[EM] Approval-Strategy article at CES website

[Michael Ossipoff]

On Fri, Nov 4, 2016 at 5:35 AM, Juho Laatu <juho.laatu at gmail.com> wrote:

>
> On 04 Nov 2016, at 03:30, Michael Ossipoff <email9648742 at gmail.com> wrote:
>
> First, two things: Most of what I've said about the top-set was meant to
> be about the strong top-set.
>
> 1.
>
> I say that, for most people, based on what they say they want, don't want,
> and don't like, there are, for them, a strong top-set and a strong
> bottom-set.
>
>
> I guess you can say that's typical for all people in all democracies. The
> strength of those groups varies of course.
>


Yes, most likely anywhere there are candidates and parties that are honest
and offer what people actually want, and there are also dishonest
candidates & parties who are financed by and serve the rich.

...And so, for everyone who isn't rich, the choice is starkly clear,
resulting in a strong top-set an a strong bottom-set.

...and so, anywhere, Approval voting is completely easy: Approve only your
top-set, the honest pro-people candidates.


>
>
>
> But, as you said, in the general case, not specific to any country, or to
> any era, but, instead, just hypothetically, a voter might not have a strong
> top-set, or even a top-set.
>
> 2.
>
> When you speak of normal Approval strategy, I presume that you're
> referring to expectation-maximizing strategy for the general case, when
> there might not be a top-set or a strong top-set.
>
>
> Yes. And even if one has a strong top-set and a strong bottom-set, the
> normal approval strategy would help in casting an efficient vote
>

But, when there's a strong top-set and a strong bottom-set, approving
(only) the strong top-set _is_ the expectation maximizing strategy, the
normal approval strategy, the efficient vote.




> , even if the person would prefer to cast a vote that is not optimally
> efficient (from the expectation-maximizing point of view) (but optimally
> best for him/her in some other way).
>

Certainly. If there's no strong top-set & strong bottom-set, then a voter
might feel like voting differently from what an expectation-maximizing
strategy says to do. And, as you said, s/he should then do so.

But, if there's a strong top-set and strong bottom-set, then I guarantee
that a voter _will_ feel like approving all of the strong top-set and none
of the strong bottom set.


>
>
>
> Of course that strategy has been described or proposed in many forms.
>
> In general, I'd say that it means approving down to the CWs, if there is
> one, and if it's known or guessed who s/he is. Or, if there isn't one, or
> if it's a 0-info election, then approving down to the expected
> winner-merit, however estimated.
>
>
> If there is a CWs (sincere Condorcet winner, as derived from the
> preferences of the voters when voting in the actual approval election), and
> all other voters are assumed to use the normal approval strategy, and all
> voters are assumed to base their voting decision in sufficiently accurate
> information on what the preferences of the voters are, then it makes sense
> to put the approval cutoff next to the (assumed) CWs.
>

Yes. Immediately below the CWs.


> (When discussing poll based strategies, one has to take into account also
> possible distortions in all the assumptions.)
>

The best and most reliable poll, free of strategic deceptive voting, is the
previous election.


>
> A simple example might be helpful here. Let's say there are seven
> candidates in a linear opinion space. A1, A2, B1 and B2 are strong
> candidates (e.g. democrat and republican), and all three C candidates are
> weak in the sense that they have very little first preference support.
>

You're assuming that the Republocrats have the most 1st preference support.
There's no reason to believe that that's true.

Whenever people get the opportunity to hear about candidates who offer
something better, the Republocrat is not their favorite. How could it be?

In the honest conditions of the Green scenario, the Republocrats would be
all finished.

Even now, with our disinformational media, Bernie was immensely popular,
while Hillary & Donald are the most strongly disliked presidential
candidates in the history of this country.

The fact that Hillary & Donald were reported to have "won" the primaries
means nothing. There's no reason to believe that Hillary actually outpolled
Bernie.

76% of the people wanted Jill & Gary (Libertarian) to be in the
presidential debates. A majority of the people want universal medical care.
All of our wars (including the Contra operation, a terrorist campaign
against the Nicaraguans) were opposed by a majority of the population.

etc., etc.

They wanted  Jill & Gary in the debates because they know that they want
something fresh, new & better.

...and that's why the Republocrat-managed debate excluded anything other
than the official state-sanctioned candidates, proposals, I& policies.

So let's not assume that the Republocrats are really the most favorite.



> The linear opinion space is A1 - A2 - C1 - C2 - C3 - B1 - B2. C2 is the
> CWs.
>

Yes, Internet polls that aren't promoted to partisan organizations don't
have a Republocrat as the CWs. They point to Jill Stein (previously Ralph
Nader, Bernie, or whoever the Green candidate is) as the CWs. (By the way,
as a progressive candidate, Stein would be in my strong top-set).


> A supporters should put their approval cutoff between C2 and C3. B
> supporters should put their approval cutoff between C1 and C2. C2 will be
> elected. I guess this is what you meant with "approving down to the CWs".
>

Yes.


>
> Note that this theory does not hold if we assume that some voters vote
> based on their personal preferences only (e.g. their top-set and bottom-set)
>

Agreed. If the CWs is in your strong bottom-set, then you shoudn't, and
wouldn't, approve down that far.

If you have a strong top-set and strong bottom-set, then approving (only)
your strong top-set takes precedence over all else.


> , or that they do not have accurate information about the preferences
> during the election day. In real life that "accurate information" can of
> course be only approximate information that the voter has derived e.g. from
> the polls and other pieces of information during the days leading to the
> election.
>

Certainly. For example, as i was saying, lacking reliable predictive
information, the election should be considered 0-info.

If someone isn't convinced when they're told that Jill is the CWs, then
they hopefully will at least accept that they don't have reliable
predictive information, and should use 0-info strategy.


>
> And then back to the poll strategies. The strategic plan of party A
> includes of course the target to make C1 or A2 the winner instead of C2 (or
> someone worse). It is difficult for the B supporters to decide where to put
> their approval cutoff. Why not tweak the poll results a bit (by supporters
> giving appropriate false information in the polls, or maybe by publishing
> false/modified polls). That might make some of them approve also C1, and
> maybe also A2. Or at least the C2 and C3 supporters might do so. Since C
> party is small, small changes in balance might lead to important changes in
> the outcome. It is quite possible that A2 will win as a result of the
> applied strategy.
>

Yes, quite so. Having a pre-election informational poll before the election
invites strategic deceptive voting, and could be a real mess.

Better to just use the previous election as the poll.


>
> . . .
>
> CES is going to do a poll, to test the various voting systems in a big
> national poll. Instead of doing it themselves, they're going to hire a
> company to do it, a company affiliated with the Associated-Press  :^ |
>
>
> A typical poll in Finland is one that has been ordered by some major
> newspaper or other media, but from from some well known neutral (possibly
> international) research company.
>

Yes, but you have to be skeptical when someone or some company claims to be
neutral. But maybe that's less so in Finland than here.

For example, there was the "Independent" 9/11 Commission, whose appointed
leadership, & choice of what to consider and what to leave out, suggested
an agenda other than independence.  :^)

And of course International doesn't necessarily mean neutral, in our
empire-based international world.


> The name of the company that implemented the poll and error margins are
> always mentioned when the poll results are published, to keep the trust
> level up.
>

Sure, the name of the company could make a difference in trust.


>
> . . .
>
>   Typical voters take into account their experience in life, results of
>> previous elections, all the polls that they have seen (with emphasis on
>> those that they consider reliable), and then make their educated guess on
>> who the potential winners are.
>>
>
> Yes.  But, here, it's best for most people (who never hear about the more
> reliable small-organization polls) to regard elections as entirely 0-info.
>
>
> If you have some semi-reliable or vague info available, you probably use
> it to the extent that its reliability level justifies.
>

Sure, but people here are far too trusting about that.


>
> . . .
>
> All points of view would be heard, and the "lesser-evil" advocates would
> be laughed off the stage.
>
>
> The normal approval strategy includes the notion of "lesser evil" since
> the voter doesn't necessarily like the frontrunners.
>

It all depends on how much the voter doesn't like a lesser-evil
frontrunner. If both predicted frontrunners are in your strong bottom-set,
then you wont't approve either frontrunner.  Nor should you.

But I was speaking of a future Utopian Green scenario in which the
lesser-evil is an unwinnable joke, not an expected frontrunner.

If neither the lesser-evil nor the greater evil is winnable, then you
certainly don't need the lesser-evil.

In no way would the normal Approval strategy say to approve hir.


>
> . . .
>
> There can be an approval election with exactly two potential winners. In
>> this situation the election is in practice just an election between two
>> candidates (if we focus only on the question who will win this time).
>>
>
> No. If a voter has a strong top-set and a strong bottom-set, and both of
> your alleged frontrunners are in the bottom-set, then the voter won't
> approve either. Hir optimal strategy is to approve (only) hir strong
> top-set.
>
> ...because polls, even if they were honest (Dream on), aren't 100%
> reliable.
>
>
> You must be saying that this voter assumes that the polls are not
> reliable, the voter knows better himself/herself
>

No, the voter needn't believe that s/he knows better than the poll. S/he
merely knows that the poll (for reasons of dishonesty or error) is less
than 100% reliable. And, because it isn't, then there's a chance that its
"frontrunners" won't really be frontrunners.  ...and that  therefore you
can elect from your strong top-set, and that your optimal strategy is to
approve only your strong top-set.

...And even if I believed that the winner had (for non-count-fraud reasons)
to be in my bottom-set, I'd probably nevertheless approve only my top-set,
because I don't really are which bottom-set candidate wins.

Of course if count fraud is the likely reason why the winner is one of the
media's 2 "frontrunners" (as is the case here, now), then the only right
vote is to boycott the election.



> ..., or guesses, and therefore put his approval cutoff in some other
> place. I'd say that this voter thinks that the two (or any number of)
> potential winners are different ones than the real ones
>

...or merely that the poll's "frontrunners" _might_ not be the actual
frontrunners.



> (that we assumed above), and therefore votes in a way that makes his/her
> vote void.
>

You're worrying too much about that. You just can't always, or even
usually, know who the frontrunners are going to be. So don't worry so much
about maybe not voting between the frontrunners.

For some reason, the following paragraph is automatially indented:


If both predictged frontrunners are in your srong bottom-set, then even the
slightest possibility that the prediction is wrong (and there always is
that possibility), then the "normal Approval strategy" still says to
approve only your strong top-set.

And so, no matter what the frontrunner prediction is, the "normal Approval
strategy" says to approve (only) your strong top-set. ...and it's right.

Of course if you don't have a strong top set, then maximizing your
expectation should be based on approving down to the CWs.

...or voting between the two candidates that you expect to be the 2
frontrunners (preferably with your Approval cutoff between the 2
frontrunners, placed so that your Approval cutoff's distance from each
frontrunner is inversely proportional to that frontrunner's probability of
outpolling the other frontrunner).


The voter did not manage to make a difference in the only relevant
> question, which one of the two potential winners to pick.
>

What's relevant to you might be irrelevant to someone else. Maybe that
other voter couldn't care less which strong bottom-set candidate wins.



> He/she was mistaken on what his/her optimal vote was
>


Certainly not.


> (continuing to assume that we now know that only the two mentioned
> candidates had any chances to win).
>

...and that assumption is where the error is. Predictive information, even
with honest polling, isn't 100% reliable.



>
> . . .
>
> Those voters that approve only one of those candidates will have a say in
>> this election.
>>
>
I couldn't care less about having a say about whether Hillary or Donald
wins.

(...even if I believed that the election was legitimate, and that Hillary
actually won the primary vote--but there's no reason to believe that.)



> Those voters that approve both or none of them will have no say in this
>> election.
>>
>
> I assure you that they couldn't care less which of two unacceptables wins.
>
>
> If that's what the "election" offers,  they won't mind not having a say in
> it.
>
>
> Quite possible. And leads to casting a vote that has no influence on who
> will be elected in this approval election.
>

Irrelevant if the winner has to be in your strong bottom-set. Anyway, even
if you think that it would be a little better to elect the lesser-evil in
your strong bottom-set, the normal approval strategy would still say to not
approve anyone in your bottom-set, because the polls aren't 100% reliable,
and, by the definition of "strong bottom-set", any merit difference between
bottom-set candidates is negligible compared to the difference between the
strong top-set and the strong bottom-set.



>
> . . .
>
> By normal strategy, you mean an expectation-maximizing strategy when there
> is no strong top-set or strong bottom-set.
>
>
> The normal strategy can be used with or without having / identifying
> strong top and bottom sets.
>

Did I say otherwise?  And the normal strategy can also be used when there
-is_ a strong top-set and a strong bottom-set. Then, it says to approve
only your strong top-set.




>
>
> But, when there is a strong top-set and a strong bottom-set, then the
> normal strategy, and the strong vote, the only strong vote, _is_ to approve
> (only) your strong top-set.
>
> So there's no conflict or distinction between normal strategy & strong
> vote, vs approving only all of one's strong top-set when there is one.
>
>
> If someone bases his/her decision on where to put the approval cutoff only
> on his/her personal preferences, without taking into account who the
> potential winners are (if any half reliable such information is available),
> then some votes are probably not strong / efficient in the sense that they
> would influence the outcome of this election as strongly as possible.
>
> Irrelevant. By the definition of the strong top-set and the strong
bottom-set, any merit-difference _within_ the strong bottom-set is
negligible compared to the merit difference _between_ the strong top &
bottom sets.

Therefore, because polls aren't 100% reliable, it's always optimal to
approve (only) your strong top-set.

You keep saying that, voting only for the strong top-set, you might not
vote between the actual two frontrunners. And then I keep repeating that
it's optimal to approve (only) your strong top-set, because, by the
definition of the strong top-set and strong bottom-set, the merit
difference _within_ those 2 sets is negligible in comparison to the
merit-difference _between_ those 2 sets.

Michael Ossipoff



> Juho
>
>
>
>
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20161104/33b501d5/attachment-0001.htm>