Philippine Elections: What (and What Not) to Automate

Halalang Marangal (HALAL) has released its Audit Report Number 4 on the May 2007 mid-term elections. It details the areas in which the COMELEC and NAMFREL fell short during the past exercise (refer to the PCIJ's blog entry for a summary) and derives lessons on how future elections can be improved. Of particular interest are its recommendations on what part of the process to automate, and more importantly, what not to automate. The current manual system can be divided into three stages:

1. Voting
2. Precinct-level counting
3. Canvass (which takes place at the Municipal, Provincial & National levels)

HALAL recommends that the precinct tally not be automated. The Audit Report states:

"The manual counting of votes at the precinct level is superior to automated counting in terms of openness, transparency, and providing an invaluable lesson in civics to all participants and witnesses. Because it is non-automated and open, it is slow enough that any citizen can actually audit in real-time the counting process. But because of its massively parallel approach of simultaneous tallies in all the two hundred thousand plus precincts in the country, is also fast enough that results are usually in within 6-12 hours."[emphasis mine]

HALAL makes a good point about the social value of the precinct level tally which will be lost if we implemented a system like the one from Mega-Pacific that was junked by the Supreme Court. Instead, it recommends automation to start at the Municipal level canvassing:

"HALAL recommends that ERs be encoded into computers as soon as they are brought to a municipal canvassing center. HALAL recommends, once the ERs are encoded, that original printouts be made available at cost to any interested party, during the municipal canvassing and afterwards, as certified true copies of these ERs...once ERs are encoded, that all subsequent consolidations, additions of row and column totals, computations of indicators like voter turnouts and ballot fill-up rates, and the printing of certified true copies of SOVs/COCs be done with computers."

We can better follow the logic of this recommendation if we look at a timeline of the recently concluded canvassing process:

Timeline of COMELEC Canvass by Region
(click on image to enlarge)

The above timeline shows that canvassing takes much longer than the average precinct tally (which was completed within 6 to 12 hours according to HALAL as quoted above), taking almost sixty days for the entire process to complete. It also reveals a weak relationship between the number of votes and the time taken to canvass those votes. For Region 4A, 36 million aggregate votes for Senators were canvassed in three days. By contrast, the much fewer 8.5 million votes of the Autonomous Region in Muslim Mindanao (ARMM) took forty six days for the same process. This means that for the recently concluded election, in terms of percentage of votes canvassed, we found ourselves in a situation where it took as long for the COMELEC to certify the remaining three percent of the Senatorial votes as it did the preceding ninety-seven percent. (Refer to diagram below showing the percentage of Senatorial votes canvassed over time.)

Percentage of Senatorial Votes Canvassed Over Time
(click on image to enlarge)

Entering the results at the initial stage of the canvassing process i.e. the municipal level, would help shorten the duration by taking advantage of the near simultaneous data entry that takes place among the country's hundreds of municipalities. More importantly, it would also make it harder to massage the data further down the line especially if as HALAL advises, it is immediately made available online:

"All ERs, SOVs and COCs should be put online or at least be made available to any interested party, together with their file checksums, at no additional cost, making the entire tabulation process completely transparent."

The attributes of transparency and auditability are emphasized because the last thing we need is an automated system that operates like a black-box. With these in mind, i believe that such an automated system must be built as Open Source software.

Automation that does not lend itself to inspection will just bring forth a younger breed of technology savvy Bedol-types. We can get a flavor of such a 'black box' type system in the results of the local absentee votes. As can be seen in the first diagram above, the canvassing of the local absentee ballots was completed relatively quickly and early (by May 20). Within these local absentee votes, Zubiri emerged as the topnotcher. Was the result fair? Who knows?

Update Aug-24-2007: Via Manolo, it's a hopeful sign that the media is giving some attention to Verzola's advocacy.