Recovering from tech failure yesterday, we have a basket of new polls including a Nielsen, a Morgan phone poll and two sets of marginal seat polling – one from Galaxy and one from Newspoll.
First up, the all pollster chart:
Every pollster has shown an increase for Labor between late July and the first two weeks of August except for Essential – however, they run a two week rolling sample so a lag like that is to be expected. Also worth noting is the continuing peculiar similarity of the 2004 vs 2010 campaign:
The Nielsen from yesterday in the Fairfax press was in the field between August 10-12 and ran from a sample of 1346, giving us an MoE that maxes out around the 2.7% mark. The vote estimates come in like this:
As we can see, the big mover appeared to be men – jumping 8 points for Labor and making up just about the entirety of their 4 point total primary vote gain. That would be surprising if it actually occurred with that magnitude, so keep a eye out on whether it’s repeated by anyone else over the next few days.
On the approval ratings and preferred PM, the results come in as:
At the moment, we’re seeing a Labor vote that is a little less than where we would historically expect it to be considering the levels coming in on approval ratings and preferred PM (or satisfaction ratings and Better PM with Newspoll). That’s not to say anything about the vote or the polls, but rather just noting that this election is a little odd in more ways than one.
Next up, we move on to the Galaxy marginal seat polling – which the media headlines and reports have sort of made a rather chunky balls up of, reporting it as some sort of weird 51/49 Coalition lead at the national level.
As Antony points out, you can’t just take the average of state results and call it a national average – it needs to be weighted first (for obvious reasons such as SA having around a quarter of the population of NSW). So saying, we shouldn’t be extrapolating marginal seat polling to derive national estimates anyway – we have proper national polls that are specifically designed for that.
However – if one felt some overwhelming need to do such a thing – the result actually comes out with an ALP two party preferred of around 51/49 to Labor at the national level, pretty much the opposite of what has been reported. As an exercise in passing nerdiness, I ran the Galaxy results through the election simulation and came up with the same figure as Antony using a completely different methodology.
Galaxy wasn’t the only marginal seat polling we’ve seen of late, as Newspoll via The Oz yesterday also produced some marginal seat polling of its own.
Galaxy polled 200 people in each of 4 marginal seats for the 5 States of NSW, Victoria, Queensland, WA and SA – for a grand total sample of 4000 people from 20 seats. But this isn’t really a sample of 4000, it’s 5 separate State based samples of 800 a pop. You can see the full Galaxy results over here.
Newspoll on the other hand polled 3351 people across 17 marginal electorates in three states –6 in NSW, 3 in Victoria and 8 in Qld. The NSW sample was 1202, the Victorian sample was 605 and the Qld sample was 1544 and the full results of that Newspoll can be seen over here.
Tallying up the results at the state level and comparing the two, we get:
What we see is that the Newspoll was more favourable to Labor than Galaxy was across the three states where their respective polling overlapped, by between 1.1 and 4.6%. On the seat level of that overlap, all 4 of the seats Galaxy polled in NSW were in the 6 NSW seats that Newspoll sampled, 3 of the seats Galaxy polled in Qld were in the 8 Qld seats that Newspoll sampled and 2 of the seats Galaxy polled in Victoria were in the 3 Victorian seats that Newspoll sampled. We can break that down into an easy graphic to see who did what here:
If we pool the results of both polls and weight by sample size, to get an aggregated figure for the state level for NSW, Victoria and Qld, we get:
Which is pretty much what we’re seeing at the state breakdowns from the national polling. In fact, if we combine all the phone poll polling I can find for the August 8-15 period, look at the approximate state level results and compare it to the marginal seat results we get:
The results are the same for all intents and purposes – and we wouldn’t really expect it to be otherwise. Seats swing with a mean of X and a standard deviation of Y for any given state, where the state swing is the best predictor of the result we have for any given seat. Marginal seats aren’t particularly magical – the behave like everything else – it’s just that parties put more resources into them and the fight is a little tougher on the ground because they are the seats that generally change hands when a government does. But their marginality doesn’t make them behave any differently from any other random bunch of seats further up or down the pendulum, it’s just that the swings we see in most elections are large enough to make those marginal seats the ones to fall.
So the aggregated marginal seat polling tells us no more about what is happening in any particular seat than the large national polling breakdowns do – the expected variation in the swing between 4 marginal seats isn’t fundamentally different from the variation we would expect to see from any arbitrarily larger number of aggregated seats at the state level, be they marginal or not. Essentially, these marginal seat polls just verify what we already know, rather than telling us anything particularly new.
So what do we know?
We known that the polling suggests that Labor are picking up a swing in Victoria of between 3 and 6 points, probably picking up a small swing in SA of a point or two, are facing a swing against them in Qld of between 3 and 6 points and a swing against them in NSW of a few points – with WA being a little all over the place. That’s what we knew last week anyway – and we also witnessed Labor reducing the size of the swings against them compared to the 2 weeks of polling that came before it.
The chart at the top tells most of the story, and tomorrow’s Newspoll will let us add a bit more nuance to that when we can combine it with the last two Nielsens and run the simulations and rerun out trend measures.
Meanwhile, the usual Nielsen charts come in like this: