4. Dopp: [IRV] Confuses voters
All the evidence shows that voters are not
confused by IRV. The rate of spoiled ballots did
not increase in any of the U.S. cities when they
switched to IRV. For example, Burlington (VT)
used IRV for the first time in a hotly contested
race for mayor in 2006, and among those casting
votes in the IRV race fully 99.9% of ballots
were valid, with the very highest valid ballot
rate in the ward in town with the highest number
of low-income voters. San Franciscos rate of
valid ballots in the most closely contested race
in its first citywide election with IRV was
99.6%. Furthermore, exit polls have been
conducted in every city having an IRV election
for the first time in the modern era. Each
survey shows that voters overwhelmingly prefer
IRV to their old method of elections.
Classic political argument, a combination of
anecdotal contradiction and actual deception.
"All the evidence." Well, then there really
shouldn't be any problem looking at that
evidence, should there? Burlinton is an anecdote,
and a confined one, implying that the most
confused voters would, of course, be in the "ward
in town with the highest number of low-income
voters." I'm not going to examine the Burlington
data today, simply for lack of time, but,
instead, will look at the allegedly non-existence
evidence in the other direction.
It's hard to get spoilage data from prior San
Francisco elections. It's not shown in the
spreadsheet, though one can derive the difference
between the "turnout" and the sum of votes for the candidates.
In 2000 Supervisorial elections, total informal ballots were
District 1: 3983/24211 (5 listed candidates). runoff: 21/14394
District 2: 11136/38206 (2 listed candidates).
won by Gavin Newsome by 26433 to 637 write-in.
let's take a look at a difficult district, District 10
District 9: 4120/23884 (12 listed candidates). runoff: 365/11014
Now, what about 2004, the first year for RCV:
District 1: 2636/31333 (1st choice, 7 listed
candidates). A total of 5393 ballots did not
contain a valid vote for the final runoff round.
District 2: 6623/40931 (1st choice, 5 listed
candidates). won by Alito-Pier with 21103/34308.
District 9: 2262/26995 (1st choice, 6 listed
candidates). won by Ammiano with 12547/24733.
What to make of these statistics. First of all,
there is no real spoilage data here, for mixed in
with true spoiled ballots are undervotes.
Apparently that information is available, because
there is an analysis of it at
http://rangevoting.org/SPRates.html. What is
shown there is an analysis of overvotes.
Overvotes spoil ballots. The analyst did not
count overvotes in lower ranks, which would
affect vote transfers. He found 0.082% overvotes
in the plurality races on the ballot (basis
223,837 votes), compared to 0.60% overvotes on
the IRV races (basis 1,294,721), which is statistically significant.
Undervotes are not necessarily errors. Overvotes
generally are. To determine the fact here would
require a far deeper study that I can do in a few hours.
There is a study at
http://www.gregdennis.com/voting/sf_irv.pdf that
goes into much more detail. I don't think one
could read this study and come away with the
conclusion, given by FairVote, that "All the
evidence shows that voters are not confused by
IRV." According to the paper, 14% of Latinos and
27% of Asian voters, in exit polls conducted by
the Chines American Voter Education Committee,
found IRV difficult to use. Apparently, poll
workers made errors in instructing voters.
The author of the paper points out that the
ballot images do not reflect the actual patterns
of votes on the ballots. In particular, marking a
candidate in more than one column was expected to
occur commonly. However, if one looks at the
ballot images, apparently, no such votes are
recorded. None. Why? The machines were programmed
that way, they record such additional marks as
undervotes. I'd think someone concerned about
election security would be less than thrilled
that the machines are "interpreting" the votes in creating the ballot images.
The other error that one might expect from the
rather forbidding ballot (see the ballot image in
the paper), is that voters would overvote,
marking their three choices in a single column
rather than in three separate columns. The 0.85%
overvote rate is "high for the OPtech Eagle III-P
precinct-based optical scan machines. In the 2002
Gubernatorial election in Florida, the state's
Division of Elections found these same machines
to allow for only a 0.5% rate of overvoting.
Mr. Dennis then notes: "Closer inspect of the
data ... seems to reveal something curious:
overvotes are only followed by undervotes, never
valid votes or other overvotes. He is referring
to his "Table SF," which reports the voting
patterns found from the ballot images: there are
three possible vote categories for each rank: a
single valid vote (V), an overvote (O), or an
undervote (U). Representing a vote for an office
by VVU, for example, means that the voter made a
valid first choice vote, a valid second choice
vote, and no third choice vote. Or VVO eans that
the voter made the same vote as VVU, except that
in the third rank the vote was overvoted, i.e,
cast for more than one candidate. The pattern OVV
does not show up in the ballot images. It turns
out that this, too, was deliberately programmed
that way, because an overvote in first choice
invalidates that and all subsequent choices.
I.e., if OVV is legally the same as OUU, why
bother putting it in the "ballot image."
(Because it is suppose to be an image of the
ballot! -- that's why!) In fact, these ballot
images are preprocessed, they are not the raw ballot data.)
Dennis also points out that the San Francisco
Poll Worker Manual contained an error which
apparently caused poll workers to tell voters
that they must vote for three candidates, or their vote wouldn't count.
Do these problems mean that we should not use IRV
or another preferential ballot system? Not
necessarily. But we should not deny that it may cause some confusion!
continued with " Dopp: 5.Confusing, complex and
time-consuming to implement and to count
"
----
Election-Methods mailing list - see http://electorama.com/em for list info
