New leaders, new poll, and a new crusade for ACT. But will it end any differently? And what of the other small parties?

If I was a cartoonist today I'd be digging out the old Don Quixote images, like this one, and placing Jamie Whyte and David Seymour on their respective steeds. The two new Act leaders (they need to be seen as a unit) are off on a vain quest for relevance, determined to champion a code of free-market purity that few still care about in a world where political reality is being defined by coalition deals.

Whyte and Seymour were chosen yesterday as the young men (by political standards) to re-energise the ACT party and reconnect it with a disinterested and disillusioned public. In that sense, they are the brave and ambitious choice. If Act want a future as more than National's puppet and beyond this year's election, then it needs the clean break this pair represent.

If you're on the ACT board or a forward-looking member, this is the 'hail mary' choice (hey, it's superbowl day). Throw the ball to these guys and see if somehow, they can get it into the in-zone. Because otherwise ACT is toast anyway.

John Boscawen, to continue the gridiron metaphor for a moment, could have moved them forward a few inches, got them another down at this election and done another deal for another three years. Perhaps that could have been a base to grow. But we're talking about Act here; a party with a reputation so tarnished it yesterday registered a fat 0 percent in tv3's first big poll of the year (though it's still on 0.4 percent in our Poll of Polls).

So good on Act for giving it a whirl. My fear however is that Whyte and Seymour are embarking on a quixotic adventure that will, as with our old Spanish friend, lead to some adventures that veer between comedy and tragedy, but end up with our heroes back where they started.

Their problem is that, like chivalry in Quixote's time, libertarianism is an idea currently sitting on the dark side of the moon. It's out of fashion, eclipsed. There's nothing like a global financial crisis with the world's largest economy dependent on quantitative easing and government stimuli par for the course nearly everywhere in the developed and developing world to make people nervous of 'small government'. (I noted that Conor Roberts has just tweeted that the Libertarians are no longer registered as a party).

TV3's poll has got a lot of commentary already today, both with New Zealand First's rise above the 5 percent threshold and the double zeroes for National's coalition partners United Future and Act. But while those are notable, they are hardly surprising.

New Zealand First's rise is confirmation of my thoughts last week (if I do say so myself), that John Key's announcement that he could consider a coalition deal with Winston Peters after all is a re-birth for the man and his party. But what is surprising is how quickly it's happened. Maybe it's simply an announcement bump that could fade between now and the campaign. But it shows that when Peters is prominent in the public eye, voters still like what they see.

Looking back at the Poll of Polls, at this time in the last election cycle he was down in the twos. If other polls back up the fact that he's already pushed his way from the 4.5 percent where he's sitting now to over five percent, then it's a hell of a launching pad for the party.

But if you thought this is all good news for New Zealand First, consider this. Yes, his own backers can vote for him happy to trust him to decide whether he goes with National, Labour or neither after the election. But what if you're a voter somewhere between the centre and the right who wants to give National a wake-up call or who perhaps wants to supply National with a coalition partner who isn't on life-support?

New Zealand First is a gamble. Peters won't disclose his preferences and could go with Labour. Some of those supporting him now might be assuming that he'll end up with the government, as many have suggested. But if they want to be really certain of a Key government, then where do they look? To the Conservatives, of course.

In a funny way, Colin Craig can afford to be quite happy with this poll. Not only does he stay over two percent while there's a swing to New Zealand First, there's room for him to pluck away at these larger New Zealand First numbers and become National's new best friend.

As for the double zeroes, well, United Future has been there for a while in our poll and Act's near-death experience has been widely discusssed. The political reality is that both parties are utterly reliant on National handing them a seat, and that hasn't changed.

Little mentioned in the discussion of this first election poll today has been the Maori Party. I've been meaning to write a separate post on them, but for now just let me say that Te Ururoa Flavell and co should not be overlooked this year. While the party's two seats are still being chalked up on National's side of the ledger, it's increasingly clear that that should not be assumed. It is not a party to be taken for granted this time round.

 

Comments (23)

by Chris de Lisle on February 03, 2014
Chris de Lisle

I can't seem to find any of the important statistics about this poll on TV3's website: what was the margin of error? Did the polling actually occur after Key's announcement?

I know that you're not at TV3, but this does seem to be a journalistic trend- how hard is it to just attach the poll report (or the primary source material more generally) to a given news story?

 

by Chris de Lisle on February 03, 2014
Chris de Lisle

Ah, I see tucked away down here (http://www.3news.co.nz/3-News-Reid-Research-poll-shows-Peters-as-kingmak...) that the margin is 3.1%. So NZ First could be anywhere between 2.6% and 8.8%.

by Tim Watkin on February 04, 2014
Tim Watkin

Sure Chris, but as you know the trick with reading any poll is the trend. The point is that the trend for Peters in this poll is up, and most importantly up above 5%. As I noted, he was unusually close to that already in our Poll of Polls but this puts him over the line early in the year. So he's relevant early.

by Graeme Edgeler on February 04, 2014
Graeme Edgeler

Ah, I see tucked away down here (http://www.3news.co.nz/3-News-Reid-Research-poll-shows-Peters-as-kingmak...) that the margin is 3.1%. So NZ First could be anywhere between 2.6% and 8.8%.

That's not quite how margins of error work. Rather, it suggests that there is a 95% chance that NZ First's true support in between 4.3% and 7.1%.

by Richard Aston on February 04, 2014
Richard Aston

What will be interesting is how much of NZ First's support base will be eaten away by the new conservatives over this year.

Do we think NZ First voters will be attracted to the Conservatives?

I imagine someone must be commissioning some polls on that .

 

 

by Richard Aston on February 04, 2014
Richard Aston

Tim I agree with the trend idea for NZ First this horizon research poll show it very clearly .

On the other hand we have yet to see what electorate seats deals National offers and to whom. If the conservative can scrap across 5 % and get an electorate seat gifted to them that will shift everything surely.

 

by Andrew Geddis on February 04, 2014
Andrew Geddis

Rather, it suggests that there is a 95% chance that NZ First's true support in between 4.3% and 7.1%.

No. It suggests that there is a 95% chance that NZ First's true support is between 2.6% and 8.8%.

The "margin of error" is +/- 3.1% of the reported result, not within 3.1% total.

by Christopher Nimmo on February 04, 2014
Christopher Nimmo

The "margin of error" is only +/- 3.1% at 50%; for smaller results the margin is smaller - polls should really report the MoE for each result (disclaimer - I'm trusting Wikipedia here)

by James Green on February 04, 2014
James Green

Christopher Nimmo (and wikipedia) are correct. The margin of error is smaller at the extremes (close to 0 and close to 100%). It is also possible to calculate an asymmetric confidence interval (the margin of error closer to the outer bound is smaller, whereas heading towards 50% is bigger).

Although strictly speaking, it would be useful for articles to always mention the MoE

1. It's always a little over 3%. The only parameter that affects the margin of error is the sample size. 800-100 participants represents the 'sweet spot' between number of people sampled and precision. 400 people gives 5%, 600 4%, 1068 3%, 2400 2%, so you can see the benefit of staying in the 800-1000 zone.

2. The margin of error is not actually particularly interesting. More interesting as Tim has noted is 'the trend', but I'd also suggest that a precision index around either the difference between two parties (ie, is there a meaningful difference) and change over time (did the proportion move in a detectable fashion between polls). And then on into more sophisticated analysis pooling polls (which should increase precision as well).

by Andrew Geddis on February 04, 2014
Andrew Geddis

Huh.

Teach me to meddle with the occult (aka statistics).

Most important, was Graeme Edgeler right? Because I really, really want him not to be ... .

by James Green on February 04, 2014
James Green

You're at least closer to being right, or right if you don't correct for the small value of NZ First's poll rating. Adding and subtracting the margin of error gives you the 95% confidence interval, not splitting it, as Graeme has done.

Random aside, to get 1% precision, you need to sample approx 9600, whereas 100 punters gives 10%....

by James Green on February 04, 2014
James Green

Useful other thoughts on interpreting MoEs.

1. For Act and UF, if nobody admits to wanting to vote for them, there could be up to 3, so 0% (CI 0-0.3%).

2. Assuming two say Reid Research polls with effectively the same sample size, the MoE for change between polls is 1.41 times the MoE, which would suggest that for the present poll, none of the big parties changed by 4.37%, so they're all effectively no change. However, some of the minor parties did drop by quite a lot (eg Mana), but you'd need to use a formula to check to see if it was meaningful.

by Peter McCaffrey on February 06, 2014
Peter McCaffrey

Graeme is definitely wrong, sorry Graeme.

Chris & Andrew vs Graeme were arguing whether a margin of error of 3.1% applies on either side, or is it 3.1% total. (ie: 6.9%-13.1% vs 8.45%-11.55%).

Chris & Andrew were correct that margins of error apply on both sides.

Graeme was correct to point out that actually a MoE means there is a 95% chance a party is within that range.

But all of you were wrong about what the MoE actually should be.

As James points out, the stated MoE applies only at 50% and shrinks as you approach 0% or 100% (ie: the size of what the MoE should be is a bell curve).

Finally, as Graeme might have been alluding to with his 95% statement, WITHIN a margin of error, not all possibilities are equally likely, a MoE is ITSELF a bell curve too. So a poll rating of 50% with a MoE of 5% means it's 95% likely the real result is between 45% and 50%. But it's FAR more likely that the result is between 49% and 51%, than it being at the outer limits of that range.

by Peter McCaffrey on February 06, 2014
Peter McCaffrey

means it's 95% likely the real result is between 45% and 55%*

Trust me to do a typo while correcting Graeme. 

by Peter McCaffrey on February 06, 2014
Peter McCaffrey

So... turns out Graeme was right all along.

He actually calculated what the MoE would be at 5.7% for a sample size of 1000, which is why his MoE is close to (but not exactly) half the full one.

It's 1.44% MoE on 5.7%.

Sigh. Graeme wins. 

by william blake on February 06, 2014
william blake

Standard deviation.

by Graeme Edgeler on February 06, 2014
Graeme Edgeler

Chris & Andrew vs Graeme were arguing whether a margin of error of 3.1% applies on either side, or is it 3.1% total. (ie: 6.9%-13.1% vs 8.45%-11.55%).

I was *not* arguing that!

The 3.1% margin of error is the maximum statisitcal MoE (with a 95% confidence) for results of 50% with a sample of 1000. Which is to say, if you get a poll of 1000 people, which gives a party 50%, then there's a 95% chance that their real result (if you asked everyone) is between 46.9% and 53.1%. You can also look at this in reverse: consider the population. Imagine if you asked everyone, and out of the say, 2,000,000 people, 50% said: 'I'm voting National'.

If, instead of asking everyone, you select 1000 of them, its possible, but very unlikely, that you will just happen to speak with 1000 National supporters. More likely, you'll speak to broadish cross-section, reasonably close to the 50% true figure. Out of all the possible groups of 1000 (could be person 1, person 2, person 3 ... person 1000; or 1, 2, 3 ... 999, 1001; etc.) and there are a lot of them (many many more than there are atoms in the universe) in 95% of all those possible groups you will find that there are between 469 and 531 National supporters.

When you're talking about a group with less support than 50%, the 95% margin is smaller in absolute terms. If the real level of support for New Zealand First (i.e. asking all 2,000,000 people) is 5.7%, if you got really lucky, then your group of 1000 would have 57 people who said: 'I would vote for NZ First', but chances are, you won't: then in 95% of all the possible groups of 1000 people, between ~43 and ~71 people will say I support NZF.

The margin of error with a sample of 1000 at 50% support is +/- ~3.1%. The margin of error at 5.7% support is +/- ~1.44% (I rounded to 1.4% in my calculation). This is the point I was making - don't use the margin of error (3.1%) that only applies to a result of 50%, when talking about a result of 5.7%. If you are going to talk about the range of likely results, then you should use the margin of error that applies to that result. 

by Andrew Geddis on February 06, 2014
Andrew Geddis

And I think we all know what the moral of this comment thread is ... statistics is hard for little brains, and don't fuck about with the Edge.

Seriously, but ... kind of makes even the boiler plate "the margin of error was 3.1%" statement a bit useless to all but the most cognescienti.

by Tim Watkin on February 06, 2014
Tim Watkin

It was a noble effort Peter, though a little like the tilting at windmills metaphor in the post! My heads hurts, but they're useful comments.

by Graeme Edgeler on February 07, 2014
Graeme Edgeler

Seriously, but ... kind of makes even the boiler plate "the margin of error was 3.1%" statement a bit useless to all but the most cognescienti.

I have long thought that it would be better for them to report the results as a range. In some respects they can, and probably do, already do this. Using a thickish line on the graph over time is a good way of representing the uncertainty. It could work for seats too, this poll shows National has a 95% chance of getting between 55 and 58 seats, etc. The false precision we are presented with seems rather non-journalistic.

Another alternative, particularly useful in a case like New Zealand First would for them to take their result and turn it into a percentage chance. The poll might give New Zealand First an 80% chance of making it over the threshold, or Labour+Greens may have a 35% chance of getting more seats overall than National.

And I think we all know what the moral of this comment thread is ... statistics is hard for little brains, and don't fuck about with the Edge.

It's times like these that I wish I had studied statistics beyond sixth form. I know enough to know where to look for some of the answers, but the calculation of things like the chance that NZF will get over 5% based on a single poll result is beyond me.

by Graeme Edgeler on February 07, 2014
Graeme Edgeler

Seriously, but ... kind of makes even the boiler plate "the margin of error was 3.1%" statement a bit useless to all but the most cognescienti.

I'd have said it was worse than that, it's not just useless, it's harmful. Anyone who understands the information doesn't benefit from being given it (the sample size is sufficient), and for most others it actively misleads them about the levels of uncertainty that are actually present.

by Pete George on February 07, 2014
Pete George

"for most others it actively misleads them about the levels of uncertainty that are actually present."

Can you explain all this to Patrick Gower et al? Like, what is the % chance that his one poll is a gamechanger and Winston rules?

by Andrew Robertson on February 10, 2014
Andrew Robertson

Dammit I need to check this blog more frequently! Always miss these discussions.

The margin of error usually cited assums a random probability sample. All polls cite this MoE because it give some comparable indication of the price paid for not taking a census. Interestingly though, only a couple of the polls in NZ actually seem to use this sampling approach.

NZ public poll methods grid | Grumpollie
http://grumpollie.wordpress.com/nz-public-poll-methods-grid/

Post new comment

You must be logged in to post a comment.