This is a revised version of a commentary posted on Newsbreak Online (http://newsbreak.com.ph) on 4 April 2009.
1. Much has been said about the ‘full’ automation of the 2010 elections by COMELEC, but initially the discussions bordered on the extremes: having it automated and having it as before. Now that some P11.2 billion has been approved for the automation, the bid is out for the acquisition of equipment and technology. But technology is a continuum from being absent to being present. I was not certain about which technology was to be adopted by COMELEC. My only sources were recent commentaries on the automation to be undertaken by the Commission on Elections (COMELEC) for the 2010 elections by two prominent citizens: Tong Payumo, former Bataan congressman and former Subic Bay Metropolitan Administration Chairman, and Gus Lagman, who is a proponent of the Open Electoral System (OES) of TransparentElections.org
Essentially both had the same idea of what it would be. On Election Day, each voter will shade the ovals opposite the names of the candidates he/she wants to vote for on a specially-printed ballot. The voter will then feed the ballot himself to the PCOS (Precinct Count Optical Scan) machine. The PCOS is an OMR (optical mark recognition) machine. COMELEC will cluster the 320,000 precincts down to 80,000 (clustering up to a maximum 5 precincts into one) and install one PCOS machine in each.
2. However the COMELEC RFP (Request for Proposals) specifies that the system should be able to recognize the following marks on the appropriate space on the ballot opposite the name of the candidate to be voted for: full shade; partial shade; check mark; and x mark. Moreover the system should be able to capture and store in an encrypted format the digital images of the ballot for at least 2,000 ballot sides (1,000 ballots, with back to back printing; 200 voters per precinct times 5 precincts).
3. Optical mark recognition (OMR) is the scanning of paper to detect the presence or absence of a mark (e.g. shaded box) in a predetermined position. It evolved from the IBM punch cards of old where punched holes were sorted to represent information. OMR is the simplest of commonly available form processing technologies (e.g. Lotto tickets). OMR equipment has been available for many years and has nowadays reached very good levels of reliability. But OMR has relatively stringent requirements for the successful processing of the paper forms. Thus, countries with very dusty or humid climates and poor transport infrastructures are discouraged from using OMR. Special questionnaire/ballot design restrictions include the quality of the paper along with precise specifications regarding the printing and cutting of the sheets.
In the numerous censuses that were conducted from 1990 to 2000 by the National Statistics Office, I did not choose the OMR technology. Then scanners and forms were imported. (I don’t know if there are any Philippine manufactures today.) The printing forms had to be precise so that the information in it could be read by the scanner. In the field it would be difficult to keep the forms, dirt-free, dust-free, and at specified humidity content. Understandably if the form was dirty, dusty or rather wet, either the input machine would reject it, or, if accepted, the form could not be read because of the ‘shading’ caused by dirt, dust, smudges, and the like. This was actually reported by the Bangladesh Bureau of Statistics in their earlier censuses and it seemed obvious that the same can happen here. The results would be incomplete; a back-up system would be needed to key the data in another computer program and merged with the results from the OMR machine. Can this be done in an election?
4. The paper form is often called an optical answer sheet or a “bubble sheet”; an example being the form used in multiple choice question examinations. OMR technology is used to detect shaded ovals or boxes that correspond to answers (or votes for a particular candidate). But shadings should completely cover each distinct oval or box and should not stray to the adjacent ones; otherwise the machine will reject the form or ballot. A more serious error is incorrect printing such that all ovals are read as filled, even if this is not the case. This happens if the outline of the ovals is too thick, or is irregular, as was the case of 19 thousand absentee ballots in the county of Gwinnett, Georgia during the 2008 U.S. presidential election. This was discovered during a test run made in late October and after around 10 thousand had already been returned. The slight difference was not apparent to the naked eye. For more details, see http://en.wikipedia.org/wiki/Optical_answer_sheet.
5. Many traditional OMR devices work with a dedicated scanner device that shines a beam of light onto a special form paper. Where the scanner detects less light in a particular area, due to shading, on the specially printed ballot, this is counted as a vote for a particular candidate by the computer program.
Would the scanners be imported? In the 1990 Philippine census the United States Agency for International Development (USAID) donated some 400 microcomputers to process the census, which was the first time for a middle-sized country. These PCs (110 volts) were provided with auto voltage regulators and generator sets to assure stability and continuity of work. The economic life of most of the PC sets was rather short; these US microcomputers were not built for the dust and humidity conditions in the Philippines.
Per the COMELEC RFP, the scanners and other equipment will be on a one-time lease basis, with option to purchase. I think it should not purchase. Where would the scanners be stored after elections? There would be need for climate- and humidity-controlled warehouses for these and other equipment. And by the next elections improved machines and technologies may become available, rendering these obsolescent and unusable in a short period of time. We should not have more of the Mega Pacific counting machines, now at the warehouse, dusty, rusty and with mold.
In contrast to the dedicated OMR device with specially printed forms, desktop OMR software allows the creation of forms in a word processor and printing these on a laser printer. The filled up forms are then fed to a common desktop image scanner and read by the OMR software. This appears to be cheaper than the arrangement where both scanners and forms are sourced from abroad, although the control over the printing of ballots will be reduced. I am not sure though if COMELEC is open to this ‘less-controlled’ OMR technology.
6. Another similar technology are optical character recognition (OCR) and intelligent character recognition (ICR) which consist of the use of scanners to read characters (alphabets and/or numbers as opposed to shaded areas handled by OMR) at specific locations in the questionnaire/ballot. OCR recognizes printed characters only, whereas ICR could recognize handwritten text as well. This would require that handwriting of text in questionnaire/ballot forms by the voters be as uniform as possible to common “model” handwriting that could be processed by the recognition software engine.
Does one remember when the use of zip code was first promoted by the Post Office and a mail sender had to enter the code in pre-printed boxes on the envelop, but with specified strokes for each number. There was only one way of writing the digits from zero to 9. But it was difficult to change the way people wrote digits; among others, our ‘1’ was different from the ‘1’ recognized by the software. I think the scanner was manufactured in Japan. Of course, without proper training of the population, this experiment failed. However OCR/ICR technology has progressed very much and it is still improving with the implementation of more sophisticated recognition algorithms and the use of neural networks for self-learning of the system.
7. Available literature will always cite the speed of processing with the OMR, but there have been reservations. The National Bureau of Statistics of China for instance reports that OMR has its advantage of faster scanning and more accurate recognition, but these advantages are based on higher quality paper and stricter accuracy for printing and cutting the census forms. But it was difficult to meet these high requirements since the required amount of forms was huge. It thus shifted to OCR for the 2000 census which involved the processing of over 400 million questionnaires.
Is COMELEC ready for a new technology?
8. The question is whether the electoral system, with some 50 million registered voters by COMELEC estimate, is ready for the radical changes? Is the COMELEC prepared to decide on and implement the technology that would be appropriate for the Philippines? Can the senior voters, among others, fill in/shade the spaces or enter the ‘x’ or check mark that is recognizable by the scanner? If the scanner rejects a filled-up ballot for various reasons, how would the voter’s choices be counted? Will the ballot be a spoiled one? What if these spoilages would not be far and few, and not be random? While many are concerned mainly with the counting of votes, I am deeply concerned at the first step: casting of ballot and electronically ‘dropping ’this in the ballot box. The technicians deployed to the precincts with the PCOS may not be able to handle the legal aspects arising from this.
9. With the foregoing and if the above concerns cannot be addressed adequately, I hope that the COMELEC goes conservative in its automation and consider the Open Election System (OES) proposed by TransparentElections.org. I understand that the OES scheme retains the manual system of voting, counting and preparation of the election returns (ERs) at the precinct level and which, therefore, requires no voter training. The reading and counting of votes during elections would be as before. What will be automated are the canvassing and consolidation of ERs from the municipal to the national level. The encoding of the ERs and posting these on the Web make the results accessible to the public here and abroad.
10. Among census administrators, we have found that how smoothly a census exercise ran was less attributable to the method chosen and more attributable to how well the method was implemented and managed. The same would be true for elections as well.
11. I stand to be corrected by the COMELEC. I have no other agenda than to raise these issues for discussion. I am not a member of any bidding team nor will be.