Partisan balance in Xinbei City

Last month, I spent two days interviewing people in Xinbei City while preparing to write a column for the United Daily News.  One of the themes I heard again and again from people on both sides is that the partisan balance basically favors the blue camp.  However, when you pressed people, they didn’t have a really clear idea of exactly how large the blue camp’s advantage is.  Perhaps the most common estimate is that it is now somewhat smaller than when Chou Hsi-wei 周錫瑋 won in 2005.


I wonder if we can do better than that.  (Spoiler alert: Not really.)


There are a couple of approaches to this kind of question.  One is to look at surveys, which give you an up to the minute snapshot of public opinion.  On the other hand, there are lots of problems with surveys.  For one, most of the sample sizes are too small to say anything about subsets of the data.  (So when you see a report about a survey claiming that one candidate is leading in areas A, B, and C while the other candidate is ahead in area D, feel free to laugh.  The sample size for each of these areas is almost always too small to get a useful estimate.)  The other approach, which I’m going to look at here, might be called the political map approach.  Basically, we look at old election results to get some idea of how this election might turn out.  This approach also has problems.  It assumes that this election will be like previous elections.  If there has been any realignment of political forces, old political maps might be completely irrelevant.  That said, let’s jump in.


There have been six county-wide single-seat elections since 1997, three county executive elections and three presidential elections.


KMT DPP blue green
2008 president 61.1 38.9 61.1 38.9
2005 executive 54.9 44.3 54.9 44.3
2004 president 53.1 46.9 53.1 46.9
2001 executive 48.2 51.3 48.2 51.3
2000 president 22.4 36.7 62.6 36.7
1997 executive 38.7 40.7 54.3 40.7
average 46.4 43.1 55.7 43.1
standard deviation 13.9 5.4 5.4 5.4


As you can see, the 2005 county executive race was a fairly typical result, with the blue camp winning by about 10%.  The only time the DPP cracked 50% was in Su Tseng-chang’s 蘇貞昌 re-election campaign in 2001.  The KMT’s best performance was in the 2008 presidential election, when it broke 60%.

Suppose we assume that the cumulative effect of the Tsai 蔡英文 and Chu 朱立倫 campaigns is that Tsai is one standard deviation above normal.  In other words, if you simulated the election an infinite number of times, Tsai (Chu) would be better (worse) than about 83% of the DPP (KMT) candidates.  To put it another way, we are assuming that Tsai is a really strong DPP candidate, though perhaps not quite a spectacular one.  In this case, the result would be 50.3 to 48.6 in favor of Chu.

If that doesn’t convince you that Tsai is fighting an uphill battle, let’s look at things from the township level.  The following are DPP votes:


    08P 04P 00P 05M 01M 97M ave s.d.
  overall 38.9 46.9 36.7 44.3 51.3 40.7 43.1 5.4
板橋市 Banqiao 42.0% 50.5% 40.2% 47.4% 54.5% 45.1% 46.6% 5.4%
三重市 Sanchong 47.0% 55.8% 44.4% 53.4% 60.5% 49.3% 51.7% 5.9%
新莊市 Xinzhuang 43.4% 52.7% 33.9% 50.1% 57.7% 47.1% 47.5% 8.2%
蘆洲市 Luzhou 47.2% 55.8% 39.2% 53.0% 60.8% 50.3% 51.0% 7.4%
中和市 Zhonghe 32.2% 38.7% 30.4% 36.7% 42.0% 34.1% 35.7% 4.3%
永和市 Yonghe 27.9% 34.2% 27.3% 31.8% 37.0% 30.6% 31.4% 3.7%
新店市 Xindian 27.7% 33.7% 41.6% 30.9% 35.8% 29.8% 33.2% 5.0%
樹林市 Shulin 42.8% 52.1% 39.7% 49.1% 56.9% 48.6% 48.2% 6.2%
鶯歌鎮 Yingge 42.2% 51.9% 51.9% 48.4% 59.8% 44.9% 49.8% 6.2%
三峽鎮 Sanxia 40.7% 49.8% 38.0% 45.5% 56.5% 42.7% 45.5% 6.7%
土城市 Tucheng 39.5% 47.6% 37.4% 44.6% 51.3% 44.7% 44.2% 5.1%
淡水鎮 Danshui 37.3% 46.1% 35.5% 43.2% 54.1% 32.5% 41.4% 8.0%
五股鄉 Wugu 44.2% 51.7% 39.8% 49.4% 57.2% 37.1% 46.6% 7.6%
泰山鄉 Taishan 40.5% 49.9% 37.5% 47.0% 54.6% 40.0% 44.9% 6.7%
林口鄉 Linkou 39.0% 48.1% 36.1% 45.5% 56.4% 42.3% 44.6% 7.2%
八里鄉 Bali 40.0% 48.1% 35.4% 46.3% 54.3% 38.7% 43.8% 7.0%
汐止市 Xizhi 35.4% 42.9% 34.3% 40.2% 44.1% 25.5% 37.1% 6.9%
瑞芳鎮 Ruifang 36.4% 45.1% 32.8% 41.9% 52.2% 40.4% 41.5% 6.8%
深坑鄉 Shenkeng 37.6% 44.3% 27.1% 43.0% 48.1% 37.9% 39.7% 7.3%
石碇鄉 Shiding 52.8% 58.9% 44.8% 56.1% 65.5% 47.3% 54.2% 7.6%
坪林鄉 Pinglin 58.6% 64.4% 50.8% 60.6% 77.4% 55.1% 61.1% 9.2%
平溪鄉 Pingxi 46.6% 53.1% 41.6% 49.0% 64.1% 41.0% 49.2% 8.6%
雙溪鄉 Shuangxi 42.6% 50.2% 43.9% 46.5% 56.6% 45.2% 47.5% 5.2%
貢寮鄉 Gongliao 55.0% 61.6% 31.8% 55.9% 64.6% 58.8% 54.6% 11.7%
烏來鄉 Wulai 31.0% 37.1% 51.9% 34.9% 52.4% 32.4% 39.9% 9.7%
三芝鄉 Sanzhi 34.5% 44.3% 24.5% 41.2% 61.4% 21.5% 37.9% 14.6%
石門鄉 Shimen 35.5% 43.3% 29.8% 40.8% 59.6% 31.1% 40.0% 10.9%
金山鄉 Jinshan 41.0% 50.2% 38.1% 44.1% 60.6% 42.2% 46.0% 8.2%
萬里鄉 Wanli 38.5% 45.3% 40.5% 40.8% 54.7% 36.8% 42.7% 6.5%


A good place to start is with the biggest city, Banqiao.  Historically, Banqiao has been a pretty good bellwether for Taipei County.  If you win in Banqiao, you’ll probably win the whole race.  In these recent years, it seems to favor the DPP slightly more than the rest of the county, so if Tsai is going to win, she probably needs to win Banqiao by at least 5%.  I don’t know why, but getting 53% in Banqiao seems more daunting to me than winning the whole county.

I sorted the townships in this order for a specific reason.  You can think of these as distinct clumps.  The weights are based on the number of eligible voters in the 2008 presidential election.  So, for example, Banqiao had 14.5% of the eligible voters in Taipei County in that election.

The averages and standard deviations are simple averages from the table above; they are not weighted for the size of the township.  This makes them a little wrong, but hopefully not by too much.


clump Includes weight Ave. S.D.
1 Banqiao 14.5 46.6 5.4
2 Sanchong, Xinzhuang, Luzhou 25.1 50.1 7.2
3 Zhonghe, Yonghe, Xindian 25.6 33.5 4.3
4 Shulin, Yingge, Sanxia, Tucheng 14.8 46.9 6.1
5 Danshui, Wugu, Linkou, Taishan, Bali 9.9 44.2 7.3
6 Xizhi 4.9 39.4 7.0
7 Ruifang, Shenkeng, Shiding, Pinglin, Pingxi, Shuangxi, Gongliao, Wulai 3.2 48.5 8.3
8 Sanzhi, Shimen, Jinshan, Wanli 2.1 41.7 10.0


Let’s start from group 3.  This group of townships south of Taipei City is the DPP’s worst area.  Moreover, it has the smallest standard deviation.  The DPP has never had the odd candidate who was able to win large numbers of votes in these three townships.  The game here is for the DPP to lose by as little as possible, except that they don’t seem to ever be able to increase their vote here by very much.  If Tsai can get anywhere near 40% in these townships, it will be a triumph for her.  35% probably won’t be enough.  One way to think about the election is that Chu will take a big lead in these areas, and Tsai will have to make up that deficit everywhere else.

She has to try to offset most of her losses from group 3 in group 2, which is the DPP’s best area.   Group 2, which has another quarter of the electorate, is just north of the Dahan River and west of the Danshui River.  These areas have a pretty large standard deviation, indicating that there are a lot of swing voters.  Tsai has to win close to 60% here; 55% probably won’t be enough.

This makes groups 1 and 4 the decisive battlegrounds.  Whatever Tsai can’t make up in group 2, she has to win in these two.  Groups 1 and 4 are west of Taipei City and to the south of the Dahan River.  Banqiao experienced fast population growth in the 1980s and 1990s and is now basically saturated.  The townships in group 4 are a little further out and are currently growing quickly.  While historically these lean a little blue, Tsai would have to win them by about 53% or so to have a chance at winning the whole election.

Areas 7 and 8 are perhaps the most interesting.  These townships, which only comprise about 5% of the total electorate, are the most rural parts of Taipei County.  Area 7 includes the mountainous areas to the southeast of Taipei City which used to have a coal mining industry and still grows lots of tea.  Area 8 includes the northern coast.  These townships have the highest standard deviations in the county.  Sometimes they swing heavily to one side, and sometimes they lurch violently the other way.

I have a notion that this is probably generalizable to the rest of Taiwan.  I am guessing that rural areas have larger swings.  I think this has something to do with more clearly defined party politics in more urban areas, perhaps because rural people have stronger party identification or perhaps because personal connections (which often work against party politics) are easier to build in rural areas.

Populations with strong party IDs might confuse this trend.  The most obvious groups is mainlanders, who tend to have stronger negative party ID toward the DPP than the population at large.  So groups 1, 2, and 3 are roughly at the same level of urbanization.  In terms of mainlanders, 3>1>2.  Not coincidentally, the standard deviations are 2<1<3.

At any rate, some of Su Tseng-chang’s in 2001 most impressive results were in small townships.  Of course, he had spent four years building those ties.  Even if Tsai were to match Su’s results everywhere else in Taipei County, she almost certainly will not be able to match his performance in these two groups.  In a very close race, this might be the difference.


So let’s sum up.  The DPP historically gets about 43% in Taipei County, and a swing of more than 5% is relatively difficult.  Moreover, we’re pretty sure that Tsai hasn’t built up the sorts of local ties that are helpful in winning over some parts of the electorate.  In short, you have to make some very optimistic assumptions in order to conclude that she is likely to win.

That doesn’t mean that she can’t win.  If you did this analysis in Taipei City, you would conclude that there is absolutely no way for Su to win.  Polls are currently telling a different story.  In 1997, I stared at these political maps and concluded there was no way the DPP could produce winning campaigns around Taiwan.  Then they did.  The 1997 numbers looked entirely different from everything that had come before them, but they materialized all the same.  The point is simply that history says it is more likely that the KMT will win this election.  Since the polls are not telling an unambiguously different story, it’s probably wise to err on the side of history.


Tags: ,

2 Responses to “Partisan balance in Xinbei City”

  1. James Chen Says:

    Thanks for all your excellent blog posts! Please keep them coming. I would say your writing is on par with Nate Silver’s FiveThirtyEight column in the New York Times — or, perhaps I should say Nate Silver’s blog is on par with yours!

  2. Jason Lee Says:

    Very good analysis. I appreciate the detail you put into the districts in all the elections. It is probably true that the 2005 Taipei County election is the most reasonable election to see what the average support rate is for both the camps because nothing out of the ordinary seemed to happen in that election.

    While your analysis is objective, things may be a little more complicated than you have indicated here because there have been reports of election fraud (although not proven) that Su Tseng-Chang cheated in the 2001 Taipei County election. Additionally, there have been reports of fraud in both the 2000 and 2004 presidential election. If you look at the invalid ballot rate in Taipei County in places like Xinzhuang and Luzhou, they are clearly irregular in that one voting station will have a reasonable invalid ballot rate of about 1-2% and the one right next to it will have a much higher invalid ballot rate. This strongly suggests that the staff in those voting stations artificially created invalid ballots from valid ballots.

    The explanation that there were new voting regulations enforced for the 2004 presidential election does not hold water because the higher rate should have been more uniform instead of fairly normal invalid ballot rates found in every election and then very very high invalid ballot rates in voting stations right next to “normal” ones.

    This phenomenon also occurred in Central, Southern, and even Jinmen island in the 2004 presidential election which makes it the most obviously “contaminated” election and this has nothing to do with the 319 incident.

    Therefore, when examining support rates, the possibility of election fraud needs to be taken into account because results can be distorted in one election and not others. 2008 presidential election was not rigged, but was a little too biased towards the KMT because the DPP was at an all time low in terms of support.

    Today’s 2010 election indicates that the DPP has a solid support base in Taipei County with over 47% of the vote. 55-45 is probably the most reasonable spread between blue and green support in the county. However, it can vary depending on the candidate. Tsai Ying-Wen did much better than I thought she would and was facing a legitimate candidate in Chu Li-Lun who has a strong record from Taoyuan County.

Leave a Reply

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

You are commenting using your 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: