Case Study
SPI Political Knowledge Lake | Campaign message-testing case
A candidate challenging within an incumbent mayor’s political territory faces a basic choice: inherit the incumbent’s achievements, promise improvements, or attack unresolved problems. This study suggests that Lee Shu-chuan (李四川) and Su Chiao-hui (蘇巧慧) face different directional signals when discussing Hou Yu-ih (侯友宜).
The short answer: Lee has a relatively stronger opening in “recognise and continue”; Su’s strongest option is “retain and improve.” But the greater limitation for both is not the spread between four frames. It is the large share who said that none would affect them or that they could not judge. The strategic signal is to move the main contest back to each candidate’s own issues—not to search for a perfect line about Hou. Between almost half (46.8% for Lee) and more than half (55.8% for Su) said no way of discussing Hou would influence them. For those voters, the next test should focus on each candidate’s own offer. In the same wave, everyday policy messages that clearly named the beneficiary, benefit and proof attracted the most attention. Discussing the incumbent is defence; building your own issue territory is offence. SPI’s companion analysis uses 1,003 complete Taiwan responses and 12 policy messages to show how teams can test the issue and frame: Same issue, different frame.
We asked 231 New Taipei respondents one single-choice question: if [candidate] discussed Hou Yu-ih during the campaign, which approach would be “most likely” to make you seriously consider supporting [candidate]? Respondents chose one of four approaches, “none of these would affect me,” or “don’t know / cannot judge.” The question was asked once for Lee Shu-chuan and once for Su Chiao-hui.
“Recognise and continue” drew 17.3%, 6.9 points above “identify problems and replace” at 10.4%, and 7.4 points above “do not raise him proactively.” (Option 1 exceeded option 3 by 6.9 points.)
“Retain and improve” drew 14.7%, 6.5 points above “do not raise him proactively” at 8.2%, and 5.2 points above “recognise and continue.” (Option 2 exceeded option 4 by 6.5 points.)
The combined “none would affect me” and “don’t know” shares were 46.8% for Lee and 55.8% for Su. These voters were not moved by how a candidate discussed Hou; the candidate’s own agenda is the remaining route.
The two approaches that treated Hou’s administration positively ranked first: “recognise and continue” at 17.3% and “retain and improve” at 15.6%, totalling 32.9%. “Identify problems and replace” drew only 10.4%. Acceptance of the two continuity-oriented approaches combined was about three times the criticism frame; that relative gap is the decision signal. A continuity frame warrants priority testing for Lee, while attacking the incumbent city government is not his stronger route. Equally important, 46.8% said no approach would affect them or could not judge. Discussing Hou may help, but it cannot substitute for Lee’s own proposition.
The priority test is therefore not a clean break with Hou. It is to inherit recognised municipal achievements, then add Lee’s own improvements.
| How Lee discusses Hou | Share |
|---|---|
| Recognise and continue | 17.3% |
| Retain and improve | 15.6% |
| Identify problems and replace | 10.4% |
| Do not raise him proactively | 10.0% |
| None would affect me | 26.4% |
| Don’t know / cannot judge | 20.3% |
The 95% bootstrap interval for every option appears in the technical note below.
Su’s highest directional option was “retain and improve.” It exceeded “identify problems and replace” by 3.0 percentage points, “recognise and continue” by 5.2 points, and “do not raise him proactively” by 6.5 points.
This signal is closer to “acknowledge effective policy, then state the improvement clearly” than either blanket endorsement of the incumbent or purely negative attack.
| How Su discusses Hou | Share |
|---|---|
| Retain and improve | 14.7% |
| Identify problems and replace | 11.7% |
| Recognise and continue | 9.5% |
| Do not raise him proactively | 8.2% |
| None would affect me | 33.3% |
| Don’t know / cannot judge | 22.5% |
The 95% bootstrap interval for every option appears in the technical note below.
Among 216 valid ratings on a 0–10 municipal-performance scale, Hou’s mean was 5.95, with a 95% bootstrap interval of 5.59–6.28. Scores of 8–10 accounted for 26.4%, while scores of 0–2 accounted for 9.1%—a 17.3-point difference. The most common single score was 5, at 18.2%.
This is directional context on the incumbent from the current New Taipei sample. It is not a mayoral-approval poll and cannot be converted directly into support for any candidate.
Lee Shu-chuan | The position fits better than the current product shelf.
The continuity position is available to him: respondents accepted “inherit and continue” more than “criticise the incumbent.” The potential problem lies in the offer. His infrastructure-oriented messages did not sit among the most attractive tested items, while one everyday policy that matched the leading-message formula was not used as the central proposition.
Suggested next tests:
Su Chiao-hui | Hou is a defensive question, not the main offensive territory.
Even Su’s best Hou frame had a small, uncertain lead, and more than half said no framing of Hou would move them. Her own everyday-policy messages occupied more of the leading group in the companion test.
Suggested next tests:
Shared interpretation | This is a contest between policy offers, not a referendum on trust in Hou.
The first side to rewrite its offer consistently as “beneficiary + benefit + proof” gains the better testable proposition for voters who remain unpersuaded. The next correct design is to test each side’s own message shelf and include the opponent’s strongest item as a defensive benchmark.
| Interpretation | Numerical basis |
|---|---|
| Lee: continuity over criticism | Option 1, recognise and continue, 17.3%—6.9 points above option 3 at 10.4%, and 7.4 points above option 4 (n=231) |
| Hou’s record: mildly positive directional context | Mean municipal-performance rating 5.95 (216 valid ratings); 26.4% rated 8–10 and 9.1% rated 0–2 |
| Lee’s infrastructure messages were not leading items | Across the nationwide n=1,003, 12-message test, rail extension, industry corridor and cross-party governance fell in “do not lead with yet” |
| Lee’s senior-card policy was an exception | The only leading-group message corresponding to a public Lee policy (correspondence follows the questionnaire’s disclosed source) |
| Su’s best Hou frame was only a defensive patch | Option 2, retain and improve, 14.7%—only 3.0 points above option 3 at 11.7%, with overlapping intervals |
| Do not overinvest in discussing Hou | “None would affect me + don’t know”: 55.8% for Su and 46.8% for Lee |
| Leading-message structure | The leading items—night childcare, young-child healthcare, denture subsidy and youth entrepreneurship—each specify beneficiary + benefit + proof |
95% bootstrap intervals for each option
Lee Shu-chuan: 1: 12.6%–22.1%; 2: 11.3%–20.3%; 3: 6.5%–14.3%; 4: 6.5%–13.9%; none would affect: 20.8%–32.0%; don’t know: 15.2%–26.0%.
Su Chiao-hui: 2: 10.4%–19.5%; 3: 7.4%–16.0%; 1: 6.1%–13.4%; 4: 4.8%–12.1%; none would affect: 27.3%–40.3%; don’t know: 17.3%–28.1%.
Research organisation: ShareParty Insight (SPI); lead: Tatt Chen
Field dates: 17–21 August 2026
Method: SPI Taiwan online research panel, mobile self-completion; by design this module was asked only of New Taipei respondents
Sample: 231 New Taipei respondents within 1,003 complete nationwide responses; the other 772 were not asked this module
Analysis: Unweighted; 2,000 respondent-level bootstrap resamples; 95% option intervals above
Precision reference: approximately ±6.4 percentage points; the interpretation is directional
ShareParty Insight (SPI) combines survey research, consented mobile-use behaviour and strategy analysis to help teams decide which issue to contest, how to frame it, who should carry it and when the strategy needs to change.
Explore SPI Taiwan Political Intelligence, or contact sales@sharepartyinsight.com.
AI assistance disclosure: Generative AI assisted the first draft. ShareParty Insight is responsible for numerical verification, editing and publication decisions.