Hart InterCivic Optical-Scan Has A Weak Spot
By John Gideon,
Information Manager for VotersUnite.Org and VoteTrustUSA
05 July 2005
Two Washington state counties use Direct Recording Electronic voting machines as their primary voting system. Optical-scan machines are their secondary systems, though they are used to count a majority of the votes in these counties due to the high percentage of absentee and vote-by-mail voters. The recount of the DRE machines was done by re-running the report tapes from the machines.
Late in November of 2004 a machine recount was done, by law, in the state of Washington, due to a close election between two candidates for Governor. This recount was to be a complete recount of all optical-scan and punch card ballots using the same machines that had counted the ballots earlier in the month. The software was changed, in nearly all counties, to cause the machines to read only the governorís race.
Yakima County uses Hart InterCivic eSlate central count optical scan and DRE systems. The optical scan system uses a Kodak scanner that scans the ballots, and then the ballot images are analyzed by the eSlate Ballot Now software and tallied.
Unfortunately, Yakima County made the decision that they would not re-scan their ballots for the machine recount. The decision was made at the county level, at least, and possibly at the state level, that it would be alright to simply re-analyze the ballot images previously generated by the scanner.
However, fortunately for the voters but unfortunately for Yakima County and Hart InterCivic, the machine recount was followed by a hand-recount requested by the candidate who came in second in the recount. In one Yakima precinct, the hand recount resulted in 24 new votes that had previously gone uncounted. This is 24 votes representing 24 voters from Precinct 3301 who had lost their votes for Governor until the hand count of the original ballots found those votes.
How did this happen? Thatís the question that the county asked Hart InterCivic. An email received in response to an Open Records Request to Yakima County answers that question. On June 15, 2005, Travis Harrell a manager at Hart InterCivic reported the results of an investigation his company had been carrying out.
The reported reason for the 24 uncounted ballots is that some foreign material, i.e. dirt or a small piece of paper, was deposited on the screen of the scanner during the scanning of the ballots. This foreign material caused a white line to run down the right hand column of votes on all 24 of those ballots. Part of an actual ballot image is shown below.
[Editor's Note: After this article was printed, John called Kodak to ask about the white line, and he was told that dirt on the lens would cause a black line, not a white line. They didn't know what would cause a white line.]
Note that there were four other races in that same damaged column, so the votes of 24 voters in four other races were permanently lost. See page 4 of the email message from Mr. Harrell:
Mr. Harrell reported the following:
This problem is indicative of a potential problem with all Hart InterCivic Optical-Scan systems. They rely on a scanner that must be kept completely clean so dirt is not deposited on the screen and then distorts the ballot images. The system can be set up not to tell the elections officials why the ballots are being rejected, so they have no warning that something may be remiss with the scanner.
1. The scanned ballot images of the 24 ballots in question (Batch 86, Pct 3301) all contain a white vertical line spanning the entire length of the ballot. The line runs directly through the left portion of all option boxes in the 4th column of each ballot. The line was most likely caused by a small foreign object (dirt or paper debris) in the scanner that subsequently dislodged. The attached image shows the line clearly and is representative of all 24 ballots.
2. Since the white line "whited out" a portion of each option box, Ballot Now was unable to detect at least 90 percent of each "target box" and therefore classified each contest in the 4th column of each ballot as a Damaged Contest.
3. Audit log data confirmed that during the processing of Batch 86, the option to Autoresolve Damaged Contests was selected. Consequently, Ballot Now (in autoresolve mode) confirmed and recorded the damaged contests as undervoted.
4. The undervotes recorded in the Cast Vote Records generated by BN were subsequently written to the BN MBB and then tabulated and reported by Tally.
5 All components of the eSlate System functioned to design and certification specifications.
The problem is also indicative of a problem with elections procedures. In this case, someone approved the use of the originally scanned images in the recount, even though the recount was clearly supposed to be a complete machine recount of all ballots cast on optical scan and punch card systems. This decision allowed those 24 votes to remain uncounted a second time.
How could the county have known that there was a problem even if there was no recount? The precinct in discussion was precinct 3301. A total of 512 ballots were cast via absentee voting and counted by the optical-scan system. A total of 31 undervotes were reported in that precinct.
Undervotes are not out of the ordinary. In fact, in Yakima County, almost all precincts had some undervotes in the gubernatorial race. However, while the average gubernatorial undervote rate for all precincts was 1.6%, Precinct 3301 reported 6% undervotes. This should have been a red-flag. Hand counting those 31 "undervotes" caused by a speck of foreign material revealed only 7 true undervotes a 1.3% percentage that is comparable to the rest of the county.
So what needs to be done in the future so this never happens again? First, the rules for machine recounts must be spelled out and followed. Short-cuts only lead to problems and should never be taken.
Second, elections officials must be aware of anomalies. They must look for red-flags and question them. Anomalies can indicate lost votes.
Third, election officials should examine all Damaged Contests rather than allowing the software to "autoresolve" such problems without human oversight.
Fourth, election officials should consider setting the software to detect a vote when less than 90% of the target box is filled in. As the ballot image demonstrates, a significantly smaller percentage can be filled in and still indicate a vote for the selected candidate.
And finally, if the hand recount had not been done, no one would ever have known that a speck of dirt could cause ballots not to be counted. A percentage of all ballots counted by the Hart InterCivic optical scan machines must be audited every election. The audit must be to hand count the original, voter marked paper ballots and compare the results of the hand count to the machine results. Without this minimal check on the accuracy of the machines, we will never know whether votes are being discarded.
[Thanks to Paul Lehto for requesting the records that brought this information to light and for calling it to our attention.]