Public opinion polls have dominated the news everyday throughout the 2010 Australian Federal election, but how accurate and ethical are they? Amy Yang reports.

Newspoll conducts telephone interviews to pick election winners. Image: sianparkles

Newspoll’s 25-year track record for picking election winners comes down to an undisclosed sampling method.

“We don’t actually use pure random sampling we use a special kind of structured sampling,” said Martin O’Shannessy, chief executive officer of Newspoll.

As part of its unique sampling method Newspoll, owned by News Limited, conducts telephone interviews on a weekly basis. It can contact households up to three times for a response.

“The average absolute error between real world election events and Newspoll over the last 25 years has been 1.5 percent, which is actually about half of what you would expect in most purely random telephone surveys.”

“Everything Newspoll does in this area is published exclusively in The Australian…because this is paid for by The Australian, it is first published in The Australian,” said O’ Shannessy.

News Limited also sponsors Galaxy polls while The Sydney Morning Herald sponsors Nielsen.

Donna Hills, Chief statistician at Data Analysis Australia said polls are fairly constructed.

“I believe that every election they [political pollsters] have got it right is when the liberal party has won…it’s not a true random selection”

Hills said that when analysing poll results the percentages used to validate political commentaries were often misleading and inaccurate.

“When you get a number from a poll you say, okay 40 per cent of people say this, however when you expand this figure to the population it’s actually between 35 and 45 per cent. So given that with election polls the numbers are often 48 per cent versus 52 per cent, add to that confidence intervals, and chances are there is no statistical difference between those two numbers,” said Hill.

The Australia Press Council (APC) sets an ethical guideline for how polls should be reported on by newspapers and magazines. Those guidelines include outlining the sample selection and data collection procedures of a survey and clearly stating any form of sponsorship or affiliations.

“The report should as far as possible include in detail the identity of a sponsor of the survey, the exact wording of the questions asked, the population for which the sample is drawn, sample size, method of sampling, which results were based on only part of the sample, and the base number of which these percentages were arrived at,” said Herman.

A distinction needs to be made between statistical surveys and those that are ring-in or internet survey polls.

Herman said that how a poll is reported on however is a very different to how that poll is analysed.

“I mean, it’s quite open to anyone analysing the data to place the emphasis on preferred prime minister as opposed to the gross figures on party preference” said Herman.

At Roy Morgan, polls are conducted face to face, via telephone interviews and online.

Gary Morgan, Chief pollster at Roy Morgan said the non-response percentage for telephone interviews is 60 per cent compared to 30 to 40 per cent for face to face interviews. But non-response can be due to a number of reasons including general unavailability and disconnected lines.

Roy Morgan does not provide data exclusively to any one news outlet, however it has been referred to on various occasions. Mr Morgan said the TV stations often misrepresent their data.

“Television stations call our polls Newspoll rather than the Morgan Poll,” said Mr Morgan.

Mr Morgan said misrepresenting data is neither in the best interests of the polling organisation nor the news outlet.

“Newspoll and Nielsen is made up of only 10 per cent in political polling, so it’s in their best interest to report correctly because more is at stake,” said Mr Herman.

So how should people engage with the results provided by political opinion polls?

Hills suggests that readers need to be aware of the purpose of the article, and who that piece of data might be targeting. Likewise, looking out for how many people participated in the survey compared to how many people were actually asked is necessary towards understanding the scope of the results produced.

“If they did a telephone call and they called 2000 people to get 100 responses, then what about the 1900 who refused to do the survey? You have no way of knowing whether that 100 truly represent the 1900 who don’t answer,” said Hills.