Chuang Po-Chung's Perspective: Understanding Polls – Has President Lai’s Approval Rating 'Surged' or Has Disapproval 'Skyrocketed'?

On August 5, 2026, the Mei-Li Island Electronic Newspaper released its July national policy poll. The same questionnaire, the same dataset, was translated into two different versions by two media outlets on the same day.民視 headlined 'Trust and Approval Surge, Shocking Trend Revealed' (later corrected to 'rise' after criticism); TVBS ran with 'Lai Qingde Polls Hit “Dead Cross”! Disapproval Soars Among This Group.' If the public sees both headlines simultaneously, they might suspect they are living in two parallel universes. President Lai himself might also be utterly confused!

Zooming in on the data itself, the absurdity becomes even more pronounced. The July poll shows Lai Qingde’s governance approval at 44.7%, up only 0.8% from June; disapproval at 47.4%, down slightly by 0.1%. Trust increased by 0.6%, distrust decreased by 0.9%, while the sampling error reached ±2.83%. In other words, all month-on-month changes fall within the margin of error—statistically, there is no significant meaning.民視’s 'surge' and TVBS’s 'skyrocket' describe the exact same 'no change' situation. Reviewing recent Mei-Li Island polls, the questionnaire design remains consistent, with monthly fluctuations mostly under 1%, all within sampling error. The same nearly stable data can generate completely opposite news narratives in the hands of media with different political leanings—an absurdity that speaks for itself.

Reading both reports side by side, the issue lies in the collapse of professional standards.民視, in an attempt to construct a 'surge' narrative, exaggerates a less-than-1% increase by calling it a 'shocking trend.' TVBS’s 'skyrocket' is equally absurd: the headline confidently claims 'disapproval soars among this group,' but the cross-analysis in the article remains identical to June—dissatisfied groups are still males, pan-blue, and Taiwan People’s Party supporters, with no group showing a statistically significant increase in dissatisfaction. One wonders if the headline and article were written by different people, or if the editor simply ignored the content and mechanically inserted exaggerated terms like 'surge' or 'skyrocket' based on political templates. Both outlets fail to do the most basic thing: inform readers that these changes fall within sampling error and lack statistical significance. When polling literacy is intentionally or unintentionally omitted, only emotional headlines remain to fight political battles.

This 'one poll, different interpretations' is not an isolated case but a shared historical problem among blue and green media. For example, a headline reads 'Disapproval spikes to 47.6%, showing a “dead cross”,' yet the article honestly states 'changes are within sampling error.' Others use question-based headlines like 'Impact of the toxic oil scandal? Approval shows “dead cross” again,' implying governance collapse, while the article only presents the fact that disapproval slightly decreased from 47.6% to 47.4%. If poll publishers have ulterior motives, and media outlets selectively amplify favorable or unfavorable fragments based on their stance, a collusive industry chain is born.

It must be acknowledged that structural limitations in the Mei-Li Island poll create ample room for loose interpretation. Take economic satisfaction: the 20–29 age group and college-educated respondents have rarely shown over half believing the economy is good. This is likely not economic recovery, but a wealth illusion from nationwide stock trading and an index breaking 40,000 points. Yet Mei-Li only asks the vague 'economic satisfaction,' condensing broad economic perceptions into a single, arbitrarily interpretable indicator. When polling agencies produce data with loosely designed questionnaires, and media consume it with stance-driven headlines, the public receives not information, but refined emotional fodder.

Therefore, polling news should return to basic principles: First, reports must state survey methods, sample size, and sampling error, giving audiences a statistical framework. Second, changes within error margins must not use trend verbs like 'rise' or 'decline,' let alone 'surge' or 'skyrocket'—fluctuations within error are no change. Third, headlines must reflect data in the body; headline claims must be supported by numbers in the text, and misinterpretations should be publicly corrected. Fourth, no pre-set stance frameworks; positive and negative data from the same poll should be presented together, not selectively extracted to favor a specific side. Fifth, cross-analyses must note the detailed error after sample segmentation; claims about 'a certain group' must prove statistical significance, not merely list static sample structures. Sixth, media should disclose their interests with poll publishers and related political parties.

Because the real problem has never been stance, but stance replacing facts, and emotion replacing evidence. If every media outlet were willing to spend just half an hour honestly analyzing data, perhaps this farce of 'one poll, different narratives' would no longer repeat monthly.