From Crikey 27 August 2007

12. Pooling the polls: can we learn anything new?

Peter Brent of Mumble Elections writes:

With today’s Galaxy poll in News Ltd tabloids, it's a fine time to mix the polls again. For those who missed it last week, here’s a rundown of how I’m doing it.

The Galaxy, taken over the weekend, gave Labor 57% two party preferred to the Coalition’s 43%. But my estimate of likely preference flows (from Galaxy’s rounded primary numbers) puts it closer to 56 to 44. [ Update: I now have the Galaxy data to one decimal place; it still comes in, for me, closer to 56 for Labor than 57.]

A little surprising, as I believe my strategy is similar to Galaxy’s, and in the past if anything their two party preferred numbers have been a little kind to the government.

Anyway, here is the aggregated graph for the fortnights ending 30 June to 26 August. It’s not a straight out averaging, but weights for both sample size and how recent the poll was. So today’s Galaxy counts for more than the polls before it. Two party preferreds are my estimates from pollsters’ primary data.

Commentators like to pick apart each survey and try to identify the "event" that moved the polls one way or another. But that’s nearly always pointless. As tempting as it is to postulate that Kevin Rudd’s New York adventure gave him a spike, it might simply be that Galaxy interviewed a few too many Labor voters on on the weekend.

Or they may indeed be measuring a further shift to Labor, but for reasons our feeble minds can't identify. Or maybe it's due to something that happened months ago that has been slow-burning in the back of the collective psyche.

All reasons to look not just at individual polls, but at trends. The polls indicate the government is still unable to extract itself from dire electoral doo-doo.

You can read all about the methodology here.