◎ Huang Bairui
There is less than a month left before the November 28th vote, and the election situation in Taipei City is still confusing. The numbers from multiple polling agencies have different directions, which has led to different interpretations by each camp. As a voter, how can I evaluate it from a relatively objective perspective, or even predict the possible results? I would like to share my views in this article.
As of the end of October, the surveys conducted by the four agencies in Taipei basically had the same trend, that is, "Chiang sinks and sinks."
TVBS's gap narrowed from Chiang's 28-point lead in May to 8 points in October; Zhen Media flipped from Chiang's 27.7-point lead to Shen's 3.8-point lead; and the Xinmin Association's lead shifted from Chiang's 1.2-point lead to Shen's 17-point lead.
The trends are basically the same, but the numbers are highly volatile. How do we know the actual situation?
Back to the basic concept, opinion polls are random sampling, and the survey agency then estimates the population through weighting based on the sampling results.
Some people may ask why weighting should be done instead of directly publishing the results of the telephone interview.
The reason is that the sample structure you drew may deviate from the parent structure. For example, we all know that Taipei may have only 20% young people, but after you made 1,068 calls (error margin ±3), you found that 50% of the young people answered.
At this time, weighting must be done to adjust the weight, reduce the proportion of young people's responses, and increase the proportion of other ethnic groups, so as not to accidentally amplify the voices of certain ethnic groups, causing the results to deviate from the matrix.
However, from the poll descriptions of each company, it can be found that each company weights the results by gender, age, region, and education level, but no one weights the results by political party.
"Party weighting" is very important, but we may not announce it because "the principle is very simple, but it is very difficult to implement", and everyone will have different views on the real political party map.
What is "party weighting"?
That is to say, we need to restore the proportion of responses from supporters of each camp in the poll to the true political party map.
For example, we all believe that the political party landscape in Taipei City is more blue than green. So the effective sample of a poll today has more green camp supporters. Does this mean there is a problem with the sampling?
Not necessarily, but maybe the blue ones think it is embarrassing to support Jiang Wanan, or the green ones have strong incentives to express their support for Shen Boyang, resulting in different answer rates.
At this time, if we want to interpret the poll numbers fairly, we must find a way to restore the "party structure."
But the problem is that the party structure does not look like there is a clear distribution of categories such as population and gender. The structure of a political party is a bit uncertain and may change at any time. For example, if a voter votes for Huang Shanshan in 2022 and Lai Qingde in 2024, which party is he considered a supporter of?
When it comes to Taipei City, everyone says that blue is greater than green, but how much greater is it? Maybe no one can say it accurately.
The Taipei City political party map given by TVBS is 58:42, which means that the sample drawn needs to be reduced to 58% blue and white camp and 42% green camp. TVBS based this number on using the 2022 Taipei mayoral election results and dividing Huang Shanshan's vote 6 points for Jiang Wanan and 4 points for Shen Boyang.
How do we know if this number is right or wrong?
The first way to deal with this is to simply "standardize" the numbers from each polling agency. It means not taking into account the "real political party map", but at least allowing all polls to have a consistent comparison standard.
For example, let us first assume that the blue camp is half and green (this is an assumption, not an estimate), and adjust the samples of each poll to be half blue camp and half green camp. Under this circumstance, the gap between TVBS and Xinmin’s October poll narrowed from 25 percentage points to about 7 percentage points (Xinmin became Shen Boyang +3, TVBS became Jiang Wanan +4.4).
After adjustment, the differences between the four polls, including Zhen Media and Lianhe Daily, all fall within the error range of ±5.4 percentage points, which is a statistical tie. The difference between TVBS and Xinmin is also within the error range of the comparison between the two polls (about ±7.6 percentage points).
That is to say, the poll results produced by four polling agencies in October were based on the same comparison basis. All results fell within the sampling error range. In other words, it was completely impossible to tell from the poll whether Shen or Chiang would win.
Therefore, the real question then is how to estimate the true political party map.
Generally speaking, there is a relatively common estimate of party territory, which is based on past election results. TVBS used the 2022 mayoral election results and made adjustments, which is intuitive and logical.
But if we consider several similar elections, the picture will be completely different.
The elections I used are "2018 North City Election", "2022 North City Election" and "2024 Presidential Election". Firstly, we will not rely on just one election to estimate the political party territory. Secondly, there are three groups of candidates in all three times, so they are more comparable.
At this time we will find that Taipei has a very stable pattern:
The Kuomintang received about 580,000 votes in the three elections of Ding Shouzhong in 2018, Jiang Wanan in 2022, and Hou Youyi in 2024, which almost did not change with the election type.
But the DPP is completely different: 435,000 for Chen Shizhong in 2022 and 588,000 for Lai Qingde in 2024.
Why is there such a big difference, and where do the votes for the DPP come from?
The answer is "turnout."
When turnout in Taipei rose from 67.7% to 74.3%, the KMT vote increased by only about 2%, while the DPP vote increased by about 35%.
To borrow a term from economics, the DPP's "voter rate elasticity" (the extent to which the party's votes change when the turnout rate changes) is relatively high.
In other words, most of the voters who are urged out will vote for the DPP, while the Kuomintang has reached its ceiling in urging votes.
Therefore, in terms of strategy, the DPP will hope that everyone will come out to vote, and the Kuomintang will hope that as long as my supporters come out to vote, it is best for ordinary people not to come out to vote.
So my own interpretation is that "Jiang Wanan has a ceiling, and it starts to decrease from the ceiling, while Shen Boyang does not have a strong foundation, but there is no ceiling." The key lies in how many "voters who might not otherwise vote" can be mobilized in the end.
So who are these voters? Most are young/middle voters.
Research by the Central Election Commission shows that the turnout rate of voters aged 20 to 35 in local elections is about 50%, and that of voters in the presidential election is over 70%, with a gap of 16 percentage points; for those aged 66 and above, there is almost no difference. The Kuomintang supporters are older and have stable voting habits, while the DPP relies heavily on bringing young presidential voters into local elections.
All in all, the higher the turnout, the better it is for the DPP.
So to answer "What is the real political party map?", I think the answer is that the political party map may be different under different turnout rates.
Next, how do we estimate the results based on this turnout rate? This time, I think the election popularity is close to the 2014 election. The turnout rate that year was 70% (the past two times were about 65-67%). The high estimate is 74% of the 2024 presidential election. Then use interpolation to estimate the basic market under different turnout rates.
At the same time, using the median voter as the X-axis, simulate the election results of the median voter's support rate for Shen Boyang under different percentages.
The two variables can intersect to give two possible scenarios for a significant win:
1. With a turnout of 70% (similar to the popularity of the 2014 Ke election), Shen won the support of more than 64% of the median voters.
2. With a turnout of 74% (similar to the popularity of a presidential election), Shen can win as long as he gets the support of more than 50% of the median voters.
Therefore, Shen may win, but the conditions are very difficult. The conditions to achieve it are to increase the turnout and win over originally non-blue and green voters to vote for Shen Boyang.
Of course, the above are all my inferences and analyzes based on existing poll numbers.
In terms of actual trends, non-blue-green votes in Taipei (voters who voted for Huang Shanshan in 2022 and Ko Wenzhe in 2024) have steadily accounted for a quarter, about 330,000 to 350,000 votes. However, only about a quarter of them actually identify with the People’s Party (based on party support), and the remaining approximately 250,000 votes are from median voters who do not identify with a party.
In the past month, this group of people has obviously shifted from being biased towards Jiang.
Shen has the advantage of slowly attracting support from non-blue-green voters who might otherwise have voted for Chiang Wanan.
Even so, Shen's challenges to being elected are many.
But isn't an election full of challenges and hopes very exciting?
