Antony Green takes exception to News Ltd & Sky's psephological howler

This was Sky News last night (14 August 2010):

















This is The Australian and News Ltd this morning (15 August 2010):

THE Coalition is leading the government and is on track to win the 17 seats it needs for an election victory, according to a Galaxy Poll.
On a two-party preferred basis, the Coalition is ahead of the ALP 51.4 per cent to 48.6 per cent, the opinion poll published by News Limited's Sunday newspapers.
News Limited also publishes of The Australian.
The poll of 4000 voters in 20 marginal electorates in five states predicts devastation for Labor in Queensland, where a potential swing of 5.4 per cent against the government could cost it 10 seats.

This is Antony Green's Election Blog, also published early this morning, telling the world that the Australian media is getting it wrong, wrong, wrong:

When I entered the swings into the ABC calculator, I got the following result

http://www.abc.net.au/elections/federal/2010/calculator/?swing=state&national=0.3&nsw=-2.4&vic=1.6&qld=-5.4&wa=-2.1&sa=0&tas=0&act=0&nt=0&retiringfactor=1

This produced a national 2PP figure of 51.1% for Labor, completely at odds with the figure of Coalition 51.4% reported above. Then it dawned on me that whoever calculated 51.4% has made a very basic error of political analysis.

The five 2PP figures for 2007 and 2010 reported above apply only to the four electorates in each state. They are not 2-party preferred figures for each state, and the Galaxy tables have been very precise in setting out the 4-seats per state nature of the poll.

What someone has done is take the five entries in the 2010 2PP column and average them to get a national figure. Wrong. Very wrong.

There are two serious errors committed here. First, the figures are for four electorates, NOT the states. Second, while the state samples are the same size, the state populations are not. To get a National 2PP figure, you need a weighted average of the state swings, NOT a simple average of the survey 2PPs by state.

The method I used in the calculator is the correct way to produce a national 2PP %. If you assume the swings in the four states are uniform in each state, the new state Labor 2PP %'s are NSW 51.3%, VIC 55.9%, QLD 45.0%, SA 52.4% and WA 44.6%.

These produce a National Labor 2PP % from this Galaxy poll of 51%, not 48.6%. Someone has made an absolute howler in trying to turn polls in 20 marginal seats into a national figure.

The Neilsen Poll published by Fairfax this morning and previewed by Lateline last night supports Antony Green's assessment that surveyed voter intentions are now favouring Labor.

Update: Channel 7's Weekend Sunrise was still repeating the wrong two party preferred percentages during news updates - up to and including the 8am recap.

Actual Galaxy Research 8-13 August poll results here showing 2pp % Labor 52% Liberal 48%.