Rectifying the “flaws” of latest Pulse Asia surveys will lead to good chances for Leni Robredo to win the presidential elections on May 9, according to a former secretary general of the then National Statistical Coordination Board (NSCB).

“I think she has a chance. Huwag lang madaya (assuming there’s no massive cheating),” Ramon A. Virola told CoverStory.ph in a phone interview on Thursday.

Results of Pulse Asia surveys in February, March and April showed Ferdinand Marcos Jr., the dictator’s son and namesake, ahead by a wide margin in the field of 10 presidential candidates, including Robredo. The April 16-21 surveys, the latest, showed no movement from the March results of a  56% voting preference for Marcos and 23% for Robredo.

Virola said his study and analyses of the surveys done by Pulse Asia in February and March revealed defects in the sampling of respondents. These included the underrepresentation of certain groups and the overrepresentation of others in the allocation of sample respondents, rendering the surveys, he said, “biased” against Robredo.  

FB posts

On May 2, the statistics expert posted his summary findings in his Facebook account, which later drew mainstream media attention for scientifically scrutinizing Pulse Asia’s survey results—the first time in 23 years since the opinion polling firm was set up and started to offer success metrics for candidates in both national and local elections.

Virola has a doctorate in statistics from the University of Michigan and is a former professorial lecturer in the graduate programs of the Statistical Center/School of Statistics and the Department of Mathematics (now the Institute of Mathematics) at the University of the Philippines.  

The NSCB, which he headed, served as the central statistical coordinating agency of the Philippine statistical system and is now part of the Philippine Statistics Authority (PSA).   

In the interview, Virola reaffirmed the “flaws” he found in the February and March Pulse Asia surveys, specifically the underrepresentation of the 18-41  age group, of those who reached college and of those  who belong to the ABC socioeconomic classes, and the overrepresentation of the 58-or-over age group, of those who did not reach college, and  of those belonging to the DE classes.

He said the “flaws” might not have been corrected in Pulse Asia’s latest survey. “It may be logistically not possible for Pulse Asia to make changes in their design” given time and money constraints.

Young supporters

He surmised that if there were relatively more Robredo supporters from the young, from those with higher educational attainment, and from those with higher socioeconomic classes, the “Pulse Asia Survey is biased against Leni!”

The bias also includes the non-inclusion of respondents from the A and B classes. 

In his analysis, Virola compared Pulse Asia’s allocations of sampling of respondents with other sources, such as the distribution of the population by 1SEC 2017, age-group data  by Philippine Statistics Authority (PSA)and the registered voters by age from the Commission on Elections. 

Pulse Asia Research Inc., in a statement, disputed Virola’s assertions, saying he had used a socioeconomic classification that was “different from the regular” one used by Pulse Asia “and most marketing firms.” 

Virola’s study showed a Robredo win, especially if the  undersampling of young voters is checked. “This alone will be sufficient to turn the tables around in favor of Leni” if this is corrected under certain assumptions, he said.

Correcting the flaws

Virola attempted to  “rectify” Pulse Asia’s March 16-21 surveys by adjusting and reweighing the results using updated databases. He recomputed the results assuming a certain sharing of the votes, which he acknowledged may be “quite arbitrary” but not without bases, considering Google Trends figures showing Robredo leading the race with 55% over Marcos’ 24%.  

Link: https://trends.google.com/trends/story/US_cu_nxB7CX8BAACbmM_en 

Citing studies, he said, Google Trends has a track record of predicting  the elections in many countries.  

When making the voting share assumptions, he also said he considered  the massive turnout in Robredo’s campaign rallies, mostly attended by young people and  his talks with other professionals.  

His findings:

• The nationwide count becomes 53.7% for Marcos and 29.3% for Robredo if the underrepresentation of those who belong to the ABC socio-economic classes is rectified by using the PSA-WB classification  and assuming a 60-40 vote for Leni among the A and B classes. (Details in table. Link: Virola tableSS v2.0 5 02 2022-01)

• The nationwide gap is reduced from 56%-24% to 48.8%-31.2%  if the  underrepresentation in the sample of those who reached college and if the PSA distribution  is used and assuming a 60%-40% vote in favor of Robredo among those who reached college. 

• Robredo takes over the lead by a close vote of 40.4% to 39.6% if adjustment  for the underrepresentation of the younger age group (18-41) is made  by using the Commission on Elections’ distribution of registered voters by age. (Virola said using registered voters as base matters a lot in election polls).

• The assumptions are 55%-45% vote in favor of Robredo among those aged 18-41 and among those aged 42-57 and the retention of the 56%-24% vote from the March survey in favor of Marcos among those aged 58 or over. (Details in table. 

Biggest source of bias

Thus, Virola said, the biggest source of possible bias of the Pulse Asia surveys in favor of Marcos is the underrepresentation of the young voters in the sample of respondents. 

He also said in the interview with CoverStory that the undersampling of those who reached college could matter.  He cited the wrong prediction of a Hillary Clinton victory in the 2016 US presidential elections due to, among other reasons, the failure was the oversampling of those  with college education.  

In the US elections, the American Association for Public Opinion Research found “the state-level polling underestimating the level of [Donald] Trump’s support, most importantly in Pennsylvania, Michigan, and Wisconsin”. 

The biggest reasons for the discrepancy were the big change in vote preference during the campaign’s final days, failure to properly adjust for an overrepresentation of college graduates, and  many Trump voters failing to reveal their preferences until after the election which could have also been the result of late-deciding Trump voters.

Virola said these could have parallelisms in the Philippine elections. The undersampling of those who have reached college, he said, could make a prediction for a Marcos win wrong.

Bigger lead for Robredo

Is  the assumption of a 55%-45% vote in favor of Robredo from those aged 18-57 reasonable?

He said the slight lead of Robredo would get bigger if  the assumptions are for a 60%-40% vote count for those aged 18-41 in favor of Leni, instead of 55%-45%, and/or

if  adjustment can be made for the underrepresentation by educational attainment and by socio economic class and for the Don’t Know/Refused votes, without double counting. 

Virola also said that polling failure had happened and  elections could go wrong for valid reasons.  

He renewed his call for Pulse Asia to “very seriously review not only its survey design but also its entire business process to assure the public of the integrity of its surveys in the future.”