This is a mildly technical e-mail, but just may save an election near
you next week. Pass this along to your favorite political party or
candidate, because they'll need to act quickly if their election is
problematic. Their database guy will know what to do with this new
method to isolate absentee votes anomalies. You can see shortcut
images and discuss this here:
http://www.bbvforums.org/forums/messages/8/81251.htmlAbsentee voting is even more opaque and high-risk than most US voting
methods. Locations with no-fault absentee voting like Marin County,
California now have nearly two-thirds of all votes cast as absentee
votes.
A recent ballot measure of great concern to environmentally-conscious
citizens was Marin County's Novato Sanitary District "Measure F",
which won by just 179 votes in June 2010. Measure F was thumped at
the polls, but absentee votes flipped it to a win.
Marin County citizens obtained public records including the Diebold
GEMS Microsoft Access databases. Using these databases, we can shed
more light on what happened. The results of this inquiry raise
questions about the integrity of the Measure F election process.
Detailed images showing new absentee vote analysis methodology:
VEOLIA
"Measure F" concerns a controversial French company trying to to land
a water contract in Novato California. The locals were up in arms
about it, citing a dubious corporate reputation and a heavy-handed
company financial investment in the election:
http://newsblaze.com/story/20100603073032zzzz.nb/topstory.html"In what is probably one of the more egregious examples in the Bay
Area of a big corporation buying an election, Veolia Water has
dumped, according to election officials, $140,000 into the Yes on
Measure F campaign," a local news article writes. And judging from
the smackdown Measure F took at the polls, the locals weren't too
crazy about Veolia. Or were they?
THE VOTE FLIPS
On Election Night June 8, Measure F was announced to be winning by a
hair on absentee ballots. Polling place votes were s-l-o-w coming in,
with not a single precinct uploaded until 10:28 p.m., long after most
election watchers went home. The precincts reversed that absentee
trend, but later, as more absentee ballots were counted, the trend
flipped again. When the election was finalized on June 25, Measure F
and Veolia Water won by 179 votes.
WHAT THE GEMS DATABASE SHOWS
The Diebold GEMS central tabulator uses a Microsoft Access database
to accumulate and calculate vote tallies. This database contains a
number of tables, and within these tables you will find detailed vote
results for each absentee batch, broken out precinct by precinct.
I examined the Marin County vote tables and here's what I found:
1. Measure F was losing on the absentee vote -- citizens were
defeating it, though it was a close race. Marin County began scanning
absentee ballots on May 28 for the June 8 election. It had scanned 67
batches by 3 p.m. Election Day. The "NO" votes were winning.
2. At 3 p.m. Election Day, just hours before the first results report
was issued, Measure F reversed its trend. In just three and a half
hours, more than a week of "NO" votes turned to enough "YES" votes to
announce that Measure F was winning -- by a close margin -- on its
absentee votes.
3. The locals waited. And waited. And waited. Polling place results
just did not come in. They were told that a police car carrying the
memory cards broke down enroute to the election center. (Not true.)
They were told that the poll workers dropped off materials late, but
when they asked to see the forms showing drop-off times, they were
told no one wrote down the time. Finally, at 10:30 p.m., those pesky
poll votes filled with resounding NO votes were reported.
4. At end of Election Night, the NO votes had it. All that remained
were the late absentee votes.
5. The GEMS data reveals that during the scanning of these
after-election absentee ballots, batches kept going missing, usually
followed by a drop in NO votes and a reversal of fortune for the YES
votes. On June 14, batch after batch disappeared from the record,
reappearing (one assumes) later on with a preponderence of YES votes.
June 14 began with Measure F losing. It ended with a Measure F
winning margin that held to the final tally of 179 YES, Measure F
passed.
WHAT DO THE MISSING BATCHES MEAN?
Actually, the GEMS database reveals that two kinds of disappearing
vote batches occurred. The first, a set of three precincts with
polling place votes uploaded on June 8, Election Night, disappeared
and went poof around quitting time June 23. They reappeared June 25
with new memory cards, changed by onesies and twosies but with no
significant new numbers.
The second missing batches were several sequentially numbered
absentee batches, which went missing in clumps on June 14, the same
day Measure F switched from NO to YES.
ANSWER TO FIRST MISSING BATCH QUESTION MAKES SENSE. ANSWER TO SECOND
MISSING BATCH ISSUE DOES NOT.
I called Marin County Elections and spoke with Melvin Briones. The
missing polling place votes, he explained, were due to the California
requirement to hand count a small percentage of precincts. When
discrepancies were found in the hand counts in these precincts, they
had to be rescanned with new memory cards. This jives with what we
know to be true: Optical scan machines don't always interpret marks
on the ballot the same way a human eye does. Discrepancies of
one-half to 1 percent are common. These will show up in hand counted
batches.
And then we get to the explanation for the missing absentee batches.
Briones explained that the batches are scanned in sets of 100 or 200,
and he said that when the machine doesn't report the ballot number
correctly, the batch is erased and re-run.
Now here's why this explanation has me confused: The batches come in
on trays containing varying numbers of ballots. The ballots are not
pre-counted on these trays. The batches may average 100 to 200 each,
but in fact, according to the GEMS databases, the actual numbers are
somewhat random: 63, 163, 61, 88, 106, 46, 47, 137, 199, 210, 137,
177, to quote actual batch sizes shown in the GEMS record.
Now, since they aren't hand counting all these varied batch sizes,
how would they know if the machine counted the wrong number of
ballots? Do they only pull a batch if the machine spits out a number
that is purely ridiculous? Why then would we believe the machine's
count on batches that are wrong but not ridiculously so?
And if the machines were spitting out wrong numbers of ballots, but
only sometimes, even if the Marin County elections staff had a 100%
accurate ouija board to divine that ballot numbers were correct,
wouldn't this occasional wrong counting pose a concern for the
overall accuracy of the count?
I don't understand.
I asked Melvin why disappearing absentee batches clumped up on June
14, and he said he'd have to check the log. That's good, then, they
log the problem batches. But when I asked Melvin if they write down
the reason they abort the batch on this log, he said probably not.
But if they don't write down the reason, why would checking the log
tell Melvin anything about why a bunch of missing batches clumped up
on June 14?
WHAT WE DO KNOW - AND THE NUMBERS JUST DON'T WORK ON THIS
Concerned Marin County citizens requested a hand recount. The number
of votes in the recount exactly, precisely matched the official Marin
County result.
Why is this exact match something that just wouldn't happen? Well,
remember that according to Marin County's own manual audit, three
polling places showed discrepant counts and had to be rescanned using
new memory cards. And of course, there are all those absentee
scanners magically reporting the wrong number of ballots, causing
ballot scanning abortions.
There were over 15,000 Measure F votes, two-thirds of them absentee.
Do you mean to tell me that EVERY SINGLE ONE of 10,037 ballots was
read exactly the same by the machine and the human eye? Normal rates
would show from 50-150 discrepancies. An exact match in Marin County?
Didn't happen. Not without some help.
NO OTHER RACE SHOWED A DOUBLE TREND SHIFT LIKE MEASURE F
I pulled every race on the ballot that was even slightly competitive,
and ran the same absentee analysis. None showed the odd double-flip
pattern seen with Measure F.
WHAT'S "NEW" ABOUT THIS ANALYSIS?
We haven't had much opportunity to examine the databases, but these
are consistently being released now, following several court cases
ordering their release. The data, of course, belongs to the public,
not to the vendor.
The databases contain information that is much more detailed than
that contained on results reports. Absentee votes, and early votes,
flow into the system in mixed batches; that is, each batch has
multiple precincts in it.
The databases, however, parse out these batches and assign a
timestamp to each batch as well, allowing a precise analysis of vote
flow. Even though Election Day wasn't until June 8, using the
database you can see vote results as of May 28, June 1, and each day
votes were scanned.
We can separate out the inflowing votes by time, precinct, race, and
ballot choice. By doing this, it becomes clear that Measure F was NOT
winning its absentee votes before Election Day, and hit a rather
dramatic trend shift just a few hours before the first results report.
By doing this, we can see exactly when batches went missing, were
aborted, whatever -- and we can compare these missing batch events
with trend shifts, to see that most often when an absentee batch went
missing in Marin, it was followed almost immediately by an upshift in
the YES votes.
WHAT TO DO ABOUT THIS
Ditching the absentee vote craze for a return to need-only absentee
voting is a good idea.
If you are in a location with lots of absentee voting -- heads up
especially for Washington state, Oregon, Colorado, California,
Florida, and Ohio -- and if the election looks like a stinker,
immediately request the databases.
With the Diebold GEMS system, the files are stored with the file
extension "gbf" -- those are passworded files that you will not be
able to open. Request both the original gbf and the mdb (MS Access)
files.
You can find out which voting system your county uses by going to
verifiedvoting.org or by looking at the state and county section of
the
http://www.blackboxvoting.org Web site. Different voting
manufacturers use different database structures, but the detailed
information referenced above is contained in all the databases. It is
your right to demand copies of the full databases in non-proprietary
format.
If you have experience with databases, you should be able to sort out
the data tables to identify precincts, races, candidates, vote types
(ie absentee, polling place, early), upload times and votes. You can
then export the information to sortable formats and perform
calculations and look at graphs.
Using these kinds of absentee vote diagnostics, you may be able to
identify exactly when the vote count got kinky. It's a start.
BUT THIS ISN'T GOOD ENOUGH
The public has a right to be able to see and authenticate each
essential part of a public election, without need for special
expertise.
Requiring citizens to obtain complicated databases, with skills in
Microsoft Access, SQL Server, SQLite, and other database forms just
to get a glimmer of a clue what went on in a public election is not
acceptable. You can't run a democracy this way.
Absentee voting conceals who voted, conceals the chain of custody,
and conceals the count. Diagnostics like these are interesting,
though excruciating to perform. They may come in handy this November
in controversial absentee-heavy races.
But we need to remember to ask for what we really want, and it ain't
this. No one should have to sift through a wheelbarrow of obscure
data tables for a smattering of circumstantial evidence as to what
really happened in an election.
What we really want is to restore the public ability to see and
authenticate each essential step in the election, without need for
special expertise.
PERMISSION TO REPRINT AND/OR EXCERPT GRANTED, WITH LINK TO
http://www.blackboxvoting.org