The PCOS accuracy requirement

4

Comelec prescribed an accuracy rate of 99.995 percent as a response to the requirement of the Election Automation Law that the automated election system perform accurately. This was included in the technical specifications of the PCOS enumerated in the Request for Proposal for the Automation of the 2010 National and Local Elections issued by the Comelec. This requirement is interpreted to mean that a discrepancy of one vote out of 20,000 vote marks between a manual count and PCOS count is acceptable.

In evaluating the PCOS eventually used in the 2010 and 2013 elections, the Comelec conducted a simple test of the machine using a few hundred test ballots with a total of 20,000 vote marks. Having found no discrepancy between the manual count and the PCOS count, the Special Bids and Awards Committee declared the PCOS machine 100 percent accurate.

As required by law, an audit of the PCOS in randomly selected clustered precincts was conducted after the close of election operations on election day. This audit activity is referred to as Random Manual Audit (RMA).

The reported PCOS accuracy rate in 2010 was 99.6 percent or a discrepancy of 80 votes out of 20,000 vote marks between the manual count and PCOS count. In 2013 it was 99.9747 percent or a discrepancy of 5 votes for every 20,000 vote marks between the manual count and PCOS count.


But in both 2010 and 2013 the RMA Team (RMAT) arbitrarily set aside the required 99.995 percent accuracy rate declaring that “Based on ninety-nine percent (99 percent) accuracy rate, the allowable margin of variance was set at one percent (equivalent to less than 10 votes’ difference in absolute value for every 1,000 valid votes counted).” This led me to think: How did the RMAT determine the accuracy rate of 99 percent? What data did the RMAT base it on?

Well, trawling the internet for information, I found a company (http://www.abbyy.com/) that specializes in mark recognition technologies, including optical mark recognition, which is the underlying technology in the PCOS. The company must have a basis to declare and publish a 99.995 percent accuracy rate of its products. That basis must have been the results gathered from exhaustive tests of its products.

In setting aside the 99.995 percent accuracy rate, the 2013 RMAT explained that “Since most of the ‘variance’ can be attributed to human errors or clerical errors, aiming for a higher accuracy rate to as high as 99.995 percent (i.e. 1 vote difference in absolute value for 20,000 valid votes counted) could be statistically improbable.” Revealing. Does it mean that rather than simply assessing the accuracy of the PCOS, the RMAT aimed to match the 99.995 percent accuracy rate? The 2010 and 2013 RMA Reports show that the RMAT went through several iterations when it found that the manual count and PCOS count differed by at least 10 votes Failing to meet its goal, it declared the 99.995 percent accuracy rate statistically improbable to match.

Let’s look at the potential impact of a 99 percent accuracy rate vis-à-vis a 99.995 percent accuracy rate. Let’s assume 52 million registered voters. Let us also assume that 80 percent went to the voting precincts to vote. This means that on election day 41.6 million voters actually voted. Let us further assume that all of them voted for a president or vice president. An accuracy rate of 99 percent, using our assumed numbers, would mean a discrepancy of 416,000 votes between the manual count and the PCOS count. On the other hand, an accuracy rate of 99.995 percent translates to a discrepancy of 2,080 votes between the manual count and the PCOS count. 416,000 vote vis-à-vis 2,080 vote is an ocean of a difference!

Let’s Face IT! The numbers have spoken and numbers don’t lie. The PCOS accuracy findings in the 2010 and 2013 RMA are obviously below the 99.995 percent accuracy rate set by Comelec. The RMAT conveniently avoided to explicitly and unequivocally declare that the PCOS counts did not meet the accuracy rate requirement. Instead, the RMAT, without any basis, lowered the accuracy bar to 99 percent to show that the PCOS accuracy rates in 2010 and 2013 RMA are better than its prescribed level.

Unacceptable!

Share.
.
Loading...

Please follow our commenting guidelines.

4 Comments

  1. you forgot one important thing – the time stamp of each machine, it was proven that the servers are getting feeds from an offline machines….

  2. What is being grossly forgotten in this long-running PCOS accuracy debate is the mandatory requirement to CERTIFY in a laboratory environment (in the Canlubang site maybe) and much ahead of the deployment of each and every PCOS unit to the field by testing each and every one of these units to determine if they pass the 99.995% accuracy hurdle or ONE scanning error in 20,000 marks. The equiv of this accuracy hurdle is ONE scanning error in any of the ballots out of 667 ballots tested (assuming 30 marks per ballot to accommodate a senatorial and local election) . All units that fail to pass this test are rejected and replaced by units that will pass the test. The rejected units MUST be put aside and NOT USED in the actual election. Now, we have been seeing discrepancies of 88 in some 300+ ballots in the case of the Nueva Ecija recount (discrepancy is any mismatch between the PCOS count and manual count on the same ballot). 88 discrepancies is equiv to being some 200 times less accurate than prescribed. Definitely this is unacceptable. Many ballot images do not even match the corresponding physical ballot’s when these must perfectly match. Ballot tampering ? Now we have a three-way mismatch – the PCOS count vs the physical (actual) ballot vs the supposed image of the same physical ballot. Ano namang kalokohan ito ?

  3. This is an impressive research. Alarming, indeed, as it is with almost half million difference that could spell out a different result specially for a presidential election winner.

  4. Take away 416,000 votes then add it (416,000) on the other = 832,000. Almost 1 million difference from true count.