#BernieOrBust, #NeverTrump, and MJ vs. Score

[Jameson Quinn]

I think that the current presidential race makes a good argument that MJ is better than Score. In both major parties, the nominee won the primary more-or-less decisively, but there are a significant number of people who are deeply dissatisfied with their party's nominee. 

Let's make some simplifying assumptions, for the sake of argument:

-Say that the current nominees would have won their respective primaries even under a better voting system. That is, despite their flaws, that each is the Condorcet and Score winner among only the voters of their own party.

-Say that none of the primary losers have a large amount of cross-party appeal. Obviously they'd have some, and I know that their supporters argue that it would be significant, but just say it's not, for the sake of argument.

-For the sake of brevity, let's say there was just one Republican alternative to Trump. I'll use Kasich, but you could substitute any or all of the Trump alternatives and the argument would be the same.

Under those assumptions, one of the current nominees is going to be the eventual winner. In that case, a rational approval strategy for somebody whose preferences are Bernie>>>Hillary>Kasich>Trump, or Kasich>>>Trump>Bernie>Hillary, is to approve all candidates from their preferred party, and no candidates from the other party.

But for somebody like this who's attracted to the #BernieOrBust or #NeverTrump movement, that's highly distasteful. They don't want to vote for the lesser evil; from their perspective, that person is still evil. Under plurality, they're willing to risk a spoiled election in order to make this point.

How would they vote under approval? Perhaps some of them will be mollified by the chance to vote for their true preference, and so with that sugar-coating, they'll be willing to swallow approving the "lesser evil". But it will still leave a bitter taste for them to cast a ballot that doesn't make the distinction they consider by far the largest, between the good candidate and the lesser evil. Some of them may refuse to do so; so there's still a risk of a spoiled election.

Under score? Now they're able to cast their honest ballot; for instance, our #BernieOrBust voter might vote Bernie 100, Hillary 4, Kasich 1, Trump 0. But that's almost as bad, from a strategic point of view, as not approving Hillary in approval. If we want to avoid a spoiled election, this person should give a nearly-strategically-optimal 99 to Hillary. But the temptation to slide her down towards the honest score is strong, and so the risk of spoiled elections is still reasonably high.

Under MJ? Switching to the #NeverTrump voter, the honest ballot would be Kasich A, Trump D, Bernie F, Hillary F. And given the assumptions we've made, that is still strategically optimal; if Republicans are a majority of voters, a Republican will win.

So MJ, unlike Approval or Score, would allow these voters to cast a ballot that is both honestly expressive and strategic.

---

Separate argument. Say that you have a jar of jelly beans, and people are guessing the count. The 5 guesses so far are: 200, 700, 825, 860, 8000. If you were going to make a guess based on those, without seeing the jar, you'd probably say 826; that is, you'd use the median. The mean is clearly a bad idea, because the massive outlier pulls it off.

My point is that if you're looking for the wisdom of a crowd, the median is better than the mean. This is true whether the outliers are due simply to bad guessing, or whether they're due to strategic voting.

---

But what about VSE (that is, BR)? Doesn't score voting beat MJ in BR? Yes, it does, in simulations with a fixed strategy percentage. But what if, instead of strategic and unstrategic voters, you have high-information and low-information voters. The low-information voters have true utilities for the candidates, but do not know what those true utilities are; they only know the utility measured imprecisely, with a symmetrically-distributed error that includes outliers (for instance, a student t distribution). In that case, the "honest" score ballots will have outliers that should be ignored, just as with the jellybeans, so in this case MJ will actually have better VSE with respect to the true, unknown utilities. This is in a model without any strategy (except, optionally, normalization).

---

Basically, I'm arguing that if people are unhappy with approval, MJ (or other graded Bucklin system) is a logical next step. I understand that some of you will still favor some other single-winner system (score or otherwise) as better than MJ. And that's OK; I think we can all still agree that approval is the first step. But I just wanted to make these arguments and see if they are convincing to anybody. 

[Warren D Smith]

On 8/11/16, Jameson Quinn <jameso...@gmail.com> wrote:
> I think that the current presidential race makes a good argument that MJ is
> better than Score. In both major parties, the nominee won the primary
> more-or-less decisively, but there are a significant number of people who
> are deeply dissatisfied with their party's nominee.
>
> Let's make some simplifying assumptions, for the sake of argument:
> -Say that the current nominees would have won their respective primaries
> even under a better voting system. That is, despite their flaws, that each
> is the Condorcet and Score winner among only the voters of their own party.

--I have poll-based evidence that assumption was false.

> -Say that none of the primary losers have a large amount of cross-party
> appeal. Obviously they'd have some, and I know that their supporters argue
> that it would be significant, but just say it's not, for the sake of
> argument.

--I also have poll-based evidence that assumption was false.

> -For the sake of brevity, let's say there was just one Republican
> alternative to Trump. I'll use Kasich, but you could substitute any or all
> of the Trump alternatives and the argument would be the same.

--Given that there were about 20 republicans running, this
assumption also seems, to say the least, suspicious.

So given that all three of your starting point assumptions seem
wrong or dubious before you've even given a single word of
any actual argument, why should I read further?

I must be a masochist.

> Under those assumptions, one of the current nominees is going to be the
> eventual winner. In that case, a rational approval strategy for somebody
> whose preferences are Bernie>>>Hillary>Kasich>Trump, or
> Kasich>>>Trump>Bernie>Hillary, is to approve all candidates from their
> preferred party, and no candidates from the other party.
>
> But for somebody like this who's attracted to the #BernieOrBust or
> #NeverTrump movement, that's highly distasteful. They don't want to vote
> for the lesser evil; from their perspective, that person is still evil.
> Under plurality, they're willing to risk a spoiled election in order to
> make this point.

--Have any actual evidence that the class of voters you just considered,
actually are a substantial chunk of the electorate? (The HC campaign
sure does not seem to be worried about this so-called chunk. They
instead seem to have the attitude they can take dissatisfied-dem
voters for granted.) There actually is some evidence supporting
your view, sort of, namely plurality polls suggesting Gary Johnson = 15% and
Jill Stein = 5%. If those results hold then you could argue, I suppose,
that the Johnson voters "were not worried about a spoiler scenario because
of the overriding desire to make their point" but I suspect a more realistic
and true interpretation would be that the Stein voters fall in that class,
but the Johnson voters do not because they are not voting for GJ to
"make a point"
but actually because they genuinely honestly consider him superior to Trump and
genuinely cannot accept President Trump. I happen to know at least one
sitting Republican senator has said exactly that. This would explain the large
difference between "15" and "5". So far, I have not seen any high Dem officials
saying HC is unacceptable so they will vote Stein.
Your hypothesis fails to explain all that.

> How would they vote under approval? Perhaps some of them will be mollified
> by the chance to vote for their true preference, and so with that
> sugar-coating, they'll be willing to swallow approving the "lesser evil".
> But it will still leave a bitter taste for them to cast a ballot that
> doesn't make the distinction they consider by far the largest, between the
> good candidate and the lesser evil. Some of them may refuse to do so; so
> there's still a risk of a spoiled election.

> Under score? Now they're able to cast their honest ballot; for instance,
> our #BernieOrBust voter might vote Bernie 100, Hillary 4, Kasich 1, Trump
> 0. But that's almost as bad, from a strategic point of view, as not
> approving Hillary in approval. If we want to avoid a spoiled election, this
> person should give a nearly-strategically-optimal 99 to Hillary. But the
> temptation to slide her down towards the honest score is strong, and so the
> risk of spoiled elections is still reasonably high.

> Under MJ? Switching to the #NeverTrump voter, the honest ballot would be
> Kasich A, Trump D, Bernie F, Hillary F. And given the assumptions we've
> made, that is still strategically optimal; if Republicans are a majority of
> voters, a Republican will win.

> So MJ, unlike Approval or Score, would allow these voters to cast a ballot
> that is both honestly expressive and strategic.


--

Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[Jameson Quinn]

These assumptions were not supposed to be true. They were simplifying assumptions for the purpose of looking at one angle of the current situation. We can argue about whether they're true enough to be useful on either side of the equation, but whatever you decide about that, I think that they constitute at the very least a plausible hypothetical future situation, and that argument that follows is worth considering on that basis. 

So, let's state it as a hypothetical, or return to the old standby, Nader/Gore/Bush. I think it's clear that honest Nader>>>Gore>Bush voters existed in significant numbers; probably at least multiple times the number who actually voted Nader. 

"Honest"/strategic ballots for these voters:

FPTP: Nader   /   Gore

Approval: Nader    /    Nader, Gore

Score: Nader 100, Gore 5, Bush 0     /   Nader 100, Gore 99, Bush 0*

MJ: Nader A, Gore D, Bush F     /   Nader A, Gore B, Bush F*

Strategic considerations for the voter:

FPTP: "People tell me that a vote for Nader is a vote for Bush. But man, voting for Gore over Nader feels so wrong."

Approval: "I don't want Bush to win, so I should probably approve Gore. But it feels a little strange to say that Gore and Nader are the same on my ballot. I guess that I have to choose one side or the other, and at least I know my options."

Score: "I don't want Bush to win, so I should probably give Gore 100. But that feels like a way bigger lie than it did under approval. Even giving him 99, which is actually more expressive than my approval ballot, feels like a bigger lie, because I could be entirely honest and give him 5. It feels like this voting system is really inviting me to vote Gore as low as I think I can afford to; in other words, to flirt with disaster."

MJ: "Well, I suppose it's possible that a bunch of centrist Gore supporters grade Bush a C, so that I can't get Gore to win without grading him at least B. But frankly, I trust Democrats to hate Republicans, and I trust Gore voters to have enough intuitive understanding of later-no-harm failure to grade Bush at F. So as long as I give Gore at least a D, Gore wins. Unless of course enough major party voters grade Nader at D or C, and then I win. Basically, even though my honest vote isn't perfectly optimal strategically in all cases, it's almost certainly good enough to be optimal in reality. Which feels pretty good!"

I think that similar arguments hold for #BernieOrBust or #NeverTrump voters this year, but even if you disagree, I think the Nader case shows that they're realistic.  

*Yes I know, that's not fully optimal strategy, but it's close enough.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Warren D Smith]

I just saw another poll which claimed Johnson 5%, Stein 2%
(as opposed to a different poll claiming 15 & 5)
so these numbers evidently are quite unreliable.

[Warren D Smith]

> These assumptions were not supposed to be true. They were simplifying
> assumptions for the purpose of looking at one angle of the current
> situation.

--well, look. If you want to argue "there exist abstract scenarios in
which MJ is
superior to plain-average-based score voting," then I already agree.
But you'd begun with

"I think that the current presidential race makes a good argument that
MJ is better than Score. In both major parties, the nominee won the
primary more-or-less decisively, but there are a significant number of

people who are deeply dissatisfied with their party's nominee..."

suggesting you were not talking about abstract scenarios, but rather
about the real world current race situation. And then you began
speaking about that present
situation by starting with 3 assumptions. All of which I attack.

Now you are saying, "who cares about the real world, and who cares
about my 3 assumptions, and who cares if my assumptions have anything
to do with the real world
-- none of that matters for my purposes."

Which is ok, but then your headline sentence was wrong, and your
several starting
paragraphs all irrelevant. (I recommend beginning your essays with
sentences that actually are about what you are going to talk about,
and starting paragraphs that are relevant. That makes the reader not
have a massive urge to strangle you, or,
more likely, just quit reading.)

[Phil Uhrich]

The polling of 3rd parties is only one factor.  There are surely plenty of Sanders supporters who just won't bother to show up to vote if not for him which is not something the polls can capture well.  If everyone with an interest in the democratic primary, voted for it Sanders would have won anyways. Poll covering the whole strech: http://polling.reuters.com/#poll/TR131/dates/20160202-20160614/type/overall  

It isn't till you filter likely voters that Clinton wins.  There is also more then a negligible amount of people who don't fit easily into D and R boxes and would vote Sanders -> Trump or vise versa.  MJ tries to reduce 3 dimensions of data into 2. I see no reason to try and infer what a Bernie or Bust person wants.  Maybe they only think Clinton is 5% better then Trump and if its between those two they just don't care as much.  Meddling with their vote based on you ability to infer their intention is really a way to get your desired candidate out of their vote.  If adults want to "flirt with disaster" who are you to try and stop them?

Phil