Hard News by Russell Brown

Read Post

Hard News: Where are all the polls at?

228 Responses

First ←Older Page 1 2 3 4 5 10 Newer→ Last

  • Bart Janssen, in reply to Russell Brown,

    New Zealand sample sizes are not small by international standards.

    The point was not that the sample size was small but rather that it is easy to sample a non-representative group.

    Auckland • Since Nov 2006 • 4451 posts Report Reply

  • mpledger, in reply to linger,

    >> for smaller parties the polls are less accurate

    > Only in terms of relative error, not in terms of the absolute error

    Only in terms of the sampling error. The non-sampling error for small parties is probably quite large.

    National's margin of error is quite close to the survey's margin of error at 3% (giving a sampling error of roughly 1.5%). I could easily believe the non-sampling error for small parties is at least 1.5% even if the sampling error is minuscule ...

    ... especially for the Maori Party and Act - the former because of phone line & demographic issues and the latter because of geographic concentration.

    Since Oct 2012 • 97 posts Report Reply

  • Russell Brown, in reply to Bart Janssen,

    The polls are accurate for parties that poll close to 50%.

    For smaller parties the polls are less accurate, hence the “surprising” results for NZ First etc.

    Not really. NZ First won 8.66% of the vote. The last five polls: 6.6, 8, 8.4, 7.1, 8.

    So three were very close and the other two fairly close. (And one was Ipsos, which seemed to consistently underrate NZ First.) That's not "horribly inaccurate".

    Auckland • Since Nov 2006 • 22756 posts Report Reply

  • linger, in reply to mpledger,

    Not disagreeing there — see the last few lines of my comment above. Theoretically extreme case: if a minor party’s support were entirely limited to a small, definable demographic or regional subset of the population, then the relevant sample size for calculating the error in that party’s estimated support would be the (unknown, but much smaller) size of the subsample obtained for its demographic, rather than the overall sample size. Hence the error for smaller parties with restricted support bases should be larger than that predicted from the overall sample size, all else being equal. But in reality, such restrictions on party support are probabilistic rather than categorical, so that approach will overestimate the error. The empirical proof is there in Lumley’s results, which show less variation (on an absolute scale) around figures for the smaller parties.

    Tokyo • Since Apr 2007 • 1889 posts Report Reply

  • linger,

    TL;DR — the expected effects on errors in support estimates arising from lower overall support rates (lower p, hence lower theoretical variance) and regional/demographic restrictions on support (hence lower effective sample size, and higher theoretical variance) seem in practice largely to cancel each other out in NZ data.

    Tokyo • Since Apr 2007 • 1889 posts Report Reply

  • Shaun Lott,

    Not happening here (I hope!) but of interest...
    https://fivethirtyeight.com/features/fake-polls-are-a-real-problem/

    Waitakere • Since Aug 2009 • 111 posts Report Reply

  • Tom Semmens,

    David Farrar is no longer sharing his polling information to various sites for Poll of Polls.

    According to Martyn Bradbury, over at the daily blog. Can anyone confirm if that is true? If so, I am getting in an extra supply of tea of biscuits for election night.

    Sevilla, Espana • Since Nov 2006 • 2213 posts Report Reply

  • Russell Brown, in reply to Tom Semmens,

    According to Martyn Bradbury, over at the daily blog. Can anyone confirm if that is true? If so, I am getting in an extra supply of tea of biscuits for election night.

    Yup. RNZ don’t get it any more for Colin James’s poll-of-polls.

    Although bless you for wanting to fact-check Bomber :-)

    Auckland • Since Nov 2006 • 22756 posts Report Reply

  • Russell Brown, in reply to linger,

    TL;DR — the expected effects on errors in support estimates arising from lower overall support rates (lower p, hence lower theoretical variance) and regional/demographic restrictions on support (hence lower effective sample size, and higher theoretical variance) seem in practice largely to cancel each other out in NZ data.

    As I understand it, the big variance in different polling firms' practice is around how – and how hard – the call-centre workers push for a preference. That and intention to vote at all.

    Auckland • Since Nov 2006 • 22756 posts Report Reply

  • Ianmac,

    Is it possible that the polls cause the voters to vote to match the polls?

    Bleneim • Since Aug 2008 • 135 posts Report Reply

  • linger, in reply to Ianmac,

    There's some degree of feedback, most often taking the form of consolidation on the poll-anointed "winner" at the expense of poll-anointed "certain-losers" — though probably just as much through encouraging voter abstention as through encouraging an actual switch in support.

    Tokyo • Since Apr 2007 • 1889 posts Report Reply

  • Bart Janssen, in reply to linger,

    encouraging voter abstention as through encouraging an actual switch in support.

    My understanding of the data is that's the major effect. When voters perceive there is no point in voting because one party already has it won then voter turnout drops.

    Auckland • Since Nov 2006 • 4451 posts Report Reply

  • simon g,

    NZ Herald has a new online poll. So it looks like Digipoll has gone.

    They've broken down the numbers into various demographics, as they used to do with Digipoll. There was always a question about the size of those samples (taken from within the overall thousand), and we're none the wiser now.

    As per my earlier post, presumably the Herald has more data from the same poll, so we can expect a party vote headline later, maybe tomorrow. (The polling dates and data don't change, only the timing of the Big Reveal).

    Auckland • Since Nov 2006 • 1324 posts Report Reply

  • simon g, in reply to simon g,

    Also the Herald online still doesn't understand how to use quotation marks. Comments (from English? We can only guess) are embedded in the article. Sigh.

    Auckland • Since Nov 2006 • 1324 posts Report Reply

  • simon g,

    As predicted ... the Herald has a "new poll" today.

    Small print: it's still the same one. Same polling period (Aug 16-21). This really is not good enough.

    Auckland • Since Nov 2006 • 1324 posts Report Reply

  • Gavin White,

    The Herald poll is by TNS, who used to do the TV3 poll (as CM and then TNS) before Reid. My recollection is that Reid split off from TNS. TNS are a big global company, so in theory they should be using a solid methodology, but the global brands haven't had a great record with NZ elections (e.g. Ipsos and before them Nielsen).

    They surely asked a vote question but I don't see the sense in delaying releasing it - if they release it tomorrow it will have been more than a week since the poll finished, so it's not very useful as an indicator. I think either they didn't think it was a credible number or they don't want to be seen as competing with the public polls for some reason. On the other hand, the fact that they haven't used the party vote as a demographic in either of the two articles to date suggests that maybe they don't have it.

    Mikaere - what you're describing was probably not a push poll. I don't think they were contacting you to try to get you to change your mind (specifically), they were trying to find out what would change the minds of people LIKE you. I don't think any NZ political party has the money to be able to afford the number of calls that would be needed to do a true push poll, and I feel certain that Len's campaign didn't. Quite possibly, another group of participants in the same survey heard pro-Banks and anti-Brown messages.

    Ianmac - yes there can be a feedback loop, and that's a big risk for the Greens in particular right now. If the polls start showing them consistently at 3.5%-4%, then people may see them as dead in the water and they could drop even further. On the other hand, if they poll 4%-6%, then those considering the Greens will know that every vote counts, and that could draw people in. Similarly, when the stories were all about Labour falling in the polls voters would have seen them as no-hopers and looked for other options, but now they're rising in the polls voters see they have a chance again and get on board.

    Some might suggest that's a problem with the polls, but I don't think so. In the absence of polls, people would come to their own conclusions based on who they thought was winning or losing. If public polls were banned, then the stories would all be about the old Tauranga Boys High poll of Year 13 students, or about Paul the Octopus, or whatever 'prediction' was out there. Even if they're not perfect, they have some science behind them, so they're a lot better and more reliable than the alternatives.

    As Bart Janssen says, there's very clearly a relationship between the closeness of the election and turnout (compare 2005 and 2014 as an illustration). Downs' theory on rational voting, if I recall correctly (although he argued that voting was only rational if your vote was almost certain to decide the result, which is of course exceedingly unlikely).

    Wellington • Since Jul 2014 • 16 posts Report Reply

  • Trevor Nicholls, in reply to Gavin White,

    As Bart Janssen says, there's very clearly a relationship between the closeness of the election and turnout (compare 2005 and 2014 as an illustration). Downs' theory on rational voting, if I recall correctly (although he argued that voting was only rational if your vote was almost certain to decide the result, which is of course exceedingly unlikely).

    I read yesterday that (due to polarisation and gerrymandering) fewer than 5% of potential voters in the USA are in even theoretically competitive congressional districts.

    Wellington, NZ • Since Nov 2006 • 310 posts Report Reply

  • Gavin White, in reply to Trevor Nicholls,

    That's both shocking and not that surprising Trevor. In a similar vein, I recall that one of the resources I used for my 1996 thesis on turnout was a book of NZ election results from 1946 to 1990. The thing that sticks in my mind from that was that there were a lot of seats that hadn't changed hands since at least 1946 - and of course those tend to be the ancestors of seats that have never changed under MMP either.

    From what I remember, I think the single best predictor of turnout under FPP was the margin of the seat at the last election. MMP lessens that variation, but it's still likely to be very important.

    Wellington • Since Jul 2014 • 16 posts Report Reply

  • David Bulog,

    I really dig this UK polling/market research company
    http://www.qriously.com/how-it-works/

    http://www.qriously.com/blog/poll-uk-general/

    WITH 24 HOURS TO GO, LABOUR NUDGES INTO THE LEAD, BUT RESULT STILL TOO CLOSE TO CALL

    http://www.qriously.com/blog/raw-data-uk-election-poll/
    2213 people polled—you can download results as a csv

    Instead of aggregated results tables, we’ve decided to share the raw non-identifiable data of our election results so that anybody can download it and play around. We hope this will encourage more transparency and openness around polling data.

    Since Aug 2017 • 10 posts Report Reply

  • Ian Dalziel,

    re gerrymandering
    they really got creative defining 'Chch Central' electorate last time they tinkered,
    there is now a tongue of voters stretching all the way from the central city to Redwood and the main north road - for whatever reason...
    ...following a rich vein of National voters it seems?

    Christchurch • Since Dec 2006 • 7889 posts Report Reply

  • Tom Semmens,

    Just curious, any up to date polling in the public arena for Ilam?

    Sevilla, Espana • Since Nov 2006 • 2213 posts Report Reply

  • Gavin White,

    I doubt there's even any private polling Tom - it's too expensive. The only time you see legitimate electorate polls these days is in seats like Ohariu, which (until recently) could affect the nationwide result. Robos may exist, but they'll be unreliable as hell.

    Wellington • Since Jul 2014 • 16 posts Report Reply

  • Rob Stowell, in reply to Tom Semmens,

    Ilam could be interesting - fingers crossed. At least, if all the green and labour voters strategically voted for Manji. He's a popular councillor, and fairly well-liked and respected locally. Brownlee isn't especially well-liked, even by National people I talk to.

    Whakaraupo • Since Nov 2006 • 2091 posts Report Reply

  • linger, in reply to Rob Stowell,

    While it’s nice to think Brownlee could be ousted, it can’t be done through strategic voting. He’s ranked 5th on National’s list, so he'll be in Parliament regardless. Better for Labour and the Greens to focus on the party vote. If they can agree to support one non-Brownlee candidate in Ilam as well, that’s a bonus, but not something that drastically changes the outcome at a national level.

    Tokyo • Since Apr 2007 • 1889 posts Report Reply

  • Bart Janssen, in reply to Ian Dalziel,

    they really got creative defining 'Chch Central' electorate last time they tinkered,

    Fortunately with MMP the effect of that is moot for the parliament as a whole, but it can still piss of locals who feel they aren't represented correctly at the electorate level.

    Auckland • Since Nov 2006 • 4451 posts Report Reply

First ←Older Page 1 2 3 4 5 10 Newer→ Last

Post your response…

Please sign in using your Public Address credentials…

Login

You may also create an account or retrieve your password.