Aggregated Presidential Polls

Comments for 12/31:

This is the final update to this chart since the blackout period starts tomorrow.

When I started this chart back in August, I had no idea it would end up looking like this.

Tsai 48.8%, Han 20.7%, Soong 7.4%.

___________________________________________________________________

Comments for 12/29:

We have a flurry of polls today. There may be a few more that come in before the deadline, but the picture probably won’t change much from here. The picture shows that the race has tightened in the last few days. This is due to two things: (1) Some (but not all) of the polls show a closer race, and (2) new polls from TVBS and ETtoday, which are consistently the most favorable pollsters to the blue side this year.

Even if the polls show a tightening race, they may not actually reflect an underlying tightening race. A story from Formosa argues that Han fans are drifting back into the polls. That is, when Han instructed them to boycott polls (and then support Tsai), a group of his supporters followed his instructions for a couple weeks. However, they gradually shifted back to answering polls sincerely. If this is the case, the current polls are a pretty accurate reflection of public opinion.

Tsai 48.5%, Han 20.9%, Soong 7.5%.

___________________________________________________________________

Comments for 12/24:

Not many new polls this last week. The MyFormosa and Apple Daily polls hint that the race might be tightening up just a tad, but the Green Party poll shows no such change.

I think most pollsters are waiting for the last week of the year to release their final polls before the blackout period starts. Remember, no new polls can be published in the ten days before the election, so Dec 31 is the last day we can see any new polls. Since you aren’t supposed to talk about polling results during the blackout period, most pollsters will probably publish their final results a few days before that deadline.

Tsai 49.4%, Han 20.2%, Soong 7.4%.

___________________________________________________________________

Comments for 12/16:

Three new polls, from TVBS, UDN, and Apple Daily, all show the race two or three points closer than their previous polls. There is also a poll from a “scholarly” foundation that shows Tsai winning by 38%. Remember, I weight the first three much more heavily than the latter poll, so it isn’t skewing the results very much.

Tsai 50.1%, Han 21.6%, Soong 7.5%.

___________________________________________________________________

Comments for 12/13:

Tsai’s weighted average has broken 50% for the first time. Even considering Han’s ploy to undermine polling, this is a monumental milestone, especially if one thinks back to March when Tsai seemingly had no hope of winning her party’s nomination, much less re-election.

Tsai 50.2%, Han 20.9%, Soong 7.4%.

___________________________________________________________________

Comments for 12/10:

A TVBS and a UDN (plus a Green Party) poll haven’t made Han’s numbers look any better. Yikes.

I updated so quickly mostly because I was writing a separate post on the state of the race. Go read that one.

Tsai 49.7%, Han 22.0%, Soong 7.6%.

___________________________________________________________________

Comments for 12/9:

Introducing the way-too-cluttered timeline! I’ve been resisting putting event markers on my aggregated polls for several months. However, now that I’ve finally given in to the temptation, I’ve gone way overboard and put far too many things on the chart. Believe it or not, this is the edited version; the first draft had even more events. However, let me know if you think I’ve missed any major events, either in Taiwan or in Hong Kong.

I was wary of putting Hong Kong events on the timeline, since it makes it look as if what is happening in Taiwan is a direct result of events in Hong Kong. Of course, Taiwan’s relationship with China (broadly conceived) would have been the focal point of this election no matter what happened in Hong Kong. However, the Hong Kong protests were, in fact, the way the China cleavage was introduced into the election this year, so it is reasonable to look at how events in Hong Kong correlated with polling in Taiwan.

As for the new polls, there have been three since my last update. Tsai leads by 30-35% in all three. To put it bluntly, Han’s polling numbers are in free fall. This might be because he has told his supporters not to participate (or to lie about their intentions if they do), but it could also be due to the recent spate of horrible news for him. We can’t tell for sure right now which is the more important cause, though I might try to sort this out in a separate post. All we can know for certain is that polls are terrible for him right now.

Tsai 49.6%, Han 21.9%, Soong 7.4%.

___________________________________________________________________

Comments for 12/5:

It still isn’t clear what the impact of Han’s call to ruin polling will be. We have had several new polls published this week, but (strangely) all but two  conducted at least part of their interviewing in the days before Han’s statement. Only an Apple Daily and a Green Party poll are entirely in the post-statement period. In both, Tsai led by more than 30%. In the Apple poll, Han went down about 4% while Tsai went up about 9%. That would imply that at least 5% of previously undecided voters turned to Tsai. In the Green Party poll, Han went down 10% while Tsai only went up 3%. This is not quite the same picture.

It could be that most Han supporters are continuing to answer surveys as normal. We political junkies tend to overestimate how much ordinary people pay attention to the latest developments in politics. It could be that even Han’s seemingly devoted followers aren’t really that sheepish. On the other hand, it could be that about 5-10% of the electorate is following his instructions, though some might be saying they have no opinion rather than supporting Tsai.

I hope you all enjoy my new-look chart. I’m debating whether adding major events makes it look overly cluttered.

Tsai 47.8%, Han 24.3%, Soong 7.5%.

___________________________________________________________________

Comments for 11/27:

I have a separate post discussing two recent developments, the dramatic increase in Tsai’s lead and Han’s attempt to destroy polling.

Tsai 46.1%, Han 25.7%, Soong 7.7%.

___________________________________________________________________

 

Comments for 11/26:

No one is publishing results for the two-way race, so I did not update the Tsai v Han chart.

The gap between Tsai and Han in my three-way chart has widened dramatically in the past few days. This might be due to the horrible (for the KMT) news over the past two weeks. However, it is also due to the fact that a couple of “academic” foundations that tend to have a fairly clear green bias published polls. The gap in the Taiwan Public Opinion Foundation’s poll was nearly 30 points. Nevertheless, it doesn’t matter who the pollster is; it isn’t good news for the KMT when two pollsters publish polls on the same day and their main talking point is that the size of the gap is unprecedented.

Tsai 44.0%, Han 27.3%, Soong 8.6%.

___________________________________________________________________

___________________________________________________________________

Comments for 11/22:

This week, there are five new polls for the three-way race. Four of the five (UDN, Apple, PinView, Green Party) show Tsai with a lead of between 16 and 19 points. The fifth poll, from TVBS, shows the lead at only 8 points. TVBS continues to do something very differently than everyone else. Maybe TVBS is right and everyone else is wrong, though that usually isn’t the best way to bet.

This might be the last time I produce the two-way chart. This week, only PinView published a result in that matchup. Now that the field is finalized, I’m afraid pollsters will stop asking the head-to-head question. Personally, I think that question is very informative, so I will be sad if I am forced to discontinue this chart due to lack of data.

Tsai 42.6%, Han 28.4%, Soong 8.9%.

Tsai 46.2%, Han 30.6%.

___________________________________________________________________

Comments for 11/17:

I’ve started a chart for the three-way Tsai v Han v Soong race. It looks like, after all these twists and turns, this is the race we will end up with. I’ll continue the Tsai v Han two-way chart as long as pollsters keep producing data since those results are probably quite indicative of how strategic voting will affect the final result.

It hasn’t been a heavy polling week, but Tsai’s numbers are trending up in the two-way race because some of her lackluster results from late October and early November are starting to fade out of the data. I expect we will get a flood of polls this week as everyone tries to determine whether the controversies over party lists, the VP choices, Han’s real estate dealings, the escalation of violence in Hong Kong, or anything else has impacted the race.

Tsai 42.3%, Han 29.1%, Soong 9.3%.

Tsai 46.3%, Han 31.3%.

___________________________________________________________________

Comments for 11/11:

Both candidate are a little higher this time, mostly due to a TVBS poll that had Tsai leading 49-40. Unlike everyone else, TVBS asks a filter question at the beginning. If respondents say they do not (or might not) plan to vote, TVBS filters them out. In their current poll, that filters out 19% of the sample. If you put those 19% back in, their undecided rates would be a lot higher — and probably in line with everyone else’s. Still, as a poll aggregator, I am respecting the numbers they present to the public.

Once again, the Han camp has promised that they will close the gap with Tsai in the near future. This time, they are supposed to catch up by mid-November. There isn’t much evidence of that happening yet. On the other hand, there also isn’t much evidence that reports about Han’s real estate dealings are driving his numbers further down. (To be fair, his numbers are so low that the DPP’s realistic goal right now should be to keep them down at their present levels rather than to drive them further down.)

Tsai 44.9%, Han 31.3%.

___________________________________________________________________

Comments for 11/5:

There isn’t much change in my polling average since the last post, but, as always, there is more than meets the eye. Last time I suggested that that gap between Tsai and Han might be closing, though the evidence wasn’t conclusive. This time, the evidence suggests that any effect, if there actually was one, may have been ephemeral. There are only three new polls, and the two more informative polls suggest that the gap between Tsai and Han is as big as ever.

Last week, the Apple Daily poll showed the gap shrinking from 17.7% to only 12.2%. This week, Apple puts the gap at 17.0%. Similarly, the gap in the Green Party weekly polls had shrunk from 19.7% to 16.3% to only 11.5%. This week their gap is back up to 18.6%.

The third poll was from the Taiwan Competitiveness Forum, and this poll placed the gap at a mere 8.7%. However, it is hard to know what this result means since they haven’t published a poll since mid-July, over a hundred days ago. For what it’s worth, that poll had Han winning by 7.0%.

It looks to me as if Tsai’s support took a small dip last week, perhaps due to the controversy over how to handle the Chen Tongjia case (involving the murder suspect in the case that eventually developed into the Hong Kong anti-extradition protests). It does not look like Han’s popularity went up significantly. Whatever the cause, the change seems to have been fleeting. However, it is a clear sign that Tsai’s support is relatively frail. Perhaps about 5% of voters became unsure about supporting Tsai in the face of this relatively innocuous news (or maybe it was some other similarly minor cause). They seem to be back in her camp now, but the DPP probably should be concerned about the shallow nature of their support.

Tsai 43.9%, Han 30.0%.

___________________________________________________________________

Comments for 11/2:

Several times I have produced a chart that looks like things are changing, but in the comments I have told you to ignore those changes because they were produced by some extraneous factor. This week, the chart looks different, but I am not going to explain why you should ignore those changes. In fact, this week the changes might be real. I’m not sure about that yet, but I am definitely sitting up and paying attention. Han might be narrowing the gap a bit.

There have been four polls published this week that suggest Han is doing better. There are reasons to doubt the importance of any single one. However, taken collectively, they suggest something might be going on.

Last week’s data included an Apple Daily poll showing the gap shrinking from 17.7% in their previous poll to only 12.2%. This is a big shift, but I tend to take sudden movements in Apple Daily polls with a grain of salt. Apple results tend to jump around more than anyone else’s, so I always mentally discount their implications until I get outside confirmation.

Last week, I stated that Tsai had led by double digits in 34 straight polls. Sure enough, the next day a poll was published with her leading by only single digits. ETtoday published a poll showing Tsai leading by 9.0%, 42.1%-33.1%. However, ETtoday has a weird methodology. According to their methodology statement, they conduct polls by “cell phone SMS” 手機簡訊. I have no idea what that means, how it works, how stable the results should be, what kinds of response rates they might get, or any biases that it might introduce. Are they exchanging multiple text messages (one for every question)??? In their late September survey, they got 4279 responses in five days; this time they only got 1110 responses in six days. I have no idea what to make of this survey result.

Storm media published a poll (conducted by Taiwan Survey Indicators Research) showing Tsai leading by only 11.5%. This is the first poll Storm has published this year so I don’t have prior results to compare it to, but this is at the low end of the range. Historically, TISR’s methodology has produced a slight green bias, so, if anything, this result probably indicates the gap is shrinking. (By the way, the former leader of TISR, Tai Li-an, is running the MyFormosa polls. Their most recent poll also showed a small decline in the gap, from 18.9% to 16.9%.)

Maybe the most compelling bit of evidence comes from the Green Party polls. I am automatically skeptical of polls published by political parties, but their results have been fairly consistent and seem somewhat credible. The Green Party polls have consistently shown a green camp bias, putting the gap at the larger end of the range. In their last three weekly polls, the gap has decreased from 19.7% to 16.3% to only 11.5%.

(There was also a PinView poll that didn’t show much change at all from previous waves. I don’t fully trust PinView’s internet survey methodology either.)

Again, I can tell you why not to believe any of these individual results. However, when this sort of evidence starts piling up, we probably need to pay attention.

Interestingly, we didn’t really see much movement in the most blue-biased polls (UDN, TVBS). The movement seems to be more evident in the green-biased polls (Green, MyFormosa). I wonder if there is some shift in a demographic group that is captured more effectively by a green-biased methodology than a blue-biased methodology.

Tsai 43.8%, Han 30.2%.

___________________________________________________________________

Comments for 10/28:

I’m updating my chart after only one day because a couple of polls were published last night that were conducted during the time period I had already produced a chart for. Is that a confusing sentence? Let me try again. Since the Wikipedia page that I use as my main source of data dates polls by the dates they are conducted (and not by the date that they hold a press conference for the media), some polls are released well after they are conducted. Yesterday (10/28), a Taiwan Public Opinion Foundation poll was published, but the interviews concluded a full six days earlier (10/22). So now I have to go back and recalculate a full week of results to account for this new data. MyFormosa also released a poll today, but that one only required redoing three days of results, which is a more normal time lag. Anyway, even though today is October 29, I’m only calculating results through October 28. I’ll probably have to revise this estimate for October 28 because someone else will publish a poll tomorrow that finished its interviews on October 26 or 27. The good news is that these adjustments don’t make that much of a difference. Including these two new green-leaning polls in the data changed the estimate for Tsai’s lead on Oct 27 from 14.5% to 15.0%. If you are worried about a 0.5% shift, you are assuming far more precision than these results merit.

Tsai 45.6%, Han 30.7%.

___________________________________________________________________

Comments for 10/27:

This week the Frozen Garlic weighted poll average shows major gains for Han Kuo-yu. In a single week, he cut the deficit from 17.0 points to 14.5 points. The comeback has begun!!!

Not so fast.

What has actually happened is that four pollsters who have consistently found the race to be closer than most other pollsters released polls. In its previous poll (9/21) UDN gave Tsai a 13% lead. This time, UDN found Tsai leading by 12%. TVBS’s previous poll (9/27) had a 12% gap; this time it was 13%. PinView’s previous two polls put Tsai’s lead at 13.7% (10/2) and 12.6% (10/15); this time it was 11.9%. Apple was the only one that showed a bigger change, but Apple’s results have been the most volatile (read: least confidence-inspiring) of any major pollster. Apple’s previous three polls showed the gap at 10.4% (9/29), 13.4% (10/6), 16.4% (10/14), and 18.2% (10/22). Today’s Apple poll had the gap shrinking to 12.2%. Only one pollster (Green Party) that has consistently given better results for the DPP published a poll this week. When the other pollsters weigh in with their next polls, the gap will probably increase a bit.

It is worth noting that even all of the blue-biased pollsters* have Tsai winning by double-digits. The last poll to find a single digit gap was a Sept 12 Pinview poll, which only gave Tsai a 6.9% lead. 34 polls have been published since then, and all have found the lead at above 10%.  (The last poll to have Han leading outright was an August 9 TVBS poll which gave him a 3% edge. Tsai has won all of the subsequent 62 polls.)

[*I use this label purely to indicate a numerical pattern rather than to imply they have any ideological preference or that they manipulate their methodology to obtain favorable outcomes.]

Tsai 45.0%, Han 30.5%.

___________________________________________________________________

Comments for 10/20:

The polls keep getting just a bit worse for Han every time I update this chart. Tsai seems to have hit a plateau over the past six weeks, but Han just keeps sliding. Yesterday’s Apple Daily poll had him at an unfathomable 22.3%. I’ll make further comments about the presidential race in a separate post. There are six new polls since my last update, so these new numbers are based on a bit more data than last time. I should probably note that it has been almost four weeks since TVBS — consistently Han’s best polls — released new data.

Tsai 45.5%, Han 28.5%.

___________________________________________________________________

Comments for 10/14:

It has been nearly two weeks since I updated this chart, and this isn’t only because I was traveling. The pollsters seem to have almost completely stopped publishing new polls. Only three new polls have been published since then, two by Apple Daily (one of which was released today) and one by the Green Party. I had almost given up on waiting for some new data. At any rate, the current aggregate poll results are based on less data right now than since mid-June, which is why today’s Apple poll (with higher undecided numbers than other polls) had such a big effect on the trend lines. Hopefully we will get a flood of new data soon and restore some robustness to these figures.

It is worth noting that each new Apple and Green Party poll is continuing to be a bit worse than the previous one for Han. His campaign has trumpeted a reorganization and has promised to turn the polls around by the end of the month, but so far there is no concrete evidence that Han’s polling numbers are improving.

Tsai 44.1%, Han 28.9%.

___________________________________________________________________

Comments for 10/2:

Another good week for Tsai. Even assuming KMT legislative district candidates will run significantly ahead of Han, this gap is getting to be so large that it will overwhelm everything. My instinct is to be skeptical and assume that the race is actually closer, but we have had a mountain of sustained polling evidence pointing to  large Tsai lead.

I am holding off on doing a separate chart for Tsai-Han-Lu because the recent round of polls haven’t polled that matchup. Lu has around 5%, and she draws fairly evenly from Tsai and Han.

Tsai 47.2%, Han 32.6%.

___________________________________________________________________

Comments for 9/27:

A couple of extremely favorable — but perhaps less trustworthy — polls for Tsai increase the gap between Tsai and Han a little. The media discourse this week was actually in the opposite direction. Because the gap in Green Party polls shrank from 17% to 14%, the talking heads were all trying to explain why the race was tightening. As usual, treat all results and explanations with a healthy dose of skepticism.

Tsai 46.4%, Han 32.1%.

___________________________________________________________________

Comments for 9/22:

There are no big changes in the Tsai – Han race in the past few weeks. This is not because there has been no new data. Rather, the polls have been remarkably steady. In the past two days, both UDN and Apple published results that are almost exactly at my average. Tsai’s long six-month ascension in the polls seems to have plateaued, while Han’s commensurate slide seems to have been arrested.

I may start a new chart to track the Tsai-Han-Lu race. However, as yet we only have three results, and all three are from lower-rated pollsters. As we saw with the last few days of the Tsai-Han-Ko chart, it’s not ideal to produce one of these aggregated polls if there are not many poll results to aggregate. I’ll comment more on the effect of adding her to the race in a separate post.

One technical note: Since the wikipedia page editors have decided to start listing the days when the poll was conducted rather than the day it was published, I will follow their lead and change my standard as well. So even though UDN published its poll today, it is now listed as being completed two days ago.

Tsai 45.4%, Han 33.2%.

___________________________________________________________________

___________________________________________________________________

Comments for 9/17:

Today was the deadline to register a petition drive to become an independent candidate, and neither Gou nor Ko registered. As a result, this is [probably] the last time I will produce a chart for any of the three-way races. There is a chance one of them will try to obtain the PFP nomination and register in November. However, even in that scenario, they won’t be included in any of the surveys over the next month or two. Without regular survey data, I can’t produce any charts. Starting next time, I will only update the Tsai-Han head-to-head race.

It wasn’t a terribly productive week in polling. There were only a few polls, and most of them were by lower rated pollsters. Now that we have passed this deadline, I expect to see a flurry of polls over the next few days.

Ko’s last chart looks like he is closing the gap on Tsai. In fact, this simply reflects the paucity of recent data on this race. Only one poll published in September provided information for this race, and that poll (from UDN) was something of an outlier for Ko. I’d ignore this chart after Sept 1 if I were me.

The two Gou charts show him trending in opposite directions. Again, this is probably more a matter of data availability than a reflection of the actual race. As with Ko, polling on Gou (by himself) has almost ceased in the past two weeks. Nearly every poll has only asked the question of Gou in alliance with Ko. So if you want to know what Ko’s support looked like near the end of his race, look at the chart of him allied with Ko, not the chart of him by himself.

Tsai 45.7%, Han 34.6%.

Tsai 32.3%, Han 27.9%, Ko 26.2%.

Tsai 33.6%, Han 25.5, Gou 25.6%.

Tsai 33.1%, Han 28.9, Gou (allied with Ko) 25.6%.

___________________________________________________________________

Comments for 9/9:

There are four new polls. I don’t put a heavy weight on the Green Party poll or the PinView polls released last week, though their results were pretty close to my polling average. Today, we got an Apple poll. This week’s Apple poll has Terry Gou falling from the previous two weeks. However, the most interesting new poll is from United Daily News. Years ago, UDN was THE authoritative voice in public polling. It doesn’t release a lot of polls these days, but I still pay close attention when they do out of respect for the good old days (read: my long-lost squandered youth). UDN results have historically skewed a bit toward the blue side, so the fact that they have Tsai beating Han by 11% in the head to head race is perhaps even more devastating to the Han camp as the 19% margin in the (green leaning) MyFormosa poll. UDN also published a result for the Tsai–Han-Ko race, which is the first data we have seen in that race for ten days. It was a pretty good result for Ko, which, combined with the lack of other recent data, sent Ko’s line shooting upward. Just remember, Ko’s spike is entirely due to this single poll, and we can’t even tell you whether that result is much different from UDN’s previous results since their last survey was conducted in late July, a markedly different context from today.

Tsai 45.7%, Han 33.6%.

Tsai 34.0%, Han 28.2%, Ko 24.6%.

Tsai 33.9%, Han 25.9, Gou 25.3%.

Tsai 33.2%, Han 27.3, Gou (allied with Ko) 26.6%.

___________________________________________________________________

Comments for 9/3:

Two new polls, one from Apple and one from TVBS. Last week’s Apple poll had a crazy result: Gou’s support went up 8% when Ko supported him. Gou always does a bit better when the question explicitly states that Ko supports him, but the difference is usually 2-3%. I don’t want to dismiss strange results, because “herding” (in which pollsters are afraid to publish results that deviate from everyone else’s) is a real danger. However, this difference didn’t seem to stem from a skewed sample (which can happen); it was just this single question that stood out as strange. Anyway, this week the weird effect is gone. Apple has Gou at 25.1% by himself and 28.1% with Ko’s support. You should probably mentally discard last week’s result (and the noticeable bump that Gou received in my aggregate trend line resulting from it).

These two polls narrowed the gap between Tsai and Han, but do not be deceived: both of these polls were good news for Tsai. Apple and TVBS are always skewed a bit blue, so it isn’t surprising that the gap was smaller than in a green-skewed poll such as MyFormosa. However, Tsai increased her lead over Han from the previous Apple and TVBS polls, from 6.8% to 12.7% in the Apple polls and from 5% to 8% in the TVBS polls.

Neither poll provided data on the Tsai-Han-Ko matchup.

Tsai 46.5%, Han 34.5%.

Tsai 36.3%, Han 28.8%, Ko 22.2%.

Tsai 34.2%, Han 26.7, Gou 24.9%.

Tsai 32.7%, Han 27.2, Gou (allied with Ko) 27.7%.

___________________________________________________________________

Comments for 8/31:

Three new polls, including one (from MyFormosa) with the heaviest weight. There is a separate post discussing the MyFormosa polls in detail. Here, let me note that this poll — which is very good news for Tsai — has a lot more influence on my ratings where I have fewer other polls (ie: Tsai-Han-Ko matchup) than where I have many other polls (ie: Tsai-Han and Tsai-Han-Gou(allied with Ko)).

Tsai 46.2%, Han 33.6%.

Tsai 35.8%, Han 28.8%, Ko 22.3%.

Tsai 34.6%, Han 27.4, Gou 24.7%.

Tsai 32.6%, Han 26.9, Gou (allied with Ko) 27.8%.

___________________________________________________________________

Comments for 8/28:

The polls this week are awful. I don’t mean they are bad news for a specific candidate; I mean they look like a bunch of low-quality and potentially misleading lousy crappy things that make people hate pollsters. There are five new polls, and all of them are problematic.

The CNews匯流 poll shows Tsai winning by a ridiculous amount. In the head to head race, she’s beating Han by a 2-to-1 ratio (53.6-26.2%). This is even more extreme than the results produced by deep green “academic” think tanks, which are already pretty far out of line with most polls. I don’t have much confidence that the CNews poll (conducted by Trend Survey, which is not exactly my most trusted company) is right and all the others are way overestimating Han’s support.

Apple published an absolute head-scratcher this week. Their head to head result — Tsai wins by 6.8% — isn’t that far out of line with everyone else. Apple has generally been a little bit less favorable to her than other outlets, so this result seems consistent. However, they report MUCH higher support for Gou than everyone else. In the three-way race where Ko supports Gou, Gou wins in a landslide (Gou 35.0%, Han 24.2%, Tsai 24.0%). No one else has Gou even leading, much less crushing both Tsai and Han by 11%. [This poll caused Gou to edge ahead of Han in my chart. Treat that change skeptically.] I’m running out of patience with Apple. I’ve loyally bought their print edition every week for several months, and they keep making me angry. They change their methodology from week-to-week, sometimes including cell phones and sometimes not. That defeats the whole purpose of having a weekly tracking poll. Their writeups are confusing, and it sometimes seems as if they are intentionally trying to mislead readers. They seem to have two goals. 1) They want to say that things are completely different this week than last week, so buy a newspaper. 2) They consistently highlight whatever number looks best for Terry Gou, often comparing it to previous results in misleading ways. They desperately wanted him to win the KMT nomination, and now they desperately want him to enter the race.

There is a Green Party poll. It’s pretty consistent with previous results from their polls. Likewise for a poll from the Taiwan Competitiveness Forum (or whatever they call themselves). Normally, I disdain to take polls from political parties or “academic” foundations very seriously. This week, they are the best of the bunch.

Finally, the Gou camp “leaked” its internal polls. A self-respecting analyst like me should probably ignore these, right. I’m treating them the same way I treat KMT and DPP polls, weighting them at .25. I’m not actually even sure these polls exist. The only methodological statement calls them 調和式民調(“mixed polls”??), which is not a term that exists in standard polling methodology. The strange thing is that Gou’s poll results are pretty close to my results; they don’t seem to be dramatically inflating his chances for publicity purposes. Maybe they are actual leaked polls. Or maybe they are doing exactly what I’m doing and aggregating polls in some way.

What a terrible week for polling.

One last note. Only one of the five polls asked about the Tsai-Han-Ko race. The pollsters seem to have collectively decided that Ko is not running. I’ll try to keep producing this chart until Ko makes an official announcement, but I might run out of data first.

Tsai 45.4%, Han 34.5%.

Tsai 33.5%, Han 29.6%, Ko 22.3%.

Tsai 32.8%, Han 27.9, Gou 24.5%.

Tsai 31.8%, Han 27.2, Gou (allied with Ko) 28.3%.

___________________________________________________________________

Comments for 8/24:

There have been four new polls, but none of them polled all four of the matchups that I am charting. The Tsai-Han and Tsai-Han-Ko matchups are affected by a deep green think tank poll that was released today and showed Tsai far ahead; for some reason this poll did not ask about Gou.

The last two polls from Pinview (品觀點) have combined telephone and internet surveys, so I reduced the pollster weight for these surveys from .75 (the value for new media outlets) to .50 (my value for new media outlets doing internet surveys). The Pinview survey results are pretty close to the overall average, so this doesn’t actually affect my results very much.

We’ve had about 10 polls over the past 10 days that consistently show better results for Tsai, both relative to the previous week’s polling averages and to the results of the previous waves of those particular polls. As such, I’m pretty confident by now that Tsai’s recent bump is not a statistical blip. I wouldn’t go so far as to say that she is “likely” to win re-election, but I think we can say that she is leading the race now (according to the public polls).

Tsai 44.7%, Han 36.4%.

Tsai 33.2%, Han 29.3%, Ko 22.7%.

Tsai 33.8%, Han 28.1, Gou 24.4%.

Tsai 31.6%, Han 28.9, Gou (allied with Ko) 27.4%.

___________________________________________________________________

Comments for 8/20:

Three new polls. They were all fairly close to the previous polling averages, so they didn’t change things much. The one exception was a Fount Media poll showing Han winning the three-way race with Ko by 3.8% over Tsai.

Tsai 43.9%, Han 37.3%.

Tsai 31.8%, Han 29.7%, Ko 23.5%.

Tsai 34.1%, Han 28.4, Gou 24.6%.

Tsai 31.0%, Han 29.4, Gou (allied with Ko) 28.3%.

___________________________________________________________________

Comments for 8/18:

I made a mistake last time by choosing to use the surveys for Terry Gou that specified Gou would be in alliance with Ko. For one thing, recent events have made that alliance less likely. For another, even if it does take shape, at some point pollsters will stop explicitly putting it into their question, and I won’t have an obvious cutting point to decide where to combine the two different sets of polls. It’s better just to use the questions that ask only about Gou. If his support spikes because he allies with Ko, we will see it in the polling results (the same way we would see his support go up for any other type of good news.)

However, since I already put the polling results for the Gou/Ko alliance, I’ll continue to update that chart as long as it seems useful. For now, presenting both charts has the advantage of illustrating how much difference Ko makes for Gou. With him, it is nearly a three-way tie. Without him, Gou is pretty clearly lagging in third place and Tsai opens up a significant lead on Han.

There has only been one poll since 8/15, so the aggregated polls aren’t really very different from then.

___________________________________________________________________

Comments for 8/15:

Three new polls are out today, and all are a bit better for Tsai than their previous results.

I’m adding polls for the three-way race with Tsai, Han, and Terry Gou. The polls often test Gou by himself and/or Gou in alliance with Ko. There is more data for Gou in alliance with Ko, so I’m using that as my standard. Also, it seems likely to me that if Gou runs, it will be in alliance with Ko. (I also included surveys in which Gou is allied with both Ko and Wang. Wang’s effect is negligible. Ko usually adds 2-3%; Wang adds around .2-.3%).

Tsai 42.2%, Han 37.0%.

Tsai 31.1%, Han 29.8%, Ko 22.9%.

Tsai 31.1%, Han 29.3, Gou 27.4%.

___________________________________________________________________

Original Post:

A few days ago, I asked myself a very simple question, “According to the polls, who is winning the presidential race right now?” To my surprise, I had no idea. New polls come out seemingly every day, and they often have very different results. Of course, most of the public polling in Taiwan these days is of low quality or is politically motivated, so you always have yourself to ask how much any given result.

What I wanted was a poll of polls, like the ones fivethirtyeight or Real Clear Politics publish for American elections. Of course, those guys are true data hounds and I’m just a hack. If they are a Ferrari, I’m a 50cc “little lamb” scooter. On the other hand, as Taiwanese know, a little lamb can be pretty useful, especially if the alternative is walking. So here is my lousy attempt at aggregating presidential poll results.

I’ll give you the results first, and, for people who want to know more, I’ll then explain the methodology in detail. I aggregated the polls for two races: (1) the head to head race between Tsai Ing-wen and Dan Han Kuo-yu, and (2) the three-way race between Tsai, Han, and Ko Wen-je. These two matchups have been heavily polled over the past four months, so there is enough data to put together an aggregate result. I know that many of you will want to know about Terry Gou. Right now, there isn’t enough data to do this sort of exercise. Maybe I’ll be able to put something together in a week or two, but for right now you should look at individual polls to see how he is doing.

In the head to head race, as of August 12, Tsai (40.2%) leads Han (38.1%) by 2.1%. This is what’s known as a close race. The race has been extremely tight for over a month; the last time either candidate had a lead of at least 3% was on July 8.

You might think of the current phase of the race as the third phase. In May, Han led by a lot, but Tsai was making up ground. June and July were dominated by the nomination contests. I did not include polls conducted during the primaries, but both candidates got a post-victory bounce. In the third phase, those bounces have faded, and the race is simply tight.

In the three-way race, Han (31.0%) currently leads Tsai (29.4%) by a narrow 1.6%. Ko is clearly in third place, with 22.9%.

The three-way race shows some of the same patterns. For example, you can see the same post-nomination bounce for Tsai and Han in this data as well. However, the most important question in any three-way race is who is in third place. This is the position of death, since strategic voting can eviscerate a third place candidate. Avoiding third place is paramount, especially for a candidate like Ko who doesn’t have a solid bloc of party voters behind him.  In early May, Han was clearly leading, and Tsai and Ko were roughly tied for second. If anything, Ko was slightly ahead of Tsai. However, about the time Tsai won the DPP nomination, she passed Ko in the polls and has been jockeying with Han ever since. Ko has been solidly in third place for two months. His support hasn’t slipped all that much, but Tsai’s modest rise combined with his slight erosion have totally changed the look of the race.

Comparing the two charts, Ko’s entry clearly hurts Tsai more than Han. As a general rule of thumb, Ko takes about two votes from Tsai for every vote he takes from Han.

To return to my original question, “Who is winning right now?” It is very close between Tsai and Han, while Ko is a bit behind. Of course, we still have a long way to go before the election, and I’m sure things will continue to change. Heck, we still don’t even know whether we will have two, three, or four major candidates. It’s all still up in the air.

Methodology

Data Inclusion

I am not collecting polls. The wonderful volunteers who edit Wikipedia are doing all that hard work for me. My data all comes from the Wikipedia page on polls for the 2020 presidential election.

I am not using all the polls for the three-way race. I did not use any polls conducted during the DPP or KMT’s polling primaries. This did not eliminate any head to head polls, but it did eliminate three polls from June 13-15 and four polls from July 10-15 for the three-way race. I am also not using poll results that ask about coalitions between Ko and Gou or Ko and Wang. I am only using the ones that assume Ko is running on his own.

Pollster ratings

Not all polls are created equal. Some polls are high quality; others are basically useless. In order to avoid having the results be skewed by lots of low-quality polls, I gave each pollster a rating. The results are weighted accordingly.

In general, I judge poll quality by who commissions the poll rather than by who executes it. Most polling organizations in Taiwan are perfectly capable of doing a high quality poll. However, the organization paying for the results controls what we learn. They decide whether to publish the results, which results to publish, and whether or not they want to fiddle with the results to make them “better.” Pollsters are in no position to contradict or correct any announced results since doing so would endanger their chances of getting business in the future.

Almost all public polls are published by media organization, think tanks, or political parties. Of these, I trust media polls the most. Taiwan’s media is heavily partisan, but polling tends to be run independently of the editorial team.

Among the various media, I trust polls from established outlets more than I trust polls published by brand-new online media groups. These days a new online newspaper seems to pop up every day, and many of them try to use polls to grab attention. In an era of fake news, I am automatically more skeptical of any news outlet that hasn’t been around for a while.

Two media groups get special attention. The China Times is an established newspaper, and its polls are historically ok. However, given its close association with Han Kuo-yu this year, I don’t consider the China Times to be a normal media outlet in this presidential election. I’m giving them the same weight I would give an interested political party. ETtoday is a new media outlet. Methodologically, their sample includes both telephone and internet respondents. I am wary of using internet surveys to obtain a point estimate for the population. I don’t think it is worthless, but this methodology earns a bit more skepticism from me.

There are a number of think tanks and “academic” foundations that publish polls. I am highly skeptical of anything coming from one of these groups. They all have policy agendas, and they often use their polling results to try to advance that agenda.

Finally, there are political parties. As a matter of principle, I should probably dismiss any poll from a political party out of hand. The two big parties do excellent internal polls, but those results are only released when they are advantageous and we have no way of knowing whether the results are real or entirely made up. As a practical matter, I don’t have any KMT or DPP results in my data now. The only two “polls” they have released are the primary results, and those were ruled out along with other results from the primary period.

The Green Party has published a number of polls over the past few months. On the one hand, the Green Party is a political player with a clear agenda. On the other hand, it isn’t directly involved in the presidential race. I consider their polls to have some information, and I give them the same rating as a think tank.

Finally, I give extra points to the organizations that I consider to have a track record of excellence. United Daily News doesn’t publish as many polls these days as it once did, but I’m giving it an “excellent” rating based on its historical reputation. TVBS polls quite a lot and generally does a good job. TVBS polls tend to have a consistent pro-blue bias, which is not necessarily a bad thing. Seeing the same bias over and over gives reassurance that they are following the same standard operating procedures. (TVBS seems to have one poll every cycle that is completely out of left field. This year, it was the June 24 poll that showed Tsai crushing everyone else.) Finally, there is my-formosa. This is a new media outlet, but I am not giving credit to them. I’m breaking my own rule to give credit to a polling organization. The my-formosa polls are conducted by Beacon, which is Tai Li-an’s new company. Tai Li-an has previously done the Taiwan Indicators Survey Research and Global Views polls. He has a long track record of excellence, and I trust him more than anyone else who regularly publishes polls. (Ironically, TVBS and my-formosa, which I have singled out as the two best polls, don’t show the same results. If you just look at TVBS polls, Han is clearly ahead. The opposite is true for my-formosa polls.)

I originally did these polling aggregations without adding pollster ratings. It’s a lot of work, but it doesn’t actually change the results very much. Almost all days changed less than 1% after adding these ratings.

category	organization	weight
Excellent reputation	TVBS	1.50
	United Daily News	1.50
	台灣指標民調	1.50
	美麗島民調	1.50
Established media	Liberty Times	1.00
	三立新聞	1.00
	年代民調	1.00
	蘋果日報	1.00
	遠見研究	1.00
	鏡週刊	1.00
Academic	世新大學	1.00
	中華傳播學會	1.00
Brand new media	優傳媒	0.75
	匯流民調	0.75
	品觀點	0.75
	放言 (山水)	0.75
Think tank	兩岸政策協會	0.50
	台灣制憲基金會	0.50
	台灣民意基金會	0.50
	台灣競爭力論壇	0.50
Uninvolved party	Green Party	0.50
New media, internet survey	ETtoday新聞雲	0.50
Involved party	KMT	0.25
	DPP	0.25
Equivalent to involved party	中國時報	0.25

Age of poll

I calculated the race on every day, starting from May 1. Polls were included if they were released in the twenty days before that date. Each day, a poll’s weight decreased by 0.05. On the day it was released, the poll was weighted at 1.00. The next day, it had a weight of 0.95, the third day 0.90, and so on. Polls thus enter the data at full strength and slowly fade away over the next 20 days.

For each day, the pollster weight is multiplied by the age weight to get a final weight. Tsai, Han, and Ko’s support is multiplied by this weight. All these results are added up and divided by the sum of weights to get each candidate’s aggregate support for that given day.

To give a concrete example, let me show you how I got Tsai’s current 40.2% in the head-to-head race. 16 polls have been published in the past 20 days. Each poll gets two weights based on the pollster rating and date of publication. Tsai’s support is multiplied by this weight.

date	org	Tsai%	W1(date)	W2(org)	W=w1*w2	wTsai
724	綠黨	.437	.05	.50	.0250	.0109
725	品觀點	.345	.10	.75	.0750	.0259
729	美麗島民調	.464	.30	1.50	.4500	.2088
729	聯合報	.320	.30	1.50	.4500	.1440
730	蘋果日報	.354	.35	1.00	.3500	.1239
731	ETtoday新聞雲	.390	.40	.50	.2000	.0780
731	綠黨	.439	.40	.50	.2000	.0878
801	品觀點	.360	.45	.75	.3375	.1215
805	蘋果日報	.375	.65	1.00	.6500	.2438
805	鏡週刊	.384	.65	1.00	.6500	.2496
807	台灣制憲基金會	.510	.75	.50	.3750	.1913
807	綠黨	.445	.75	.50	.3750	.1669
808	品觀點	.366	.80	.75	.6000	.2196
809	TVBS	.450	.85	1.50	1.2750	.5738
810	ETtoday新聞雲	.378	.90	.50	.4500	.1701
812	蘋果日報	.382	1.00	1.00	1.0000	.3820

If you take the sum of those last two columns, you get sum(W)=7.4625 and sum(wTsai)=2.9977. Divide the second by the first: 2.9977/7.4625= 0.4017.

Individual polls, especially low-quality polls, do not have much of an impact on the trend lines. It usually takes several polls with results that are markedly different from preceding results to move the trend lines up or down. For example, Han’s support shows a marked drop at the end of July. Six polls – including several high-quality polls – were published over three days (July 30-Aug 1), and this enormous amount of new, unfavorable data moved Han’s trend line down quite a bit.