IRV in SF

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I attended a meeting at the San Francisco Department of Elections
today. It was a brown bag seminar for mayoral candidates and their staff;
I attended for Michael Denny's campaign. The main purpose of the meeting
was to introduce ranked choice voting to the candidates and their
campaigns; we had a very frank presentation from Eugene Tom, who I
understand is the ballot designer.

First, they *are* going ahead with RCV. Barring any legal challenges, this
will be the method used for the November 2003 elections for mayor, district
attorney, and sheriff. They have a worst-case plan prepared to do a manual
count and calculation.

Second, the DoE is calling this "ranked choice voting" to dispel any notion
that there will be instant results. Because they need to know not only who
got the most first-place votes, but more importantly, who got the *fewest*,
they have to have *all* ballots in before they can proceed. Unless the
winner has a clear and overwhelming majority, that means they have to tally
all absentee and provisional ballots, which will take until Saturday or
Sunday after the election.

If they receive full vendor support from the ballot machine company, they
can then produce results in a day or two after that; it's just
number-crunching. However, without that support, they will need to tally
each ballot manually under the rules set by the Secretary of State, which
involve the stuff you saw in Florida: an official, an observer, and two
recorders go through a pile one at a time, holding up the ballot,
declaiming its number, and reading off the votes. Under this scenario, the
actual number-crunching will still only take a day, but it may take five
weeks to declare a winner. Note that they expect about 240,000 votes for
over twenty candidates for mayor. (For comparison, the two eastern cities
that use choice voting have about 19,000 ballots to deal with.)

The law as passed in March 2002 requests that voters be able to rank all
choices, but mandates a minimum of three choices in the face of logistical
difficulty. The current plan is to go with three choices. The vendor's
Eagle Optech machines can really only handle one-of-many votes; they are
certainly not equipped for tallying standardized-test-type bubble
fill-ins. So the plan is to have three lists of candidates for each race:
Mayor, first choice, pick one: Brown, Smith, Jones; Mayor, second choice,
pick one: Brown, Smith, Jones; Mayor, third choice, pick one: Brown, Smith,
Jones.

The reporting problem is that the software is currently designed to tally
the votes in each race separately; i.e., it could report that Brown got
100,000 first-place votes, 70,000 second-place votes, and 20,000
third-place votes, but it can't say that there were 30,000 ballots with
Brown-Smith-Jones, 22,000 with Smith-Brown-Jones, etc. The actual
correlation between the votes on each ballot is critical for tracking which
second-choice votes get used.

So the machine vendor is working on this reporting capability, but there's
a Catch-22 situation (as reported in Caleb Kleppner's earlier
e-mail). Eugene Tom was very frank about this. There's not a new contract
to sign; they're already locked into ES&S, per "our wise predecessors"
(Tom's words). However, ES&S says that this isn't in their contract, and
they want more money to develop it. The city doesn't want to pay unless
they can guarantee certification; ES&S doesn't want to work on it unless
they're going to get paid; and the Secretary of State can't guarantee
certification of something they haven't seen.

So the good news is that RCV *is going to happen*. The bad news is (a) it
might be a cluster****, and (b) it's not complete IRV because the voters
only get three choices.

The Department of Elections is looking to the candidates, campaigns, and
parties to publicize this and to educate the voters on the new
system. Please help them out as best you can when talking with friends.

And finally, I have had nothing but positive dealings with the folks at
DoE. They are a very patient bunch, very good at explaining procedures to
candidates and potential candidates, and they seem to be doing their best
in the face of a bad situation with regard to the logistics here. Anyone
who isn't directly involved in a campaign should offer to help out with the
manual count, should it become necessary; they're going to need a lot of
staff to get it done.

~Chris
- --
"We are truly sleepwalking through history. ... This war is not
  necessary at this time." - Sen. Byrd on Iraq, 12 February 2003
Political gadfly and freelance nerd: <URL: http://crism.maden.org/ >
PGP Fingerprint: BBA6 4085 DED0 E176 D6D4 5DFC AC52 F825 AFEC 58DA

I mentioned to my co-workers that it looked like IRV was a "go" for the
mayoral election, and one of them sent me the following link:

http://electionmethods.org/evaluation.htm

Anyone care to comment? I wish I'd seen this before I voted on it.

Rob

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I mentioned to my co-workers that it looked like IRV was a "go" for the
mayoral election, and one of them sent me the following link:

http://electionmethods.org/evaluation.htm

Anyone care to comment? I wish I'd seen this before I voted on it.

Well...

They're right that IRV isn't perfect. To use their example: let's say A,
B, and C are running for office. There are twenty voters, and their ranked
votes are:

8: A,B
7: C,B
5: B

Under IRV, B is eliminated; with no second-choice votes for the group of 5,
A wins outright. [They perpetually ignore the diversity of votes in their
analyses, though at the national level there may be something to that, as
one likely wouldn't see many Nader-Bush-Gore ballots.]

They point out that B is really the ideal winner in this vote; twelve
voters prefer B to A, and thirteen prefer B to C. Their ideal circumstance
seems to be a sort of round-robin binary tournament; here, B beats A 12-8,
A beats C 8-7, and B beats C 13-7. How this plays out with a larger pool
(like, say, 24 candidates for SF mayor), I'm not sure. They discuss the
Condorcet method, which may be exceedingly accurate, but is also
exceedingly complicated.

But leaving the ivory tower for the real world, what was the case before we
implemented IRV? Let's assume the votes above were in people's heads; the
voting would go like this:

8: A
7: C
5: B

Now there's a run-off election between A and C. The five who voted for B
get a chance, at least, to vote for A or C; apparently, they are
sufficiently disdainful to abstain, and the same 8 and 7 votes are cast
again, resulting in victory for A. In this scenario, IRV changes nothing,
and is certainly no worse. However, assuming some of those five actually
wanted to consider a second choice, or even third, than IRV is an
improvement if only in the saved expense of running a second full election
and in the higher voter turnout for the (virtual) runoff.

However, consider the difference in the communication from the
electorate. The *result* is no worse and no better under IRV in this
scenario; however, the published results will indicate that 15 people
listed B as a second choice and 5 as a first choice. This gives B a sort
of mandate, and certainly lends B's platform some legitimacy, in a way that
the old system did not.

Now the reality is that the voting spread is going to be more varied than
this Web site's examples indicate; for example, you'd be more likely to see
something like:

5: A,B,C
4: A,C,B
4: B,C,A
3: B,C
2: C,B,A
2: C,A,B

In this case, the old system would set A and B in a run-off, which
presumably A would win. Under IRV, A still wins, but B's and C's support
is noted. Under Condorcet, uh... (crunches some numbers)... I believe A
wins, but it's a little harder to tell. Condorcet does tell us that people
prefer B over C 12-8, but we could also determine that from the IRV
published results.

So in summary: IRV is an improvement over the two-election-cycle system
previously in place. It isn't perfect. There may be better systems. I
would still have voted for Prop A, but if this is successfully adopted,
implemented, and embraced, maybe we can push for something even better.

~Chris
- --
"We are truly sleepwalking through history. ... This war is not
  necessary at this time." - Sen. Byrd on Iraq, 12 February 2003
Political gadfly and freelance nerd: <URL: http://crism.maden.org/ >
PGP Fingerprint: BBA6 4085 DED0 E176 D6D4 5DFC AC52 F825 AFEC 58DA