Poll of polls

As loyal readers know, I’ve been a bit out of the loop for a while. It seems as if the only really important thing to have happened in the last couple of weeks was China’s military parade and the accompanying kerfuffle in Taiwan over whether it was appropriate for various people to attend. What kind of effect might that have?

In the USA, Real Clear Politics and fivethirtyeight.com (starring Nate Silver) frequently publish polls of polls. That is, they take all the recent polls and average them together. I thought I’d try doing this. I got my raw data from the wonderful volunteers who edit the Wikipedia page on the 2016 presidential election. Please note that I am including all the polls from Wikipedia without making any judgment about their accuracy. Each poll was weighted according to its recency, with a half-life of 7 days. That is, on the day it was finished, it got a weight of 1.00; on the 7th day, it got a weight of .500; on the 14th day it had a weight of .25, and so on. I dropped data when the weight dropped below .05, which happened on the 31st day. It only took me about 3 hours to figure out how to do this, so don’t expect this chart to be a daily feature on my blog. Anyway, here is what the 2016 presidential race has looked like over the past three months.

prez poll of polls 20160915

This chart smooths things out quite a bit, but the basic outlines are quite clear. Hung passed the KMT’s polling primary on June 13, right at the beginning of this time series. That was her high point, and it may have been artificially high. Recall that some green supporters strategically expressed support for her in order to ensure that the KMT nominated a weak candidates. Hung was duly nominated on July 19, but her support had already started eroding. By this chart, she has been steadily drifting downward for three months, and while the pace of her decline has slowed, she is still going in the wrong direction. Soong announced his candidacy on August 6, and he immediately opened up a clear gap between himself and Hung. However, in the past two weeks Soong has also been trending downwards, and he doesn’t seem to be very far ahead of Hung at this point. In the past few days, three polls have shown Tsai Ing-wen opening up a bigger lead. At this point, Tsai is further ahead than she has been at any point since mid-June. In fact, biggest competition is from “don’t know,” which is well ahead of both Soong and Hung. It hasn’t been a good summer for the blue side of the spectrum.

11 Responses to “Poll of polls”

  1. Greg (@greghao) Says:

    So at what point would you feel comfortable calling this election for Tsai? Usual caveats aside?

    • frozengarlic Says:

      It would take a major shakeup for someone else to win. That has been my position for at least a year, and I expect it will continue to be my opinion until the votes are counted.

  2. Mike Says:

    As a primary contributing author of the page, I thank you for your insightful post, yet again. As you will be aware, a founding principle of Wikipedia is neutral point of view, and together with the need to apply reporting objectivity (wherever possible), it does necessitate us to report all polls that are publicly released, regardless of reliability. I’m hoping that the results are not skewed too drastically as a result of polls that may appear to be less than reliable.

    • frozengarlic Says:

      Thanks for helping to put all those polls together.

      With meta-analysis such as this poll of polls, it is assumed that the various individual studies will each use different methodologies and will come up with slightly different measures. However, that is supposedly one of the strengths. Some of the polls look favorable to the KMT, and green supporters don’t trust them at all. China Times? What a joke! Similarly, my blue friends all sneer at the polls from Liberty Times. I’m skeptical of some of these poll results as well, such as the Sept 10 KMT poll that shows Tsai simply crushing the rest of the field. The hope is that by averaging all these together, we can cancel out some of the noise. That said, if I had my own polling house that I could trust completely, I wouldn’t care a whit what all the other polls say.

  3. Allister Says:

    Pardon me, who or what is DK?

    • frozengarlic Says:

      “Don’t know.” This is a catch-all for all the non-response categories that different survey organizations have, such as “they’re all good,” “they’re all bad,” refusal to answer, “haven’t decided yet,” and “don’t know.”

  4. R Says:

    Great work!

    In Canada there’s actually several sites that make uses of poll of polls to predict seat allocations. It would be great if someone in Taiwan can come up with a similar model. One of the most prominent of the Canadian site:
    http://www.threehundredeight.com/

    • frozengarlic Says:

      Did you read their methodology for predicting the individual ridings? On the one hand, it is a big mess of (often unclear) assumptions and adjustments. For example, they explain that incumbents and star candidates are given a bonus, but they don’t explain exactly how this works. I also have no idea how today’s polls would translate into actual voting numbers. Are you comfortable inferring that a poll showing Tsai 45, Hung 15, and Soong 15 can be simply translated into a national vote of Tsai 60, Hung 20, and Soong 20? And then what does that mean for the legislative vote? We don’t really believe that all the extra votes for Tsai will translate into extra votes for DPP legislative candidates, do we? On the other hand, after all their massaging, they still aren’t very good at predicting the close races. Predicting the winners of the races where the eventual margin is over 20% is not that much of a challenge.

      I might try something like this if I am bored sometime, but I certainly won’t put too much stock in the results.

      • R Says:

        On the polling input to a hypothetical legislative race simulator, I was not suggesting that the presidential race numbers be used, but rather the actual polls on the legislative race in Taiwan be used. As few as they are, they do exist, & there is a compilation on Wikipedia as well. Yes the problem is the very small number of these, but recently it looks like more are getting commissioned. I won’t be surprised if even more get released in the future, if the focus shift to the relatively more competitive legislative races.
        https://zh.wikipedia.org/zh-tw/2016%E5%B9%B4%E4%B8%AD%E8%8F%AF%E6%B0%91%E5%9C%8B%E7%AB%8B%E6%B3%95%E5%A7%94%E5%93%A1%E9%81%B8%E8%88%89#.E6.B0.91.E6.84.8F.E8.AA.BF.E6.9F.A5

        As for the model of the Canadian site, I don’t doubt that there are some questions that can be asked of it, & many assumption made. In the end predicting is a messy business. Nonetheless the site been around since 2008; along with 7 years of tinkering & re-weighting to the model, I think the model’s track record is generally decent, especially if inputted with accurate & up to date polling.

  5. csempere109 Says:

    Come back, stuff is happening!

  6. Updated poll of polls | Frozen Garlic Says:

    […] A Blog on Elections in Taiwan « Poll of polls […]

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 )

Google+ photo

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

Connecting to %s


%d bloggers like this: