Poll Roundup: Shorten's Latest Shocker - Or Was It?

2PP Aggregate: 53.8 to Coalition (updated to 53.4 on 1 February)
Coalition would win election "held now", probably with increased majority.
(Newspoll update added at bottom 1 February)

Pollsters are gradually emerging from their summer hibernation and over the next few weeks we will get a clearer picture of how the Turnbull Coalition government is placed as it kicks off the election year.  I am not sure exactly when Newspoll will emerge but enough data have come out in the last few weeks to make some quick comments about the overall state of play.  The 2PP estimate above will be updated and any further comments added tomorrow night following Essential, but it never alters the picture all that much.

So far this year we've had two Morgans, one Essential and one ReachTEL.  Morgan and Essential were the most strangely behaving polls late last year, with Morgan showing a massive swing to the Coalition immediately following the replacement of Tony Abbott with Malcolm Turnbull, but Essential showing a much more modest change that other posters soon stopped replicating.  Anyway the first Morgan was 55.5% two-party preferred to the Coalition by last-election preferences while the second was 54% (the closest since just after Turnbull was installed - and this off primaries that would normally have been good for only 53%).  Last week's Essential reading was just 51% to Coalition.  The Morgan had a 15% primary for the Greens, which I'm certainly not taking seriously.



I have been regarding Morgan as skewing to the Coalition and Essential as skewing to Labor compared with other polls in the Turnbull era (while Abbott was PM, Morgan skewed to Labor) and after considering the primaries I counted the Morgans as 53.8 and 52.3 to Coalition and the Essential as 52.7.  As Mark the Ballot notes, it is possible Morgan is coming back to the field, but there is no such sign so far for Essential.   However another pollster with a less clear record of skew to either side came out with a thumping result for the government (a 55:45 from ReachTEL, which I counted at 55.1) and this latter result has driven my estimate of the government's standing to new highs.  Last week it ended the week at 54.3 to Coalition, my highest figure for the government this term.

These early figures are not based on a lot of data, and it's possible that when Newspoll and Ipsos fire up the lead will moderate.  Here's the smoothed tracking graph:

There's no sign yet that the government's travails over summer (the resignation of Jamie Briggs and the standing-aside of Mal Brough) have done it the slightest harm.

Morgan reported that they have the Nick Xenophon Team (which they now include in the readout/textout, polling 2% nationwide) on 22.5% and ahead of Labor in South Australia.  It remains to be seen whether this holds up at an election when voters are voting for actual NXT candidates rather than Xenophon himself, but at this stage the result suggests NXT is more or less a purely South Australian factor, with below 1% support in the rest of the country.

Leaderships

Last week's Essential came out with unsurprising leadership figures.  Malcolm Turnbull's net approval was down from +33 (56-23) to +26 (51-25) while Bill Shorten's was unchanged at -20 (27-47).  Turnbull led Shorten as better Prime Minister by 33 points (51-18) down from 39 (54-15) a month before.  In the context of a poll recording more Labor-friendly readings than others, these were still very good figures for the PM.

ReachTEL, however, delivered atrocious results for the Opposition Leader despite showing little net change on voting intention.  (Don't read anything into that alone - Opposition Leader ratings have little to do with voting intention.)  Shorten's net personal rating crashed to a terrible -43.6 (13.8% good/very good, 28.8% satisfactory, 57.4% poor/very poor) compared to a previous worst for him of -29.8.  Likewise the Better Prime Minister forced-choice answer saw Turnbull jump to an enormous 80.8 to 19.2% lead (as if the previous 71.3% to 28.7% lead hadn't been massive enough.)

But are these new shock results correct?  PollBludger member Doyley compared the by-party breakdowns in the November and January ReachTELs and made the following innocent but interesting observation:

"More labor ( + 0. 7 ) and greens ( + 2.7 ) voters prefer Shorten as PM in this poll than in the November poll with a rise of 1.8% in coalition voters supporting Turnbull.

For what, if anything that is worth."

Doyley was not suggesting any error here, but it soon occurred to me that for such a massive shift to have happened in the overall preferred-PM result with such a small shift in preferred-PM scores by party, then something truly amazing must have happened with the preferences of Others voters.

ReachTEL usually don't publish the percentage of Others voters preferring either leader as PM but an estimate of it can be derived from the primary vote totals and the breakdowns for Coalition, Labor and Green voters.  Furthermore, because the pollster publishes figures to one decimal place, the impact of rounding errors on these derived estimates should be relatively small.

Here's what I get for ReachTEL national polls since May 2015:

The figures in grey on the right are my derived estimates of the proportion of Others voters who preferred the PM to the Opposition Leader.  We can see that while Tony Abbott was PM, Others voters apparently preferred Bill Shorten to Abbott, with Abbott's preferred-PM share languishing in the mid-high 30s on my estimates.  (In the one case in which the share of Others voters provided was published, it is fairly close to my estimate, with a difference of 6.3 points).  After Turnbull takes over, the Others voters swing increasingly to the Coalition's side, with nearly two-thirds preferring Turnbull to Shorten in both October and November.

All this makes perfect sense, but then based on the January data the same method projects a mathematically impossible 166% of Others voters preferring Turnbull to Shorten.  (Another way of putting it is that the published figures for Coalition, Labor and Greens suggest Shorten should be on at least 25% even if Others voters all prefer Turnbull.)

Now, there might well be more to ReachTEL's methods than I am aware of, and some unpublished change might have happened - there might be unpublished excluded "undecided" voters on one question or both (though this isn't consistent with what I understand of their methods so far), there might be different scaling methods applied to the totals as to the raw party figures (though I am unsure why anyone would do this) or there might be something else strange I hadn't thought of.

But basically there is this: when Others voters are just 9% of the total, if Shorten's preferred-PM ratings among Coalition, ALP and Greens voters are barely changing, and if the overall primaries are not changing much, then there is just no way changes among Others voters will cause a loss in support of 9.5%.  It appears that either the 80.8%-19.2% total, or one or more of the party breakdowns for preferred PM, is incorrect.  (It's also in theory possible that the primary votes are incorrect, but I would really prefer not to go there!)

When we look at Shorten's leadership ratings the same issue emerges.  If Shorten is considered to be going well by 37.3% of the 31.8% who vote Labor, 6.1% of the 48.5% who vote Coalition, and 16.4% of the 10.8% who like the Greens, that should mean that at least 16.6% (give or take a bit for rounding) of the overall sample rate his performance as Good or Very Good, even if Others voters uniformly agree that he should never have been leader in a million years.  So 13.8% of all voters seems too low.  And if 57.4% think Shorten is a dud, but this includes only 69.4% of Coalition supporters, 23.8% of Labor supporters and 43.6% of Greens supporters, then Shorten would have to be getting the thumbs down from about 126% of Others.  That's a bit high, even for Bill Shorten.

I find the same issues again with the supposed net rating rise for Malcolm Turnbull from +33.4 to +41.1 - this also seems too high given the published breakdowns of Coalition, Labor and Greens supporters.  For example his Good/Very Good rating is supposedly up from 48.2 to 53.6 despite being down among Coalition supporters (67.1 to 66.1), Labor supporters (31.3 to 25.7) and Greens supporters (33.7 to 27.3). This doesn't make sense, and it's not consistent with the run of other polling either.

On this basis I'm really doubtful about all the leadership totals in the January ReachTEL specifically - I don't doubt Shorten's doing badly, but I'm unsure it's that badly.  These doubts don't extend to any other poll by this pollster. There may be some innocent explanation and I will very happily publish anything I find out on this that I am at liberty to reveal, so please check back for updates later.

A Problem With Australian Polling - Lack Of Public Oversight

Let's assume that this is actually an error.  If that is so then I don't want to pick on one pollster specifically, since I suspect calculation errors in published Australian polls may not be super-rare, but that some pollsters are easier to check on than others.   In ReachTEL's case it is relatively easy to spot dubious figures because of the amount of data they release and because they release it to one decimal place (greatly reducing the impact of rounding), yet this is the first time I've seen something of this sort in their figures.  Other pollsters (and I'm looking especially at Newspoll here) do not release nearly enough data  with enough precision for the public to detect possible calculation errors should they exist. Often, the evidence for an error (if there was one) would be lost in their rounding.  Polling is a mathematically complex job with often tight turnarounds between the receipt of data and the need to present results to a commissioning media source for publication.  Programming or data entry mistakes are bound to happen in a small minority of cases.

Let's compare with the sort of thing that pollsters release overseas (at least, some of them).  Suppose one thought that Public Policy Polling had an error in their recent poll of US Presidential voting intentions in North Carolina.  Well, one could go to PPP's website and download an 87-page PDF with crosstabs for everything under the sun. (Admittedly they're not to one decimal place, but they're frequently exhaustive, unlike those released by Australian pollsters).   This is also common in the UK polling industry, not that it saved a lot of UK polls from being hooey at the recent election.  For Australia, publishing breakdowns to the nearest whole number for just Coalition, Labor and Greens supporters (as many pollsters are doing) and leaving readers to derive a ropey estimate for Others if they want to check for mistakes is just not good enough.  All pollsters should release crosstabs that include Others voters, or at least a combined Greens/Others. Ideally, all pollsters should follow ReachTEL in releasing figures to one decimal place, even if this is not done in the initial public release of figures.

With the clear potential for polling figures to make or unmake political careers, I don't think the current scanty level of supporting data release from Australian pollsters generally is good enough.

Other Polls

I feel some hesitation in running through other polling results after all that because I'd like to know I'm reporting facts about samples of public opinion, but for all I know for sure I might be reporting somebody's spreadsheet glitch.  Anyway Galaxy attracted a lot of attention with a seat poll of Fairfax showing incumbent Clive Palmer on an embarrassing 2% of the vote, and with a net personal rating in the electorate of -76 (7% approving).  This comes on the heels of revelations that Palmer's nickel refinery donated money to Palmer's political party shortly before sacking workers, which might be the explanation for the dire result.

It's long been expected that Palmer's 26.5% primary from the last election would collapse to something like 10-12% at the election leading to him losing his seat back to the LNP, but these claims that his support has sunk to statistical-blip level are new.  (The LNP have wasted no time claiming their internal polls have Palmer even lower.)  Seat polling being what it is, and the sample size being modest, it might be that Palmer is really on several percent of the vote, but this year I will be modelling Fairfax as a vacant LNP seat off its last-election 2PP of 61.7% to Coalition, and ignoring the incumbent.

Essential provided some findings on the US Presidential election, showing that Australian voters overwhelmingly prefer Hilary Clinton, with the exception of Greens voters who also like Bernie Sanders, and Others voters who are less keen on Clinton and about equally keen on Donald Trump.  The interesting thing is that results for Liberal and Labor supporters are nearly identical but this isn't new; Australian Liberals do not identify strongly with US Republicans and also strongly supported the election of Barack Obama.

ReachTEL released some results about support or opposition for the Turnbull government's decision to decline a US request for additional military support for Iraq and Syria (strong support from all parties except Others), and support for increasing the GST to 15% to "provide more funding for services" (31.6% in favour, 52.5% against, with non-Coalition voters heavily against.)  In these cases they did provide full breakdowns including Others, but I have the same issues with the overall figures  for both questions not squaring with the primaries and breakdowns as I did for the leadership questions.  I have checked some results for the November poll issues questions which also provided full breakdowns, and in that case there is no such problem.

Redistribution

The federal redistribution is effectively done and you can see a new notional pendulum at Antony Green's site.  Some of these figures will change slightly, and I am waiting for the late February release of full demographic transfer figures before I fire up my seat-prediction model and assessment of the 2PP vote Labor needs for a 50-50 chance of winning.

What I do want to draw attention to right away is that while Labor makes a net notional gain of two seats, it also starts with a greater proportion of close seats (eleven under 2% compared to five under 2% for the Coalition).  Also a lot of the Coalition's close seats are double-sophomore seats which it won from Labor at the last election and therefore should expect an advantage based on the change of sitting member.  A notable exception here is Solomon (NT).  What all this means is that if the swing between the parties is relatively small, the Coalition should do better relevant to the swing than uniform-swing based projections expect.  This is one reason why, though I don't rule out Labor being surprisingly competitive at the election based on the polls now, I am very pessimistic of their chances of actually winning.

Essential Update (Jan 27):  Essential came out with a 52:48 this week, but the primaries were still on the 51 side.  After considering Essential's recent behaviour and the primaries I aggregated it at 53.2, which had no impact on the bottom line.

Essential found that its respondents have a negligibly improved view of the state of the Australian economy compared to in September (28% good 31% poor cf. 26-32) but that the proportion believing it to be heading in the right direction has dropped from 34% to 30%.  The drop comes from Coalition voters, and goes to the unsure camp rather than the pessimistic one.  Essential finds no real change in its question on trust to handle the economy with Don'tKnow on 55% streets ahead of ScoMo on 26 and ChriBo on 19.

Essential also finds a big flip in attitudes towards increasing military involvement in Syria and Iraq.  Previously increase led decrease 32-19 but now decrease leads increase 34-18.  Essential also shows that voters for both major parties strongly regard Australia Day as a source of national pride and that it is mainly Invasion Day for Greens voters (even then, not overwhelmingly).

Non-Update (Jan 28, 9:30 am): At this stage there is no further word on the ReachTEL issues, nor have the figures on their site been amended.

Newspoll Update (1 Feb):  

Normally I start a new piece when there's a Newspoll out but in this case time is short because of work commitments, which may also result in reduced attention to this site for the next week and a bit.  The first Newspoll of the new year is similar to the last one of last year, with a fourth consecutive 53:47 2PP, which I've aggregated as 52.8 after accounting for the primaries.   Malcolm Turnbull's netsat is unchanged at +22, Bill Shorten is up three to a still miserable -35, and the biggest change is a shift in the Better Prime Minister readings from 60:14 to Turnbull in early December to 59:20 in Turnbull's favour now.  This is Shorten's second best reading since Turnbull became PM, and suggests that something (maybe making noises about the GST) might be moving some voters who dislike both leaders off the fence in Shorten's favour.  All up, no massive surprises here, and nothing of the calamity implied by the ReachTEL figures.

My aggregate has come down to 53.4 now, though it's rather Newspoll-heavy at the moment with all the other data a week and a half old, and Newspoll does seem to lean slightly to the ALP.