Polls effectively tied with 'others' up

This week's ReachTEL and Newspoll have Others gaining, at Labor's expense in ReachTEL and at the expense of all other parties in Newspoll. Ipsos was the exception this week, with Others unchanged on last fortnight. The gains for Others have delivered the Coalition a slight two party improvement, except in Ipsos. Here is this week's poll table.

The ReachTEL headline figure had the Coalition gaining 2 points, but this used respondent allocated preferences. The primary vote changes implied a 50-50 tie by last election preferences, a 1 point gain for the Coalition. As with ReachTEL, Ipsos' respondent allocated preferences were the same as the previous election method, at 51-49 to Labor.

ReachTEL had total Others at 13.5%, including Nick Xenophon Team (NXT) on 5.0%. In Newspoll, Others were at 15% (up 3 since last fortnight). The linked article says that NXT was at 3%, but Newspoll does not include NXT in the initial party readout, so its support is likely to be underestimated.

A ReachTEL single seat poll of Longman, held by the Liberal Nationals by 6.9%, shows a 50-50 tie. This is a seat without a sophomore surge factor. In addition, Palmer United Party (PUP) won 12.8% in 2013, and their How to Vote cards were hostile to Labor. Without PUP, Labor is likely to do better.

If Others represent a surge for the NXT in SA, the NXT could win some of the 11 SA House seats, and make a hung Parliament more likely. If the high Others vote represents general disillusionment with both major parties, it makes the two party figure unreliable, since many Other voters will not decide which major party to preference until they actually vote.

Once again there is little change in the poll aggregates' Two Party Preferred (2PP) estimates. Kevin Bonham's aggregate is at 50.1% 2PP to Labor, and the Poll Bludger's BludgerTrack was at 50.3% 2PP to Labor prior to Newspoll. Primary votes in BludgerTrack were 40.9% for the Coalition, 35.0% for Labor, 10.7% for the Greens and 3.7% for the NXT.

Leaders' ratings and other polling

In Newspoll, Turnbull's satisfied rating was down one point to 37%, and his dissatisfied rating was up one point to 51%, for a net approval of -14. Shorten's net approval slumped seven points to -19, and he has returned to his position four weeks ago.

In Ipsos, 45% approved of Turnbull's performance (down 3), and 42% disapproved (up 4), for a net approval of +3. Turnbull's Ipsos ratings have been far higher than in other polls. Shorten's net approval was steady at -6. 55% thought the Coaltiion would win the election, with only 22% backing Labor.

Turnbull leads as better PM by 56-44 in ReachTEL (55-45 last week), by 49-31 in Ipsos (47-30 last fortnight) and by 45-30 in Newspoll (46-31 last fortnight).

In ReachTEL, 52% thought that Shorten was performing better in the election campaign, while 48% thought Turnbull better.

Why do many polls use only the previous election method for their 2PP estimates?

There are two widely used methods to estimate the two party vote: last election preferences and respondent allocation. But Newspoll, Galaxy and Essential currently use only the last election method.

Respondent allocation, asking voters which way they will allocate their preferences, is intuitively appealing, but it has its drawbacks. Minor party voters can be influenced by their party's How to Vote cards; these cards are distributed at polling places, and poll respondents are not usually aware of their party's cards.

An additional problem is that the number of minor party respondents in a poll is about 20% of the total poll sample. Estimates of minor party preference flows are thus much more prone to error than the overall poll, and this can mean the 2PP estimate is well off, even if the primary votes are accurate.

According to Kevin Bonham, the previous election method has performed well at elections since 1983. If the exact primaries were known, last election methods would have produced errors of more than one point only in 1990 and 2013; in both those elections Labor overperformed.

The worst case of respondent allocation failing occurred in 2004, when most polls greatly overestimated Labor's share of preferences. Galaxy used the previous election method, and was shown to be correct, and Newspoll has since adopted that method.

Although Labor's share of preferences improved greatly at the recent Queensland and NSW elections, these elections used optional preferential voting, under which minor party preferences are more volatile than using compulsory preferential voting.

When Turnbull replaced Abbott as PM, respondent preferences had the Coalition doing about one point better than the previous election. However, in the last month or so, this has flipped, and Labor is now doing a little better on respondent preferences. But the latest Ipsos and ReachTEL show no difference between respondent allocation and previous election methods.

In 2013, PUP handed out pro-Coalition How to Vote cards at all seats it contested. PUP will not be a factor at this election, and the NXT currently says it will hand out open How to Vote cards, with neither major party preferenced. Labor could thus do better than the previous election method suggests, but the previous election method has a strong track record.

Adrian Beaumont, PhD Student, School of Mathematics and Statistics, University of Melbourne

This article is republished from The Conversation under a Creative Commons license. Read the original article.