IMAGINE two photographers standing on the same street corner at almost the same hour, aiming at the same crowd. One captures the faces in front; the other frames a different arrangement behind them. Both pictures may be genuine, but neither tells the whole story. That is what it can feel like when reputable polling firms publish Senate preference numbers within days of one another and arrive at sharply different results.
The May 2025 midterm elections give us an answer key. Bam Aquino finished second and Kiko Pangilinan fifth, while Publicus Asia’s May 1 survey ranked them second and third. By contrast, Pulse Asia, Social Weather Stations, OCTA Research, and WR Numero placed Aquino and Pangilinan much lower in their final preelection surveys. This is not enough to crown a permanent winner among pollsters. It is a reason to ask what each survey measured, what it missed and how its performance should be judged after votes are counted.
The same puzzle is visible in early polling for the 2028 Senate race. Pulse Asia’s Sept. 17–24, 2026 survey put Grace Poe at 50.2 percent. OCTA Research’s Sept. 28–Oct. 4 survey recorded her at 36 percent, down from 44.5 percent in the pollster’s July survey — an 8.5-point drop. Pulse Asia’s September result is not evidence of a rise from its own July baseline; Poe was not included in the previous Pulse Asia list. OCTA also cautioned that the candidate lists differed between its survey rounds. These figures raise a question worth investigating, but they are not a direct head-to-head experiment.
A headline percentage is the end of a chain of choices about whom to ask, what to ask, and how to adjust the answers. The 2025 results and the 2026 Poe gap both sit somewhere in that chain. The task is to examine the choices, not to decide by instinct which pollster must be right.
What the 2025 elections tell us
Aquino received 20,971,899 votes and Pangilinan, 15,343,229. Using approximately 57.35 million voters who turned out as the denominator, those totals equal about 36.6 and 26.8 votes per 100 voters, respectively. This is a useful comparison with a survey question asking respondents which Senate candidates they would choose, because voters could select as many as 12 candidates. It is not the candidates’ share of all senatorial votes counted, which uses a different denominator. Because voters could choose up to 12 names, these survey percentages and the final “votes per 100 voters” figures are related but not identical measures. The comparison is useful for seeing direction and magnitude; it is not a claim that any poll should have matched the final tally point for point.
Pulse Asia’s May 6–9 survey recorded Aquino at 32.8 percent and Pangilinan at 25.1 percent. Its report placed both within the statistical range of candidates who could still enter the winners’ circle. Using the same turnout denominator described above, their final totals work out to approximately 36.6 and 26.8 votes per 100 voters. On that basis, the survey’s figures were lower, particularly for Aquino. This is an illustrative comparison, not proof that a poll should precisely predict the final ranking; the survey measure and election benchmark are not perfectly interchangeable.
Publicus Asia’s May 1 Senatus 2 survey ranked Aquino second and Pangilinan third. It used a purposive sample of 3,000 respondents drawn from an online market-research panel. Publicus reported a national margin of error of plus or minus two percentage points. Because respondents were drawn from a nonprobability panel, that figure should not be interpreted as equivalent to the conventional sampling margin of error for a probability sample. Its ranking is useful evidence about this particular survey, not proof of overall superiority.
The measures should remain distinct: whether a poll identified eventual winners, how closely it ordered candidates, and how far its estimates differed from an appropriately defined election benchmark. A survey can perform well on one measure and less well on another.
Why surveys diverge
First, samples reach different people. Face-to-face probability samples and online panels can attract respondents with different habits and political interests, and weighting cannot automatically remove every difference between people who answer and those who do not.
Second, question wording changes what is measured: favorability, trust and voting preference are not interchangeable, nor is a single-choice question equivalent to one allowing respondents to select up to 12 candidates.
Third, candidate lists and presentation matter. Which names appear, their order and whether respondents see a prompted list can affect responses. OCTA’s Poe comparison, for example, needs the caveat that its lists differed between rounds.
Fourth, weighting corrects known imbalances only to the extent that the benchmarks and assumptions are sound. Fifth, a margin of error describes only specified sampling uncertainty, not every source of error or bias.
Finally, a poll is a snapshot, not a forecast. Pulse Asia’s post-election analysis found that 25 percent of respondents decided the day before voting and another 11 percent on election day itself — late decisions that can change a race after fieldwork ends.
The US lesson, and what we should demand
The American Association for Public Opinion Research’s review of 2024 examined 611 publicly released polls across presidential, Senate, gubernatorial and congressional contests in the final two weeks of campaigning. Average absolute error in the two-party margin was 3.3 percentage points in 2024, compared with 5.3 in 2020 and 5.2 in 2016. The review also found that polls underestimated Republican support relative to Democratic support in presidential contests for a third consecutive cycle. Its value is not that American methods can simply be imported here, but that systematic post-election evaluation is possible.
Philippine pollsters should routinely publish questionnaires, sampling and weighting details, fieldwork dates, candidate-list presentation and clear explanations of what their uncertainty estimates do — and do not — cover. Media organizations, universities and polling firms should establish a standing public post-election audit, using common measures but reporting them separately. The audit should assess every firm for which sufficient comparable data are available, rather than selecting only the most striking successes or failures.
Readers, meanwhile, should ask who was interviewed, how respondents were selected, how many candidates they could choose, who commissioned the poll and whether the technical details are public. A survey’s reputation is no substitute for those answers. Readers can still ask a short set of practical questions: Who was interviewed and how were they selected? How many candidates could they choose? Who commissioned the poll, and are the questionnaire, weighting, and fieldwork dates public? A firm’s reputation is not a substitute for those answers.
The goal is not to crown a permanent champion. It is to make every poll’s strengths and weaknesses visible. When two photographers show us different pictures of the same crowd, the responsible response is not to pick a favorite, but to ask how each frame was made — and what each one left outside it.
Bong Habana Ruado is a former public-sector official whose career has spanned government, politics, public policy and the private sector. His current interests include public opinion research, governance data, and explanatory journalism, with a particular focus on interpreting political surveys beyond their headline numbers. He holds degrees in political science and law (Juris Doctor) and completed the Strategic Business Economics Program at the University of Asia and the Pacific.