Archive for October, 2019

Overview of the LY race

October 31, 2019

What should we expect to happen in the upcoming legislative elections? Will the DPP keep its (fairly sizeable) majority, or will it lose a lot of seats in January?

The worst way to think about the legislative elections is to think about small parties and the party list vote. To put it bluntly, there is way too much focus on small parties. They will not win many seats because (1) they don’t have that much support in society and (2) the system is heavily rigged against them. This year, the small parties’ support bases overlap significantly, as they all draw on the same group of alienated voters who think the two big establishment parties are thoroughly corrupt. If one does well, it will cannibalize votes from other small parties. My guess is that the best case scenario for all the small parties combined is winning about 20% of the party list votes or roughly eight seats. The worst case scenario is that none of them will pass the 5% threshold, and they will get zero party list seats. As in years past, the two big parties will continue to win the overwhelming majority of the 34 party list seats.

However, if you are thinking about the party list seats, you are focusing on the wrong thing. The election will be decided by the 73 single seat districts. Let me make a not-terribly-bold prediction: all of the 73 districts will be won by a candidate nominated by (or supported by) the DPP or the KMT. Of the 73 seats, about a quarter should be safe KMT seats, a quarter should be safe DPP seats, and nearly half are fairly close to the national balance of power. Barring a razor-edge election, these 30-35 seats will mostly go to the same side. As such, it is highly likely that either the KMT or DPP will win two-thirds of the 73 district seats and therefore an outright majority in the legislature.

[Aside: In a close election, the system is slightly biased toward the KMT. That is, if the two big parties win about the same number of district votes, the KMT will probably end up with a small majority in the legislature. This is mostly because indigenous voters, who are a solid KMT constituency, are significantly overrepresented. However, it is also because in a neutral environment the KMT would probably also win most of the smaller (read: overrepresented) districts, including Lienchiang, Kinmen, Penghu, Taitung, Hualien, and the six Nantou, Miaoli, and Hsinchu County districts. The only smaller districts that the DPP should be expected to do better in are the three Chiayi districts. However, if the DPP wins by a lot, as in 2016, they can win some of these smaller seats and mostly undo the systemic bias.]


In this post, I want to try to think about the partisan balance of each district. That is, if we didn’t know anything about the candidates running in the specific race, what would we expect the result to be in that district?

In past elections, district elections have been highly correlated with the presidential results. That is, in most precincts, the number of votes won by the blue or green presidential candidates was very similar to the number of votes won by the blue or green legislative candidates. Of course, there was some local variation, but you might be surprised by just how little. I once ran a simple correlation of vote shares for some paper was writing. I don’t recall offhand the exact results, but the correlation between the presidential vote shares and the district vote shares at the precinct level was about .95, give or take .02 (and I think it has been increasing over the past three elections). Weng Chung-chun beat Ma Ying-jeou’s support in 2008 and 2012 by over 10%, Wang Hui-mei beat the blue presidential vote in 2016 by about 15%, and just about everyone else was within 10% of their presidential vote.

The other critical bit of historical background is that Taiwanese elections are extremely stable. The DPP’s and KMT’s strongest areas are the same from year to year. Geographical changes in the distribution of votes are small and gradual. Tsai Ing-wen’s 2012 presidential vote was almost the same as Frank Hsieh’s 2008 presidential vote, except that she increase his vote share by 5% in nearly every precinct. There is evidence that the coalitions might change a little more this year than in years past, but, overall, past geographical patterns should be a fairly accurate guide to what to expect in January.

Given this history, the easiest way to think about the state of each legislative race is to use our relatively ample polling data about the presidential race and extrapolate those numbers to each district. Tsai Ing-wen seems to have opened up a fairly sizeable and steady lead in the presidential race. In my most recent aggregated poll results, I found her lead at 15.0%.

Let’s think about two districts, New Taipei 10 (Tucheng and Sanxia) and Taoyuan 6 (Bade and Daxi). In 2016, Tsai Ing-wen won 56.1% nationally. New Taipei 10 was almost exactly at the national average, with Tsai winning 56.5%. Subtracting the two, we rate New Taipei 10 as a +0.4% district, meaning the DPP should be 0.4% better than it is overall nationally. If the two parties were roughly tied this district would be a tossup, and the party that could win it would probably win a majority in the legislature. In fact, in 2016 I calculated that NT10 was the 57th seat – the median seat in the entire election. As it turned out, the two parties were not tied nationally. The DPP won far more votes nationally, and it won this seat (and the legislative majority) easily. Taoyuan 6 was a very different sort of district. Tsai won 48.6%, so this is a -7.5% district. Needless to say, the KMT needs to win all the -7.5% districts if it is going to win a majority. However, it turned out that the 7.5% cushion was not quite big enough. Tsai’s majority made up almost all of that deficit, and the remaining deficit was small enough for the DPP-allied candidate to (just barely) overcome. A 56.1%-43.8% presidential race implies that the cut point is at -5.6%, so a -7.5% district like Taoyuan 6 should be rated as having a very slight blue lean in that environment.

What if the DPP actually wins the vote count by 15.0%, as the polls currently suggest in the presidential election? A simple application of this model indicates that the DPP should be favored in 59 of the 73 seats. What if the race tightens and the DPP wins by a more modest 54-46%? In that case, the DPP should still be favored in 49 of the 73 seats. Here’s a table of various national results applied to the individual districts:

DPP vote% KMT vote% DPP favored KMT favored
45 55 22 51
46 54 24 49
47 53 25 48
48 52 26 47
49 51 31 42
50 50 37 36
51 49 40 33
52 48 43 30
53 47 48 25
54 46 49 24
55 45 53 20
56 44 56 17
57 43 57 16
58 42 60 13
59 41 62 11
60 40 63 10

In a tie vote, the DPP’s 37-36 edge in district seats might produce a legislature with no absolute majority. Remember, the KMT will probably win at least five of the six indigenous seats, so this might actually be a 38-41 split before the party list seats are distributed. The small parties would then have to win at least three or four party list seats to keep the KMT under 57 total seats. If the KMT wins the national (district) vote 51-49, it will be favored in 42 district seats. If you then add five indigenous seats, it is hard to imagine the KMT not getting at least 10 list seats and an outright majority. The math is not quite as kind to the DPP. Even if it wins the national vote by a 52-48 margin, its 43 district seats might not be enough to get to 57 total seats. The DPP probably needs to beat the KMT by at least 5% in the national vote to be assured of a single party majority in the legislature.


Obviously, this is a crude and simplistic model. It assumes that this time will be like last time. The advantage of this assumption is that it does not require any subjective judgments on my part. However, we can all be sure that this time will be different in some important ways and this model will inevitably be wrong. The difficulty lies in predicting how it will be wrong. We are now entering into the realm of subjective judgments, so the rest of the post is highly vulnerable to my own prejudices, wishes, fears, and biases. More realistic assumptions are not always better assumptions. [Gulp.] But here we go…


What if the legislative race is not like the presidential race? In the presidential race, it looks as if Tsai and Han will take over 90% of all votes, even if Lu and Soong enter the race. (Only Ko could change this structure, and it doesn’t seem the chances of the PFP nominating Ko are very high right now.) As far as I can tell, the polling seems to indicate that Tsai’s vote is a grand coalition of everyone who is not a fervent Han supporter. That is, there are a lot of people who will unhappily vote for Tsai’s re-election, but this does not necessarily imply that they will also vote for DPP legislative candidates.

I’m particularly thinking of the 10-15% of the electorate that are anti-establishment and disgruntled. I think most will end up voting for Tsai, but there is a good chance they will also look to the other two votes to express their unhappiness. In fact, I assume that they will take full advantage of the party list votes to vote for the TPP, NPP, PFP, Greens, SDP, Statebuilding Party, or some other small party competing for this block of voters. The real question is whether that will satisfy their anger, or if they will also vote against the DPP in the district elections. In the past, most split ticket voting has taken place within the two big camps, and relatively few voters have crossed the central dividing line. That is, relatively few voters will cast a party list vote for a blue party and a district vote for a green party. (Those who do tend to be people with a localist orientation and who often have a neutral or even positive attitude to both big parties, not those who are disaffected and hate both big parties.)  I don’t think many Ko fans who vote begrudgingly for Tsai will cast their district vote for the KMT nominee. However, they might vote for a third-party candidate with little chance of winning rather than the DPP nominee. If that is the general trend, the DPP share of the legislative district vote could be significantly lower than its presidential vote.

To some extent, the previous model already accounts for such a phenomenon. While I defined the relative partisan balance in each district by the 2016 presidential vote, I was careful to suggest that we should think of the spread in the 2020 district vote. For example, after taking into account the behavior of alienated Ko supporters, if the DPP wins the two-party vote by 54-46%, the DPP should be favored to win 49 district seats. However, it won’t be that easy. For one thing, alienated Ko supporters are not evenly distributed. (I think) they are more numerous in urban areas. For another, they will face different sets of candidates in the various districts. Some will have a TPP candidate, some will have a NPP candidate, some will have several minor candidates, some will only have the two major party candidates, etc. Depending on how this plays out, one +5.0 district might not end up looking anything like another +5.0% district. Right now, I don’t have a strong sense of how alienated voters will behave in the district races, and this is probably the biggest wild card.


A second consideration is that the geographical patterns have probably shifted a bit during Tsai’s presidency. Chen Shui-bian was a southerner, and there was a noticeable southern shift during his presidency. The DPP also extended its strength into more rural areas in the early 2000s. As an urbane northerner, Ma Ying-jeou reinforced this pattern during his presidency. However, Tsai, like Ma is an urbane northerner, and I noticed a small geographical shift in the polls this time. Relative to DPP support in previous years, Tsai seems stronger in the north and weaker in the south. I don’t have any evidence, but I’ll bet there is also an urban/rural divide. My guess is that Tsai has attracted some new urban support while alienating some previous rural support.

If the DPP’s support is shifting toward a more northern and urban coalition, what does this mean? Some districts that were previously solid KMT areas might be becoming swing seats. For example, Taipei 7 (Xinyi and part of Songshan) is rated as -6.5%, and – absent a DPP landslide – that should make it a blue-leaning seat. We are certainly accustomed to thinking of it as a safe KMT seat. However, if being a northern seat is worth an extra 2% and being an urban seat is also worth an extra 2%, maybe this year Taipei 7 will actually be a -2.5%. A -2.5% district is a clear target for the DPP to attack. However, the opposite process might be at work in other areas. Changhua 3 (Erlin) is a rural district in the southwestern part of the county. By the 2016 results, it is a +1.4% district. If the DPP can manage even a small edge in the national vote, a +1.4% should be relatively safe. However, if we subtract 1% for being in central Taiwan and 2% for being a very rural district, suddenly we are looking at a -1.6% district. A strong KMT candidate might be able to win that seat if the DPP can only manage a small edge in the national vote.

I went through the 73 districts and assigned them regional and urban adjustments. This was a quick and dirty process; I pulled these numbers out of thin air. I didn’t actually look at population density data or anything like that. My only limitation was that they all had to add up to zero. My quick and dirty Regional and Urban (RU) model shows a slight benefit to the DPP for these geographical shifts. At any given level for the national trend, the DPP tended to win one to three extra seats compared to the original model. In one case (DPP wins 54-46%), the RU model yielded five extra seats. If the DPP coalition really is becoming more northern and urban under Tsai, it looks like that might slightly help the DPP.


Third, the field of candidates in each district will also affect the race. I am not going to discuss the unique coalitions of individual candidates in this post, but some candidates are certainly better than others. As noted above, the distribution of third-party candidates will also matter. In this post, let me focus on a more systematic factor. Incumbents tend to do better than newcomers. In the 2016 elections, incumbency was worth about an extra 2%. For the first time, this factor will work in the DPP’s favor. It has 46 incumbents running for re-election, while the KMT only has 18 incumbents running for re-election (not counting a few incumbents who are trying to switch districts). There are only 9 seats with no incumbent running. Four years ago, the DPP’s national edge overwhelmed the KMT’s incumbency advantage. However, in a closer race, this could be a decisive factor in the DPP’s favor.


Where does this leave us? The biggest question concerns whether the DPP can transform soft Tsai Ing-wen supporters into DPP district voters. If they can do that, I think they are overwhelming favorites to win an outright majority – perhaps nearly as large as their current majority – in the next Legislative Yuan. Even if a large number of anti-establishment Tsai voters do not vote for DPP district candidates, I’m not sure how large that voting block will be. The DPP might still have enough votes to win most of the 30-35 swing seats. I hope to go through the races district by district in order to look at the unique factors in each race. For now, after looking at the overall picture, my gut tells me the DPP has a better than 50% chance to hold its legislative majority.


Appendix: The 73 districts

District Tsai 2016 District rating for DPP, relative to national average Rating with region and urban adjustment
Taipei 1 0.553 -0.8 3.2
Taipei 2 0.616 5.5 9.5
Taipei 3 0.521 -4.0 0.0
Taipei 4 0.513 -4.8 -0.8
Taipei 5 0.534 -2.7 1.3
Taipei 6 0.470 -9.1 -5.1
Taipei 7 0.496 -6.5 -2.5
Taipei 8 0.442 -11.9 -7.9
New Taipei 1 0.535 -2.6 0.4
New Taipei 2 0.629 6.8 9.8
New Taipei 3 0.614 5.3 9.3
New Taipei 4 0.606 4.5 8.5
New Taipei 5 0.585 2.4 5.4
New Taipei 6 0.581 2.0 6.0
New Taipei 7 0.576 1.5 5.5
New Taipei 8 0.490 -7.1 -3.1
New Taipei 9 0.430 -13.1 -9.1
New Taipei 10 0.565 0.4 3.4
New Taipei 11 0.439 -12.2 -8.2
New Taipei 12 0.531 -3.0 0.0
Taoyuan 1 0.545 -1.6 0.4
Taoyuan 2 0.544 -1.7 -1.7
Taoyuan 3 0.478 -8.3 -6.3
Taoyuan 4 0.532 -2.9 0.1
Taoyuan 5 0.472 -8.9 -7.9
Taoyuan 6 0.486 -7.5 -5.5
Taichung 1 0.576 1.5 0.5
Taichung 2 0.552 -0.9 -1.9
Taichung 3 0.565 0.4 0.4
Taichung 4 0.523 -3.8 -2.8
Taichung 5 0.514 -4.7 -3.7
Taichung 6 0.556 -0.5 0.5
Taichung 7 0.567 0.6 0.6
Taichung 8 0.563 0.2 -0.8
Tainan 1 0.703 14.2 10.2
Tainan 2 0.722 16.1 12.1
Tainan 3 0.677 11.6 10.6
Tainan 4 0.657 9.6 7.6
Tainan 5 0.645 8.4 8.4
Tainan 6 0.650 8.9 6.9
Kaohsiung 1 0.652 9.1 5.1
Kaohsiung 2 0.660 9.9 5.9
Kaohsiung 3 0.573 1.2 0.2
Kaohsiung 4 0.682 12.1 9.1
Kaohsiung 5 0.638 7.7 7.7
Kaohsiung 6 0.608 4.7 4.7
Kaohsiung 7 0.613 5.2 5.2
Kaohsiung 8 0.660 9.9 9.9
Hsinchu County 1 0.437 -12.4 -12.4
Hsinchu County 2 0.413 -14.8 -14.8
Miaoli 1 0.509 -5.2 -8.2
Miaoli 2 0.406 -15.5 -18.5
Changhua 1 0.570 0.9 -2.1
Changhua 2 0.545 -1.6 -3.6
Changhua 3 0.575 1.4 -1.6
Changhua 4 0.568 0.7 -2.3
Nantou 1 0.509 -5.2 -8.2
Nantou 2 0.534 -2.7 -5.7
Yunlin 1 0.621 6.0 2.0
Yunlin 2 0.646 8.5 4.5
Chiayi County 1 0.645 8.4 4.4
Chiayi County 2 0.662 10.1 6.1
Pingtung 1 0.619 5.8 1.8
Pingtung 2 0.651 9.0 5.0
Yilan 0.621 6.0 3.0
Hualien 0.369 -19.2 -22.2
Taitung 0.384 -17.7 -20.7
Penghu 0.508 -5.3 -8.3
Keelung 0.482 -7.9 -4.9
Hsinchu City 0.512 -4.9 -1.9
Chiayi City 0.599 3.8 2.8
Kinmen 0.180 -38.1 -41.1
Lienchiang 0.165 -39.6 -42.6




Are there hordes of latent Han supporters?

October 29, 2019

Pundits have been obsessed with the idea that all the undecided votes are actually latent Han supporters. There is no evidence for this, but they need some rationalization to argue that the race is closer than it appears since everyone has an interest in claiming the race is close and exciting. The recent Apple Daily, United Daily News, and TVBS polls all showed similar leads of 12-13% for Tsai. However, Apple had 32% undecided, UDN had 28% undecided, and TVBS had 9% undecided. The TVBS results suggest that those 28-32% undecided in the Apple and UDN polls are not overwhelmingly latent Han supporters. In fact, a better guess is that they have about the same proportion of latent Han and Tsai supporters (which — imagine this!! — is always the best assumption for voters who tell you they can’t make up their minds).


One reason TVBS gets 91% of respondents offering an opinion is that they employ a filter question. 13% of their full sample say that they do not plan to vote, so TVBS doesn’t ask those voters who they plan to vote for. This suggests that most of the undecideds are going to be non-voters rather than latent Han supporters. If we multiply the 52-39% results by .87 to put those voters back into the sample, we get a 45-34% Tsai lead. That leaves 21% of all voters undecided, which is still considerably lower than the Apple or UDN estimates. Since Tsai’s lead is still roughly similar, we still aren’t seeing much evidence of latent Han support.


There is another reason that pundits (especially those who prefer the KMT) are convinced the race is actually closer than the polls say. The results for the party list vote consistently paint the KMT in a better light. For example, here are the results from the recent TVBS poll on who respondents expect to vote for in the party list category:

KMT 37
DPP 25
TPP 12
New .8
Green .8
TSU .2
Other 1.5
undecided 12

The KMT has a 12 point lead in this race!! It must be extremely popular!! And the DPP is way down at 25% Clearly voters don’t actually like the DPP!

Let me reproduce that table, but this time I’m going to cut a few of the rows and add a few columns.


  Party List Vote Party ID Presidential Vote
KMT 37 30 39
DPP 25 22 52
TPP 12 13  
NPP 8 6  
PFP 2 ?  
Others/none 26 29 9

The TVBS numbers for the KMT are a lot less impressive in context. The TVBS sample includes 30% KMT identifiers and only 22% DPP identifiers. This distribution is not what most other polls are finding. For comparison, the recent Formosa poll includes 19.4% KMT identifiers and 27.1% DPP identifiers. We cannot know which one is closer to the actual number of KMT or DPP identifiers in the full population of voters, but the TVBS sample is clearly a better-case scenario for the KMT. It looks as if the 30% of people who identify with the KMT expect to vote for the KMT in both the presidential and party list races – and very few other voters will join them. The DPP also wins its identifiers and not many more in the party list vote, but it has an enormous coalition of all anti-Han (anti-KMT?) votes in the presidential election. If anything, this table makes me think that those 26% undecided in the party list section and the 29% undecided in the party ID category are likely to break disproportionately for the DPP. That is, the undecideds might actually have more latent DPP sympathizers than latent Han votes.


Ok, here’s another angle. TVBS asks a couple policy questions. One says, “Han Kuo-yu has criticized the [DPP’s revision of the] Labor Standards Law as harmful to both bosses and labor, and he wants to discuss further revisions to this law. Generally speaking, do you favor exploring further revisions to [the current] system?” 50% are in favor, and 27% are opposed.

The second question reads, “Han Kuo-yu has recently criticized the Executive Yuan’s decision to extend the high speed rail line to Pingtung as an ‘Appendix Line,’ and he promises to reconsider this extension if he is elected. Do you agree with scrapping the Executive Yuan’s decision to extend the high speed rail line to Pingtung and reconsidering the route? 42% agree and 28% disagree.

If you want to see public support for the KMT’s policy positions in these answers, you certainly can. Far more people agree with the KMT’s positions than with the DPP’s positions. However, there are a few reasons to be skeptical. First, these questions are designed to capture all dissatisfaction with the current policies. They only ask whether you would like to reconsider the policies, not whether you want to a specific alternative. A more neutral question would have stated the KMT’s position and asked whether respondents preferred the KMT position or the DPP position. Second, a pork-barrel project like the high speed rail extension isn’t designed to get mass approval. Of course people in Taipei are against it: they have to contribute taxes but don’t get very much benefit. However, people in Taipei are fairly unlikely to base their vote on the high speed rail extension. People in the south, in contrast, will be much more affected, and it might actually sway some of their votes. This is a classic case of diffuse costs and concentrated benefits. Third, why did TVBS choose these issues? Why didn’t they ask questions about whether respondents prefer the KMT or DPP versions of pension reform or whether respondents support the purchase of F-16 fighter jets from the USA? There are a lot of policy questions out there, and these results are only a small fraction of the relevant ones.


To sum up, if it makes you happy to insist that the race is actually a lot closer than the polls suggest, go ahead. However, you are either cherry picking the data or ignoring it completely.