Friday, 23 August 2013

Australian Federal Election 2013: Robopollster opens door for survey fraud?

A number of Lonergan polls have been conducted during the 2013 Australian Federal Election campaign.

These polls seem to indicate that the Coalition is set to take a number of high profile seats, including the seats of Forde and Griffith in Queensland and Lindsay in New South Wales.

The Guardian Australia 16 August 2013:

A Guardian Lonergan poll taken in the seat Thursday night shows the Liberal sitting member Bert van Manen’s support soaring from the 44% he polled in 2010 to 56% in 2013. Beattie’s primary vote was a dismal 34%, three percentage points lower than the 37% achieved by Labor’s candidate Brett Raguse in 2010.

The Guardian Australia 16 August 2013:

Today Guardian Australia reports on an exclusive poll in the marginal seat of Lindsay, in Sydney's western suburbs. The startling results: the Liberal candidate, Fiona Scott, polled 60% of first preferences, a 17-point improvement over her 43.4% performance in 2010. The Labor incumbent (David Bradbury, the Assistant Treasurer) is looking at a 13-point decline on first preferences: 44.6% in 2010 to the poll's 32%.

The Guardian Australia 22 August 2013:

Kevin Rudd is trailing Liberal rival Bill Glasson in his apparently safe Brisbane seat of Griffith, in alarming news for Labor from the latest Guardian Lonergan poll.
Glasson, who is running an intensive local grassroots campaign, leads Rudd on a two-party preferred basis by 52% to 48%. The poll's margin of error is 4%, but its findings raise the possibility that without a big effort on his home turf, Rudd could become the third prime minister in Australian history to lose his seat, behind John Howard in 2007 and Stanley Bruce in 1929.

However, the polling has an allegedly exploitable flaw – the ability for any individual survey respondent to double dip.


Guardian Australia responds later that day:

Contrary to some comments on Twitter, we keep only one interview per household. We conduct an interview with whoever answers the phone. This methodology alone results in an oversampling of older respondents relative to younger respondents, as older respondents are more likely to answer the phone.
To correct this, at the end of the survey, we ask if there are any younger voters in the household. If there are, we conduct the survey with the younger respondent and discard the response from the older respondent.
This response merely confirms that any individual survey respondent can in the second round deliberately misrepresent their age in an effort to be counted twice in the survey (eg. at the press of a telephone button a sixty year-old man turns into a twenty-three year old woman) and while this may not be successful with regard to a household being counted twice, it would lead to some responses included in final data not actually belonging to the gender/age cohorts to which they are assigned - thus still skewing results. 


No comments: