Added Votes and the Accuracy of
Los Angeles County Scanners
1% MANUAL TALLY FOR Nov. ’05 SPECIAL ELECTION
And SNAP TALLIES FOR June “06 PRIMARY ELECTION
by Judy Alter
In a study of the exactness of the vote count in Los Angeles County,
the scanners not only do not count votes accurately they also add
votes not found by hand counting. Trained employees carry out the
manual tally with great care, thus, the hand count is the accurately
counted result. I studied how often the computer and hand count matched
in the Los Angeles County report of the mandatory manual tally of
1% of precincts (62) from November 2005 Special Election and in the
12 snap tallies completed in the June 2006 primary election.
The ballot for the Special Election of November 2005 contained:
eight state-wide initiatives, 123 local races, and 16 bond measures.
I found in the totals for each contest in each precinct
that the hand count and computer count only matched exactly: 28%
in the 496 initiatives (62 precincts times 8), 12% in the 123 local
races, and 44% in the 16 bond measures. For this 1% manual tally
I did not calculate the error rate. In the snap tallies for the
June 2006 Primary Election, completed at 12 precincts after the
polls closed on election night, poll workers hand counted three
races: governor and attorney (for all political parties), and proposition
82. Poll workers completed this hand count of all ballots for these
three contests before they were scanned at Los Angeles County election
headquarters. No total snap tally exactly matched the scanner count.
In separate totals for each candidate running in the primary: i.e.,
eight Democrats and four Republicans for governor, the hand
counted snap tally and computer count matched for individual candidates
from 27% to 70% of the time. For Proposition 82 no snap tally (hand
count) computer comparison matched exactly for both yes and no votes
but in two (17%) of the 12 snap tallies the no votes matched.
Total Ballots Cast
I compared computer and hand tally for the total ballots cast (range
18-1206 ballots per precinct) for the initiatives in the 62 precincts
and for the 139 local races and bond measures in the small cities
where these elections were held in Los Angeles County. The computer
and hand tally matched for the total ballots cast for the initiatives
in 49 (79%) of the 62 precincts. In the other 13 precincts, the
total differences between the computer and the hand count included
one to three (total nine) ballots added in the computer count (and
needed to be subtracted from the hand-counted totals) and one to
three (four total) ballots not counted. In 120 (86%) of 139 local
elections and bond measures the manual tally matched for the total
ballots cast. In the remaining 19 contests the total differences
included one to three (20 total) ballots added by the computer
and one to three (eight total) ballots not read by the computer.
The snap tally precinct report did not include the total ballots
cast although the official election night reports included them:
No Differences for individual totals
I also counted each specific match for yes, no, or candidate votes
for the November 2005 Special Election. Of the 496 initiative reports,
the computer and hand tally matched for 95 yes votes and 66 no totals.
Among the 544 individual candidates in the 123 local elections,
175 (32%) matched. Among the 16 bond measures (requiring yes or
no responses), five yes or no vote counts (16% of 32 responses)
Votes that human eyes read but scanners
Votes that human eyes did not find but
the scanner ADDED
I identified two separate problems when analyzing the accuracy of
the computer counting of votes compared to human beings counting
of votes: votes that humans see and votes that they do not see because
the votes are not on the ballots. When people examine a ballot for
a mark that indicates a vote, people can discern even faint marks.
Thus, from the manual tally a logical result will be that the people
find more votes than when the computer scanned them. In the manual
tallies of the initiatives, the computer missed a total of 436 yes
votes and 479 no votes. This total number of missed votes does not
consider the error rate per contest: the error rate is low. In the
local races, the people doing the manual tally found 643 votes that
the computer did not count, and in the bond measures 108 votes.
The range of votes missed per contest ranged from 1 to 23. In the
snap tallies the computer did not read 98 votes that the poll workers
counted, range one to eight per contest.
A more serious finding is when the 1% manual hand tally and snap tally
reports showed that the computer counted (added) votes that humans
did not find on the ballots. Somehow, in the counting process by the
electronic scanner of the ’05 initiatives, the computer added
110 yes votes, and 99 no votes, range one to 17 per contest. Although
these numbers, 110 and 99 constitute a very small percentage of the
hand tallied votes, the fact that computers can add votes not written
by a voter is very disturbing. In local races the computer added 354
votes. In the bond measures, the people found only one vote added.
In the computer count of the hand counted snap tallies, the computer
added 145 votes, range one to 18 per contest.
The error rate for 12 snap tally precincts for all races is 6.5%.
The highest three error rates were for the Republican gubernatorial
primary (8%), the Republican attorney general race (7.9% with only
one candidate running!) and Prop 82 (7.8%).
I asked Deborah Wright, the Information Liaison Officer for the Los
Angeles County Registrar, how the election officials work with the
results of this 1% manual tally. Ms. Wright said the employees conducting
the manual tally sometimes recount some batches of ballots on another
scanner when the manual tally shows high discrepancies. Officials
neither notify the candidates when the manual tally shows different
winners and losers than the computer count nor do they publicize the
findings of the manual tally. The exit poll expert in charge of the
snap tally project at the Los Angeles Times used the snap tally information
to double-check their exit polls, thus, the accuracy of the vote count
was not the goal of conducting snap tallies.
While employees and election officials carry out the mandated manual
tally of 1% of Los Angeles County’s precincts with great care,
the officials do not use the results to examine the contests further.
And the California Election Code offers no guidance about what to
do with the results. I only could study and report these results
because I requested them.
To conclude, both opportunities to hand count ballots and thereby
check the accuracy of the electronic vote count—the manual
tally and snap tallies--showed that the electronic scanner/ computer
counts produced inaccurate results. If people conducting the manual
tally found computer added votes that are not on ballots in 1% of
precincts, then I expect that they would also find added votes not
on ballots in the remaining 99% of the precincts. This is especially
disturbing when election officials include non-existent votes in
certified election results. Every real vote counts. Remember that
recent elections were won or lost by small vote totals, i.e., 527
for Bush in 2000 in Florida.