Report on the Conduct of the Random Manual Audit

Below is an excerpt from the full report, which is viewable from this link:  http://bit.ly/YR6XoY

On page 7:

Some Questions for the Comelec and Smartmatic

What could explain these error rates?  What possible underlying root causes might   explain these?  How should the candidates and political parties interpret these discrepancies?  How are the various RMA results consolidated? If a variance of 16.67 percent, similar to the one in this RMA, is noted in the 2013 polls, say in the senatorial position, what would the Comelec and Smartmatic do?

How are checks made on the canvassing process?  How will the public or candidates and parties know if the canvass results are accurate?  How should the audit of the canvassing results look like?

Discrepancy Between AES and RMA

Section 13 of the G.I. BEI states that “In the event of  discrepancy between the AES and RMA results still exceeds the allowable margin of ten (10) votes per  candidate per position (emphasis supplied), the RMA committee shall...”.

What does “allowable margin” mean? What is the basis for the “10 votes per candidate per position”?

The Random Manual Audit

Results

The highest error rate registered between the machine-counted and the RMA-appreciated votes is 16.67 percent. The lowest is -1.64 percent. The Commission on Elections (Comelec), through its machine supplier, Smartmatic, guarantees an error rate of 0.005 percent or less.

The photo is from a section of the Comelec’s RMA Minutes/Report (form No. 0916). The forth column (from the right) shows the variance between the machine-counted and the RMA team-appreciated results.

Positions Audited

The audit of the results of the precinct count optical scan (PCOS) machine was done for  40 senatorial posts, 15 district representatives posts, and 12 mayoral posts. Of the total 67 posts audited, 14 of these showed a variance.

Conduct of the RMA

At the close of the polling at 1900h on 02 February 2013, the board of election inspectors transmitted the results of the machine count to three servers, namely: the central server; to the canvassing board of Quezon city; and to the KBP server.  After which the ballot box was  resealed (with packing tape and signed by the board, to make tamper-evident). After the 27 copies of the election returns were printed, the board turned the ballot box over to the RMA team. This was around 2100h, and with the presence of observers and the media.

Immediately after opening the box, the team counted the ballots. They counted 159 in the box. But the election return showed that the machine counted 158 ballots. After a query by the observers and the media, the team revealed that there was one diagnostic ballot that was the first one fed into the machine before the voting started. This ballot was later identified as having a handwritten mark  “diagnostic”, and contained no votes on it.

One of the five-member board read the votes on the ballot, and the one seated beside him checked that the reading was correct. Two members recorded the votes on the audit returns. Another did the same using the tally board that was posted on the wall for the observers to track the progress of the counting.

Tallying of 67 posts from 158 ballots took five hours, with the RMA closed at 0200h the following day.

The audit was conducted after the mock polls held at the University of the Philippines Integrated School in Diliman, Quezon city.  The Philippine elections scheduled on 10 May 2013, will be the second such exercise where the PCOS of Smartmatic, Inc., will be used.

Photos (unannotated) of the RMA are available for download at this link: http://goo.gl/zLWtE.

(About the author: Telibert is a member of the national council of the National Citizens' Movement for Free Elections (namfrel.com.ph). The organization was formed in 1983 and is recognized as the first in the world to field volunteer domestic election observers. He served as Namfrel's executive director from 1996 to 2003, and became a member of the board in 2009. He has served in various international assignments as election observer, organized similar domestic election observer groups in Afghanistan (2004) and in Timor Leste (2006). He has also served as resident country director in Timor Leste and in Papua New Guinea (2012) for the National Democratic Institute (ndi.org).)

When do we see the Comelec in court?

The automation of the 2010 polls had many other benefits, which the Philippine electoral management body failed to appreciate. The efficiencies that could be had in using machines, instead of having humans do, to count ballots and consolidate results were many and strategic, and easy and cost-effective to implement. Yet they were not capitalized on. In migrating from the manual to the automated process, the metrics of performance have changed. The inability to tap the efficiencies inherent in the modernized process, that could have been used to improve the administration of the elections, is an inefficiency. For this, we are penalized.

Let me just use the current ballot recount process (discussed in first three parts of this four-part piece) to reveal some of the burden upon the various election actors that the Commission on Elections (Comelec) themselves have caused.

Because the actual accuracy of the precinct count optical scan (PCOS) machines significantly differed from that which the Comelec had committed, it egged election contestants to call for a recount. The random manual audit done by the Comelec showed that the machine was accurate to only 99.6 percent as against the 99.995 percent promised. Even the number of individual ballots counted, which is a basic test of accuracy, differed. The protester in the recount for mayor of Manila cited unreliability of the PCOS as reason why they had asked that the ballots be reviewed manually.

The impact is that there are close to 100 election protest cases questioning the results before the Comelec and before the House of Representatives Electoral Tribunal.  There are surely others with the trial courts.

Each protester provides a list of 20 percent of the polling stations in her/his jurisdiction, a sample number with which after the recount, to prove that a total recount would be merited. If there are over a thousand polling stations, 200 would be the maximum sample size. The protester bears all the recount cost, including honoraria for the recount committees. The protestee, too, as as result is penalized because s/he has to pay for his/her own representatives in the recount committees, as well as for her/his counsels.

There is no logic behind the 20 percent-of-polling stations recount sample because fewer precincts could have achieved the purpose using statistics. The Comelec could use a random sample of less than five percent of polling stations to establish cause for a jurisdiction-wide recount. Unfortunately, Comelec sees the recount as a fight between contestants even if they themselves had provided the raison d’etre for such fight. This is inefficiency and neglect of responsibility.

Because the automated system failed to make a report on the rate of over votes, no votes (abstentions), under votes and rejected ballots, election contestants and and others are deprived of crucial information. Without this information, the non-winner who has reason to believe that s/he might have won, is better off calling for a recount to play safe. If the Comelec had provided these pieces of information, the contestants would have been able to apply basic tests of statistics to determine if there is indeed a chance to recover votes and change the results. These pieces of information would have been easily captured by the PCOS if Comlec had really appreciated what automation was capable of doing. Who, therefore, pays for this inefficiency? Everyone does.

Because the automated election system used by the Comelec does not have a 100 percent accuracy, then, there should be an automatic recount when the difference between winning and losing is outside of the committed accuracy rate. In this case the Comelec should pay for the cost of the recount. By not having an automatic recount provision, the Comelec is neglectful of the fact that they are responsible to verify results that fall outside of the accuracy rate of the system that they commissioned.

It looks like there are a lot that the Comelec needs to be made accountable for. They do contribute to unnecessarily increasing the cost of getting elected.  So, when do we see them in court?

Earlier posts are: Recount Redux, Circular Reference Error and Rule Change, and Tangibles and Intangibles of Poll Recount.

Circular Reference Error and a Rule Change

Recently the new chairperson of the (Philippine) Commission on Elections (Comelec) said that a manual recounting procedure would be “‘in uniform’ with the system already put in place by other courts” and tribunals handling election protests. (Philippine Daily Inquirer) The Comelec has the details of this new procedure in resolution 9164, passed on March 16, 2011.

The officers of the Philippine Judicial Academy, in my meeting with them in mid-March 2010, said that they were then awaiting Comelec’s guidelines in order to train judges on handing disputes in the automated election system. It is interesting that the Comelec’s most recent guidelines on recounts is the one now being presented as consistent with those that were in place with the courts.

If this form of argument were in an electronic spreadsheet there would be a prompt saying "circular reference error".

While some argue that the presidential (PET), senatorial (SET), and the house of representatives (HRET) electoral tribunals are independent of the Comelec and may promulgate their own rules, it will be untenable to have a situation where one loses in the Comelec count but wins in the tribunals’ own by virtue of difference in the rules.

Therefore, did the Comelec avert a potential contentious situation with the latest revision of the recount rules? Looks like they did indeed. Instead of being the ones to set the rules, it looks though that the tribunals had instead prevailed upon them.

To my knowledge, the first ever recount of protested ballots in the 2010 polls will be conducted next week or 10 months after the winners have been proclaimed. Contestant Lito Atienza sought a recount of ballots of the May 10, 2010 election won by Alfredo Lim in the mayoralty race for the city of Manila. Recounting will start on March 29 for ballots coming from 200 polling stations. I look forward to observing this one and the National Citizens' Movement for Free Elections' (Namfrel) request for an observer status has been sent to the Comelec.

But before then the question becomes, what orientation or training did the Comelec give to the members of the recount committee and candidates’ representatives to ensure proper and uniform appreciation of the ballot and the votes?

Item (h) of Rule 15 of resolution 9164, distinguishes between “voting marks” and “identifying marks.” The former are “markings placed beside the ovals that may appear to show intent of the voter to vote for a party,” while the latter is supposed to be “intentionally placed to identify the ballot or the voter”. Voting marks may be subject to claims, while identifying marks could be subject to objections. Does the Comelec have clear guidelines on these for the recount participants?

The precinct count optical scan (PCOS) machines were not programmed to read voting marks outside the ovals unless that mark smudges the timing tracks or the bar code on the ballot. In both cases the machine will reject the ballot because it can neither locate the mark on the oval nor authenticate it.  Accidentally smudging or soiling any other part of the ballot, however, will not cause it to be rejected and the voter will not be disenfranchised.

In the same news article (above), the Comelec chairman said that a manual recount of the ballots “will be a tool to validate the results of the PCOS count last May, if they are correct or not“.

By allowing voting marks to be possibly counted as a vote, using the principle of honoring voter intent, when this could not have been determined with the use of the PCOS machine, the Comelec has changed the rules.  Therefore, expect different results between the count of May 10, 2010 and that of the recount. Whether issues concerning voting marks will be many enough to change the result would be anybody's take though.

(Part 2 of 4. Link to Part 1 is at: http://telibert.blogspot.com/2011/03/recount-redux.html.)

Recount Redux

That there is a strong indication to suspect that errors in the counting process are many enough -- and when corrected would change the results of the election -- is the reason why a losing party would seek a recount. Right?

In the May 2010 local and national elections in the Philippines, machines were used to “read” voter marks on the ballot and aggregate polling station results. Voters were presented with paper ballots that had names of all candidates pre-printed on them and they voted by shading the oval beside the name of each of the candidate they chose. They fed their ballots into the machine that read all the shaded ovals and tallied the votes. The results were known when the polls closed and the machine printed copies of the results.

Precincts (with 200 voters each) were clustered and one machine was allocated for every 1,000 voters. The precinct count optical scan (PCOS) machines were manufactured by Smartmatic and the software that “read” the shading on the ovals was reportedly provided by Dominion.

On March 22, 2010, nine weeks before the elections, the Commission on Elections (Comelec) passed resolution 8804 titled, Rules of Procedure on Disputes in an Automated Election System in Connection with the May 10, 2010 Elections. Item l of Section 6, Rule 15 states, “...marks or shades which are less than 50% of the oval shall not be considered as valid votes. Any issue as to whether a certain mark or shade is within the threshold shall be [sic] determined by feeding the ballot on the PCOS machine, and not my human determination”. Counting would be manual unless the recount committees are unable to determine if a shade meets the 50 percent mark or not, then ballot would be fed into the PCOS.

At the end of November, six months after the polls, the Comelec passed resolution 9104 titled, Rules of Procedure for the Recount of Ballots Subject of Election Protest Cases Filed with the Comelec. Amending Section 6, Rule 15 in the previous resolution, it entirely uses the PCOS in the recount. If there is a discrepancy between the original and recounted results, the original ballots images generated in 2010 from the compact flash (CF) card would be compared against the ones just generated at the recount.

However, more than 10 months after the election, on March 16, the Comelec yet again passed another resolution numbered 9164, Reinstating and Reimplementing Comelec Resolution 8804 with Amendments. This reverted the recounting to manual and the PCOS only to be used, if at all, to authenticate the ballots because Comelec could decide on other methods of checking the bar code and the ultra-violet ray code on the ballot. Determining whether an oval the size of a pencil eraser head is shaded 50 percent or more is a now entirely up to humans.

What could explain the corrections made by the Comelec and what might be the impact?

I can think of three and all point to the uncertainty of the accuracy of the machine. Complainants do not want the “suspect” PCOS to do the re-counting (first and second resolutions); change in Comelec chairpersonship; and failure to appreciate new context brought about by automating the vote appreciation process.

Comelec specified in the bid documents that machines should have a certified accuracy of 99.995 percent. This means that the machine may miss a mark for every 20,000 that it has read, or an error rate of 0.005 percent.

Assume 650 voters in a clustered precinct and 73% came out to vote. This yields 475 voters. Assume further an average ballot fill-up rate of 60% (there are between 26 and 32 positions for election on each ballot, but let’s assume 28; and voters only fill up 60% of that, meaning they opted not to vote for some positions). These are 17 marks on every ballot. Multiplied by the number of voters, the number of marks read by the machine in this case is 8,075. So there is a possibility of eight erroneous reads. Whether there will actually be eight or one error or whether there will be any errors at all are all in the realm of probability and statistics.

But the random manual audit commissioned by the Comelec recorded an accuracy of 99.6% or an error rate of 0.4%; or one in every 1,000 marks (Center for People Empowerment and Governance). The National Citizens’ Movement for Free Elections’ (Namfrel) also reported accuracy rates of just 99.56 percent and 99.53 percent in the random manual audits they had witnessed.

Another startling discovery is that there was a discrepancy between the number of ballots counted by the machine and the ones counted by the random manual audit committee. I wrote in the Namfrel report that in Precinct Cluster 8 composed of Precincts 32-A, 33-A, 34-A, 35-A and 36-A at the San Perfecto Elementary School (San Juan, metro Manila) the PCOS machine counted 616 ballots although there were only 614 in the box (Namfrel Report 2010). The committee checked and rechecked to ensure that they themselves counted correctly. How could the machine guarantee accuracy in the counting of votes when it could not guarantee accuracy in counting the number of ballots?

A snapshot of the polling station result (known as election return) from the random manual audit of Cluster 8 of San Perfecto Elementary School in San Juan, metro Manila.

The polling station result (known as the election return) prepared at the random manual audit to check the accuracy of the precinct count optical scan machines. This return is from Precinct Cluster 8 of the San Perfecto Elementary School in the city of San Juan in metro Manila. The machine recorded 616 voters but the ballots actually found inside the box was only 614.

Indeed there is reason to suspect the accuracy and reliability of the PCOS. The Comelec is caught between a rock -- reliability of judgment and consistency of human appreciation of degrees of shading of ovals; and a hard place -- the recorded inaccuracies of the PCOS. What to do?

(Part 1 of 4)

Lessons From The Mock Elections - Taguig

At the mock elections in Taguig today, voters had to wait an hour until they were able to feed their ballots once again into the counting machine. One voter who was a member of the board of election inspectors (BEI) in the last elections said that the message "Results Have Not Been Read" kept appearing on the machine screen every time a ballot was fed. The ballot does not drop into the ballot box and the machine ejects it.  Word circulated that there was an "ambiguous marking" on the ballot. No one gave any explanation as to how the problem was remedied but after an hour the machine started to accept ballots again.

Holding several mock elections is good because it draws significant media attention and as a result enhances public awareness on the new system. More than that, it is supposed to give the Commission on Elections (Comelec) and contractor Smartmatic a chance to preview the scenarios and learn of the system and procedural strengths and weaknesses in a close-to-election day setting. It also allows the Comelec to monitor Smartmatic's systems and procedures and its response turnaround time should any of the machines exhibit problems like in the case of Taguig.

But today's mock elections was more of a show than a learning exercise. If it were a learning exercise, there would have been informed or expert observers whose tasks would be to document the process, capture data and input these into an information database to improve actual election day processes. Polling officials, voters, and even spectators would have been systematically interviewed for their observations and inputs on the process and how it could be bettered.

Smartmatic had two technical support personnel at the Taguig mock polls in Marharlika elementary school. During the period when the machine had stopped accepting ballots, one of them was constantly on the phone reporting what had happened and getting instructions on how to solve the problem. However, at one point when the waiting seemed too long, one of the Comelec officials insisted that the intended spare precinct count optical scanner (PCOS) machine be used saying, "kawawa naman ang mga botante" because the voters had waited long enough.

After an hour's wait and to the delight of the voters, the machine finally accepted their ballots. But of the 50, three were rejected.

The eight copies of the election returns were then printed out and voters and watchers carefully examined the printed results. Then the results were electronically transmitted through the mobile phone provider, to the Comelec central server in Intramuros, the board of canvassers in Taguig, and to the common server for the political parties and the citizens' arm in Pius XII compound in Paco in Manila.

Sharing the learning

It would certainly be good for the Comelec to be upfront about the challenges and issues that they faced during the mock elections, especially the ones encountered in Taguig and in other places. An explanation on what had caused the problem, and what is an "ambiguous marking" and how it should be prevented or avoided are important information for the public.

Comelec should also inform the public of the instances that ballots were rejected.  A rate of 6 percent is very high. This or even smaller numbers could mean winning or losing as seen in previous elections both at the local and national levels.

They should also assess the quality of the response of Smartmatic and the time it took to resolve the issues. Comelec should also require the contractor to commit to a maximum response time and report issues and their responses in real-time and in an orderly fashion. If Comelec plans to put up control center, it would have to track and receive, among others, these and all information related to the deployment of the resources of Smartmatic in ensuring that the machines will work as specified.

Unless it is a deliberate attempt to sabotage the elections, it is not acceptable to blame voters for a machine malfunction because they had marked the ballot wrongly. In fact a wrong marking should never cause a glitch.  A member of the Comelec team interviewed after the mock polls drifted the blame to the voters and insisting that they should learn the proper way of marking their vote.

A voter who marks his/her ballot wrongly will be penalized because the votes will not be counted.   That's the severe price to pay because other systems allow voters to correct any voting mistakes.  There is no such provision in the system that Comelec had opted for.  Unfortunately, the poor voter will pay dearly for not educating her/himself on the new process of voting.

If the machine bogs down because of an "ambiguous mark," the systems designers and programmers will have to answer for it.

Public simulation needed

In a 1,000-voter precinct (one PCOS machines for 1,000 voters), the issue that appears to be of import is voter flow management. But more fundamentally, the size and layout of the voting center, which are the public school classrooms in most cases, pose as a  more challenging constraint. A procedure that brings voters to go back-and-forth to the BEI chairperson would necessarily create a bottleneck.

Process flow management experts, looking at the different cycle times within the voting process, will be able to compute how many voters under process would create a queue.  These are significant inputs to consider because increasing the voting hours is not necessarily the right response to ensure that 1,000 voters are able to vote within the prescribed period.  By carefully examining the internal processes, the comings and goings of voters in a polling station could be managed well.  And even when a queue appears, it would be more to bring order rather than a manifestation of an ill-designed system.

Last year the Comelec junior staff along with a process-flow expert conducted extensive experiments. Their study apparently had yielded valuable data that could inform polling station preparation, voter flow management and voting process design.  Many school principals and BEI members assigned in multi-storey voting centers in the urban areas where voters are in their thousands could very well benefit from that study.

Voter flow management was never apparent in a 200-voter precinct. Now it seems that this needs to be seriously looked into.

The blindside of automating the 2010 Philippine elections

What concerns me most is that voters might commit a high number of mistakes in filling up the ballot and a good number of votes would not be counted. If voter error is too high, it could negate the benefits of automating the count.

In the 2004 presidential elections, 36.6 million voters came out to vote or 84 percent of 43.5 million registered¹. However, there were only 32.2 million votes cast for all the five candidates for president, or 88 percent. There is no accounting by the Commission on Elections (Comelec) of the 4.3 million vote difference, or 12 percent (36.6 million less 32.3 million).

There are a number of explanations to this. Not all voters really vote for a president. Some votes could have been invalidated because a name of a non-candidate was written on the ballot. The handwritten name of the candidate was illegible or entire ballots could have been invalidated due to the presence of markings or what is referred to as a “marked ballot”. It is also possible that for some reason the votes cast for president were omitted and were not read at all.

But 12 percent or 4.3 million is too large a number and the Comelec neither offered any explanations nor any quantification to attribute this case to.

An unfamiliar ballot

Most of the 50 million voters in 2010 would have never seen a ballot of the type that will be used in May. For the first time in the national electoral history of the Philippines paper ballots would have the names of all the candidates pre-printed. In all previous elections, the ballots contained blank lines above which voters wrote the names of the candidates. In the machine-readable ballot of the new automated system, there will be an oval beside each name of a candidate which the voter shades or blackens in order to register a vote (see photo of the 2010 election demonstration ballot).

The paper ballot to be used in the automated election system (AES) would be between 13 and 17 inches long, and each side would contain the names of the candidates. According to the Comelec, each side could accommodate 300 names.

In the Philippines-styled synchronized elections, there will be 11 elective positions: president, vice-president, senators, district representatives to congress, party-list representatives to congress, governor, vice-governor, provincial board members, city or municipal mayor, vice-mayor, and city or municipal councilors. Voters can vote up to 12 senators, between two and seven board members by district within a province², and between four and 12 councilors depending on the classification of the municipality or city³.

The ratio between the elective slots and candidates has been between five and seven. This means that the ballot would contain from 250 to over 400 names, or even more. There were a number of elections when the number of organizations vying for party-list seats alone exceeded a hundred.

Degree of difficulty in accomplishing the ballot

There is an inherent degree of difficulty in accomplishing this type of ballot. If the ballot would be too crowded with candidate names and names of organizations, voters would have to be more careful and ensure that they are shading the right oval of the candidate of their choice.

The longer the ballot paper the more difficult it would be for the voter to follow the list of candidates and track how many ovals have s/he shaded. This is especially a concern for the positions where multiple candidates are allowed like for senators and members of the the provincial and municipal or city councils.

The more intuitive the ballot design is, like grouping and laying out the names of the candidates of the same elective post in a right-to-left fashion (as against a top-to-bottom manner), the lesser the chances of committing errors. Of course, the bigger the fonts, the better and friendlier it would be for the voter.

Familiarity of the ballot layout before the elections; the clarity of instruction on the shading or the types of marking on the ovals that are going to be accurately read by the machine; and assuring the that voter may take as much time as necessary to cast her/his vote could decrease the chances of committing mistakes.

Voters need to be also conscious and reminded of the maximum number of the provincial board members and municipal or city council members that they could vote for. This number is not uniform and it varies according to the locality. Therefore, giving the voter the right information before voting, the moment s/he receives the ballot, and reminding him/her to review the votes cast before feeding the ballot into the machine, could be critical.

There are a number of other things that need to be considered in designing the ballot. What is important is for several ballot designs to be professionally field tested considering a variety of relevant voter demographics, like education and age. The one that would present as easier for the voter to use and accomplish, backed up by field-test data, should be adapted.

Even with simple ballot designs and voting processes, voters still commit mistakes

Ballots in Indonesia, East Timor, Bangladesh, Afghanistan, South Africa and in many other places that I have observed elections in, are simpler. They contain the names, photos and the party or individual symbols of the candidates. The voter only needs to put any mark inside a blank box, which is also pre-printed on the ballot, beside the name of the candidate of his/her choice. In Indonesia and in East Timor, voters could punch a hole in that box to indicate a vote.

But even with this simple design and presumably a more voter-friendly ballot, there was still a six percent invalid vote rate in the 2007 presidential elections in East Timor. In the second round of voting in this same elections, the invalid vote rate dropped to less than three percent. By then voters were already familiar with the ballot and there were only two candidates to vote for during the run-off.

Errors situations where votes would not count

An “overvote” is when votes cast exceed the number of candidates allowed for a particular position. Two ovals shaded or marked for the lone position of president or mayor invalidates the vote for that position. However, all other votes for all other positions would be counted. Where an “undervote” condition exists, like voting only for seven instead of 12 senators, all the seven votes would be counted.

The absence of a shading or a mark in any of the elective positions is a non-vote and there are no penalties for it. Any marks placed placed outside any oval is not going to be recognized by the machine and would constitute a non-vote as far as the machine is concerned.

If parts of the bar code or similar code/s on the ballot, which the machine is programed to detect to indicate that the ballot is authentic and indeed intended for a particular machine is defaced or altered, the entire ballot and all the votes on it would be rejected.

Election results reporting

Imagine an election where the vote difference between the top two candidates for president is about one million. It is not actually hard to do because the margin between Mr Ramos and Ms Santiago in 1998 was 874,000⁴; and between Ms Arroyo and Mr Poe in 2004 was just 1.2 million votes⁵.

If the election for president in 2010 would be close, as others have anticipated them to be, an 11 percent differential between the turnout and the total votes cast for president, as shown in 2004, would be material. The Comelec would have to account, as it conducts a progressive tally, how many of the votes were not counted and under what specific conditions were they not counted. This type of report in computer programming parlance is referred to as an exception report.

There have been many instances of local races being won only by a handful of votes. Some had been won by less than a few hundred votes or even with far less. In the same concern expressed earlier, it is, therefore, important for the Comelec to do an exception report, not just for the presidential race, but also for all other positions.

(The author was the executive director of the Philippines' National Citizens' Movement for Free Elections, Namfrel, from 1996 to 2003 and was involved in efforts to modernize Philippine elections since 1994. He has organized or assisted in setting up Namfrel-like organizations in many other countries and has observed many elections worldwide.)

1 http://www.electionguide.org/results.php?ID=225

2http://www.comelec.gov.ph/statistics/2010natl_local/number_of_seats_for_panglalawigan.html

3 http://www.comelec.gov.ph/statistics/2010natl_local/number_of_seats_for_panlungsod.html

4 http://en.wikipedia.org/wiki/Philippine_general_election,_1992

5 http://en.wikipedia.org/wiki/Philippine_general_election,_2004