Is autodeterrence bad? (+STAR fails hyperchicken catastrophically)
I think the EM mailing list made a wrong turn a while back in
misunderstanding "autodeterrence" as a positive attribute of a voting
system, which discourages burial. I'd like to put forth an argument (which
I'm still somewhat hesitant about) that it's a very bad property. Given
the chance, political machines are likely to reach out and grab this third
rail as hard as they can, even if it's terrible for their constituents,
because it maximizes their chances of election.
Start with a 2.5-candidate race between Gore, Bush, and a Nazi (who has a
small, but slightly above zero, level of support). Gore doesn't know
whether he or Bush is more popular in a runoff, but he's certain he and
Bush will make it to the runoff with honest voting. However, he realizes he
can use the Nazi as a bludgeon to increase his chances of winning. He tells
his supporters to cast votes as follows:
Gore – 5/5
Bush – 0/5
Nazi – 4/5
Gore's hope is that the Nazi is polarizing enough to defeat Bush for second
place with Gore's support (at which point he's a weak candidate in the
runoff). Risky? Yes. But it's still plausibly strategic if you think Bush
will back down.
But if Bush's faction thinks the same thing, the Nazi ends up winning.
STAR punishes burial by blowing up the country, creating an extremely
high-stakes game of chicken (hyperchicken?). This game has a mixed Nash
equilibrium that involves blowing up the country with some small (but
positive) probability. The issue isn't that burial is incentivized; it's
that it can work, but when it fails, it's so strongly disincentivized
that it can be catastrophic.
This can be especially bad since incentives are even stronger for
candidates and campaigns. Campaigns coordinate strategy; voters take cues
from campaigns and political elites (which is why the two major-party
nominees are always the top-2 winners).
The strategy above would be bad for society, and ambiguous for individual
voters (it could elect either Gore or a Nazi). On the other hand, great for
Gore's probability of winning, if Gore
Empirically, this kind of turkey-raising happens all the time. Adam Schiff
spent millions trying to boost the Republican in California over Katie
Porter. The DNC keeps intervening in Republican primaries to help nominate
extremists. They keep doing this because they think it's good for their own
personal chances of winning the election, not because they think it's good
for the country overall. And generally, they're right—even though it risks
electing Nazis, it probably helps Democrats win a few more seats.
That post is thoroughly bizarre:
Autodeterence is defined as a property for Condorcet methods, & Closed
thinks he’s somehow discrediting it by telling about a STAR situation.
We’re getting leakage & spillover from the Voting Science Forum.
On Wed, Apr 3, 2024 at 11:09 Closed Limelike Curves <
closed.limelike.curves@gmail.com> wrote:
I think the EM mailing list made a wrong turn a while back in
misunderstanding "autodeterrence" as a positive attribute of a voting
system, which discourages burial. I'd like to put forth an argument (which
I'm still somewhat hesitant about) that it's a very bad property. Given
the chance, political machines are likely to reach out and grab this third
rail as hard as they can, even if it's terrible for their constituents,
because it maximizes their chances of election.
Start with a 2.5-candidate race between Gore, Bush, and a Nazi (who has a
small, but slightly above zero, level of support). Gore doesn't know
whether he or Bush is more popular in a runoff, but he's certain he and
Bush will make it to the runoff with honest voting. However, he realizes he
can use the Nazi as a bludgeon to increase his chances of winning. He tells
his supporters to cast votes as follows:
Gore – 5/5
Bush – 0/5
Nazi – 4/5
Gore's hope is that the Nazi is polarizing enough to defeat Bush for
second place with Gore's support (at which point he's a weak candidate in
the runoff). Risky? Yes. But it's still plausibly strategic if you think
Bush will back down.
But if Bush's faction thinks the same thing, the Nazi ends up winning.
STAR punishes burial by blowing up the country, creating an extremely
high-stakes game of chicken (hyperchicken?). This game has a mixed Nash
equilibrium that involves blowing up the country with some small (but
positive) probability. The issue isn't that burial is incentivized; it's
that it can work, but when it fails, it's so strongly disincentivized
that it can be catastrophic.
This can be especially bad since incentives are even stronger for
candidates and campaigns. Campaigns coordinate strategy; voters take cues
from campaigns and political elites (which is why the two major-party
nominees are always the top-2 winners).
The strategy above would be bad for society, and ambiguous for individual
voters (it could elect either Gore or a Nazi). On the other hand, great for
Gore's probability of winning, if Gore
Empirically, this kind of turkey-raising happens all the time. Adam Schiff
spent millions trying to boost the Republican in California over Katie
Porter. The DNC keeps intervening in Republican primaries to help nominate
extremists. They keep doing this because they think it's good for their own
personal chances of winning the election, not because they think it's good
for the country overall. And generally, they're right—even though it risks
electing Nazis, it probably helps Democrats win a few more seats.
Election-Methods mailing list - see https://electorama.com/em for list
info
On 2024-04-03 20:09, Closed Limelike Curves wrote:
I think the EM mailing list made a wrong turn a while back in
misunderstanding "autodeterrence" as a positive attribute of a voting
system, which discourages burial. I'd like to put forth an argument
(which I'm still somewhat hesitant about) that it's a very /bad/
property. Given the chance, political machines are likely to reach out
and grab this third rail as hard as they can, even if it's terrible for
their constituents, because it maximizes their chances of election.
Start with a 2.5-candidate race between Gore, Bush, and a Nazi (who has
a small, but slightly above zero, level of support). Gore doesn't know
whether he or Bush is more popular in a runoff, but he's certain he and
Bush will make it to the runoff with honest voting. However, he realizes
he can use the Nazi as a bludgeon to increase his chances of winning. He
tells his supporters to cast votes as follows:
Gore – 5/5
Bush – 0/5
Nazi – 4/5
Gore's hope is that the Nazi is polarizing enough to defeat Bush for
second place with Gore's support (at which point he's a weak candidate
in the runoff). Risky? Yes. But it's still plausibly strategic if you
think Bush will back down.
But if Bush's faction thinks the same thing, the Nazi ends up winning.
STAR punishes burial by blowing up the country, creating an extremely
high-stakes game of chicken (hyperchicken?). This game has a mixed Nash
equilibrium that involves blowing up the country with some small (but
positive) probability. The issue isn't that burial is incentivized; it's
that it /can/ work, but when it fails, it's so strongly
/dis/incentivized that it can be catastrophic.
I would expect that DMTCBR methods would pass (be unaffected by) this as
long as there's a sincere CW.
Characterizing the Nash equilibria of DMTCBR/DMTBR/resistant set methods
would be interesting, but it's not like I'm lacking in work already :-)
-km
I think one problem of burial-resistant methods is that they assume the electorate are aware of the consequences of it and will act accordingly. I think it might be a bit optimistic to expect the average voter to behave any differently using any method that uses a specific ballot type. Using a ranked ballot, if A and B are the frontrunning candidates, then supporters of A might rank B bottom because it's the obvious thing to do (which has been pointed out on here before I believe). Do you think the adoption of a specific Condorcet method will prevent that? I'm not convinced.
Also, if there are two frontrunning candidates, A and B, it's quite likely anyway that supporters of A will see B as the worst candidate anyway, below the ones they know very little about. So it wouldn't really even be an act of burial, and therefore honest voting behaviour could cause a non-entity to win, because this is what burial-resistant methods do.
I've said this before, but possibly the best solution for a Condorcet method would to be to use rated ballots. In this case B is less likely to be buried by the A supporters, because they would be likely to score the non-entity candidates 0 as well.
Toby
On Wednesday, 3 April 2024 at 19:09:50 BST, Closed Limelike Curves closed.limelike.curves@gmail.com wrote:
I think the EM mailing list made a wrong turn a while back in misunderstanding "autodeterrence" as a positive attribute of a voting system, which discourages burial. I'd like to put forth an argument (which I'm still somewhat hesitant about) that it's a very bad property. Given the chance, political machines are likely to reach out and grab this third rail as hard as they can, even if it's terrible for their constituents, because it maximizes their chances of election.
Start with a 2.5-candidate race between Gore, Bush, and a Nazi (who has a small, but slightly above zero, level of support). Gore doesn't know whether he or Bush is more popular in a runoff, but he's certain he and Bush will make it to the runoff with honest voting. However, he realizes he can use the Nazi as a bludgeon to increase his chances of winning. He tells his supporters to cast votes as follows:
Gore – 5/5Bush – 0/5Nazi – 4/5
Gore's hope is that the Nazi is polarizing enough to defeat Bush for second place with Gore's support (at which point he's a weak candidate in the runoff). Risky? Yes. But it's still plausibly strategic if you think Bush will back down.
But if Bush's faction thinks the same thing, the Nazi ends up winning.
STAR punishes burial by blowing up the country, creating an extremely high-stakes game of chicken (hyperchicken?). This game has a mixed Nash equilibrium that involves blowing up the country with some small (but positive) probability. The issue isn't that burial is incentivized; it's that it can work, but when it fails, it's so strongly disincentivized that it can be catastrophic.
This can be especially bad since incentives are even stronger for candidates and campaigns. Campaigns coordinate strategy; voters take cues from campaigns and political elites (which is why the two major-party nominees are always the top-2 winners).
The strategy above would be bad for society, and ambiguous for individual voters (it could elect either Gore or a Nazi). On the other hand, great for Gore's probability of winning, if Gore
Empirically, this kind of turkey-raising happens all the time. Adam Schiff spent millions trying to boost the Republican in California over Katie Porter. The DNC keeps intervening in Republican primaries to help nominate extremists. They keep doing this because they think it's good for their own personal chances of winning the election, not because they think it's good for the country overall. And generally, they're right—even though it risks electing Nazis, it probably helps Democrats win a few more seats.
Election-Methods mailing list - see https://electorama.com/em for list info
Definitely agree on rated ballots. Besides what you said, I think ratings
are probably easier to fill out than rankings. As a bonus they also open up
a lot of good ways to resolve cycles:
- James Green-Armytage's approach lets you weight defeat strength going
by the ratings difference between the candidates. - I haven't seen analysis of this, but you could try defining pairwise
margins as the median difference in ratings X - Y (which is
positive/negative when most voters prefer X/Y). - Smith//Score is easier to explain than most resolution methods and a
top-tier resolution mechanism. - Landau//Quadratic voting—QV is generally a bad system for
single-winner elections, but it might work well if you pair it with the
Landau set because QV is honest in zero-info settings. If all candidates in
the Landau set are close to tied, QV gives honest voting.
On Thu, Apr 4, 2024 at 8:38 AM Toby Pereira tdp201b@yahoo.co.uk wrote:
I think one problem of burial-resistant methods is that they assume the
electorate are aware of the consequences of it and will act accordingly. I
think it might be a bit optimistic to expect the average voter to behave
any differently using any method that uses a specific ballot type. Using a
ranked ballot, if A and B are the frontrunning candidates, then supporters
of A might rank B bottom because it's the obvious thing to do (which has
been pointed out on here before I believe). Do you think the adoption of a
specific Condorcet method will prevent that? I'm not convinced.
Also, if there are two frontrunning candidates, A and B, it's quite likely
anyway that supporters of A will see B as the worst candidate anyway, below
the ones they know very little about. So it wouldn't really even be an act
of burial, and therefore honest voting behaviour could cause a non-entity
to win, because this is what burial-resistant methods do.
I've said this before, but possibly the best solution for a Condorcet
method would to be to use rated ballots. In this case B is less likely to
be buried by the A supporters, because they would be likely to score the
non-entity candidates 0 as well.
Toby
On Wednesday, 3 April 2024 at 19:09:50 BST, Closed Limelike Curves <
closed.limelike.curves@gmail.com> wrote:
I think the EM mailing list made a wrong turn a while back in
misunderstanding "autodeterrence" as a positive attribute of a voting
system, which discourages burial. I'd like to put forth an argument (which
I'm still somewhat hesitant about) that it's a very bad property. Given
the chance, political machines are likely to reach out and grab this third
rail as hard as they can, even if it's terrible for their constituents,
because it maximizes their chances of election.
Start with a 2.5-candidate race between Gore, Bush, and a Nazi (who has a
small, but slightly above zero, level of support). Gore doesn't know
whether he or Bush is more popular in a runoff, but he's certain he and
Bush will make it to the runoff with honest voting. However, he realizes he
can use the Nazi as a bludgeon to increase his chances of winning. He tells
his supporters to cast votes as follows:
Gore – 5/5
Bush – 0/5
Nazi – 4/5
Gore's hope is that the Nazi is polarizing enough to defeat Bush for
second place with Gore's support (at which point he's a weak candidate in
the runoff). Risky? Yes. But it's still plausibly strategic if you think
Bush will back down.
But if Bush's faction thinks the same thing, the Nazi ends up winning.
STAR punishes burial by blowing up the country, creating an extremely
high-stakes game of chicken (hyperchicken?). This game has a mixed Nash
equilibrium that involves blowing up the country with some small (but
positive) probability. The issue isn't that burial is incentivized; it's
that it can work, but when it fails, it's so strongly disincentivized
that it can be catastrophic.
This can be especially bad since incentives are even stronger for
candidates and campaigns. Campaigns coordinate strategy; voters take cues
from campaigns and political elites (which is why the two major-party
nominees are always the top-2 winners).
The strategy above would be bad for society, and ambiguous for individual
voters (it could elect either Gore or a Nazi). On the other hand, great for
Gore's probability of winning, if Gore
Empirically, this kind of turkey-raising happens all the time. Adam Schiff
spent millions trying to boost the Republican in California over Katie
Porter. The DNC keeps intervening in Republican primaries to help nominate
extremists. They keep doing this because they think it's good for their own
personal chances of winning the election, not because they think it's good
for the country overall. And generally, they're right—even though it risks
electing Nazis, it probably helps Democrats win a few more seats.
Election-Methods mailing list - see https://electorama.com/em for list
info