[EM] non-monotonicity in SODA

Tue Jan 10 12:06:27 PST 2012

2012/1/10 David L Wetzell <wetzelld at gmail.com>

> JQ, you gotta also recognize that every extra wrinkle you add to a rule to
> deal with a contrived irregularity makes it disproportionately more
> difficult to market to the general public.

Yeah, I'm not actually going to add the partial-assignment fix to the
official rules for SODA. In the unlikely event that the nonmonotonicity
occurred (something which requires a careful balance of 1 minor and 3 major
candidates, and can't happen unless only one of the 3 majors expresses a
preference between the other two), it wouldn't be too obvious, and it would
only "harm" a small minority of voters. The issue only came up in the
context of a live, auto-delegation-assignment web poll; in that case, with
one vote added at a time, nonmonotonicity would be an obvious blemish.

(The other reason I mentioned the fix is to show how natural it is. In a
real sense, it is the nonmonotonicity which is not natural to SODA, only a
factor of the simplified rules.)

Jameson

>
> When you get around to the marketing of SODA, you're going to realize that
> a lot of FairVote's simplifications of IRV were necessary to streamline the
> product.
>
> dlw
>
> On Tue, Jan 10, 2012 at 12:37 PM, <
> election-methods-request at lists.electorama.com> wrote:
>
>> Send Election-Methods mailing list submissions to
>>        election-methods at lists.electorama.com
>>
>> To subscribe or unsubscribe via the World Wide Web, visit
>>
>> http://lists.electorama.com/listinfo.cgi/election-methods-electorama.com
>>
>> or, via email, send a message with subject or body 'help' to
>>        election-methods-request at lists.electorama.com
>>
>> You can reach the person managing the list at
>>        election-methods-owner at lists.electorama.com
>>
>> When replying, please edit your Subject line so it is more specific
>> than "Re: Contents of Election-Methods digest..."
>>
>> Today's Topics:
>>
>>   1. Re: A problem with IRV3/AV3 (Jameson Quinn) (David L Wetzell)
>>   2. CLDMM{O (fsimmons at pcc.edu)
>>   3. Re: A problem with IRV3/AV3 (Jameson Quinn) (Jameson Quinn)
>>   4. SODA: polls via "like/+1/reddit"; resulting nonmonotonicity;
>>      natural fix (Jameson Quinn)
>>   5. AERLO in conditional voting. Speculations. (MIKE OSSIPOFF)
>>
>>
>> ---------- Forwarded message ----------
>> From: David L Wetzell <wetzelld at gmail.com>
>> To: election-methods at lists.electorama.com
>> Cc:
>> Date: Mon, 9 Jan 2012 14:28:53 -0600
>> Subject: Re: [EM] A problem with IRV3/AV3 (Jameson Quinn)
>>
>>> ---------- Forwarded message ----------
>>> From: Jameson Quinn <jameson.quinn at gmail.com>
>>> To: EM <election-methods at lists.electorama.com>
>>> Cc:
>>> Date: Sun, 8 Jan 2012 19:03:29 -0600
>>> Subject: [EM] A problem with IRV3/AV3
>>> Imagine a scenario of an ABCD one-dimensional continuum:
>>>
>>> 41: A>B>C
>>> 19: B>A>C
>>> 20: B>C>D
>>> 20: C>B>D
>>>
>>> If the A voters vote A>>D then A will win. By raising the turkey D over
>>> the true CW B, they have stolen the win. Even if their strategy fails to
>>> keep B out of the top 3, they lose nothing; B will still win.
>>>
>>
>> thanks for doing this.  In the first stage wouldn't B and C tie for 3rd
>> place if only the first set of voters all voted strategically together in
>> the same way?     They'd both get rankings from 59 of the voters.  So if it
>> came down to a coin-toss, there'd be a 50-50 chance of the CW winning vs
>> the 2nd place candidate given a massive coordinated strategic vote by only
>> a subset of the sample (We assume none of the 3rd or 4th set of voters
>> decide to strategically leave off D rankings)?
>>
>>>
>>> To be honest, it was harder to tune this scenario than I thought it
>>> would be. Thus, having taken the time to write this down, I am no longer
>>> opposed to IRV3/AV3. (For IRV2/AV2, it's easier to get this problem. It's
>>> also easier to get the problem if there are clones involved, but real-world
>>> clones beyond 3 candidates are unlikely.)
>>>
>>
>> Thank you again.
>>
>> The MSM+relevant portion of the Blogosphere shd be helpful in identifying
>> clones in real world.
>>
>>>
>>> Since I'm now not opposed to IRV3/AV3, I consider it one of the 3
>>> reforms (along with SODA and IRV) that would be most acceptable to
>>> incumbents, because it avoids the weak Condorcet winner problem.
>>>
>>
>> remind me what is the weak Condorcet winner problem?
>>
>>
>>> Still, it is basically just as bad as IRV for nonmonotonicity and
>>> spoilers; all the spoiler scenarios I consider realistic are essentially
>>> 3-candidate anyway. As such, I see no reason to believe that it would not
>>> lead to lesser-evil voting and 2-party domination, as IRV does. Since I see
>>> 2-party domination (as opposed to just having 2 strongest parties, a
>>> logical necessity) as a source of the most-serious problems with Plurality,
>>> I still feel that SODA is a much better option than IRV3/AV3.
>>>
>>
>> dlw: And our difference is that I see the near exclusive use of Plurality
>> voting rules as the source of my country's current evils, since it's not
>> hard to imagine 2 party dominated system that is a lot better.  All it
>> takes is for there to be better checks and balances between them and for
>> there to be two quite different major parties plus scope for
>> outsiders/dissenters to express themselves via minor parties and LTPs.
>>
>> dlw
>> ps, I'm going to repost this on my blog.
>>
>>>
>>> Jameson
>>>
>>
>>
>>
>> ---------- Forwarded message ----------
>> From: fsimmons at pcc.edu
>> To: election-methods at lists.electorama.com
>> Cc:
>> Date: Mon, 09 Jan 2012 21:12:04 +0000 (GMT)
>> Subject: [EM] CLDMM{O
>> Mike,
>>
>> I wonder if it is possible for a CL to win three slot MMPO when the
>> number of ballots on which X appears
>> in the bottom slot is counted as an oppsitions to X.
>>
>> In other words, I wonder if the CL disqualification is redudant in that
>> context.
>>
>> Also, how does the CLD rule affect the FBC in general?
>>
>> Forest
>>
>>
>>
>> ---------- Forwarded message ----------
>> From: Jameson Quinn <jameson.quinn at gmail.com>
>> To: David L Wetzell <wetzelld at gmail.com>
>> Cc: election-methods at lists.electorama.com
>> Date: Mon, 9 Jan 2012 18:20:02 -0600
>> Subject: Re: [EM] A problem with IRV3/AV3 (Jameson Quinn)
>>
>>
>> 2012/1/9 David L Wetzell <wetzelld at gmail.com>
>>
>>> ---------- Forwarded message ----------
>>>> From: Jameson Quinn <jameson.quinn at gmail.com>
>>>> To: EM <election-methods at lists.electorama.com>
>>>> Cc:
>>>> Date: Sun, 8 Jan 2012 19:03:29 -0600
>>>> Subject: [EM] A problem with IRV3/AV3
>>>> Imagine a scenario of an ABCD one-dimensional continuum:
>>>>
>>>> 41: A>B>C
>>>> 19: B>A>C
>>>> 20: B>C>D
>>>> 20: C>B>D
>>>>
>>>> If the A voters vote A>>D then A will win. By raising the turkey D over
>>>> the true CW B, they have stolen the win. Even if their strategy fails to
>>>> keep B out of the top 3, they lose nothing; B will still win.
>>>>
>>>
>>> thanks for doing this.  In the first stage wouldn't B and C tie for 3rd
>>> place if only the first set of voters all voted strategically together in
>>> the same way?
>>>
>> Right, though of course it would be easy to fix that by changing some
>> fraction of the B>A>C voters to B>A.
>>
>>
>>>     They'd both get rankings from 59 of the voters.  So if it came down
>>> to a coin-toss, there'd be a 50-50 chance of the CW winning vs the 2nd
>>> place candidate given a massive coordinated strategic vote by only a subset
>>> of the sample (We assume none of the 3rd or 4th set of voters decide to
>>> strategically leave off D rankings)?
>>>
>>>>
>>>> To be honest, it was harder to tune this scenario than I thought it
>>>> would be. Thus, having taken the time to write this down, I am no longer
>>>> opposed to IRV3/AV3. (For IRV2/AV2, it's easier to get this problem. It's
>>>> also easier to get the problem if there are clones involved, but real-world
>>>> clones beyond 3 candidates are unlikely.)
>>>>
>>>
>>> Thank you again.
>>>
>>> The MSM+relevant portion of the Blogosphere shd be helpful in
>>> identifying clones in real world.
>>>
>>>>
>>>> Since I'm now not opposed to IRV3/AV3, I consider it one of the 3
>>>> reforms (along with SODA and IRV) that would be most acceptable to
>>>> incumbents, because it avoids the weak Condorcet winner problem.
>>>>
>>>
>>> remind me what is the weak Condorcet winner problem?
>>>
>>
>> A polarized electorate, 49% A, 49%B. 2% support an unknown centrist X.
>> Now, 25% each from A and B decide, "X couldn't possibly be worse than the
>> other side", so add a second choice, without really looking into whether X
>> really is better or worse. Most systems would then make X win, even if they
>> would be crushed by either in an actual runoff where the voters actually
>> took a serious look at them.
>>
>>
>>>
>>>
>>>> Still, it is basically just as bad as IRV for nonmonotonicity and
>>>> spoilers; all the spoiler scenarios I consider realistic are essentially
>>>> 3-candidate anyway. As such, I see no reason to believe that it would not
>>>> lead to lesser-evil voting and 2-party domination, as IRV does. Since I see
>>>> 2-party domination (as opposed to just having 2 strongest parties, a
>>>> logical necessity) as a source of the most-serious problems with Plurality,
>>>> I still feel that SODA is a much better option than IRV3/AV3.
>>>>
>>>
>>> dlw: And our difference is that I see the near exclusive use of
>>> Plurality voting rules as the source of my country's current evils, since
>>> it's not hard to imagine 2 party dominated system that is a lot better.
>>>  All it takes is for there to be better checks and balances between them
>>> and for there to be two quite different major parties plus scope for
>>> outsiders/dissenters to express themselves via minor parties and LTPs.
>>>
>>> dlw
>>> ps, I'm going to repost this on my blog.
>>>
>>>>
>>>> Jameson
>>>>
>>>
>>>
>>> ----
>>> Election-Methods mailing list - see http://electorama.com/em for list
>>> info
>>>
>>>
>>
>>
>> ---------- Forwarded message ----------
>> From: Jameson Quinn <jameson.quinn at gmail.com>
>> To: EM <election-methods at lists.electorama.com>
>> Cc:
>> Date: Tue, 10 Jan 2012 11:59:05 -0600
>> Subject: [EM] SODA: polls via "like/+1/reddit"; resulting
>> nonmonotonicity; natural fix
>> I'm designing a SODA poll that would use facebook "like", google+ "+1",
>> and/or reddit upvotes, along with automated delegated vote assignment, to
>> give live-updated results. In thinking about this, I've realized that SODA
>> can be nonmonotonic in the following (highly contrived) scenario:
>>
>> (delegated preferences in parentheses)
>>
>> 35: A(>C)
>> 30: B
>> 25: C
>> 10-n: X
>> n: Y(>B>A)
>>
>> With n=4, A wins. With n=6, Y's votes are enough to make B win, so A
>> approves C to prevent that from happening, and C wins; a worse result from
>> the perspective of the Y voters.
>>
>> The natural fix is to allow A to approve C with only some of their
>> delegated votes. Then, when n=6, A can approve C with 12 votes. Now Y's
>> votes cannot make B win, so Y approves A, and the nonmonotonicity is gone.
>>
>> Of course, in order for this to work like that in a live poll, I have to
>> make the logic for automatically updating assigned approvals much, much
>> more complex. In fact, off the top of my head, I can't even prove that the
>> general problem isn't NP-hard. But in real life, it's very unlikely that
>> the scenario would be even this complex, so I'm not too worried about that.
>>
>> Jameson
>>
>>
>> ---------- Forwarded message ----------
>> From: MIKE OSSIPOFF <nkklrp at hotmail.com>
>> To: <election-methods at electorama.com>
>> Cc:
>> Date: Tue, 10 Jan 2012 18:37:00 +0000
>> Subject: [EM] AERLO in conditional voting. Speculations.
>>  I want to emphasize the distinction between speculations and proposals.
>>
>> AOC, MTAOC, MCAOC, AOCBucklin, and their automatic
>> (non-optional)-conditionality versions
>> AC, MTAC, MCAC and ACBucklin are proposals, not speculations.
>>
>> I prefer the optional-conditionality versions to the
>> automatic-conditionality versions.
>>
>> Because the conditional voting methods are offered as optional, offering
>> those ways of using one's
>> Approval vote as voting options in the Approval election doesn't wrong
>> anyone who doesn't choose
>> to use them.
>>
>> I sometimes mention speculations too. I'll mention a few at the end of
>> this post.
>>
>> AERLO in conditional voting:
>>
>> 1. Obviously, if you want your middle rating for a candidate to be
>> conditional, then you certainly wouldn't
>> want to give hir unconditional AERLO status. So, plainly, if a ballot
>> marks an above-AERLO candidate
>> as "(conditional)", the AERLO status should only apply if the candidate
>> qualifies for receiving your
>> conditional middle rating.
>>
>> 2. I suggested that the default assumption for designation of
>> coalition-suitable candidates should
>> be "Above AERLO (or top-rated or top-ranked if the ballot doesn't use
>> AERLO)".
>>
>> But, for use in conditional voting, I suggest, for that default
>> assumption, two additional requirements
>> for a coalition-suitable candidate:
>>
>> a) must not be a conditionally-rated candidate
>>
>> b) must be ranked over the candidate being considered for actually
>> receiving the conditional vote listed
>> for hir on the ballot.
>>
>> I mentioned that, in MTA or MCA, when AERLO is used, a ballot's
>> middle-rated candidates could be
>> listed vertically, as a ranking, for AERLO purposes, even though they're
>> all middle-rated. That
>> qualifies as the ranking referred to in b).
>>
>> 3. Though I wouldn't unilaterally suggest it, AERLO could fairly be
>> automatic at the bottom of any
>> ranking that doesn't choose AERLO.
>>
>> After all, you like all of your ranked candidates better than any of your
>> unranked ones. So, in the event
>> that none of your ranked candidates wins, it could only be beneficial to
>> you to move them all up to
>> 1st place, for a 2nd count. That could only be beneficial.
>>
>> Still, it amounts to changing someone's ranking without their permission.
>> Based on that principle,
>> and wanting to offer AERLO as an _option_, my inclination is to not make
>> AERLO automatic
>> at the end of ballots that don't specify use of AERLO. If others
>> advocated that, then sure. But
>> I wouldn't unilaterally suggest any automatic, non-optional application
>> of AERLO, or any
>> other non-optional modifications of a voted ballot.
>>
>> A speculation, regarding #1, above:
>>
>> Maybe, for top rating, there could and should be a stronger mutuality
>> requirement, a top-mutuality
>> requirement, before moving the conditional, but above-AERLO, candidate to
>> top, but I haven't
>> looked at if, or how, that could work. It's a speculation. I emphasize
>> that I don't suggest that complication for the
>> poll. And, for public proposals, that would be a _later_ proposal. If
>> such a requirement were workable
>> and desirable, it could be implemented exactly as conditional middle
>> ratings are dealt with in MTAOC.
>>
>> Some FBC/ABE Condorcet-like rank method speculations:
>>
>> I mentioned that maybe Condorcet could be modified for FBC/ABE methods.
>> Before that, of course Kevin's
>> tied-at-top pairwise comparisons had been suggested for use in such
>> methods.
>>
>> Yesterday I suggested MMPO, but with an initial disqualification of
>> Condorcet losers. But that wouldn't avoid
>> Kevin's bad-example, because we could add a Candidate D, whom no one
>> ranks. C pairwise pair-beats hir, and
>> so isn't Condorcet loser.
>>
>> Maybe Condorcet (Tied-At-Top), MMPO2 is more promising. I don't know if
>> it would have the desired
>> properties. It's only a speculation at this point.
>>
>> Other such speculations:
>>
>> 1. MinMax Tied-at-Top pairwise defeat? A Tied-at-Top counterpart to MMPO.
>>
>> Of course I mean that its pairwise comparisons would be Kevin's
>> tied-at-top pairwise comparisons.
>>
>> 2. Greatest Tied-at-Top pairwise win?
>>
>> 3. Condorcet(Tied-at-Top), Top.    Among the candidates who don't have a
>> tied-at-top pairwise defeat, elect the one
>> who has the most top ratings.
>>
>> I haven't examined those speculations.
>>
>> Mike Ossipoff
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> Election-Methods mailing list
>> Election-Methods at lists.electorama.com
>> http://lists.electorama.com/listinfo.cgi/election-methods-electorama.com
>>
>>
>
> ----
> 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/20120110/cd702828/attachment-0004.htm>