Scanning of ballot papers
At every election, the paper ballots are counted in each polling place or counting centre after the close of polling at 6pm. At each ACT election from 1989 to 1998, the recheck of ballot papers and distribution of preferences that takes place after polling day, at a central scrutiny centre, was also undertaken by hand. This manual counting of all ballot papers after polling day was a very slow and painstaking process. At the 2001 and 2004 elections, after the first manual count of ballot papers at polling places, the preferences shown on all paper ballots were data entered at the central scrutiny centre, rather than being manually rechecked.
At the 2008 election the ballot papers were scanned and intelligent character recognition (ICR) software used to identify preferences shown on every formal paper ballot. Any preferences that could not be identified by the software or that did not meet business rules were verified by electoral officials.
All informal ballots continued to be manually rechecked. Scrutineers were entitled to observe the process and seek rulings on interpretations placed on ballot papers.
The scanning system will be used again at the 2012 election.
The scanning process
The scanning process worked as follows:
All formal ballot papers were parcelled at each polling place or counting centre in batches consisting of no more than 50 papers. Each batch related to first preferences counted to a particular candidate.
Each batch was allocated a number that uniquely identified the batch, the relevant polling place and electorate.
During the conduct of the scanning process it became clear that batching of ballot papers to bundles of 50 was unnecessary and slowed the scanning process. Ballot papers were subsequently batched into bundles of up to 200 for each polling place.
At the central scrutiny centre, starting on the Monday after polling day, each batch was scanned and imaged, and the image read by intelligent character recognition (ICR) software.
The ICR software interpreted the preferences shown, and the Robson rotation version number printed, on each ballot paper.
The scanning system used a range of strategies to ensure that preferences are captured with 100% accuracy.
After each batch of ballot papers was scanned, Elections ACT operators were presented with an image of each preference number on each ballot paper, and conducted an initial validation check on the accuracy of the scanned results (for example, by checking every “1” on every ballot paper , every “2” on every ballot paper, and so on).
After this initial check, the ICR software applieed a set of business rules to each ballot paper. Those ballot papers that the ICR software interpreted with a high level of certainty passed the business rules check were automatically admitted for counting. Any ballot papers which had unclear numbers, or did not pass the business rule check (for example having a break in the sequence of numbers, appearing to be informal, or appearing to show a first preference for a candidate other than the candidate named on the batch card), were flagged as requiring the verification of an Elections ACT officer.
The Elections ACT officer investigated each ballot paper requiring verification by comparing the on-screen image of the ballot paper with the ICR interpretation of the preferences on the original ballot paper to determine whether there has been any error in scanning, interpretation or validation.
The Elections ACT officer corrected any identified errors in scanning, interpretation or validation on the computer system. If necessary the officer viewed the original ballot paper if the scanned image was not sufficient to resolve any doubt.
If an Elections ACT officer changed a record of a vote on-screen, that change had to be verified by a second Elections ACT officer.
Scrutineers were entitled to observe the scanning, validation and verification processes, and where a scrutineer believed that a ballot paper has been incorrectly interpreted, the interpretation could be challenged. Challenged ballot papers were re-examined by a senior electoral official.
The above process continued until all formal paper ballots had been scanned and all ballot papers verified.
Once all scanned ballot papers were verified, the preference data from those ballot papers was transferred to the electronic voting and counting system (eVACS).
All ballot papers identified at the first manual count as informal were manually rechecked at the central scrutiny centre. Any papers ruled at that stage to be formal were scanned. Ballot papers confirmed as informal were not scanned.
The October 2004 ACT Legislative Assembly election saw the continued use of electronic voting and vote counting in the ACT, with a 70% increase in the number of electronic votes recorded. However, 176 340 ballot papers required data entry. The recruitment of skilled data operators capable of completing this task in a timely manner was a difficult and costly undertaking.
In its report into the use of electronic voting and counting following the 2004 ACT Legislative Assembly election, the ACT Electoral Commission identified this part of the election count as open to human error, which could in turn reduce the accuracy of the count. In its report, the Commission recommended the investigation of scanning technologies for the electronic capture of data from ballot papers at the next election. The investigation was undertaken with a view to replacing the data entry of ballot papers.
Elections ACT identified the basic requirements for an electronic data capture system as being one that will scan ballot papers, be able to read and interpret hand written numbers, allow for correction and verification of the captured data and provide an output file that is able to be read by the electronic voting and counting system (eVACS).
Given these basic requirements, Elections ACT released a consultation paper in May 2006 to determine if it would be possible to scan ballot papers for an ACT election. The responses from industry to the consultation were very positive, indicating that the technology existed to provide accurate scanning and capture of ballot information. Information gained from the consultation process enabled comprehensive requirements for the system to be prepared, leading to the calling for quotes for the development and implementation of a ballot paper scanning system.
In June 2007 a contract was awarded to Secure Vote Pty Ltd, now known as SEMA Group Pty Ltd, to develop a system for the scanning of ballot papers for ACT Legislative Assembly elections. Secure Vote/SEMA Group has extensive experience in the use of scanning technology and in particular its application to electoral related operations. From then until the 2008 election, the system was developed and extensively tested to meet the exacting standards of accuracy and confidence set by the ACT Electoral Commission. An independent audit of the system, conducted prior to the 2008 election, concluded that the software was free of any code that could fraudulently or inadvertently alter the preferences of a voter.
The scanning system was used successfully at the October 2008 ACT election. Scanning of ballot papers was completed and the final result announced on Saturday 25 October 2008, 7 days after polling day. A report on the operation of the scanning system is included in the Report on the ACT Legislative Assembly Election 2008 published by the Commission in September 2009. The report noted that the Commission was confident that the count of preferences using the scanning system was conducted at a very high level of accuracy.
A review of the accuracy of the scanning system was conducted in 2011 and reported in the ACT Electoral Commission Annual Report 2010/2011. This review concluded that the system demonstrated a very high degree of accuracy. A few minor areas for improvement were identified. These will be addressed before the 2012 election.