types of voters

Like many people, I was shocked by the 2018 election. I did not expect such a ferocious anti-DPP wave, and I had no idea what to make of the Han Kuo-yu phenomenon. I was planning to conduct an internet survey for one of my research projects (on an unrelated matter), and I thought I might use it to learn something about the state of the electorate.

THIS IS NOT A REPRESENTATIVE SAMPLE OF TAIWAN’S ELECTORATE. In fact, my sample is very different from Taiwan’s electorate. It is useless to look for anything more specific than very big and crude trends. I will say things such as, “there is a very large group that …” and “there is a small but noticeable group that…” Don’t worry about exactly how big each group is; it isn’t that size in the overall electorate. The goal here is to look for groups of people who don’t follow the traditional party lines. If we can identify big, broad groups of voters who don’t follow the standard voting patterns, maybe we will get some insight into what happened last year – and what may happen next January.

At the end of an already lengthy questionnaire, I added eleven more questions. The first four were about the major political parties, and the other seven were about specific politicians. Each one was of the same format, “How much do you like XXX? On a scale of zero to ten, where zero means you dislike it very much and ten means you like it very much, how many points would you give XXX?” In a telephone or face-to-face survey, we typically allow respondents to refuse to answer or to say they don’t know. However, in internet surveys we are paying respondents, and we don’t let them more on to the next page until they give an answer. As a result, all 1000 of my respondents gave a valid answer to each of the eleven questions.

Before I show you any results, let me tell you a bit about how my sample is biased. Over three-fourths of my sample has at a university or higher education, and almost no one has a junior high or less. You should probably think of this as a non-representative sample of highly educated people rather than a non-representative sample of all Taiwanese. There are too many people aged 30-49, and not enough aged under 29 or over 60. There are too many public employees, white collar executives, and office clerks, and not enough blue collar laborers, farmers, student, or homemakers. Politically, the deviations from society’s mean are smaller. There are slightly too many mainlanders and people who identify as both Chinese and Taiwanese. However, on the question of Taiwan’s future status, there are not enough people who want to maintain the status quo indefinitely but too many who want to move toward eventual independence. About one-third of my sample identifies with a green camp party, one-third with a blue camp party, and one-third expresses no party identification or identifies with an unaligned party. Demographically, this sample is extremely different from the overall population; politically it is reasonably close. The survey was conducted in mid-April.

I asked how much respondents liked eleven parties and individuals. We finalized the questionnaire before Terry Gou announced his candidacy, so he is not included. Here are overall average scores for each party or person.

name name mean Stand. Dev.
KMT 國民黨 3.68 2.70
DPP 民進黨 3.56 2.59
NPP 時代力量 3.87 2.78
PFP 親民黨 2.84 2.11
Tsai Ing-wen 蔡英文 3.97 3.12
Lai Ching-teh 賴清德 3.99 2.80
Wu Den-yi 吳敦義 2.25 2.24
Chu Li-lun 朱立倫 3.61 2.57
Ko Wen-je 柯文哲 5.13 2.88
Han Kuo-yu 韓國瑜 4.46 3.53
Wang Jin-pyng 王金平 3.31 2.33

Educated Taiwanese are a pretty skeptical bunch. The only one of the eleven to break 5.00 is Ko Wen-je, and he barely manages it. Han Kuo-yu comes in second at 4.46, and the two DPP presidential aspirants are both a hair under 4.00. The leading candidates are all more liked than their parties. Anyway, these overall scores are not that useful with a non-representative sample.

I took these eleven variables and put them into a hierarchical cluster model. Cluster models calculate the distance between each case and group more similar cases together. When running the model, you specify how many groups you want. I looked at as few as four groups and as many as 25. From 11 to 25 clusters, there were five big groups and the rest of the clusters had only one to eight cases. It was pretty obvious that I was interested in those five big groups. I used the results from 11 clusters. The five big clusters held all but 17 of the cases, and I manually recoded those other six clusters into what looked to me like the best fit. This left five big groups.
















Cases 221 86 284 71 338
KMT 6.05 3.03 5.39 1.77 1.26
Wu 3.81 .94 3.71 .38 .74
Chu 5.37 2.17 5.33 1.76 1.77
Han 8.43 3.23 6.04 4.18 .92
Wang 2.97 .94 4.77 2.76 3.02
PFP 2.85 1.60 4.13 1.80 2.28
Ko 4.00 2.00 5.99 7.73 5.40
NPP .81 1.58 4.06 5.18 6.01
DPP .83 1.71 4.25 1.32 5.70
Tsai .48 1.31 4.16 2.54 7.06
Lai 1.04 1.53 4.77 2.51 6.21

My old statistics teacher used to say that the hardest part of running a cluster analysis is not doing any of the statistical work. The crucial step is naming the clusters so that you capture the essence of each group.

Cluster 1 and cluster 5 are pretty straightforward. Cluster 1 is Solid Blue. This group of respondents likes the KMT, Chu, and Han, and it dislikes the DPP, the NPP, and all of the DPP candidates. It doesn’t hate Ko, but it clearly prefers Han and Chu to him. Pay special attention to Han; this group absolutely adores (8.43) Han. This group does not like Wang, so I think of this group as having fairly orthodox KMT preferences.

Cluster 5 is Solid Green. This is the biggest single group. It clearly prefers all the green options to all the blue options. Predictably, among the blue options, it dislikes Wang the least. However, Wang isn’t likely to get any votes from this group. Among the two DPP candidates, there is a slight preference for Tsai. Notably, Ko is relatively well-liked in this group, and he will probably siphon a few votes away.

There are three large groups and two small groups. The third large group is cluster 3, which I have labeled the “battleground.” This group doesn’t really adore or despise anyone, and it generally likes the blue options a little more than the green options. This group likes Han, but it is much less passionate about him than the Solid Blue group. In fact, unlike cluster 1, this group likes Wang quite a lot; it is easily his best group.  Ko is basically tied with Han. I think of this as the amorphous middle in Taiwan politics that isn’t rooted to any particular party or ideology. In the current atmosphere they lean a bit more blue than green, but I suspect they leaned slightly to the green side in 2016. Everyone will pull some votes from this group. If this group does end up voting mostly blue, that will tilt the overall balance toward the blue camp.  Alternatively, this might end up being the biggest source of votes for Ko.

Clusters 2 and 4 are significantly smaller than the first three. Cluster 2 is disillusioned with politics. It doesn’t like anyone or anything. The only two options it doesn’t absolutely hate are the KMT and Han, and even they barely break 3.00. I suspect a lot of this group won’t bother to vote, and some who do will cast protest votes.  Of those who do cast useful votes, most will probably vote for the KMT candidate, assuming it isn’t Wang or Wu.

Cluster 4 is the anti-establishment party group. These voters dislike the two big parties. However, they are not totally alienated. They like the NPP, and they love Ko (7.73). However, you should not think of this group as green camp voters. If Ko doesn’t run, their next option is Han. Apparently, this group likes political outsiders.

With these five groups, you can start to see the outlines of what happened in 2018. We might imagine that in 2016 most of group 3 voted for Tsai (and DPP district legislative candidates). Group 4 was probably much smaller or much more similar to group 5 back in 2016. In 2018, however, group 4 probably did not turn out for the DPP. They may have stayed home, or they may have voted for third party candidates. In Kaohsiung, they probably voted for the outsider, Han. Even more devastating, the DPP lost group 3, the enormous battleground group. This group doesn’t strongly prefer the KMT to the DPP, but a vote is a vote. The good news for the DPP is that this group won’t automatically vote for the KMT in 2020. It might be able to do better in 2020, and Ko will siphon away large numbers of voters who would otherwise vote for the KMT.

The Han phenomenon is interesting. Han has figured out how to simultaneously appeal to three very different groups. The orthodox KMT people in group 1 absolutely love him, so he must speak KMT gospel fluently. The people in the battleground group 3 like him, so he must be able to speak to the broad non-ideological masses in that group. And group 4 is willing to consider him since he has figured out how to make anti-establishment appeals. The strange thing is that he can do these three things simultaneously. If he is the KMT nominee, the DPP strategy should be to paint him into the first box. That is, they should hammer home that he is just another orthodox KMT figure; he really isn’t the representative of the common people, much less the protest candidate. Unfortunately for them, Ko might be the primary beneficiary of such a strategy.

Ko is surprisingly strong across all categories, with the exception of the (small) group 2. In this sample, it looks not only like Ko is well-positioned to win first preferences, it also looks like he is ready to scoop up strategic voters if either the KMT or DPP attacks against each other succeed. However, keep in mind that this sample probably overestimates Ko’s support, since it doesn’t include low-educated voters. Without organizational muscle, Ko will have a hard time with that demographic. Still, you can see from this breakdown of educated voters why Ko thinks he has a good chance to win.

There isn’t much difference between Tsai and Lai in this analysis. They look pretty much the same in all five groups. I tried looking for the Lai primary voters who supposedly are fueling his challenge to Tsai. I looked for people who preferred Lai by at least three points over Tsai and also gave Lai at least a six. I found 48 such respondents. However, by the same standards, I found 84 people who preferred Tsai to Lai. I simply couldn’t find a large group of deep greens who supposedly are fed up with Tsai but love Lai. I’m sure they exist at the elite level, but they might be louder than they are numerous.

At any rate, both Tsai and Lai have a clear claim on group 5. Group 5 is big, but it probably isn’t big enough to win, even in a three-way race. The problem is that they don’t have any other good groups. They will win a few votes in group 3, but both the KMT and Ko are more popular there. Ko will eviscerate them in group 4. Other than mobilizing group 5, their best bet is try to squeeze a few more votes out of group 3. Lai might be better positioned to do this than Tsai, but either will find this a difficult task.

There seems to be a consensus in the punditry that the DPP is better off if Ko does not run. I am not so sure about this. Most of the votes that Ko wins in groups 3 and 4 would otherwise go to the KMT. It might be better for the DPP if Ko runs and siphons away those votes. Of course, the pundits seem to be assuming that if Ko doesn’t run, he will endorse the DPP candidate. I don’t know why they would make this assumption after the bitter 2018 campaign. Nevertheless, if he does endorse Tsai and campaign hard for her (it seems nearly impossible to me that he would enthusiastically endorse Lai), it is possible that she could win over a large chunk of group 3. Ko’s influence would be most critical for group 4, where he might be the key to swinging that significant voting block to her. However, I suspect that Ko would rather be the king than the kingmaker.





2 Responses to “types of voters”

  1. cassambito Says:

    It’s amazing how well defined these groups are, one can almost “see” these voter profiles materialize.

  2. Mike Says:

    What an absolutely fascinating breakdown of voter “profiles” – and, indeed, a very insightful and vivid classification of the “middle-ground” voters of Taiwan.

    Might you potentially be able to provide weighted proportions of those five groups (e.g. using age, sex, and region as you mentioned in the preceding post) to get a slightly more “representative” idea of how the highly educated voters of Taiwan break-down, and how big each voter profile group is?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: