Reproduced from Crikey, 10 April 2007

14. Forget the election contest, look at the pollsters

Peter Brent of Mumble Politics writes:

There's movement on the Australian political polling front, with repercussions for the big federal year ahead.

Last Tuesday, Newspoll CEO Sol Lebovic packed his pens and paperclips and left the building forever. Since starting the company in 1985, Lebovic had built it into possibly our most famous political pollster (the other contender being Roy Morgan Research).

Newspoll, half-owned by News Limited, appears regularly in The Australian, and interestingly, given a hint of boardroom shenanigans involved, that seems to have been the only newspaper to cover the departure.

Another recent event was young tearaway market researcher Galaxy finally launching its website - see the brash public polling page.

The connection is that the guy who runs Galaxy's political polls, David Briggs, used to work at Newspoll.

Galaxy first emerged before the 2004 federal election, and since then, across one federal and three state contests, its final two party preferred numbers have been, on average, closest of all pollsters to the actual results. Why?

Well, the pollsters have different ways of doing things - sampling and weighting - that are too complicated for us to fathom. But other variables are discernible to the naked eye.

Preference allocation is one. At last month's NSW election, Newspoll used the tired old preference strategy it abandoned at the federal level after it caused grief at the 2004 poll. It asked for second preferences only and extrapolated from there to get the two party preferred vote. This is one of the worst approaches around, even more so under NSW’s optional - rather than compulsory - preferential voting system.

Newspoll’s final NSW survey overstated Labor's 2PP by more than four percent.

At Galaxy, meanwhile, they allocated preferences per minor party/independents, estimating likely flows and exhaustion. Galaxy was easily the closest, within one point of the actual result.

Preference allocation methodology can make as much as a two percent difference to the headline 2PP.

There are other differences. As far as I know, Newspoll is the only major outfit not to include the Greens (along with Labor, Liberal and National) in their voting intention question. The others do (and most also had Family First for last year’s Victorian election). This, too, impacts on the two party preferred figure. (There are, it must be said, legitimate arguments both for and against prompting of support for minor parties.)

And then there's the "secret formula" stuff - sampling and weighting – which impacts on the primary vote data.

With ever-growing mobile phone use, and a general public increasingly reticent towards telephonic probing, today’s pollsters can’t rest on past practices.

The way Briggs tells it, his suggestions for fine-tuning met with resistance at his old firm, but he now has the freedom to implement his crazy ideas.