The Courier-Mail reports a Galaxy poll conducted over the previous two days shows the Liberal National Party with a narrow lead on the primary vote (43 per cent to 42 per cent), and a dead heat on two-party preferred. There is a popular view among Labor partisans that Galaxy is not so much a market research outfit as a Virtucon-style empire of evil, whose first order of business at all times is to promote the political interests of the Coalition. This is partly because it consistently showed federal Labor with slightly less gigantic leads than its rivals during 2007, notwithstanding that those numbers ended up being closer to the actual result. No doubt this latest finding will be grist to the partisans’ mill. Not for the first time, I feel compelled to lay out Galaxy’s record in final pre-election polls going back to the organisation’s foundation in 2004. This list shows Galaxy’s final two-party results followed by the actual election results in brackets:
|WA 2008:||51-49 (52-48 approx)|
|Federal 2007:||52-48 (52.7-47.3)|
|NSW 2007:||53-47 (52.3-47.7)|
|Victoria 2006:||55-45 (54.4-45.6)|
|Queensland 2006:||56.5-43.5 (54.9-45.1)|
|Federal 2004:||52-48 (52.7-47.3)|
No doubt the naysayers will point to Galaxy’s first poll going into the 2006 campaign (which was in fact conducted before the election was announced, unlike the current poll) which had the Coalition leading Labor 51-49. That result probably did flatter the Coalition even at the time, but the crucial fact was that their campaign had yet to be sent off the rails by the Liberals’ confusion as to who would be the premier if the Coalition was elected. There is no real reason to doubt that Galaxy’s latest survey was conducted in a competent and professional fashion, and that it should accordingly be analysed objectively once we have a sample size from which to calculate the margin of error. That isn’t yet the case, but the Courier-Mail will presumably be forthcoming with more information shortly. Past experience suggests it would have been 800, in which case the MoE would be about 3.5 per cent.
UPDATE by Possum:
As we don’t know the sample size of the Galaxy poll, in order to look at what it means we have to assume a size – so let’s go a sample of 1000. When the info is released later I’ll rerun the charts – but unless the poll is below 700 in size, it won’t really make that much difference.
A new morning brings us the sample size – 800, so we’ll update the figures and charts below.
Remembering back to our key chart that tells us the probability of the ALP reaching 45 seats with a given election result.
If this poll were to come literally true on election day, the ALP would have a 94% probability of gaining the 45 seats it needs to form government. However, since this poll has known uncertainty built into it – the sampling error – we have to adjust the probability curve to accommodate this additional uncertainty.
So what is the Galaxy poll telling us after we do all that? It’s telling us that if the true state of public opinion is within the margin of error of the Galaxy poll (and that MoE is distributed normally), there is currently an 83.2% probability of the ALP gaining the 45 seats it needs to form government. The simulation results look like this.
To summarise: If the election result was 50/50 on the TPP, then the ALP has a 94% probability of retaining government, if the result on election day is somewhere within the Margin of Error of this poll (assuming a sample size of 1000) of 3.5%, then the probability of the ALP retaining government is 83.2%.
UPDATE 2 (Possum):
While we are all waiting for the details of the poll – particularly the sample size – here’s a quick chart that shows the Margin of Error for a 50/50 polling result, for sample sizes between 0 and 2000.
UPDATE 3 (William):
Further reporting by the Courier-Mail confirms that the sample was 800. Bligh continues to lead Springborg as better premier 50 per cent to 37 per cent, marking little change from Galaxy’s last survey in September, at which Labor led 52-48.
UPDATE 4 (Possum)
All the charts and probabilities above have now been updated for the 800 sample size that comes with an MoE of 3.5%. Also note that William makes a very fine Nostradamus on these things!