At some point in the previous fortnight, I resolved to abandon my practice of reporting Morgan face-to-face results by highlighting the previous-election measure of two-party preferred, rather than the respondent-allocated method favoured by Morgan. For those not familiar with this issue, there are two methods pollsters can use to determine how minor party and independent voters would allocate preferences: asking them, or making a distribution according to how voters for the relevant parties divided between Labor and the Coalition at the previous election. While the former would appear to pass the common sense test, it is in fact the latter which has consistently given more reliable results. It would seem that asking respondents places them in a position which is not replicated in the polling booth, where many follow how-to-vote cards or otherwise contrive to avoid engaging mentally with how they order their preferences the significance of which many would not appreciate. As a result, the previous-election method has come to be favoured by every company other than Morgan, with Newspoll having adopted the practice after its final pre-election poll in 2004 was broadly accurate with regard to the primary vote, but awry on two-party preferred. My policy of favouring the previous-election measure was adopted for the sake of consistency in a period when Morgan seemed to be jumping around from one method to the other. However, Morgan has since settled upon the respondent-allocated measure, despite its poor track record. Highlighting a different result from Morgan’s was thus creating confusion, notwithstanding that I believe it to be the superior method.
Today Morgan has published its first face-to-face poll results since I made this decision, and they have rather awkwardly produced the biggest divide yet between the two measures. The respondent-allocated figure highlighted by Morgan has the Coalition with a thumping 56.5-43.5 lead, much the same as the 56-44 result from the previous poll (which was conducted on the weekend of July 9/10, with the carbon tax announcement coming on the latter date; the current poll combines the weekends of July 16/17 and July 23/24). However, the previous-election method gives the government a far happier result of 53-47. The primary votes are in fact little changed: Labor is up a point to 34.5 per cent, the Coalition down one to 47 per cent and the Greens up half to 12 per cent. What has happened is an exacerbation of the recent trend where Labor’s share of minor party and independent preferences has gone in the same direction as its level of direct support. However, I remain unconvinced that this will be replicated on polling day. The Morgan figures for non-major parties are essentially identical to those recorded from the previous election, when the Greens polled 11.8 per cent and others 6.6 per cent (compared with 12 per cent and 6.5 per cent). If we take the Morgan respondent-allocated figure at face value, this suggests Labor’s share of all non-major party preferences has slumped from 66 per cent to 49 per cent. Since nearly two thirds of these voters are Greens supporters, this seems very hard to credit. The question nonetheless remains as to why poll respondents who favour minor parties and independents have become so much less likely to nominate Labor than the Coalition, to which I don’t have an answer but keep in mind that the solid swing against Labor in the 2010 election was not reflected in the share of preferences it received.
Nonetheless, the record should note that Morgan has published a figure of 56.5-43.5, and has done so using a method that other pollsters were happy with until about half a decade ago. Equally though we should note that the alternative and apparently more reliable method has produced a result solidly more favourable to Labor than other pollsters have been producing of late. This brings us back to the old issue of the strong lean to Labor which has traditionally been evident in Morgan face-to-face polls, which the recent anti-Labor trend of respondent-allocated preferences has obscured. This point is illustrated by a chart I produced last month showing how Morgan face-to-face results (along with Essential and Nielsen) have differed from Newspoll since the start of 2009. As you can see from the two measures provided by Morgan, the issue of which preference method used was largely academic until the start of this year, when the present gap began to open. On this basis Morgan had become less favourable to Labor than Newspoll using the respondent-allocated method; its previous-election results remained more favourable, though only to the tune of about one point rather than the traditional three.