The Dim-Post

June 12, 2012

What next?

Filed under: Politics,polls — danylmc @ 9:15 am

Peter has modified the tracking poll script a little, using an adaptive smoother, with the comment:

It gives slightly more credible results (at the cost of some oversensitive end-effect weirdness)

The page is here. Rob Salmond has also updated his poll o’ polls.

It’ll be interesting to see what happens next. I think we’re about to see National’s bid to change the conversation, win back the political initiative, set the agenda, all that good stuff – coming at us in the next few days. This may consist of the standard diversion tactics: ‘cracking down’ on whoever, provoking a fight with the unions or Maori or civil liberty groups, or announcing that they’re considering something crazy which they won’t actually do – but I suspect we’ll see something more substantive.

About these ads

15 Comments »

  1. It seems that this method weights the election results much more strongly than any poll results. Fair enough — they have the largest sample size. But the result is that the trend lines for National and the Greens tend to track the bottom of their polling range – their election results are typically lower than polled – and Labour and especially NZ First are more towards the top of their polls, as they do a bit better at election time than elsewhere.

    That suggests a couple of things that could be going on. One option that there is a systematic bias in polling, though opposite to the ways typically suggested. It’s often claimed that polls miss the youth vote, i.e. those with only a mobile phone, or less likely to respond to a survey, but if this were the case, you might expect the Green vote to be typically underestimated, and the older catchment of NZ First to be overestimated, but the polls are almost universally the opposite. The other option is that people change their minds come election time, in particular becoming more conservative (-Green, +NZ First) while not overly rewarding either of the main parties.

    Comment by Greg — June 12, 2012 @ 9:37 am

  2. Hmmm..they have advanced the asset sale timetable…which should produce some heat…

    Comment by Peter Martin — June 12, 2012 @ 10:19 am

  3. I think we’ll see nothing in the near future beyond more class war rhetoric about the danger posed by the underclass and mean spirited penny pinching. This is a government that with every passing day reveals more and more that it’s true animating spirit is that of the pinch faced, hectoring authoritarianism of Shiplyism. So expect more nickel and dime eroding of services designed as much to prove a Thatcherist point about the immorality of social provision as save money.

    In the medium term, in keeping with the enthusiastic class war of this government, I expect Bill English will have wrenched enough money out of ACC payments, DPB mums and other beneficiaries and users of government services to declare victory on the debt problem sometime before the next election and announce nice, fat tax cuts aimed squarely the middle class.

    Comment by Sanctuary — June 12, 2012 @ 10:32 am

  4. The extrapolation of the Nat (sp)line still looks dodgy – heading down at a steep rate from the most recent Roy Morgan result, completely rejecting the more recent (and higher) TVNZ and TV3 polls…

    FM

    Comment by Fooman — June 12, 2012 @ 11:20 am

  5. ‘more substantive’? from THIS gummint? gimme some-a your kool aid.

    Comment by petronious — June 12, 2012 @ 1:21 pm

  6. @Fooman The model estimates about a four and a half percentage point “house effect” for National in the ONCB and 3NRR polls, compared to about one percentage point in the RMR poll. Have a look at the interactive version, and you’ll see that the steep decline is consistent with the TV polls.

    Technical note: there is no ‘extrapolation'; the smoothed curve only extends as far as the most recent poll.

    @Greg The smoother tracking along the bottom of poll results is because of the bias correction rather than the large election weights. Again, the interactive version makes this more clear.

    Comment by pete — June 12, 2012 @ 2:16 pm

  7. Thanks Pete,

    The derivative of the Nat spline seems very high – is there a limit on the rate of change, e.g. the sum of all the absolute derivatives of the curves = 0 (unless somebody is making up poll results, and ignoring don’t knows) – the rate of decline in (Nat + NZF) seems very much higher than the rate of increase in (LAB +Green).

    Almost a basic checksum to prevent changes beyond that available as a percentage (sum of results must = 100, so therefore sum of derivatives must = 0)

    Cheers,
    FM

    Comment by Fooman — June 12, 2012 @ 3:25 pm

  8. Something more substantive?

    Well, Key has just revealed he likes Katy Perry. There you go.

    Comment by Fuzzy Dunlop — June 12, 2012 @ 3:52 pm

  9. 8.Something more substantive?

    Well, Key has just revealed he likes Katy Perry. There you go.

    Comment by Fuzzy Dunlop — June 12, 2012 @ 3:52 pm

    Is that with or without make-up?

    Comment by MeToo — June 12, 2012 @ 5:04 pm

  10. @FM At the moment each party is treated separately. I want to move from separate Gaussian models to a single multinomial model, but I have to deal with missing values then (some polls don’t have a result for all the parties of interest). That’s coming in version 2.0 (fingers crossed). So there’s no constraint on the sums or the sums of the derivatives.

    Note that the four parties shown don’t have to add to 100% — some of the share is with the smaller minor parties.

    Comment by pete — June 12, 2012 @ 5:09 pm

  11. Also: The derivatives look pretty high because we’ve got 2002–2012 on one graph. The changes look much more gradual on a less compressed scale.

    Comment by pete — June 12, 2012 @ 5:13 pm

  12. Thanks once again Peter.

    Any thoughts on applying a FFT to the curve to try and predict future election cycles (3 year between elections is suspiciously close to π…)?

    But seriously – by inspection, extrapolating the splines over the next couple of months sees Nat losing ~20%, Lab gaining ~7%, Greens +3%, and NZF losing ~2%. Where is all the remainder of the support (~12%) for Nat going?

    I guess the obvious answer is Australia…

    FM

    Comment by Fooman — June 12, 2012 @ 5:21 pm

  13. Hi there.

    This is a really interesting chart. A question though: What about when polling companies review and tweak their approaches in the period following the election. I imagine that most worthy pollsters do this, because after three years they finally have some benchmark data and they can test different theories about what produces the error for certain parties, and make little adjustments to try and correct for it. Case in point – I was slightly suprised to see how close the most recent TV1 and TV3 polls were (although they were a week or so apart so that may explain it).

    There are only really four points that can be used to estimate the error for each poll, but aspects of their polling approach may change (possibly a few time) between each election (so there may actually be less than one valid point at which this error can be estimated). The error for some of the parties in some of the poll curves looks larger than the difference between those polls and the last Election Day result, so does this chart simply assumes that all methodolgies stay the same all of the time?

    Perhaps some assumptions about this at the bottom of the chart would be good (just prevents it from being a tad misleading, and detracting from how interesting that chart really is). Just my two cents.

    Cheers

    Comment by Andy — June 12, 2012 @ 6:33 pm

  14. I wouldn’t recommend trying to extrapolate these results based on the end point derivative. I might have a look and see if there’s something I can do to the graph to discourage that!

    Comment by pete — June 12, 2012 @ 6:42 pm

  15. @Andy: Yes, that is likely to be a problem. There are a few checks that could be done, and they’re on my to-do list for this analysis. I know of at least one instance (h/t Rob Salmond) where a polling firm added weekend calls to increase its sampling coverage. The increase in mobile phone use probably plays havoc with the results too.

    Comment by pete — June 14, 2012 @ 1:11 pm


RSS feed for comments on this post. TrackBack URI

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

The Rubric Theme. Blog at WordPress.com.

Follow

Get every new post delivered to your Inbox.

Join 404 other followers

%d bloggers like this: