Can a mayoral contender win after going from ahead to behind? We selected ten mayoral races in Taiwan, South Korea and New Zealand since 2018. Most saw large reversals in about four months. In each race we compare two surveys by the same pollster. In every case the incumbent, a successor from the same party or a candidate backed by the outgoing mayor went from leading the main opponent to trailing.
Taipei had the third-largest reversal of the ten. In Shan Shui’s June poll, Chiang Wan-an led Puma Shen by 27.7 percentage points. 119 days later, Chiang trailed by 3.8 points, a swing of 31.5 points. In the eight races with results, seven of the candidates who had led lost. The only one to hold on was in Seoul, South Korea, in 2026.
Can Chiang be the next exception? In SPI’s two October Taipei surveys combined, Chiang still led and nearly one in five respondents remained undecided. The questions now are whose supporters will turn out and whom the undecided will vote for.
Scroll the table horizontally
| Rank / country and city (year) | Pollster | Fieldwork end dates / interval | Margin | Swing | Outcome for incumbent side |
|---|---|---|---|---|---|
| 1 South Korea · Ulsan (2018) | Korea Gallup | Feb 3 → May 29 · 115 days | +15.6 → −20.7 | 36.3 | Lost |
| 2 South Korea · Daejeon (2022) | ChoWon C&I | Jan 17 → May 17 · 120 days | +23.5 → −10.2 | 33.7 | Lost |
| 3 Taiwan · Taipei (2026) | Shan Shui | Jun 22 → Oct 19 · 119 days | +27.7 → −3.8 | 31.5 | Vote on Nov 28 |
| 4 South Korea · Ulsan (2026) | MediaTomato | Jan 17 → Apr 26 · 99 days | +18.4 → −11.4 | 29.8 | Lost |
| 5 South Korea · Seoul (2021 by-election) | Realmeter | Dec 6 → Mar 30 · 114 days | +5.0 → −23.8 | 28.8 | Lost |
| 6 South Korea · Sejong (2026) | Hankook Research | Dec 24 → Apr 29 · 126 days | +1.0 → −27.0 | 28.0 | Lost |
| 7 South Korea · Seoul (2026) | KSOI | Oct 26 → Feb 10 · 107 days | +14.4 → −10.9 | 25.3 | Re-elected |
| 8 Taiwan · Kaohsiung (2018) | National Credibility | Aug 17 → Nov 9 · 84 days | +2.6 → −17.4 | 20.0 | Lost |
| 9 Taiwan · New Taipei (2026) | Beacon Market Research | Apr 10 → Oct 8 · 181 days | +1.0 → −13.1 | 14.1 | Vote on Nov 28 |
| 10 New Zealand · Auckland (2022) | Curia | Jul 10 → Sep 18 · 70 days | +12.0 → −2.0 | 14.0 | Lost |
A plus sign indicates a lead for the incumbent side and a minus sign a deficit, in percentage points. The one win and seven losses describe this set of cases and cannot be used to estimate Chiang’s chances. The appendix explains the comparison.
In the two Shan Shui polls, Chiang’s support dropped 16.2 percentage points, from 55.2% to 39.0%. Shen’s rose from 27.5% to 42.8%.
SPI followed the same Taipei respondents from early August to late October. About one in four changed their answer.
Two questions for Taipei: who turns out and where undecided votes go
Shan Shui’s two polls give snapshots of the race. SPI followed the same people to see whom they came to support or whether they became undecided.
In the late October survey alone, Chiang had 35.90%, Shen 30.77% and 19.23% were undecided. For the combined October results, we kept the more recent answer for anyone who took both surveys. After weighting to match Taipei’s age and sex distribution, Chiang had 37.40%, Shen 29.45% and 18.49% were undecided.
Three paths for Puma Shen to draw level
SPI’s combined October survey put Chiang at 37.40% and Shen at 29.45%, with 18.49% undecided. Each of the three scenarios focuses on one variable and calculates what it would take for Shen to draw level. The turnout rates and vote splits are assumptions for the scenarios.
Path one — supporters turn out. If undecided voters who turn out split evenly and declared supporters stick with their choice, turnout among Shen’s supporters would have to be about 27% higher in relative terms to tie Chiang. For example, if 70% of Chiang’s supporters vote, about 89% of Shen’s would need to vote, a gap of about 19 percentage points.
Path two — win more undecided votes. If Chiang supporters, Shen supporters and undecided voters all turn out at the same rate and all votes from the undecided go to one of the two candidates, Shen would need about 71.5% of those votes to draw level. If his supporters’ turnout is 5% higher in relative terms, for example 84% against 80% for Chiang supporters and the undecided, the threshold falls to about 68%.
Path three — win over Chiang supporters. With equal turnout and an even split of undecided votes, about 4 percentage points of overall support would have to move from Chiang to Shen to tie. Each vote that switches closes the gap by two votes.
Across the six cities. In SPI’s October 15–24 survey, 38.3% of women were undecided, compared with 24.4% of other respondents. These are pooled, unweighted figures.
In SPI’s October survey across the six cities, 39.8% of respondents said they had recently encountered campaign news about the cost of living and inflation, the topic they mentioned most often. A follow-up could ask the same people which municipal policies would change their choice.
A comparison with New Taipei
Chiang needs to win back lost support. Li needs to turn an even race into a majority. At the starting points in the table, Chiang led by 27.7 percentage points and Li Ssu-chuan by only 1.0 point.
Public polls currently favor Su Chiao-hui in New Taipei. In October, the Taiwan Trust in Democracy Association estimated a 4.1-point lead for Su in vote share. In Mirror News’s November survey, respondents who said they would definitely vote also favored Su, 55.5% to 34.5%.
We put more weight on the agreement between those public polls, especially among respondents who said in November that they were certain to vote. (Taiwan Trust in Democracy Association and Mirror News)
Among those in the November poll who said they would definitely vote, even if everyone who did not state a choice voted for Li, he would reach only 44.5% against Su’s 55.5%. To overtake her, Li would need turnout from his existing supporters and votes from people who have not expressed a choice or could still change their minds.
With about twenty days until the vote, candidates have to judge which group is most likely to decide the outcome. Will it be supporters who might stay home, people who have not said whom they favor, or voters who could still switch?
Methods and sources
SPI followed 41 Taipei respondents, comparing their August 1–12 answers with their latest responses through October 27. Eleven changed their answers. SPI’s minimum for publishing the direction of such changes is 30 people. The latest wave alone supplied 78 unweighted responses. Combining the two October waves and keeping each person’s latest answer gave 84 respondents and a weighted effective sample of about 82.
The combined Taipei results are weighted by age and sex using the city’s household registration figures for people aged 20 and over as of October 31. The age bands are 20–34 and 35 or older. No answer carries a weight above three.
The six-city undecided figures use 399 valid answers to the mayoral question. The campaign topics use 400 valid responses, with multiple selections allowed.
ShareParty Insight (SPI) conducted and funded the survey. Tatt Chen was the principal investigator. The Taipei panel comparison uses the August 1–12, 2026 survey and latest answers through October 27. The six-city figures use the October 15–24 survey.
SPI members volunteered to complete an online questionnaire. The nonprobability design does not provide a conventional margin of sampling error. Eligible respondents were members aged 20 or older with household registration in one of the six cities. The population benchmarks are the October 31, 2026 registered populations aged 20 and over, 2,041,978 in Taipei and 13,710,727 across all six cities. The combined Taipei results are weighted by age and sex. The pooled six-city figures are unweighted.
See the original releases for the cited polls’ field dates, sampling methods, sample sizes, funding and margins of error.
Appendix — how we compared ten mayoral races
SPI used AI to help locate and organize public polls. We ranked the cases by the change in the incumbent side’s lead or deficit against the main opponent. The incumbent side includes the sitting mayor, a same-party successor or, in Auckland, a candidate backed by the outgoing mayor.
Question wording and candidate lists changed in some cases. In Daejeon and the 2021 Seoul by-election, the first survey asked who was most suitable among several contenders before nominations. The second asked whom respondents would vote for. The difference in wording may have widened the change between the surveys. Both Taipei polls were by Shan Shui. The June survey listed only Chiang and Shen, while October’s listed six candidates, which may also affect the gap.
The Sejong surveys were 126 days apart, beyond the original limit of 120 days. Auckland’s 14-point swing was below the original 15-point minimum. Both New Taipei surveys were designed by Tai Li-an and conducted by Beacon Market Research. My Formosa’s April poll listed only Li Ssu-chuan and Su Chiao-hui. The Taiwan Trust in Democracy Association’s October survey also listed independent Su Hui-huang. The 181 days between the polls exceed the original 120-day limit, and the 14.1-point reversal falls below the original 15-point minimum.
TVBS offers another comparison. Before Li declared his candidacy, a TVBS survey in January asked how people would vote if the KMT and Taiwan People’s Party jointly backed Li against Su. Li led by 15 points, 47% to 32%. From June to October, the KMT caucus of the New Taipei City Council commissioned four TVBS polls. Over 119 days, Li’s lead stayed between 6.2 and 6.9 points. TVBS funded the January hypothetical matchup. The council’s KMT caucus commissioned the October voting-intention survey.