Figures released by The Australian of Newspoll results for 17 marginal seats in Victoria, New South Wales and Queensland tends to confirm the broad Nielsen results in terms of the seats that might be affected.

The Australian has produced tables, but the tables are hard to read because they include seats that weren't polled and they don't include the results from individual seats, only averages across all of them. I don't think you can make the extrapolations that they do from this evidence.

For what it's worth, their tables suggest Labor will win McEwen, Latrobe, Dunkley, McMillan, Aston, Gippsland, Goldstein, Menzies and Casey in Victoria; that the Coalition will win Robertson, Macquarie, Macarthur, and Gilmore in New South Wales and Herbert, Dickson, Longman, Flynn, Dawson, and Forde in Queensland.

Their commentary suggests that the seats changing hands to the Coalition will be Macquarie, Macarthur, Bennelong, Dickson, Flynn, Dawson, Brisbane and Leichardt; with McEwen, Latrobe and Dunkley going in the other direction.

They say these contradictory things because in one case they are applying the average swing across all seats, and in the other they are looking at the results in the individual seats.

To see which seats they actually polled go to the bottom of the table. It also tells you that 3351 voters were polled making an average of 200 a seat. This is too small a figure to make the calls that they do on the seats. The sample error is 6.9% per seat, which is why they have combined them. A sample of 1,200 produces an error of 2.8%.

But while this method might reduce the sampling error one has to ask what have these seats got in common that makes them sufficiently homogenous to warrant this treatment?

In the case of Queensland for example the sample includes seats in Cairns, Townsville, Brisbane and Central Queensland.

Share this article on your favourite social bookmarking sites: