofer Munsterhjelm
Sun, Jan 2, 2022 5:48 PM
On 02.01.2022 18:32, Richard Lung wrote:
Yes, Robert, Condorcet methods are not my specialty.
But 'What is "an exclusion count"? Or "an election count?"' That is
a good question, not at all pedantic.
Roughly speaking, existing voting methods are election counts helped
out by ad hoc exclusion rules. All traditional stv, including Meek,
works this way, by getting rid of the candidate least in the way, when
the transfers of surplus votes run out. -- "Premature exclusion."
If I understand correctly, then only methods that actually do candidate
elimination (and fail LIIA) make use of exclusion counts. Minmax (the
Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
doesn't either: the winner is the candidate with the most first
preferences, and there are no eliminations as part of the process.
Did I get that right?
-km
On 02.01.2022 18:32, Richard Lung wrote:
>
> Yes, Robert, Condorcet methods are not my specialty.
>
> But 'What is "an exclusion count"? Or "an election count?"' That is
> a good question, not at all pedantic.
> Roughly speaking, existing voting methods are election counts helped
> out by ad hoc exclusion rules. All traditional stv, including Meek,
> works this way, by getting rid of the candidate least in the way, when
> the transfers of surplus votes run out. -- "Premature exclusion."
If I understand correctly, then only methods that actually do candidate
elimination (and fail LIIA) make use of exclusion counts. Minmax (the
Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
doesn't either: the winner is the candidate with the most first
preferences, and there are no eliminations as part of the process.
Did I get that right?
-km
FS
Forest Simmons
Mon, Jan 3, 2022 1:42 AM
How about this for inclusion/exclusion:
Each voter scores candidates between minus 100 percent and positive 100
percent.
The candidates' total scores (over all ballots) are tallied.
The list is sorted pairwise with priority swaps going to pairs with minimum
absolute difference in their scores.
The sorted list determines the finish order.
El dom., 2 de ene. de 2022 9:48 a. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
On 02.01.2022 18:32, Richard Lung wrote:
Yes, Robert, Condorcet methods are not my specialty.
But 'What is "an exclusion count"? Or "an election count?"' That is
a good question, not at all pedantic.
Roughly speaking, existing voting methods are election counts helped
out by ad hoc exclusion rules. All traditional stv, including Meek,
works this way, by getting rid of the candidate least in the way, when
the transfers of surplus votes run out. -- "Premature exclusion."
If I understand correctly, then only methods that actually do candidate
elimination (and fail LIIA) make use of exclusion counts. Minmax (the
Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
doesn't either: the winner is the candidate with the most first
preferences, and there are no eliminations as part of the process.
Did I get that right?
-km
Election-Methods mailing list - see https://electorama.com/em for list
info
How about this for inclusion/exclusion:
Each voter scores candidates between minus 100 percent and positive 100
percent.
The candidates' total scores (over all ballots) are tallied.
The list is sorted pairwise with priority swaps going to pairs with minimum
absolute difference in their scores.
The sorted list determines the finish order.
El dom., 2 de ene. de 2022 9:48 a. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
>
> On 02.01.2022 18:32, Richard Lung wrote:
> >
> > Yes, Robert, Condorcet methods are not my specialty.
> >
> > But 'What is "an exclusion count"? Or "an election count?"' That is
> > a good question, not at all pedantic.
> > Roughly speaking, existing voting methods are election counts helped
> > out by ad hoc exclusion rules. All traditional stv, including Meek,
> > works this way, by getting rid of the candidate least in the way, when
> > the transfers of surplus votes run out. -- "Premature exclusion."
>
> If I understand correctly, then only methods that actually do candidate
> elimination (and fail LIIA) make use of exclusion counts. Minmax (the
> Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
> doesn't either: the winner is the candidate with the most first
> preferences, and there are no eliminations as part of the process.
>
> Did I get that right?
>
> -km
> ----
> Election-Methods mailing list - see https://electorama.com/em for list
> info
>
KM
Kristofer Munsterhjelm
Mon, Jan 3, 2022 10:04 AM
On 03.01.2022 02:42, Forest Simmons wrote:
How about this for inclusion/exclusion:
Each voter scores candidates between minus 100 percent and positive 100
percent.
The candidates' total scores (over all ballots) are tallied.
The list is sorted pairwise with priority swaps going to pairs with
minimum absolute difference in their scores.
The sorted list determines the finish order.
I really ought to make a program where one can just slot in methods and
see their criterion compliances, strategy resistance, etc.
Quadelect was sort of meant to be this, but C++ isn't exactly the most
compact language as far as extensibility goes.
-km
On 03.01.2022 02:42, Forest Simmons wrote:
> How about this for inclusion/exclusion:
>
> Each voter scores candidates between minus 100 percent and positive 100
> percent.
>
> The candidates' total scores (over all ballots) are tallied.
>
> The list is sorted pairwise with priority swaps going to pairs with
> minimum absolute difference in their scores.
>
> The sorted list determines the finish order.
I really ought to make a program where one can just slot in methods and
see their criterion compliances, strategy resistance, etc.
Quadelect was sort of meant to be this, but C++ isn't exactly the most
compact language as far as extensibility goes.
-km
FS
Forest Simmons
Mon, Jan 3, 2022 5:51 PM
Which reminds me that "sorted margins" methods are unlikely to be monotone
in general.
We've been using sorted margins for symmetry purposes, but it's probably
better to sort in both directions and then use the stronger beatpath as the
finish order. At least in this example it would yield monotonicity (I'm
pretty sure).
El lun., 3 de ene. de 2022 2:04 a. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
On 03.01.2022 02:42, Forest Simmons wrote:
How about this for inclusion/exclusion:
Each voter scores candidates between minus 100 percent and positive 100
percent.
The candidates' total scores (over all ballots) are tallied.
The list is sorted pairwise with priority swaps going to pairs with
minimum absolute difference in their scores.
The sorted list determines the finish order.
I really ought to make a program where one can just slot in methods and
see their criterion compliances, strategy resistance, etc.
Quadelect was sort of meant to be this, but C++ isn't exactly the most
compact language as far as extensibility goes.
-km
Which reminds me that "sorted margins" methods are unlikely to be monotone
in general.
We've been using sorted margins for symmetry purposes, but it's probably
better to sort in both directions and then use the stronger beatpath as the
finish order. At least in this example it would yield monotonicity (I'm
pretty sure).
El lun., 3 de ene. de 2022 2:04 a. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
> On 03.01.2022 02:42, Forest Simmons wrote:
> > How about this for inclusion/exclusion:
> >
> > Each voter scores candidates between minus 100 percent and positive 100
> > percent.
> >
> > The candidates' total scores (over all ballots) are tallied.
> >
> > The list is sorted pairwise with priority swaps going to pairs with
> > minimum absolute difference in their scores.
> >
> > The sorted list determines the finish order.
>
> I really ought to make a program where one can just slot in methods and
> see their criterion compliances, strategy resistance, etc.
>
> Quadelect was sort of meant to be this, but C++ isn't exactly the most
> compact language as far as extensibility goes.
>
> -km
>
KM
Kristofer Munsterhjelm
Mon, Jan 3, 2022 10:01 PM
On 03.01.2022 18:51, Forest Simmons wrote:
Which reminds me that "sorted margins" methods are unlikely to be
monotone in general.
We've been using sorted margins for symmetry purposes, but it's
probably better to sort in both directions and then use the stronger
beatpath as the finish order. At least in this example it would yield
monotonicity (I'm pretty sure).
Strategy resistance (for various definitions thereof) without
monotonicity for purely ordinal (UD) methods seems to be pretty easy,
while strategy resistance plus monotonicity is very hard.
My optimization efforts suggest that it's possible, but the shape of the
solution still evades our grasp.
-km
On 03.01.2022 18:51, Forest Simmons wrote:
> Which reminds me that "sorted margins" methods are unlikely to be
> monotone in general.
>
> We've been using sorted margins for symmetry purposes, but it's
> probably better to sort in both directions and then use the stronger
> beatpath as the finish order. At least in this example it would yield
> monotonicity (I'm pretty sure).
Strategy resistance (for various definitions thereof) without
monotonicity for purely ordinal (UD) methods seems to be pretty easy,
while strategy resistance plus monotonicity is very hard.
My optimization efforts suggest that it's possible, but the shape of the
solution still evades our grasp.
-km
RL
Richard Lung
Tue, Jan 4, 2022 8:02 PM
KM,
I suspect those comments stem from (an unsurprising) unfamiliarity with
binomial stv. It has an exclusion count but it does not exclude or
eliminate candidates, during the binomial count (count of elections and
count of exclusions, before an over-all deciding count) and so does not
fall foul of irrationalities from "premature exclusion" etc.
Binomial STV is both a rational election count and a rational exclusion
count. Conventional STV, including Meek, Has a rational election count
but only a less powerful ordinal exclusion count. FPTP has only an
ordinal (more-than) election count.
Fptp is ambiguous as to whether it is an election count or an exclusion
count. it's of historical interest, that HG Wells began his publicity
for electoral reform, in 1911, with the observation that we no longer
have elections. We only have Rejections.
The first lesson in scientific method, I learned over fifty years ago,
included: Avoid ambiguity in a scientific test.
Regards,
Richard Lung.
On 02/01/2022 17:48, Kristofer Munsterhjelm wrote:
On 02.01.2022 18:32, Richard Lung wrote:
Yes, Robert, Condorcet methods are not my specialty.
But 'What is "an exclusion count"? Or "an election count?"' That is
a good question, not at all pedantic.
Roughly speaking, existing voting methods are election counts helped
out by ad hoc exclusion rules. All traditional stv, including Meek,
works this way, by getting rid of the candidate least in the way, when
the transfers of surplus votes run out. -- "Premature exclusion."
If I understand correctly, then only methods that actually do candidate
elimination (and fail LIIA) make use of exclusion counts. Minmax (the
Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
doesn't either: the winner is the candidate with the most first
preferences, and there are no eliminations as part of the process.
Did I get that right?
-km
KM,
I suspect those comments stem from (an unsurprising) unfamiliarity with
binomial stv. It has an exclusion count but it does not exclude or
eliminate candidates, during the binomial count (count of elections and
count of exclusions, before an over-all deciding count) and so does not
fall foul of irrationalities from "premature exclusion" etc.
Binomial STV is both a rational election count and a rational exclusion
count. Conventional STV, including Meek, Has a rational election count
but only a less powerful ordinal exclusion count. FPTP has only an
ordinal (more-than) election count.
Fptp is ambiguous as to whether it is an election count or an exclusion
count. it's of historical interest, that HG Wells began his publicity
for electoral reform, in 1911, with the observation that we no longer
have elections. We only have Rejections.
The first lesson in scientific method, I learned over fifty years ago,
included: Avoid ambiguity in a scientific test.
Regards,
Richard Lung.
On 02/01/2022 17:48, Kristofer Munsterhjelm wrote:
> On 02.01.2022 18:32, Richard Lung wrote:
>> Yes, Robert, Condorcet methods are not my specialty.
>>
>> But 'What is "an exclusion count"? Or "an election count?"' That is
>> a good question, not at all pedantic.
>> Roughly speaking, existing voting methods are election counts helped
>> out by ad hoc exclusion rules. All traditional stv, including Meek,
>> works this way, by getting rid of the candidate least in the way, when
>> the transfers of surplus votes run out. -- "Premature exclusion."
> If I understand correctly, then only methods that actually do candidate
> elimination (and fail LIIA) make use of exclusion counts. Minmax (the
> Condorcet method) doesn't. For that matter, plain old FPTP/Plurality
> doesn't either: the winner is the candidate with the most first
> preferences, and there are no eliminations as part of the process.
>
> Did I get that right?
>
> -km
KM
Kristofer Munsterhjelm
Tue, Jan 4, 2022 10:55 PM
On 04.01.2022 21:02, Richard Lung wrote:
KM,
I suspect those comments stem from (an unsurprising) unfamiliarity with
binomial stv. It has an exclusion count but it does not exclude or
eliminate candidates, during the binomial count (count of elections and
count of exclusions, before an over-all deciding count) and so does not
fall foul of irrationalities from "premature exclusion" etc.
Well, when you're saying that:
By the way, the point is that an election method should make use of
an exclusion count, as well as an election count.
You're saying something about what's desirable for any election method,
not just binomial STV. My point is that the very concept "exclusion
count" might not even make sense in methods that are sufficiently
different from binomial STV.
I might be misunderstanding the concept of just what an exclusion count
is, hence my question. But if the concept is to be applicable in full
generality, so that we can say that a method with an exclusion count is
better than one that isn't, then there must be some way of unambiguously
determining whether some arbitrary given election method has an
exclusion count or not. And I don't quite understand how that is to be
decided.
In other words, suppose that someone gives me an election method (like
FPTP or Borda or Kemeny or Approval). How do I determine if it has an
exclusion count?
-km
On 04.01.2022 21:02, Richard Lung wrote:
>
> KM,
>
> I suspect those comments stem from (an unsurprising) unfamiliarity with
> binomial stv. It has an exclusion count but it does not exclude or
> eliminate candidates, during the binomial count (count of elections and
> count of exclusions, before an over-all deciding count) and so does not
> fall foul of irrationalities from "premature exclusion" etc.
Well, when you're saying that:
> By the way, the point is that an election method should make use of
> an exclusion count, as well as an election count.
You're saying something about what's desirable for any election method,
not just binomial STV. My point is that the very concept "exclusion
count" might not even make sense in methods that are sufficiently
different from binomial STV.
I might be misunderstanding the concept of just what an exclusion count
is, hence my question. But if the concept is to be applicable in full
generality, so that we can say that a method with an exclusion count is
better than one that isn't, then there must be some way of unambiguously
determining whether some arbitrary given election method has an
exclusion count or not. And I don't quite understand how that is to be
decided.
In other words, suppose that someone gives me an election method (like
FPTP or Borda or Kemeny or Approval). How do I determine if it has an
exclusion count?
-km
FS
Forest Simmons
Wed, Jan 5, 2022 3:35 AM
Kristofer and All,
What I get from this is to use a Coombs/STV hybrid. Perhaps mostly Coombs
for Elimination and STV quotas for election ... something like that.
El mar., 4 de ene. de 2022 2:55 p. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
On 04.01.2022 21:02, Richard Lung wrote:
KM,
I suspect those comments stem from (an unsurprising) unfamiliarity with
binomial stv. It has an exclusion count but it does not exclude or
eliminate candidates, during the binomial count (count of elections and
count of exclusions, before an over-all deciding count) and so does not
fall foul of irrationalities from "premature exclusion" etc.
Well, when you're saying that:
By the way, the point is that an election method should make use of
an exclusion count, as well as an election count.
You're saying something about what's desirable for any election method,
not just binomial STV. My point is that the very concept "exclusion
count" might not even make sense in methods that are sufficiently
different from binomial STV.
I might be misunderstanding the concept of just what an exclusion count
is, hence my question. But if the concept is to be applicable in full
generality, so that we can say that a method with an exclusion count is
better than one that isn't, then there must be some way of unambiguously
determining whether some arbitrary given election method has an
exclusion count or not. And I don't quite understand how that is to be
decided.
In other words, suppose that someone gives me an election method (like
FPTP or Borda or Kemeny or Approval). How do I determine if it has an
exclusion count?
-km
Election-Methods mailing list - see https://electorama.com/em for list
info
Kristofer and All,
What I get from this is to use a Coombs/STV hybrid. Perhaps mostly Coombs
for Elimination and STV quotas for election ... something like that.
El mar., 4 de ene. de 2022 2:55 p. m., Kristofer Munsterhjelm <
km_elmet@t-online.de> escribió:
> On 04.01.2022 21:02, Richard Lung wrote:
> >
> > KM,
> >
> > I suspect those comments stem from (an unsurprising) unfamiliarity with
> > binomial stv. It has an exclusion count but it does not exclude or
> > eliminate candidates, during the binomial count (count of elections and
> > count of exclusions, before an over-all deciding count) and so does not
> > fall foul of irrationalities from "premature exclusion" etc.
>
> Well, when you're saying that:
>
> > By the way, the point is that an election method should make use of
> > an exclusion count, as well as an election count.
> You're saying something about what's desirable for any election method,
> not just binomial STV. My point is that the very concept "exclusion
> count" might not even make sense in methods that are sufficiently
> different from binomial STV.
>
> I might be misunderstanding the concept of just what an exclusion count
> is, hence my question. But if the concept is to be applicable in full
> generality, so that we can say that a method with an exclusion count is
> better than one that isn't, then there must be some way of unambiguously
> determining whether some arbitrary given election method has an
> exclusion count or not. And I don't quite understand how that is to be
> decided.
>
> In other words, suppose that someone gives me an election method (like
> FPTP or Borda or Kemeny or Approval). How do I determine if it has an
> exclusion count?
>
> -km
> ----
> Election-Methods mailing list - see https://electorama.com/em for list
> info
>
