I heard it said recently that Galaxy had a strategy of skimping on sample sizes for early campaign polls and then going to town at the end of the campaign, in the hope that their final polls will prove more accurate than the competition – which, by and large, they have done. Sure enough, today’s News Limited tabloids have unloaded a giant survey of 4000 voters covering four marginal electorates in each mainland state. The graphic as published in Perth’s Sunday Times can be viewed here. The surprise is the mild 5 per cent swing across Bonner, Herbert, Longman and Moreton, pointing to a disappointing result for Labor in Queensland. However, it has been widely reported that Labor is on course for bigger swings in the north of the state than in Brisbane, which is over-represented in this sample of seats (though this was also true of Newspoll). Newspoll and Galaxy are also at odds over Victoria, respectively pointing to swings of 8 per cent (a likely gain of five seats) and 4.5 per cent (a likely gain of zero seats). Three of the four seats sampled were the same, the fourth being McMillan in Newspoll’s case and McEwen in Galaxy’s. Where Newspoll surveyed the four most marginal seats in South Australia, Galaxy has ditched Kingston as a lost cause to take on Sturt. One can only speculate how much that accounts for the difference between the two results: a 4.5 per cent swing in Galaxy’s case and 8 per cent in Newspoll’s. It should be noted that two Sturt polls appeared in the week suggesting Galaxy’s swing is nearer the mark than Newspoll’s. Perversely, the state where the two agencies picked the most differing sample of seats, New South Wales, is the one where they produced the most similar results – a 7 per cent swing in Newspoll’s case, slightly higher in Galaxy’s.
Then there is Western Australia, which Galaxy surveyed and Newspoll did not. Here Galaxy points to a swing of barely 1 per cent, less than would be needed for Labor to win Stirling and Hasluck, and perhaps not enough to hold Cowan given the loss of Graham Edwards’ personal vote. For more on Western Australia, see the post below on the Westpoll survey, to which I have now added a graphic from The West Australian. It gives Labor a slightly bigger swing of 3.5 per cent.
UPDATE: Chris Hammer of The Bulletin lambasts Galaxy for not waiting until the campaign launches were done before conducting its poll.
Ther is no intentional bias – the is too much credibility at stake. But there is still something suspect about this polling. I think it showed the majors getting primaries of 42 and 43 – I’ve forgotten which was higher.
This alone looks doubtful. The coalition haven’t been closer than 5 points on the primary – whether marginal or state-wide (except WA). Nothing seems to have changed to suggest a 4 or 5 point fall in the Labor primary. MOE? Maybe one of the psephs could tell us more.
Has anybody done a wrap yet of whom the Sunday papers are endorsing?
The Sunday Mail is SA is endorsing the Coalition. Which isn’t really surprising given recent form, but I’m curious as to what to other states have done?
If there is a wrap somewhere and I have missed it, a link would be most appreciated. One hopes Crikey will remove themselves from the Labor bandwagon tomorrow and post a decent overview, but I am a tad impatient to wait for that.
I read some posters say suggesting the ALP will win over 90 seats as dream stuff and as someone who thinks the ALP could win up to 96 seats know that is unlikely but all year the polls have been showing the ALP heading for a swing greater than what Howard received in 1996.
While its true the marginal seat polling has appeared less then what the headline national polls are saying, but even these marginal seats polls are showing a swing of approx 5% and the evidence is that safer seats are swing by more.
The ALP have been polling around 48% all year, with the Liberals polling around 40% this appears similar to what has been occurring at state level for the past 10 years.
I think we can take it that the result will make the federal map look similar to the state maps this is why I’m willing to suggest seats like Casey may be in play, Casey on paper shouldn’t be mentioned until the call of the board.
I also think If this is right then seats like Hughes may fall before Wentworth.
Putting this poll to one side. The way the pollsters are getting the 2PP is wrong. At the last election there was a protest vote against Latham by Labor supporters. This lead to a very large number of minor party votes which then went back to Labor. At this election Rudd is very popular and unlikely to loose many vote to minor parties. This will therefore lead to less preferences coming back.
Sunday paper endorsements : The Sydney Sun Herald editorial is headed ‘Labor gets our approval’ – mainly on the strength of its education policy.
In other words, the overall 2PP of (I am told) 53/47 ALP/Coalition, is probably accurate to within 2%. But the data on any individual seat is bollocks?
netvegetable @ 106
Is there any data about individual seats? The 2PP vote is useless because the seats start out bias however the swing is likely to be correct for each state with whatever the MOE is for 800 people.
My wife was polled by Galaxy in Stirling yesterday afternoon (Saturday) so presumably there will be another “marginal” result published during the week. That could be the accurate poll to come out just before the election. She said she would be voting ALP but I think I can convince her to vote Green 1, ALP 2.
They wanted someone 18-34, we are both 60+ but the quizzed her anyway. Does anyone know if the pollsters cull out anyone who is outside their demographic target.
They also asked questions such as “do you know who sets Interest rates”. Apparently the girl was chuffed to find someone who knew it was the Reserve Bank!
Stephen @107,
the Galaxy data on state swings, based on just 4 seats, is likely to be much less reliable than that obtainable from averaging the state breakdowns of the major national polls.
4000 voters over 24 marginals. Hmmm. I am no statistician, but even with elementary stats of means and SDs, we can see there is a problem. Just taking the mean over 4000 individuals obviously defeats the purpose, as we are interested in the result for each seat. Therefore we have 24 numbers to work with, not 4000 as such. So where are they, and what is the variation amongst them? I have seen no published list of the 24 measures. With such small sample sizes contributing to each measure, the swings could vary hugely, say from 3% to 10% over the 24. So the standard of error will be high, as stated in an earlier post, and the result useless for present purposes.
Since what is relevant is the result each individual seat, unless each is sampled adequately (as did not happen) combining is no help, on my reading of the situation, ie, GIGO. Someone with expertise might have a different take on this. As I said, elementary.
There are also seats in all states that actually have local issues in the mix as well. Last Federal election for example, the liberal candidate in Corangamite in Victoria got zero/zilch/zip swing to him. This despite nearly every other seat having some sort of swing to the liberals as part of the anti-Latham backlash.
The liberal candidate has held the seat for 28 years and turned 70 several weeks ago. He first held the seat with a 12% margin. At the last election he won with a margin of 5.3%. The incumbent is not very popular, especially amongst Gen-Y and Gen-X voters. He’s probably 50/50 with the Boomers.
The seat has also had an influx of 7000 first time voters in the seat. It also has a large number of first time voters full stop!
If labour get some-type of “Ruddslide” going, expect this seat to fall! The incumbent will definately not be able to use his own personnel popularity to save himself!