Tuesday, September 8, 2015

Poll Roundup: The Dog Ate My Recovery

2PP Aggregate: 53.4 to ALP (+0.3 since last week)
ALP would easily win election "held now"

It's over a month now since Bronwyn Bishop resigned as Speaker, more or less ending an entitlements scandal that had provoked a polling blowout of about one and a half points against the Abbott Government.  It might have been expected (and indeed I rather did expect) that this was a blip and that once all the shouting had died down the government's polling would improve.  However this week's results show that it hasn't improved at all, and again we see bad, but not quite panic-station, results for the government continue to hang around. Ahead of the Canning by-election (which is covered on a separate rolling thread) this is not what the government wants to see.

A significant debacle for the government in the last fortnight was the Border Force press release which stated that Border Force intended to stop people more or less randomly for visa checks.  This was apparently never the actual intention but social media outcry resulted in the intended joint operation with Victoria Police having to be cancelled.  This epic stuffup can't have helped the government's latest attempts to repair its image.

Since the last roundup we've had a ReachTEL, a Newspoll, a Morgan and two Essentials.   The ReachTEL was 53:47 to Labor, the Newspoll was 54:46, the Morgan was 55.5:44.5 by last-election preferences (55:45 respondent-allocated), and the Essentials were both 52:48 (Essential again responding slowly or incompletely to a blowout picked up by all other pollsters).   As is fairly often the case the Morgan had primaries that seemed quite a lot less friendly to Labor than the published 2PP.  This time they would normally have given a 2PP of "only" 54.7.  My methods assume (though sometimes I wonder) that Morgan know what they are doing when calculating their 2PPs and that such differences are caused by distribution issues between the states, and perhaps also rounding.

After considering the primaries and house effects I aggregated the ReachTEL at 52.9 to ALP, the Newspoll at 53.9, the Morgan at 54.3 and both Essentials at 52.2. This comes out to a 53.4% 2PP for Labor, the same value as three and four weeks ago.  Here's the smoothed tracking graph:

The Morgan had a new record high Green vote for said pollster of 16.5%.  This is too high, but may well reflect the presence of Green-friendly issues in the news cycle, including Border Force (on which Labor were seen as being too slow to grasp what was going on) and the Syrian refugee and war situations. It's also notable that the new Newspoll continues to be slightly more Labor-friendly than the aggregate, though the average 0.6-point difference is not yet enough for me to introduce a specific house effect for it.

The methods of the robopoll half of the new Newspoll attracted a brickbat from Possum aka Scott Steel this week after he happened to be randomly polled by it.  At this stage I'm not aware of any response from anyone involved to this criticism of the robopoll's methods; if I see any response I will report it (and add it to the Storify link as well).  Whether the comments about Newspoll from a fellow polling figure are accurate or not as a reflection of the general quality of the polling, there is the suggestion that while the Newspoll brand is now run by Galaxy, the mechanics of the polling may in turn be being outsourced to Telereach.  Telereach has also previously conducted mass robopolling for JWS in 2010 (the national 2PP from which was very accurate), and a robopoll of the NT in 2014.

Whatever, it is looking like the new Newspoll is much more stable than the old one, as expected given Galaxy's history of very stable polling.  Whether it is right or not is another question.

Government's Second Birthday

The Abbott Government turned two on Monday.  I cannot remember much of being that age and am not a parent myself, but apparently the age of two is frequently marked by "mood changes, temper tantrums and use of the word "no"."  The Abbott Government might be considered precocious in having gone through such stages well before this age (and even before it was born) but in other respects has been considered a slow developer.  Its second birthday provoked a slew of articles arguing that it had done very little in that time, the authors of such pieces uniformly unconcerned that someone out there might even come to view that as a strength.

In polling terms, this government is clearly the runt of the litter:

The chart shows that at age two, four of the previous seven first-term governments had already gone to and won their first election.  While the Abbott government has trailed in polling for 21 of its 24 months, none of the others trailed for half of their first two years (two did not trail at all), and only the Whitlam government had a comparably bad average 2PP polling performance (mainly down to very bad readings late in 1974).


A few weeks back we saw big rises for ALP leader Bill Shorten in all polls, but this week has him back in the doghouse with a Newspoll netsat of -28 (30-58), compared with Tony Abbott's -33 (30-63).  The combined netsat of both leaders reaches a new low of -61, the twelfth worst ever.  Worse totals were recorded (noting Newspoll's recent methods shift) by Keating-vs-Hewson seven times in mid-late 1993, Keating-vs-Downer twice in December 1994 and Gillard-vs-Abbott in February and July 2012.

The reversion for Shorten is just the seventh time an Opposition Leader has gained 10 or more points one Newspoll only to lose 10 or more points in the next.  The last such cases involved Tony Abbott during the ascension of Julia Gillard and again immediately after the 2010 election.  There have been only three such cases involving Prime Ministers.   Shorten continues to lead Abbott as better Prime Minister in Newspoll (41-37), about the expected outcome for a 54:46 2PP.

Essential's monthly leadership polling bypassed Shorten's Dyson Heydon bounce almost entirely and had him just up slightly to -21 (29-50) with Abbott on -24 (33-57) and Shorten preferred Prime Minister (35-32).  Essential also polled Joe Hockey's approval rating and found him at -22 (30-52), back to about pre-Budget levels.

Other Polling

Newspoll reported mild support ("More are in favour of China free-trade agreement than oppose it") for the China-Australia Free Trade Agreement (43:35), with Coalition voters 64:20 in favour, Greens voters 26:53 against.  Labor voters are reportedly 42:35 in favour, but that doesn't add up without extreme rounding assumptions, so I wonder if the Labor figures were transposed.  If they were then Others come out as roughly 14% in favour, 45% opposed.

Newspoll has also found that only 30% want the Trade Unions Royal Commission to continue with Dyson Heydon in the chair (it would be interesting to know the overlap between those and the 30% who are satisfied with Tony Abbott), but 38% want it to continue without him and 15% want it scrapped entirely.

Essential showed slight improvements for the government on "party trust" to handle issues, but since the baseline is February 2015 when the government was polling terribly, that's not much to write home about.  There were polarised responses to a polarised wording in a question about the NBN.  Voters asked about political alignment tended to find Labor roughly in the centre, and even Coalition supporters were slightly more likely to find their party too right-wing than too left-wing.  I really can't make any sense of 18% of Greens voters thinking the Greens have become more left-wing, save that these are probably the sort of Greens voters who are actually not attracted to the Greens' ideology and say they'll vote for them in disgust with the major parties only.  There was a significant increase in concern about a range of economic issues, even compared to during the February slump.

This week Essential finds opposition to increasing the GST to 15% while cutting income tax, with 27% supporting and 52% opposed - Coalition voters more or less even, other voters against.

The missing ReachTEL same-sex marriage result from late July was finally released and, as expected, the robopolling method produced a weaker but still supportive response compared to other polling.  53.8% in favour, 32.8% against.

A Fabulous Resource

Finally I should draw readers' attention to the recent launch of David Barry's Australian election statistics page.  A massive amount of work has gone into this site, which among zillions of other useful things includes aggregated estimates of 2PP preference flow for all parties at each election back to 1919!  At some stage I intend to use all this to have a comprehensive look at the question of preference flow changes from election to election, especially as this applies to the ever-ongoing respondent-vs-last-election preferences debate.

I intend to have some kind of on-the-night comments about Canning on September 19. The next Poll Roundup will probably be in a fortnight's time.

Update: Refugee Poll (Sep 9)

The Guardian carried reports of a Lonergan poll supposed to show:

* 57:43 support for increasing refugee intake.
* 54:46 support for Operation Sovereign Borders.
* 55:45 saying the Liberal Party's stance on asylum seekers made them less likely as opposed to more likely to vote Coalition.
* 54:46 rating the Coalition's response to the Syrian and Middle East refugee crisis poor as opposed to good.

Notice something here?  The figures always sum to 100.  This suggests that these questions were asked by forced choice, so the voter has to pick one option or the other with no don't-know option allowed.  (If they hang up they are excluded).

Forced-choice is used by other robopollsters such as ReachTEL for basic voting intention and leader rating questions, but usually not for issue questions, in which a "don't know" or similar option is given.  While it's possible that there actually was a don't know option and the undecided have been redistributed (and that the reporting of the poll is incorrect), the quote starting "A clear majority of Australians [..]" suggests otherwise, unless the don't-know rate was really low.

There are two really good reasons not to use forced-choice on issue questions.  Firstly, it may increase the chance of people hanging up, but those who hang up may not be representative in their answers to other questions.  Secondly, in a lot of cases you're forcing the views of voters who may genuinely have no opinion.  If a voter is anti-Coalition but has no view on the Coalition's response to an issue, then if you force them to they'll most likely assume it is poor.

The report also doesn't give any information about party breakdowns.  If voting intention was asked, it would be great to avoid discarding value by publishing it.  If it wasn't, and the poll is just about a specific set of issues, then depending on how those issues were framed in the poll lead-in, there's a risk of motivated-response in who decides to take the poll at all.

I don't find the for/against balances to this sort of question implausible, but I really don't see why an outlet like the Guardian, which presumably sees itself as some kind of alternative to the Courier-Mails of this world, can't show the way by publishing the full details of its polling when reporting on it.  That is, assuming they commissioned it. Isn't that sort of detail what web coverage is especially useful for?  Even the verbatim wording of the questions would be nice.

Perhaps more will come to light on this poll in coming days.  Whatever, one thing I would be absolutely sure of is that a majority of Australians do not have any specific contentious opinion about the current refugee crisis.  A majority of those who have a view, now that's another thing ...


  1. Kevin, I can't find the two Essentials at 52-48 to Labor; neither Essential's site or GhostWhoVotes's tweets mention this week's report. I can only find last week's report (1st September). Have you got this week's report early?

  2. A summary of each week's Essential is contained in the Crikey subscriber email. It tends to come out a few hours before the results are released on the site.

  3. Love your work Kevin, and that of Williams PB

  4. another informative article.

    Shows why you are a must read!

  5. Thought I'd ask this first before going ahead and doing it: As long as I reference you, do you mind if I post your graphs and tables on Facebook?

  6. Hi Kevin. I'm curious how the Keating govt would look in the table of first term govts. Technically he was re-elected but it was his first election as PM and I seem to recall his popularity was not stellar, then he lost the following election. Was he behind in the polls for as long as Abbott has been? Were the poll gaps similar? Thanks...

    1. The Keating-Hawke government 1990-3 was behind for 28 months in a row up til November 1992 yet still won, but the first 17 of those months were Hawke, and they were more behind during the Hawke phase than the Keating phase. Keating's 2PP average up to the 1993 election was about 48.5, and most of Labor's worst polling in his first term came just after he took over from Hawke.

      For Keating's first two years as PM (to compare with the table) the ALP had an average 2PP of about 48.4, was re-elected once, and trailed for about 18 out of 24 months. Polling was very volatile in 1992-3, so that at times Labor were ahead at least 55:45 and at times behind 43:57 or worse (with one Newspoll translating to 40:60).

      Through Keating's whole prime ministership of just over four years, Labor were ahead on 2PP (which was not as much talked about in those days, but can be recalculated from primary vote polling) for a total of about a year in at least four separate chunks.