Granny Herald: from flawed data, false conclusions

Written By: - Date published: 11:21 am, March 23rd, 2013 - 17 comments
Categories: john key, national, polls - Tags:

Three of the Herald’s political staff has a crack at trying to explain why National polled 48.5% in the latest Herald-Digipoll. Consensus seems to be that John Key’s an angel sent to earth to do very little but keep enough of the middle class happy enough who’ll never lose. May I humbly suggest an alternative theory – that the Herald-Digipoll is broken.

Danyl at the Dimpost has the most mathematically poll of polls that, crucially, is anchored in reality – ie the last election result. Whereas political commentators and pollsters seem to shamelessly ignore how badly their poll got it wrong at the last election when breathlessly reporting the latest numbers, the Dimpost poll of polls corrects for their error.

What the Dimpost poll shows is that:

a) the Herald Digi-poll consistently over measures National – for example, the final Herald poll before the last election had National on 50.9%, that’s outside the margin of error above the 47.3% they got. The final Herald poll in 2008 had National on 47.9%, they got 44.9%. Statistically, the Herald isn’t just missing the bullseye, it’s missing the dartboard altogether. Only the 3news poll is worse. (in the image below, the light blue line is the Herald trend, the dark one is the poll of polls trend).

Dimpost poll of polls highlighting Herald Digipoll

b) even the Herald poll shows that National has lost a significant amount of support. From a pre-election high of 54.2%, a final pre-election number of 50.9%, to 48.5% today. That’s a big drop. Adjust to the Herald’s 3% over-polling of National at the last two elections, and it signals National is at around 45% – in the range that the Dimpost poll of polls has it. Remember, the Right has a one-seat majority in the House. Losing 2-3% since the election, especially when Banks will never get back in, is no laughing matter for National. If it were to translate into an election result, it would mean National would need New Zealand First (or a very large wasted vote) to govern.

c) more regular polls, like the Roy Morgan show a clear downward trend for National. You can’t possibly draw a trend from the five data points in 17 months that the Herald provides because the statistical noise that results in variation between every poll can’t be sensibly separated from the trend when you have so little to work with. The Roy Morgan has polled 29 times in that period and its trend is clear. (incidentally, the Roy Morgan was also the most accurate for National in 2011).

roy morgan labour greens v nats 3

To look at the Herald poll and conclude that National is now more popular than at the election, you have to ignore the Herald’s record of over-polling National, the fact that National’s support in the Herald poll has fallen significantly since the election, and the fact that more reliable polling trends tell us the opposite.

Only if you make all these logical errors (and there’s a strong incentive to do so because our paper spent a lot of money on the poll, which means its going to be treated as gospel no matter what, goddammit), can you then conclude National’s rule continues unblemished thanks to Bonny Prince Grinny.

But when’s a few fallacies ever stopped Granny?

[Note: none of this should be taken as a ‘if these trends continue’ argument. One current accurate polling, the Left needs Winnie too making him Kingmaker and he would have much more incentive to go with National, and there’s the risk of a Shearer campaign meltdown too. Put National at 40% and I’ll be happier].

17 comments on “Granny Herald: from flawed data, false conclusions ”

  • Comments are now closed

  1. Tiresias 1
    23 March 2013 at 11:40 am

    “Figures often beguile me, particularly when I have the arranging of them myself; in which case the remark attributed to Disraeli would often apply with justice and force: “There are three kinds of lies: lies, damned lies and statistics.”
    – Mark Twain’s Own Autobiography

    Presently Labour will only be picking up those sufficiently dismayed by National’s duplicities and incompetence not to want them in Government again. Labour will only start picking up genuine support when it starts presenting an alternative programme worth genuinely supporting, and six-weeks out from an election is far too late to start thinking of doing that.

  2. Bob Tambling 2
    23 March 2013 at 12:17 pm

    The worry for me then is if the polling trends continue and it looks like Shearer can just get over the line the possibilty of changing the leader will further diminish and Shearer will fight the 2014 election.
    Unfortunetly during the ensuing campaign, Key and the Nats will crucify him and we will be left with a National Government or a very weak LAb /NZF /Green one term Goverment
    I wonder from a Green party prospective (and a leftish pespective) , would it not be better if the Labour Party lost the next election with Shearer as Leader?
    If NZF play a part as may be likely, the Green can’t afford to go into a NZF /Lab/Green Gov.
    Shearer simply won’t cut the mustard and Peters is toxic. It would be a crap government and the well honed National party machine would again tear it to shreds

    The leaves the Greens in an interesting position, the lure of ministerial positions versus the pragmatic long term strategy of the Greens. A one term Lab/Nzf/Green Gov seems to serve no ones interest.
    Labour needs to have a strong leadership team and comitted activists , The reality is they have a diminishing group of activists and a crap leadership.
    The Greens on the other hand have a very strong leadership team and an ever growing group of activists ( let us not forget they are the second biggest party in Wellington).
    Labour? I can’t think of anyone in the present caucus who could take most of the country with them in an election. They need a good clean out and new blood. They don’t appear to be listening to that line of argument.
    In conclusion:
    I am committed to working very hard to getting as many Green MP’s elected as I can in 2014. I am not so sure though if I want those MP’s in a Government with the current infighting mob of the Lab party who it seems “could not pour piss out of a boot if the directions were on the heel”.

  4. tc 4
    23 March 2013 at 12:36 pm

    The more granny and her shills like Johnno, o’shillivan, rortney etc keep banging the NACT drum the more irrelevant and bias they show themselves to be.

    Labour needs to grow a pair and get out there now however with the mallarfia and their front man it seems it’ll come down to the hollowmen blowing it more so than Labour winning it.

    They could easily blag another term if they weren’t such psychos given the current state of the opposition, thankfully they are what they are.

  5. ghostrider888 5
    23 March 2013 at 1:15 pm

    article in todays Herald acknowledges the popularity of Key is slipping

  6. pete 6
    23 March 2013 at 1:34 pm

    “Only the 3news poll is worse.”

    That’s not quite true: all five polls overestimate National’s election results, and the Digipoll isn’t too much worse than Fairfax or Roy Morgan. One News Colmar Brunton, as well as 3 News Reid Research both overestimate by more than the Herald.

    • lprent 6.1
      23 March 2013 at 1:45 pm

      Points to a systematic flaw in their sampling doesn’t it…

      • pete 6.1.1
        23 March 2013 at 3:39 pm

        That’s one possible explanation, although some have theorised that National lose a lot of votes close to the election once people actually start looking at their policies.

        There’s also a theory that NZF underpolls because their voters are embarrassed to admit that to callers.

        • lprent 6.1.1.1
          23 March 2013 at 4:13 pm

          Possible. But less likely than the simpler explanation.

          Canvassing companies are using listed landlines*. When I was analyzing some somewhat larger samples of both door-knock canvassing and phone canvassing across several electorates, there was a statistically significant correlation between people who canvassed right and left and if they had a listed landline.

          For that matter there was an even stronger correlation between age group and having a listed land line – and it looks like if people don’t have a listed phone line in their mid 20’s then they never get one.

          * There is a common myth that goes around about auto-dialing random numbers. Asked a few people who work at or have worked at polling companies. It doesn’t happen and a few minutes thought about number ranges will give an explanation about why. It is cheaper to buy/get a phone list.

  7. Chris 7
    23 March 2013 at 1:47 pm

    So,will Granny Herald be ramping up lots of little good news, happy smiley,relaxed all singing, dancing reports/photos of our poor excuse for a pm then. Got to keep him relevant.

  8. Centaur 8
    23 March 2013 at 4:48 pm

    Give Key *some* credit. He may decide to call it quits. It is not as if he actually needs this gig ..

  9. Murray Olsen 9
    23 March 2013 at 7:50 pm

    I’m not even sure that the margins of error as reported are very meaningful at all. From what I know, they seem to rely on unbiased sampling and the validity of the Central Limit Theorem, which then leads to a percentage presumably taken from 100 times √N/N. The concern about landlines is just one thing that could skew this.

    • lprent 9.1
      23 March 2013 at 10:26 pm

      They do sample distribution normalising for many of the other factors like age, gender and geological distribution. They don’t adjust for factors like people working during the periods that they phone.

      But they don’t publish their techniques and criteria so it is pretty hard to see why there if it was unbiased sampling there would be any differences between the polls within the confidence limits. But there clearly is.

      • Centaur 9.1.1
        24 March 2013 at 8:50 am

        lprent: geological distribution ?
        Such as, where there are proven mineral resources ?

        • lprent 9.1.1.1
          24 March 2013 at 8:56 am

          I blame the iPad, it’s overly smart autocorrect, and that I use Geological at least 10x more often than Geographical.

    • Lanthanide 9.2
      24 March 2013 at 8:38 am

      No, they’re not. Even assuming they are following absolutely perfect statistical methods and the statistical methods work as advertised by the literature, the margin of error they’re reporting is presumably the 95% confidence interval. Which means 1 out of 20 polls are going to be wrong, but we won’t ever be able to tell which one without some other corroborating information (eg other polls taken at the same time).

      That’s assuming they’re using perfect methods, which of course they aren’t.