[EM] California (Re: Two round methods)

Sun Oct 10 20:42:36 PDT 2021

Hi Kevin

Thanks for the thoughtful reply.  I'm going to take a moment and give
my hot-take now, and then give your method some more thought later.
More inline below:

On Sun, Oct 10, 2021 at 12:16 PM Kevin Venzke <stepjak at yahoo.fr> wrote:
> I mostly like the jungle primary, or top-two runoff, but it's true that the finalists might
> get chosen arbitrarily, and we're only assured that a Condorcet Loser won't win.

Interesting point about the Condorcet Loser criterion.  That's
mathematically true, but I don't know if that actually works in two
successive elections held several months apart.  A candidate can game
the system by avoiding being the Condorcet Loser in the primary, and
then appealing to their base voters in the general election (in hopes
that a weak-sounding candidate is the only other top-two winner)

Kevin, here's what you wrote after my chirpy "Welcome to California!"
comment about how we don't separate Democrats and Republicans in our
primary elections:
> I knew this, but understood you as being surprised that a two-round system could ever sensibly
> end after one round.

Nope.  I was referring to California's system, which only has a
single-round primary, followed by a single-round general election.
Qualifying for the ballot in one of our 53 electoral districts is easy
in California, because the number of signatures to qualify is only a
fraction of the people who voted in the prior election in the given
electoral district.  Qualifying for a statewide initiative (e.g. U.S.
Senator from California) is more difficult (and expensive), which is
why we had two Democrats to choose from in the 2018 race for the U.S.
Senate:
<https://en.wikipedia.org/wiki/2018_United_States_Senate_election_in_California>

The Democratic Party here in California is really well-funded, and
both candidates had loyal followings.  One of them became mayor of San
Francisco the day that Harvey Milk and George Moscone were shot in
their offices at City Hall, and many people learned about their deaths
from her.  This was only a few days after my current representative
(Jackie Speier) was shot and left for dead on a airplane runway down
in South America.  A lot of people from San Francisco died (or were
gravely wounded) in November 1978.  We have a lot of history here.

Back to more recent history, the Republican Party had a tough year
here in California in 2018:
* <https://en.wikipedia.org/wiki/2018_California_elections>
* <https://en.wikipedia.org/wiki/2018_United_States_House_of_Representatives_elections_in_California>

The Democratic Party had a tougher time of things in 2020:
* <https://en.wikipedia.org/wiki/2020_California_elections>
* <https://en.wikipedia.org/wiki/2020_United_States_House_of_Representatives_elections_in_California>

I suspect it was because the Republicans here in California figured
out how to play Democrats off of each other.  It wasn't difficult.  As
Will Rogers famously put it: "I am not a member of any organized
political party — I am a Democrat.".

Source:
https://www.goodreads.com/quotes/14365-i-am-not-a-member-of-any-organized-political-party

> [Then Rob listed the results from the 2018 primary]:
> > What if TWO people won a majority approval in the primary election?
> > Let's keep using the CA-10 example from above, and take all of the
> > candidates that received more than 10,000 votes in the primary:
> >
> > * Jeff Denham (incumbent, Republican) 45,719 votes (37.6%)
> > * Josh Harder (Democratic) 20,742 votes (17.0%)
> > * Ted D. Howze (Republican)  17,723 votes (14.6%)
> > * Michael Eggman (Democratic) 12,446 votes (10.2%)
> > * Virginia Madueño (Democratic) 11,178 votes (9.2%)
> > * Others (11.4%)
> > * Total: 121,757 votes (100.0%)
>
> But the voters had their favorites. Would they really approve the entire list, if they
> knew there would be a second round that doesn't eliminate the alternatives? I don't think
> so. So for my method proposal I think it is fine to end the method if there is majority
> approval. (For a top-two, some remarks are below.)

If I had been voting in that district, then YES, I would have voted
for Harder and Madueño and even Eggman.  I might have even voted for
TJ Cox, because I like the guy and he was on the ballot, as I recall.
I'm trying to avoid profanity when I describe what I think of Jeff
Denham.  Many of the people I canvassed with would not.  Jeff Denham
was (and probably still is) an asshole, and voted like one:
https://projects.fivethirtyeight.com/congress-trump-score/jeff-denham/

You have to remember, this was a midterm election; Donald Trump was
president, and there was no way to vote against him, other than in the
U.S. House of Representatives and in a third of U.S. Senate races.  In
California (as I pointed out), the U.S. senate race was between two
Democrats, so we couldn't vote "against" Trump with our Senate vote.

In each of the hundreds of House districts all over the United States,
most Democrats (myself very much included)  would have voted for a
department-store mannequin wearing a "TRUMP SUCKS" t-shirt in the 2018
election, if their opponent was a Republican.  In CA-10 in particular
(where Republican Representative Jeff Denham was running against
Democratic challenger Josh Harder), the Democrats were desperate.
Jeff Denham was a three-term incumbent.  I really, sincerely do not
like Jeff Denham, and it seemed many people in CA-10 didn't either.
Apparently, according to Wikipedia, he's a lobbyist for the law firm
that Bill Gates' dad practiced at:
https://en.wikipedia.org/wiki/Jeff_Denham

> My thoughts on the design constraints here, for a second round with two candidates only.
>
> 1. I would want to retain the non-partisan nature of the first round. I don't see a way of
> preserving this while having any concept of party included in the method definition.

I don't think that's possible in a real-world election.  Technically,
the first round of California elections are "non-partisan", but the
candidates are allowed to state what their "preferred" party is on the
ballot.  That was a workaround to prevent a well-funded party from
running a lot of candidates with confusing names.  For example, did
you know that John Kennedy was a Republican?
https://en.wikipedia.org/wiki/John_Kennedy_(Louisiana_politician)

Things like ^that happen a lot in the United States.

> 2. If we advance exactly two candidates to a second round, we want to use a method that is
> good at estimating the two most likely "best" candidates. Your concerns show that
> vote-for-one is inadequate. And I dare say your hypothetical here actually shows why
> approval is inadequate.

After seeing the results in St Louis I've come around to the idea that
approval voting may actually be okay for primary elections:
https://en.wikipedia.org/wiki/2021_St._Louis_mayoral_election

I know very little about local politics in St Louis, and know almost
nothing about whether the national gender and race politics during
2020 and early 2021 played a role in that election.  I have a hard
time believing that they didn't.  That election took place less than a
year after George Floyd's murder up in Minneapolis (less than 500
miles away in a city that is also on the Mississippi River and is
culturally very similar to St Louis) and in the living memory of the
shooting of Michael Brown in a convenience store between St Louis and
its airport (which is named after an idiot white guy who flew his
airplane across the Atlantic Ocean).  Michael Brown's death also led
to protests:
https://en.wikipedia.org/wiki/Ferguson_unrest

...but unfortunately, Michael Brown's death didn't lead to much
justice for black people here in the United States.  I'm hoping that
George Floyd's death was enough of a wake-up call for this country.

Anyway, after seeing how the election went in St Louis, I'm going to
say that I "tentatively approve" of approval voting as a top two
mechanism.  Perhaps the St Louis Airport ("STL") should be named after
Michael Brown, even if Brown's most memorable moment (for the rest of
us) was just as a dumb kid who possibly stole some Swisher Sweet
cigars from a suburban convenience store.  18-year-olds do some stupid
things, and perhaps they shouldn't get airports named after them.  But
I digress....

> If a major party (say 40%+ of the voters) has the stance "let's elect any candidate from
> our party, to ensure the winner is not from the other party", they may be able to achieve
> this even without majority approval. Two candidates with 45% approval could both advance.

Well, this could be true, but lets say that we have two dominant
political parties in the United States (because.... well.... we do).
An approval-based primary election campaign is going to be all about
name recognition.  That's part of the reason why I'm saying that all
candidates who get 50% (or more) should advance, even if that means
three (or more) candidates.  My hunch, based on the approval-based
results in St Louis (for the election on 2021-03-02), is that getting
over 50% approval is incredibly difficult:

* Tishaura Jones -- 25,388 votes -- 57.0%
* Cara Spencer -- 20,659 votes -- 46.4%
* Lewis E. Reed -- 17,186 votes -- 38.6%
* Andrew Jones -- 6,428 votes -- 14.4%

Total ballots 44,571

> Maybe the best (eliminative) option is the one mentioned in 2018, to pair the approval
> winner with the candidate posing the greatest "approval opposition" to him, meaning that
> the same 45% of the voters can't pick both finalists.

Perhaps.  Back in 2018 when I was really thinking about the elections
here in California, that's when I came up with my complicated systems
("MATT" and "MAF")
https://electowiki.org/wiki/MAF
https://electowiki.org/wiki/MATT

Back in 2018, I preferred MAF, even though it was more complicated.
However, if I had to choose today, I'm not sure which one I would
choose.

> In some strange scenarios this could create a bad second round: Suppose that 80% of the
> voters approve A and B, and the other 20% approve C. Then C is in the second round. That
> would be wasteful.

Yeah, that scenario is pretty bad for both MATT and MAF.  MAF has
better safeguards, and would advance A and B (and drop C), but MAF
seems to be too complicated to evangelize.  I've tried to explain MAF
to many people for the past few years (tweaking the rules as scenarios
were described to me), and I eventually gave up on it.

> The right question to ask is a bit elusive. The right question is not "which candidate is
> most likely to beat A pairwise?" because that will not actually give a different answer
> as far as we can tell from the first round approval ballots. It's going to be the second
> most approved candidate.

Not necessarily.  Elections involve humans, and humans are
complicated.  There truly is no telling which candidate is the most
likely to "beat A pairwise"

> But should top-two selection always imply a concept of proportional representation; i.e.
> that the two finalists "represent" as many voters as possible? I am not sure why,
> theoretically. It would help maximize the policy difference between the two finalists,
> which might be good for participation rate and the appearance that the vote was worth
> holding. But maximizing the policy differences between two finalists is certainly NOT
> generally music to my ears: It makes it sound like center squeeze is actually a goal.
>
> The approach of my non-eliminative suggestion, to say it again, is to pick just one best,
> and then go to the voters again and check if there's any way to show it was the wrong
> pick.

I'll have to write some code for your method at some point for my own
analysis.  My hunch is that we're going to have to see how things go
in Fargo and in St Louis after a few elections, and hope that we get
more real-world elections using approval voting in order to do more
trustworthy analysis.

Rob