What Polls Didn't Reveal: Emotional Distribution of Chiang Wan-an and Shen Po-yang Shows Shocking Gap

On September 22, the Taipei-based Xinmin Cross-Strait Research Association released a poll showing Chiang Wan-an at 41.7% support and Shen Po-yang at 40.5%, a difference of 1.2 percentage points—within the margin of error. During the same period, social data analytics platform QSearch monitored 334,308 posts mentioning the two figures across four channels—Threads, Facebook, media outlets, and forums—from June 1 to September 20, covering exactly 16 full weeks.

Breaking down these 330,000+ posts reveals that the real difference isn’t about who has higher volume. Instead, three consistent structural gaps emerge: where the buzz originates, how others talk about them, and what issues they’re associated with—differences that remained stable over four months, barely shifting with events.

Overall, Chiang Wan-an generated 201,022 posts versus Shen Po-yang’s 133,286—a 1.5x difference. But this average masks significant monthly fluctuations: ratios were 1.27x in June, surged to 2.96x in July, then dropped to 1.25x in August and 1.24x in September.

July’s peak divergence occurred because their activity curves moved in opposite directions. Chiang’s visibility spiked due to the toxic oil scandal and the July 25 Kaodao rally, rising for four consecutive weeks from 9,592 to 17,982 weekly posts. Meanwhile, Shen focused on a U.S. visit, causing his mentions to decline for six straight weeks—from 7,003 to 3,627 weekly posts. This opposing momentum pushed their ratio above 2x—the only time during the period.

The largest single-week surges also corresponded to specific events: Chiang rose +82.8% in the week of August 31 (triggered by controversy over his son’s exchange student status and conflict-of-interest allegations), while Shen jumped +60.6% the week of September 14 (driven by crowds during night market walks in Shihpai and Gongguan, plus candidate registration). That week, Shen reached 20,825 posts—the highest single-week volume in the entire period.

Thus, focusing solely on the '1.5x total difference' overlooks the dramatic opposing trends within.

Another stable structural difference lies in the source of buzz. While both candidates generate the most mentions on Threads, the proportions differ significantly: 38.7% for Chiang versus 53.5% for Shen. Media coverage shows the reverse: 33.8% for Chiang and 23.0% for Shen. Crucially, this distribution remained nearly unchanged across all three-month segments—even as weekly volumes fluctuated by up to 80%.

In plain terms, roughly one-third of Chiang’s buzz stems from media reports and related discussions, whereas over half of Shen’s comes from organic user-generated content on Threads. Facebook mentions hover around 20% for both, and forums range between 4–6%, offering little differentiating power.

While data alone can’t explain causation, context suggests that as an incumbent mayor, Chiang’s official duties naturally generate news, while challengers like Shen must proactively create headlines to gain media attention. It’s important to note that higher Threads representation doesn’t equate to broader public opinion—it reflects engagement within that specific platform’s user base.

Emotional sentiment toward the candidates diverges even more sharply than polling numbers. Analyzing 'others’ discussion sentiment'—excluding the candidates’ own official accounts—we find that from mid-August to mid-September, negative sentiment toward Chiang was 65.8%, compared to 38.7% for Shen—a 27.1-point gap.

Trends also differ. Chiang’s negativity rose from 62.3% → 73.1% → 65.8%, peaking in the second segment before slightly declining—but remaining above 60% throughout. Shen’s declined steadily: 50.8% → 46.4% → 38.7%. Thus, the gap in negative sentiment widened from 11.5 to 27.1 points.

Two caveats are essential. First, sentiment and volume don’t equal support; this analysis does not predict election outcomes. Platform user demographics vary, and AI sentiment detection may misclassify sarcasm or irony. Second, the figures here are aggregated from over 100,000 posts per segment, where individual errors cancel out—making the 'gap' more reliable than absolute values. When analyzing specific events later, sentiment won’t be cited, as single topics often involve only hundreds of posts, and it’s hard to determine whom criticism targets when multiple parties are involved.

A key distinction also appears in agenda control. By categorizing Facebook discussions into themes, those exceeding 8% of discourse in any period are labeled 'main events.' Over four months, Chiang had seven such events, Shen had four.

Notably, the nature of these events differs. Five of Chiang’s seven main events were triggered externally: legislative questioning, the Neihu flash flood on June 25, Typhoon Bavi work suspension decisions, July’s food safety controversy, and the late-August family-related dispute. Only two stemmed from his own policy proposals (abolishing the Control Yuan, no-confidence motion). In contrast, all four of Shen’s main events arose from campaign activities or extensions: grassroots street walks, the U.S. visit, Shihpai-Gongguan-Jingmei district outreach, and the Dongfa Hao signature event.

Concentration levels also differ. From early June to early July, Chiang had three main events accounting for 32.2% of discussions, while Shen had none—his largest single theme reached only 7.5%, with over 70% scattered across unnameable long-tail topics. After mid-August, however, both converged into a typical late-campaign pattern: each dominated by a single major event, both reaching the largest volume levels of the entire period (Chiang: 2,741 posts; Shen: 2,678 posts).

It should be noted that 'main events' don’t represent all discussions. The top ten themes per period cover only 29.0% to 62.6% of Facebook posts, and the 8%-threshold main events cover even less.

Switching platforms reveals almost entirely different worlds. For the same period, the top ten themes cover 29.0%–62.6% of Facebook posts but only 6.0%–29.6% on Threads. Over 70% of Threads discussions fall outside named events, consisting of one-liners, personal anecdotes, and fragmented吐槽.

Yet cross-platform comparison shows high overlap in actual events covered. The difference lies in additional topics: Threads features more persona-driven and meme-style content—e.g., surname change controversy, humorous autograph stories on luxury items, deep-blue voter endorsements, father-daughter interactions. Facebook, meanwhile, emphasizes institutional and policy topics—abolishing the Control Yuan, amending food safety laws, supporting Chen Shih-chung’s past vaccine procurement case.

This explains a widely felt phenomenon: during the same period, your Threads feed and your friend’s Facebook feed may seem completely disconnected—yet they’re discussing the same events, just packaged differently.

Finally, we must clarify what this data cannot answer. Social media volume counts 'support' and 'criticism' equally—as mere mentions, the metric itself carries no directional valence.