"Ultimately, the impact of polls is closely linked to the media that carry them."And there are more sources of statistical noise in polls than usually acknowledged. Most political tragics would be aware of sampling error -- around 3% for a poll with a sample of around 1000, and 2.2% for polls of around 2000. But Essential Research’s pollster Andrew Bunn points out "there are sources of error which are probably more significant than the theoretical sampling error, including question wording, question order, weighting, late swing (for election forecasts), response errors (or false reporting by respondents), treatment of the 'don't knows' (we assume they give answers similar to those who respond)". In Australia, the relationship between the media and pollsters is even closer than it is elsewhere. Newspoll is half-owned by News Limited (the other owner is global marketing firm Millward Brown). While News Ltd tabloids tend to use Galaxy polls (established by former Newspoll executive David Briggs), Newspoll is without doubt the most influential poll, partly because of its accuracy (it almost perfectly predicted the 2010 two-party preferred result, for example), partly because other media, and especially the ABC, accord it influence, but mostly because of the effort invested by The Australian in attributing significance to its results even when it is negligible. Newspoll allows an outlet openly committed to promoting right-wing political agendas to significantly influence the media cycle and national agenda. Newspoll is also fortnightly, compared with the AC Nielsen poll (Nielsen is a global marketing firm) carried by Fairfax, which is at best monthly. Newspoll thus provides a better assessment of trends; Fairfax journalists are often required to explain apparently dramatic shifts in voter sentiment that have as much to do with the gaps between polls as actual voter behaviour. However, because of the media platform to which it is linked, Nielsen, too is influential, particularly given Fairfax will run the results across the Sydney Morning Herald, The Age and The Australian Financial Review, meaning the same poll is dissected repeatedly by different journalists. In comparison, Essential Research, which Crikey carries, conducts a weekly online poll (Newspoll and Nielsen are phone polls), but averages its voting intention results over two weeks, giving it around 2000 respondents and further smoothing out trends; Essential was the most accurate of polls in predicting party primary votes in 2010. Roy Morgan polling does both weekly phone polls and face-to-face polls, and keeps track of them separately; both Morgan and, especially, Newspoll, provide excellent interactive online tools for looking at poll results over an extended period. "Australian polls have an excellent record on elections," Bunn told The Power Index. "Polling is certainly much easier in Australia due to compulsory voting. We don't have to try to estimate the voter turnout, which can be very difficult to estimate -- it can be significantly affected by weather, for example." He offers some simple advice to keep polls in perspective. "Sample size is still important. You'd have to be careful of samples of only a few hundred ... Also, who has sponsored the poll -- if it was done for a political party or other interest group, you'd have to be a bit wary. The other thing I'm a bit wary of is large short-term shifts between polls -- given that a shift of 4% (for example) means that about half a million people have changed their mind, which over a few weeks is not really credible," he said. Ultimately, the impact of polls is closely linked to the media that carry them. It's not polls and pollsters that wield direct influence over voters; it's the media outlets that use them.
The Power Index: election deciders, the pollsters at #10
Polling will definitely influence the election outcome on September 14 -- but not necessarily in the way you think. Evidence that polling directly influences how people vote is mixed -- especially in Australia where nobody wants to be the front-runner.