[EL] (Not Really) ranked-choice voting in SF

[Joseph Lorenzo Hall]

On Tue, Nov 8, 2011 at 3:13 AM, David A. Holtzman <David at holtzmanlaw.com> wrote:
> The eventual availability of better equipment will relax the
> hardware limitation.

Ah, yes, thanks for the clarification!

> For ballot style samples and ideas, please have a look at this page I put
> together years ago.  Mechanized optical character (numeral) recognition is
> not necessary. The Cambridge ballot is exemplary, but I think the hardware
> to use it is no longer on the market, and it never got California or U.S.
> certification anyway.  The Portland, Maine, ballot in today’s election may
> not be federally certified but is of the “matrix” form you’re looking for
> --see the last page of the official sample ballot at
> http://www.portlandmaine.gov/voter/11082011ballot.pdf .  It looks a lot like
> the Cambridge ballot.

I cannot stress enough how important it is to have resources like the
page above with sample ballots made available.  Thank you.  I know
people conducting usability testing of IRV/RCV and this will be a very
important resource for them.

> Each full certification costs hundreds of thousands of dollars, I’m told.
> And new certification may be required for revisions/improvements, including
> software upgrades.  That’s what I meant by regulatory barriers to market
> entry.

Hundreds of thousands of dollars is, unfortunately (or fortunately,
depending on your perspective) at the lower end of the spectrum, and
often that's just for federal cert. It does seem to be getting better
and cheaper... and maybe even a bit faster. ::)


> The counting is easy, not at all a “total nightmare,” because once you’ve
> looked at each ballot and put it in a counted stack for its first choice,
> the only ballots you have to look at again are those in the smallest piles
> -- the ones whose top choice just got eliminated.  And then all you have to
> look for on each ballot is the top choice who’s still in the election.  If
> you’d like someone to conduct a mock RCV/IRV election & count for you or
> that advisory board you’re on, just let me know.]

Ah, that does not seem like a total nightmare.

The total nightmare comes in doing risk-limiting audits of IRV/RCV
races, but that's getting some good attention from math/stats/CS types
(I can point to three technical papers that each make it a bit
better).  The problem with being able to say, "we can say with x%
confidence that if we hand counted all the IRV ballots the result
would not change" from a sample in IRV/RCV is that it is very
difficult to bound the margin between winners and losers (very
difficult = it becomes computationally impossible with arbitrary
amounts of rankings possible).  Anyway, there just may be fundamental
limits to how we can audit IRV/RCV, but we'll see.  The CA SOS and the
lead statistician (Berkeley's Philip Stark) in the AB 2023 pilots are
observing San Francisco this week, so let's wish everyone well!

best, Joe

-- 
Joseph Lorenzo Hall
Postdoctoral Research Fellow
Media, Culture and Communication
New York University
https://josephhall.org/