DPP presidential primary results

[updated, see below]


The DPP announced the results of its polling primary for the 2020 presidential nomination today. President Tsai emerged victorious, defeating former premier Lai by 8.2%.


This is by no means a landslide victory, but neither is it a razor-thin margin. Tsai’s victory is clear, and that will make unifying the party for the general election a much easier task. That is, if she had only won by one percent, Lai’s supporters might have felt that Tsai stole the election by choosing a particular question wording or some other technical choice. With this margin, however, she almost certainly would have won no matter how the choice was presented.

Lai held a press conference after the results were announced, and he made a very gracious statement accepting the result and promising to support Tsai for re-election. As with my previous sentence, most media reports will focus on Lai’s acceptance of defeat. However, I thought the way that Lai delivered that statement was quite significant. He spent his first minutes talking about the general state of things, especially Hong Kong. He ended this by noting that the PRC had prepared soldiers to use there. He transitioned directly from the image of PRC troops to the primary result, saying that it was important to unify the party to re-elect Tsai. I think this was a clear message to independence fundamentalists: the stakes are extremely high, and they shouldn’t mull any reaction that would lead to the DPP losing the election. Lai is not going to lead or support any rebellion. Indirectly, he was probably telling the independence fundamentalists to shelve any ideas of putting forward their own presidential candidate.


This result is a huge victory for Tsai. After last November’s election disaster, I wondered if she would survive. The party seemed to rally around her, and by March it was expected that she would be unopposed for the nomination. However, Lai’s shock entrance into the race put her presidency into extreme and immediate danger. She was clearly behind in the polls, and there was a very real possibility that she would become a disgraced and repudiated lame duck with a full year left in her term. She has now avoided this nightmare scenario. Moreover, the competition with Lai has forced her to explain to the country why she deserves another term, something she had not bothered to do. It also focused the thinking of DPP supporters, forcing them to think about whether they really valued her record in office or not. She emerges from this process with much higher approval ratings and a much stronger sentiment of support within the green camp.

One thing that is under-appreciated is that Tsai managed this come-from-behind victory without going negative. The contest between Tsai and Lai was remarkably restrained. Their supporters sometimes grumbled that the other wasn’t respecting the rules or was engaging in personal attacks. However, it never spun out of control, and the two principals mostly stuck to the high road. Winning while eschewing negative attacks was only possible because Tsai’s allies changed the timetable. If the polls had actually been held three weeks after registration, the only way to change public opinion that much in that short of a time period would have been for Tsai to go negative. Positive messages need time to sink in, and it helps if the outside world reinforces those messages. Tsai’s discourse about building a foundation and starting only now to see some results needed two months plus a series of external events to shift perceptions. Negative attacks are quicker. All you have to do is present some evidence that demonstrates your opponent isn’t actually the type of person that voters thought. Of course, going negative also makes your opponents’ supports dislike you more, but that is something to worry about after you win. Tsai’s team figured out how to stay positive and not have to face that post-victory dilemma.


Let’s turn to the polling primary results.

There were lots of worries that Han supporters or Ko supporters might try to manipulate the DPP’s polling primary to get the weakest opponent possible. Taiwan was the first country to formally use telephone polls to determine nominations, and every time I discuss this system with other political scientists, they inevitably gravitate toward the question of manipulation by the other party. Indeed, the logic of microeconomics and institutions suggests that such behavior should be widespread. We can’t actually do rigorous research on this topic in any straightforward way. The parties would never share their polling primary data with academics, and, even if they did, those results might not contain enough information to provide definitive answers.

However, my gut tells me that manipulation by the other party is rarely widespread enough to be noticeable, much less decisive. The main reason for this is that it usually isn’t obvious who the weakest candidate is. It is useless to support an extremely weak candidate. For example, in the upcoming KMT polling primary, it won’t do much good for DPP identifiers to support Chou Hsi-wei or Chang Ya-chung since neither of them has a realistic shot of winning. You have to help someone who is already strong enough to be viable and who, almost by definition, is popular enough to win the general election. In the DPP’s case, Lai and Tsai have been fairly close in recent polls, though Lai’s consistent lead in earlier polls and his repeated insistence that he was the stronger candidate may have made an impression on some voters. Tsai has the advantages of incumbency, but she also has the baggage of her incumbency. I have seen arguments that Ko supporters might support Lai, thinking that Ko would be more likely to run if Tsai were in the race because he might be deterred by the more popular Lai. I suspect the opposite: Ko is more likely to run against Lai, since the independence fundamentalists were the ones who insisted on running a DPP candidate against him in the 2018 mayoral race. The point here is not that any of these ideas are right or wrong. The point is that there are lots of compelling ideas that different people might have about the state of the race, and they point in different directions. Moreover, what if you support a particularly odious person who appears to be weaker and that person eventually wins the presidency? How would you feel then? If the USA had had telephone polling primaries in 2016, a lot of Democrats might have supported Trump since he was widely seen as unelectable. That would not have turned out well. I think what usually happens is that people who support the other party just stay out of it. Either they don’t answer their phones, or they answer sincerely that they support the other party. “Don’t get me involved in your lousy party’s business. A pox on you all!”

The only case in which I think widespread manipulation by the other party might have been significant was the KMT’s nomination of Hung Hsiu-chu in the 2016 election. Hung was running unopposed, so she had to pass a popularity threshold rather than beat another person. She was widely seen as unelectable, and many people thought she wouldn’t pass. In that case, the KMT was expected to draft a more formidable candidate to contest the election. In this relatively unique case, most of the arrows pointed in the same direction for DPP supporters. I don’t have any concrete evidence that they helped her, but her poll results were shockingly strong.


The DPP asked respondents two main questions: who do you support among Tsai, Han, and Ko, and who do you support among Lai, Han, and Ko? The winner was determined by comparing the number of people who supported Tsai in the first question with the number of people who supported Lai in the second question. Five survey organizations were to each poll 3000 respondents. In fact, 16,051 interviews were completed, with just over half (8056) from landlines. The results from the five different organizations were similar; Tsai’s margin of victory ranged from 7.2% to 8.8%. I’m going to focus on the final average instead of worrying about those minor difference.

DPP Han Ko None
Tsai 35.7 24.5 22.7 17.1
Lai 27.5 23.5 27.4 21.7

The DPP did not report the final column; I created it by subtracting the other columns from 100%.

How did this result come about? It is important to remember that this is not a normal public opinion poll. This was a polling primary poll, and everyone knew it was happening. DPP supporters were much more eager to participate than other citizens. We have lots of anecdotal evidence of how they mobilized themselves, so I expect them to be highly overrepresented. We don’t have a breakdown of cell phones and landlines, but I would expect the DPP identifiers to be much more overrepresented among landlines. You have to stay at home to answer the landline, whereas you can carry your cell phone anywhere and continue life (mostly) as normal. In addition to the sample being skewed by partisanship, there will also be a lot of respondents who did not answer sincerely. If you want to influence the outcome, you cannot say that you would support both candidates in the general election. So we shouldn’t expect these results to look like normal polling results.

Fortunately, we have a pretty good idea of what a normal poll might look like. Apple Daily published a poll on June 11 that was conducted on June 8-9. (The DPP polling primary was held June 10-12.) This poll also used 50% cell phones and 50% landlines.

Apple poll DPP Han Ko None
Tsai 30.0 30.4 27.6 12.0
Lai 27.6 28.9 29.8 13.7

This poll is quite a bit different from the DPP polling primary result. Han and Ko are significantly higher, and the proportion of respondents with no opinion is considerably lower.

How hard would it be to get from this poll to the DPP results? I propose two steps. Step one is to skew the sample by inflating DPP responses and deflating everyone else. If you multiply all the Han, Ko, and none cells by 0.8 (which implies that the two DPP cells are increased by about 50%), you get the following table:

Skew sample DPP Han Ko None
Tsai 44.0 24.3 22.1 9.6
Lai 42.1 23.1 23.8 11.0

Step two is to adjust for strategic voting. Campaigns coach supporters to say that they “only support” a particular candidate. In the second question, if you answer that you only support Tsai, this will be coded as “none.” Not surprisingly, the DPP polling results have a high proportion of non-responses. If we shift 8% of Tsai’s support and 10% of Lai’s support to the “none” column, you can get pretty close to the level of non-responses in the primary result. A smaller number of respondents might have strategically (or sincerely) expressed support for a non-DPP candidate. Ko’s result against Lai is quite a bit higher than our expected value. We need to shift another 4% from Lai to Ko to account for this. These shifts produce the following table:

strategic DPP Han Ko None
Tsai 36.0 24.3 22.1 17.6
Lai 28.1 23.1 27.8 21.0

This result is very similar to the polling primary results. In this scenario, about 18% of Tsai’s potential support strategically decided not to vote for her, while about 33% of Lai’s potential support strategically decided not to vote for him. To put it another way, about half of DPP voters voted sincerely (supporting both) while about half voted strategically (for only one). This seems reasonable, given that various pre-election polls showed that far more DPP identifiers supported Tsai than Lai.

Please remember, these last two tables are not real data. I just made them up.

The point is not that this is exactly what happened. Instead, I want to suggest that we don’t need to make any complicated assumptions about KMT or Ko supporters to get to the actual results. These results are perfectly compatible with a world in which those people all vote sincerely. Of course, in the real world, I’m sure that some of them voted strategically for Lai or Tsai. However, the number of strategic KMT or Ko voters was small, or they cancelled out their effect by voting for different candidates, or both.

In this election, the system worked pretty much as intended. The general public had an opportunity to weigh in, but DPP supporters made up a disproportionate part of the overall sample and their preferences drove the final outcome.



Update (several hours later the same day): A friend pointed me to a breakdown of the results by age and sex, so now we have a bit more data to look at. WordPress doesn’t like wide tables, so let me break their table into several parts.

Category Sample size %
Full sample 16051 100.0
Age group    
  20-29 2600 16.2
  30-39 3023 18.8
  40-49 3066 19.1
  50-59 3007 18.7
  60-69 2430 15.1
  70&up 1763 11.0
  Male 7882 49.1
  Female 8169 50.9

The first thing to note is that they weighted the data, something I wasn’t sure about. The age breakdown is a very close match to the overall population. It would be nearly impossible to get this close without weighting. In a previous post, I presented a table from a TISR poll that showed that their unweighted sample using both cell phones and landlines was closer to the population than either exclusively cell phones or exclusively landlines, but it was still a bit off. For example, their combined sample only had 10.4% in the 20-29% age range. Weighting is standard procedure, so I don’t mean to imply that this is controversial. I am simply interested to see clear evidence that they did it (since they don’t publicize their methodology). I assume the results were weighted by age, sex, and city/county.

  Tsai Han Ko None
Male 38.2 24.4 23.7 13.7
Female 33.3 24.6 21.8 20.4
  Lai Han Ko None
Male 30.6 23.2 28.5 17.7
Female 24.5 23.7 6.3 25.4

Tsai beat Lai by 7.6% among men and 8.8% among men. I’m a little surprised that the gender gap isn’t larger. 1.2% is barely noticeable. The gender breakdown isn’t stunning, but the age breakdown is.

  Tsai Han Ko None
  20-29 38.9 10.9 36.9 13.3
  30-39 34.2 18.0 35.3 12.5
  40-49 31.0 26.2 27.2 15.6
  50-59 34.6 35.0 16.1 14.4
  60-69 41.9 32.6 8.7 16.9
  70&up 36.7 24.8 4.8 33.7
  Lai Han Ko None
  20-29 14.1 10.1 53.3 22.6
  30-39 18.9 16.9 45.5 18.7
  40-49 27.5 25.4 28.1 19.0
  50-59 34.7 33.8 15.3 16.1
  60-69 39.5 31.1 9.4 20.0
  70&up 34.3 24.0 4.3 37.4

Eyeballing those results, among people over 40, Tsai won by about 2%. Among people 39 and under, Tsai won by roughly 20%. Her comfortable margin came entirely from the youngest third of the electorate. It bears repeating that she did win the older voters. However, if today’s overall result had been a narrow 2% victory, the losing side would not have been nearly as gracious and she would face a much tougher task in uniting the party for the general election.

The staggering gap among younger voters makes me wonder if the fracas over marriage equality actually helped Tsai. It seems plausible to me that some younger voters who are disillusioned with both establishment parties took another look at her after that fight and concluded that maybe she wasn’t just another unprincipled, conservative, corrupt establishment politician. Admittedly, this is just speculation. What is clear is that Lai (along with Han) is definitely not popular among younger voters.

One Response to “DPP presidential primary results”

  1. cassambito Says:

    Thanks for explaining the results. Reading your line of thought helps me develop my own.

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 )

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: