Condorcet meeting
I was browsing Reddit when I found this:
Apparently Tideman is organizing a discussion about Condorcet methods in
September. I thought it might be interesting to EM members, although I
don't know if they'd admit more people this late!
-km
One of the agenda items is "rules for reducing a large field of
candidates to a field of 2 to 5". This seems to me an important topic,
since voters cannot be expected to vote in the way ranked preference
methods assume if the number of candidates is large. Presumably
proposals have been made for addressing it; unfortunately I haven't seen
them.
One which occurs to me is this. There are two rounds, the first of which
may have up to (say) 20 candidates and reduces the number to (say) 6,
and the second of which uses ranked preferences to find a single winner
from the 6.
In the first round, each candidate nominates up to 3 "alternates"
from the rest of the field; these are listed in order against his name
on the ballot. The ballots use FPTP format - choose one option - but
are processed as multiwinner STV ballots in which a vote for A is
treated as a vote A>B>C>D, where B...D are his alternates. Surplus votes
are not transferred. Choose the 6 winners as the survivors to the second
round.
I dare say there exist better ideas, but I thought I'd mention it.
CJC
On 24/08/2023 12:33, Kristofer Munsterhjelm wrote:
I was browsing Reddit when I found this:
Apparently Tideman is organizing a discussion about Condorcet methods
in September. I thought it might be interesting to EM members,
although I don't know if they'd admit more people this late!
-km
Election-Methods mailing list - see https://electorama.com/em for list
info
On 2023-08-24 14:29, Colin Champion wrote:
One of the agenda items is "rules for reducing a large field of
candidates to a field of 2 to 5". This seems to me an important topic,
since voters cannot be expected to vote in the way ranked preference
methods assume if the number of candidates is large. Presumably
proposals have been made for addressing it; unfortunately I haven't seen
them.
My intuitive response is to do some type of open Approval primary. The
incommensurability and chicken dilemma problems hit a lot less when
there are 5 candidates to spread the error over.
Of course, the naive STAR approach is not cloneproof, which is a bit of
a bummer.
From one perspective, we'd want the "ranked general" to have a pool
of differing shades of centrist, because the general's purpose would be
to determine which out of the reduced pool is actually the best
candidate. But this approach seems to be fundamentally clone vulnerable,
i.e. it's not something you could patch up, because clones of the winner
would all seem to be reasonable centrists in their own right. There's a
tension between doing most of the median-finding in the primary, and
clone independence.
So there may still be room for proportional voting type patches, even
though they draw the general further from the idea of closely
scrutinizing similar candidates, and even though they are somewhat ugly
kludges.
Or perhaps there exists a better method :-)
-km
On 24/08/2023 14:02, Kristofer Munsterhjelm wrote:
My intuitive response is to do some type of open Approval primary.
The incommensurability and chicken dilemma problems hit a lot less
when there are 5 candidates to spread the error over.
I wouldn't have thought that this would work very well. One of the
problems is that most voters will not initially know much about most
candidates, so they won't give them approval.
It seems to me that voter ignorance is the main problem with ranked
voting if the field is large. Voters will truncate out the candidates
they don't know much about, and this will be misinterpreted as a low
preference. So the first round should give obscure candidates a chance
of the spotlight.
CJC
On 2023-08-24 14:29, Colin Champion wrote:
One of the agenda items is "rules for reducing a large field of
candidates to a field of 2 to 5". This seems to me an important
topic, since voters cannot be expected to vote in the way ranked
preference methods assume if the number of candidates is large.
Presumably proposals have been made for addressing it; unfortunately
I haven't seen them.
My intuitive response is to do some type of open Approval primary. The
incommensurability and chicken dilemma problems hit a lot less when
there are 5 candidates to spread the error over.
Of course, the naive STAR approach is not cloneproof, which is a bit
of a bummer.
From one perspective, we'd want the "ranked general" to have a pool
of differing shades of centrist, because the general's purpose would
be to determine which out of the reduced pool is actually the best
candidate. But this approach seems to be fundamentally clone
vulnerable, i.e. it's not something you could patch up, because clones
of the winner would all seem to be reasonable centrists in their own
right. There's a tension between doing most of the median-finding in
the primary, and clone independence.
So there may still be room for proportional voting type patches, even
though they draw the general further from the idea of closely
scrutinizing similar candidates, and even though they are somewhat
ugly kludges.
Or perhaps there exists a better method :-)
-km
Also, I'm not sure if cardinal voting (such as Star) solves anything.
For most voters, sorting a list of candidates has quadratic work, and
has to be avoided if the field is large. But is cardinal voting any
better? If the scores had an intrinsic meaning, you could work through
the list and assign the correct value to each candidate. But they don't,
so you'll start by assigning tentative scores until you find an
inconsistency - eg. two candidates have the same score but vastly
different acceptability - and then you have to adjust the numbers you've
already given. You'll probably end up sorting and grouping into quintiles.
CJC
On 24/08/2023 14:31, Colin Champion wrote:
On 24/08/2023 14:02, Kristofer Munsterhjelm wrote:
My intuitive response is to do some type of open Approval primary.
The incommensurability and chicken dilemma problems hit a lot less
when there are 5 candidates to spread the error over.
I wouldn't have thought that this would work very well. One of the
problems is that most voters will not initially know much about most
candidates, so they won't give them approval.
It seems to me that voter ignorance is the main problem with ranked
voting if the field is large. Voters will truncate out the candidates
they don't know much about, and this will be misinterpreted as a low
preference. So the first round should give obscure candidates a chance
of the spotlight.
CJC
On 2023-08-24 14:29, Colin Champion wrote:
One of the agenda items is "rules for reducing a large field of
candidates to a field of 2 to 5". This seems to me an important
topic, since voters cannot be expected to vote in the way ranked
preference methods assume if the number of candidates is large.
Presumably proposals have been made for addressing it; unfortunately
I haven't seen them.
My intuitive response is to do some type of open Approval primary.
The incommensurability and chicken dilemma problems hit a lot less
when there are 5 candidates to spread the error over.
Of course, the naive STAR approach is not cloneproof, which is a bit
of a bummer.
From one perspective, we'd want the "ranked general" to have a pool
of differing shades of centrist, because the general's purpose would
be to determine which out of the reduced pool is actually the best
candidate. But this approach seems to be fundamentally clone
vulnerable, i.e. it's not something you could patch up, because
clones of the winner would all seem to be reasonable centrists in
their own right. There's a tension between doing most of the
median-finding in the primary, and clone independence.
So there may still be room for proportional voting type patches, even
though they draw the general further from the idea of closely
scrutinizing similar candidates, and even though they are somewhat
ugly kludges.
Or perhaps there exists a better method :-)
-km
Election-Methods mailing list - see https://electorama.com/em for list
info
On 2023-08-24 15:55, Colin Champion wrote:
Also, I'm not sure if cardinal voting (such as Star) solves anything.
For most voters, sorting a list of candidates has quadratic work, and
has to be avoided if the field is large. But is cardinal voting any
better? If the scores had an intrinsic meaning, you could work through
the list and assign the correct value to each candidate. But they don't,
so you'll start by assigning tentative scores until you find an
inconsistency - eg. two candidates have the same score but vastly
different acceptability - and then you have to adjust the numbers you've
already given. You'll probably end up sorting and grouping into quintiles.
My idea with approval was just that voters would think (at least I
would) "I'll just approve anyone who I think is reasonably good, we'll
sort it out in the general". While having to use approval to elect a
single candidate would be very hard (what's the other voters' scalings?
what should mine be? how do I deal with the Burr dilemma? etc), it's
easier when the particular ballot is much more accommodating of noise.
This would fail if we get to a multiparty/candidate system that's
vibrant enough that there are more than six viable candidates. At that
point we'd be kind of out of luck; at least the overhead of being
careful with one's vote would return.
-km
On 2023-08-24 15:31, Colin Champion wrote:
On 24/08/2023 14:02, Kristofer Munsterhjelm wrote:
My intuitive response is to do some type of open Approval primary.
The incommensurability and chicken dilemma problems hit a lot less
when there are 5 candidates to spread the error over.
I wouldn't have thought that this would work very well. One of the
problems is that most voters will not initially know much about most
candidates, so they won't give them approval.
It seems to me that voter ignorance is the main problem with ranked
voting if the field is large. Voters will truncate out the candidates
they don't know much about, and this will be misinterpreted as a low
preference. So the first round should give obscure candidates a chance
of the spotlight.
That's a good point. I was thinking of a different setting, where there
are say, ten established parties with good track records, each of which
fields a candidate, and you have to narrow down the field to something
manageable before doing proper ranked voting.
But now that I think about it, in such a setting, voters could probably
just rank and truncate. E.g. a left-wing voter ranks all the left-wing
parties' candidates and then truncates.
About your setting, I think that points to a more fundamental problem of
electoral democracy: only those who can make themselves heard or somehow
get popular have a chance of being elected. That's kind of your
argument, but with number of candidates equalling the number of voters;
somewhere in the mass that is the population, there's someone whose
political ideas would appeal to enough people that he'd be elected if
they knew about them. But since marketing is costly, they don't.
The solutions I most commonly know of for that problem are sortition,
asset voting, or some way of dynamically discovering the good candidates
by parallelism (e.g. Gohlke's triad method).
Your predeclared orders are kind of like asset. Sortition could either
be done directly, or by having a randomly selected group choose the five
finalists after examining candidates' platforms in detail. And parallel
methods would replace elections with something different.
I'm wary of predeclared orders because they could lead to obscure deals
(like Chris mentioned regarding above-the-line voting in Australia).
Picking five candidates instead of a single winner would mitigate the
impact, but still. I'd think after-the-fact public asset trading, with
the voters being able to see who their candidate is contributing to,
would be better. But that would require a deeper change which could be
infeasible.
-km
Kristofer - I certainly agree with your last point. A weakness I'm
conscious of in my proposal is that a TV personality popular among
silver voters might put himself forward for the sole purpose of
directing his second preference votes to a sinister candidate unwelcome
to his personal supporters. But at least they're told up front what
they're voting for.
CJC
On 24/08/2023 17:52, Kristofer Munsterhjelm wrote:
On 2023-08-24 15:31, Colin Champion wrote:
On 24/08/2023 14:02, Kristofer Munsterhjelm wrote:
> My intuitive response is to do some type of open Approval primary.
> The incommensurability and chicken dilemma problems hit a lot less
> when there are 5 candidates to spread the error over.
I wouldn't have thought that this would work very well. One of the
problems is that most voters will not initially know much about most
candidates, so they won't give them approval.
It seems to me that voter ignorance is the main problem with
ranked voting if the field is large. Voters will truncate out the
candidates they don't know much about, and this will be
misinterpreted as a low preference. So the first round should give
obscure candidates a chance of the spotlight.
That's a good point. I was thinking of a different setting, where
there are say, ten established parties with good track records, each
of which fields a candidate, and you have to narrow down the field to
something manageable before doing proper ranked voting.
But now that I think about it, in such a setting, voters could
probably just rank and truncate. E.g. a left-wing voter ranks all the
left-wing parties' candidates and then truncates.
About your setting, I think that points to a more fundamental problem
of electoral democracy: only those who can make themselves heard or
somehow get popular have a chance of being elected. That's kind of
your argument, but with number of candidates equalling the number of
voters; somewhere in the mass that is the population, there's someone
whose political ideas would appeal to enough people that he'd be
elected if they knew about them. But since marketing is costly, they
don't.
The solutions I most commonly know of for that problem are sortition,
asset voting, or some way of dynamically discovering the good
candidates by parallelism (e.g. Gohlke's triad method).
Your predeclared orders are kind of like asset. Sortition could either
be done directly, or by having a randomly selected group choose the
five finalists after examining candidates' platforms in detail. And
parallel methods would replace elections with something different.
I'm wary of predeclared orders because they could lead to obscure
deals (like Chris mentioned regarding above-the-line voting in
Australia). Picking five candidates instead of a single winner would
mitigate the impact, but still. I'd think after-the-fact public asset
trading, with the voters being able to see who their candidate is
contributing to, would be better. But that would require a deeper
change which could be infeasible.
-km
I agree with Kristofer that Approval is plenty good for the narrowing down
phase.
Your favorite pundits and candidates will definitely make known their
recommendations. Trust your own judgment and gut, as you collate and cull
out their llists of recommendations.
If there are going to be only six finalists, that doesn't mean you can only
approve six or that you have to approve more than one.
My rule is to approve my favorite as well as everybody else that I like
almost as much.
Here's an idea for deciding on n, the number of finalists after the
approval ballots have been tallied:
For this purpose, temporarily count the ballots fractionally, and let f(X)
be the fraction of the total that X gets in this tally ... so that the f(X)
values sum to unity.
The value of n should be the reciprocal of the sum of the squares of the
f(X) vslues... the standard formula for the minimum number of seats that
would be acceptable for proportional representation of a diverse population.
fws
On Thu, Aug 24, 2023, 11:11 AM Colin Champion <
colin.champion@routemaster.app> wrote:
Kristofer - I certainly agree with your last point. A weakness I'm
conscious of in my proposal is that a TV personality popular among silver
voters might put himself forward for the sole purpose of directing his
second preference votes to a sinister candidate unwelcome to his personal
supporters. But at least they're told up front what they're voting for.
CJC
On 24/08/2023 17:52, Kristofer Munsterhjelm wrote:
On 2023-08-24 15:31, Colin Champion wrote:
On 24/08/2023 14:02, Kristofer Munsterhjelm wrote:
My intuitive response is to do some type of open Approval primary.
The incommensurability and chicken dilemma problems hit a lot less
when there are 5 candidates to spread the error over.
I wouldn't have thought that this would work very well. One of the
problems is that most voters will not initially know much about most
candidates, so they won't give them approval.
It seems to me that voter ignorance is the main problem with ranked
voting if the field is large. Voters will truncate out the candidates they
don't know much about, and this will be misinterpreted as a low preference.
So the first round should give obscure candidates a chance of the
spotlight.
That's a good point. I was thinking of a different setting, where there
are say, ten established parties with good track records, each of which
fields a candidate, and you have to narrow down the field to something
manageable before doing proper ranked voting.
But now that I think about it, in such a setting, voters could probably
just rank and truncate. E.g. a left-wing voter ranks all the left-wing
parties' candidates and then truncates.
About your setting, I think that points to a more fundamental problem of
electoral democracy: only those who can make themselves heard or somehow
get popular have a chance of being elected. That's kind of your argument,
but with number of candidates equalling the number of voters; somewhere in
the mass that is the population, there's someone whose political ideas
would appeal to enough people that he'd be elected if they knew about them.
But since marketing is costly, they don't.
The solutions I most commonly know of for that problem are sortition,
asset voting, or some way of dynamically discovering the good candidates by
parallelism (e.g. Gohlke's triad method).
Your predeclared orders are kind of like asset. Sortition could either be
done directly, or by having a randomly selected group choose the five
finalists after examining candidates' platforms in detail. And parallel
methods would replace elections with something different.
I'm wary of predeclared orders because they could lead to obscure deals
(like Chris mentioned regarding above-the-line voting in Australia).
Picking five candidates instead of a single winner would mitigate the
impact, but still. I'd think after-the-fact public asset trading, with the
voters being able to see who their candidate is contributing to, would be
better. But that would require a deeper change which could be infeasible.
-km
Election-Methods mailing list - see https://electorama.com/em for list
info
On 2023-08-25 01:50, Forest Simmons wrote:
I agree with Kristofer that Approval is plenty good for the narrowing
down phase.
Your favorite pundits and candidates will definitely make known their
recommendations. Trust your own judgment and gut, as you collate and
cull out their llists of recommendations.
If there are going to be only six finalists, that doesn't mean you can
only approve six or that you have to approve more than one.
My rule is to approve my favorite as well as everybody else that I like
almost as much.
Here's an idea for deciding on n, the number of finalists after the
approval ballots have been tallied:
For this purpose, temporarily count the ballots fractionally, and let
f(X) be the fraction of the total that X gets in this tally ... so that
the f(X) values sum to unity.
The value of n should be the reciprocal of the sum of the squares of the
f(X) vslues... the standard formula for the minimum number of seats that
would be acceptable for proportional representation of a diverse population.
Another option is to use the exponential of the Shannon entropy:
https://electowiki.org/wiki/Effective_number_of_parties#Entropy_measure
-km