99.995 Percent

Comelec Chairman Jose Melo told ABSCBN News that the poll body is "ready for manual elections if poll automation tests fail" -- almost as if he expects that very disappointing outcome. He also revealed the key TECHNICAL SPECIFICATION against which all bidding systems will be tested: a required 99.995 percent accuracy rating in reading and counting ballots. How exactly does one perform a Quality Control test to measure such a capability in an automated ballot reading and counting system? And what are the Constitutional implications of a formal Comelec Request for Bid from private companies whose main technical specification contemplates as acceptable the erroneous reading and counting of 5 votes out of every 100,000, or 2000 votes out of the 40 million to be cast next year for President.

Here is the basic scenario. On Election Day 2010, up to 40 million ballots could conceivably be cast for President, Vice President, 12 Senators, Congressman, Party List Representative, and provincial and local government positions. For the sake of simplicity, assume there are twenty-five candidates selected on each ballot so that a possible one billion votes must be read off 40 million ballots and accurately credited to the indicated candidate.

What 99.995 percent accuracy rating really means is that When the thousands of ballot reading and counting machines finish their work on all the ballots that are actually cast on an Election Day in which one billion votes for national and local positions are cast, they will have made ERRORS IN READING AND COUNTING not more than 0.005 percent of the one billion, or not more than 50,000 votes cast for national and local candidates. That is how many votes the 99.995 percent accuarcy rating contemplates as tolerable, spread out over all the thousands of individual positions actually available. For a single position, such as President, the specification contemplates as acceptable the erroneous reading and counting of up to 2,000 votes out of 40 million votes cast.

This appears to be an unavoidable cost of automation because of the statistical nature of error and malfunction in electromechanical devices of all kinds. Of course, it is probably a legal and constitutional fiction, that the manual counting of ballots is 100% accurate., especially where the main technical qualification is handwriting analysis. The Omnibus Election Code condones unreadable or spoilt ballots, but the present specification is being justified by the notion that these errors are small in number and that it is unlikely they would affect the final result of any election. However unlikely, there is now the possibility of some kind of constitutional crisis should it ever occur that the winning margin be comparable to the margin of error in some future very close election. For the Constitution even contemplates the notion of a tie that would be broken by Congress. But does such a provision apply to a "statistical dead heat" that could occur?

Although Comelec has not yet released the formal Bid Specs or the $20,000 Bid Documents it is widely anticipated that among the leading bidders will be systems offering Optical Mark Readers. Public testing of these machines has been promised and much speculation is being ventilated about the security of these systems. There is also the question of how to proof-test a given machine or set of machines as being capable of 99.995% accurate reading of the votes on a ballot. I sure hope the Bid Specs give not just the required level of accuracy but also the statistical confidence interval--otherwise it could be a meaningless spec!

But here is something quite definite for those with a good background in Statistics and especially those with Quality Control backgrounds. If I gave you 100,000 machines and told you to prove that they have 99.995% accuracy how would you go about testing them rigorously to do that? Would you test them all? Would you do a sampling test? How many ballots would you test? How many votes? You have less than a year!