So far as the opinion polls are concerned, the past three months have been a period of substantial recovery for the Abbott government. The BludgerTrack poll aggregate records the Coalition increasing its two-party support since the start of June by nearly 3% to its current level of 49.2% — taking it, as Montgomery Burns might have put it, from a potential Chernobyl to a mere Three Mile Island.
As ever, though, the signal that can be discerned through the fog of scattered poll results has been apt to get lost amid the noise. For those of a mind to see it that way, the present picture is of Morgan telling the true story in having Labor’s two-party vote northwards of 55%, while Newspoll and Galaxy have distorted the picture at the behest of their publishers by as much as four points in favour of the Coalition.
Happily, what’s known in the polling trade as “house bias” is a subject on which we need not take social media and blog commentary at its word. In consolidating all published polling since the 2010 election, the trends recorded by BludgerTrack provide a more-or-less objective benchmark from which the peculiarities of the individual pollsters can be measured. This is illustrated in the table below, which shows the extent to which six pollsters have deviated from the trend on average over the past four years.
By this reckoning, out of the five polling agencies still in business, the friendliest to Labor by some distance has been the one whose chairman openly identifies with the Liberal Party, while the pollster that emerged from the union movement has been the most favourable to the Coalition.
Morgan’s lean to Labor of around 2% relative to most other pollsters is the biggest of any of the observed idiosyncrasies, and its influence on the overall trend result has a lot to do with Newspoll, Essential and Galaxy being recorded as slightly favouring the Coalition.
Pollsters other than Morgan show variability on the primary vote, with Newspoll indeed showing a tendency to understate support for Labor, but in almost every case the differences wash out on two-party preferred. Newspoll is in fact notable for coming in low for both major parties, the reverse being true of Essential Research.
There are some who would respond that this exercise overlooks a strategy of unscrupulous pollsters in rationing their credibility by running clean results during quiet periods, only to cook the books on special occasions such as election campaigns. However, observation of changes in the trend over time suggests that the truth is, as always, quite a bit more prosaic than that.
The biggest shift over the past four years has been a moderation in Morgan’s bias to Labor since the period after the 2010 election — from 3% relative to Newspoll to around half that now. This reflects a change in its methodology since March 2013, when it began to supplement its face-to-face polling with an SMS component (and also, for a time, online polling). When the poll was conducted face-to-face only, Morgan overestimated Labor’s two-party preferred vote in final pre-election polls by 1.9% in 2010, 3.8% in 2007, 4.2% in 2004 and 5.5% in 2001, causing it to pick the wrong winner on the latter two occasions.
Morgan continues to report headline results that are substantially more favourable to Labor than the results of any other pollster, but as was covered in detail here a month ago, this is as much to do with its unusual method of determining its two-party preferred result as its primary vote numbers.
The other pollster that has proved inconstant in its biases has been Essential Research, which is distinguished by its slow response to upheavals recorded by other pollsters. This has lately been manifest in a relative lean to the Coalition in a period where the rest of the pack recorded a big post-budget move to Labor. Nielsen, which shut up shop a little over a month ago, seemed to lose a long-standing Coalition bias late in its life, and instead became the best poll for the Greens.
As for Newspoll, it’s been largely set in its ways since the 2007 election, when it changed its questionnaire to include the Greens up front in the initial question on voting intention, rather than relegate it to a secondary list of minor parties offered to those who choose “other” over Labor and the Coalition. This appears to cause it to over-report Greens support among respondents whose natural inclination is to waver between the Greens and Labor, where previously it had been under-reported.
Newspoll will again be along to make its regular contribution to the background noise of Australian politics on Sunday or Monday evening, to be followed by the usual frenzy of over-analysis. When that happens, keep its historic peculiarities in mind — but even more so, keep even more in mind its 3% margin of error, and the limitations of any individual result as a guide to where the parties truly stand.