Wednesday, September 16, 2020

Curiouser And Curiouser: Recent Queensland Poll And Poll-Shaped Objects Roundup

The Queensland election is six weeks away and there has been no mainstream polling for it since the Newspoll in late July showed the LNP leading 51-49.  There are, however, a number of minor polls flying around and it's time to round them up and put them in a box.  In at least one case, the lid should then be taped firmly shut.

Polls "reported" recently have been:

Australian Institute for Progress (statewide)
YouGov (Currumbin, Redlands and Mansfield)
Omnipoll (Ipswich, Keppel, Mackay, Thuringowa)
Lonergan (Maiwar, McConnell, South Brisbane)

A further poll by AskAustralia Market Research has been reported in field but no results have yet been seen.

Australian Institute for Progress state poll

This poll was "reported" by the Courier-Mail and I decline to link to the reporting owing to the very low quality thereof.  The report fails to note whether the poll was conducted by the AiP itself, fails to note the sample size or dates of the polling and fails to publish important information about the commissioning source.  Indeed the report fails to state whether the primary vote data it reports are fresh data and, remarkably, it turns out that actually isn't the case. 

The Australian Institute for Progress does conduct its own online panel polling, albeit of such a low-profile nature these days that I don't recall noticing its 2019 federal election polling before.  This panel polling is especially worth having a look at to make sense of what the Courier-Mail has reported, which is the LNP leading 56-44 off primaries of Labor 32, LNP 38, Greens 12.  

The LNP would win the 2PP off those primary votes, but 56-44 is a big stretch.  It would require the LNP to get 60% of all preferences, compared to around 49% in 2017.  Assuming that Greens preferences split about three to one to Labor as is normal in Queensland, the LNP would need 83% of remaining preferences - these preferences coming partly from a range of right-populist parties (One Nation, United Australia, Katters Australian Party) but also to some degree from unaligned independents and minor left-wing outfits.  

This seems implausible even if those preferences came solely from right-populists, because even in the 2019 federal election where Queensland voters turned on Labor to such an extent that the party won only one Senate seat, the flows from these parties to the LNP in Queensland were still no higher than 67.0% (KAP), 67.2% (PHON) and 68.9% (UAP).   However, unusual preference flows can occur in polls that use respondent preferencing, especially if the number of voters sampled for the minor parties is small.  

In this case though, there could also be something else going on.  A window to this comes from a published AiP poll released just before the 2019 federal election.  This poll found 100% of its Greens' voters preferencing Labor, and 100% of its "Nationalists" voters (this is One Nation plus Australian Conservatives combined) preferencing the Coalition.  The sample size for this preference sampling was 781, so let's say there were 70 Greens voters.  Every single one of them must preference Labor for Labor to get 100% after rounding.  Given the actual break of Greens voters at the election (82.21% to Labor), the probability of this happening is .8221^70  = about 1 in 900,000.  Similar calculations would apply to "Nationalists" (given that they would probably have had less support, but also that One Nation preferences naturally flow more weakly), and the probability of both samples flowing 100% to a party would then be something around 1 in 100,000,000,000.  Unless one assumes that voters for the Greens and One Nation suddenly became monolithic Labor and LNP preferencers at the time of this poll having not been so at any other time or in any other poll, which is of course silly, there is clearly something amiss with this federal preferencing sample.

And indeed it turns out from the actual poll report for this new state poll that not only does the poll display unnaturally strong preference flows from both sides of the minor party equation (around 93% for both the Greens and "Nationalists") but the preferencing data was based on a sample of just 198 voters.  This means the number of voters sampled from each of Greens and "Nationalists" would have been a few dozen, which could create massive errors even if the poll's preference sampling did not have the underlying issue above.  (It is possible the strong preference flows in this case could partly result from dropping voters who were undecided as to their preference flows, since the sample drops from 247 votes for "first preference" to 198, and the party totals change slightly in the process.)  

But the old AiP poll report also gives a hint that something else strange could have happened with this state poll (and it turns out that it did, at least according to the AiP poll report).  The AiP federal poll report did not present itself as a fresh unfiltered survey of primary voting intention.  Rather, "The following tables and graphs are based on a sample which has been selected for voting intention based on the results of the most recent Newspoll to the date of the questionnaire, as published in The Australian."  I was suspicious of whether the AiP state poll was making a fresh voting intention finding of the sort reported by the Courier-Mail given the above and that its primary votes as reported by the Courier-Mail (ALP 32 LNP 38 Green 12) are exactly the same as the early June Queensland YouGov poll.  It seemed the Courier-Mail lacked the skills to know the difference, yet it turned out this was exactly what the new AiP poll said too: 

"The following tables and graphs are based on a sample which has been selected for voting
intention based on the results of the YouGov poll published on June 7 in The Courier Mail."

(Very confusingly, despite this opening statement that the "tables and graphs" are based on a sample selected in this manner, the opening table contradicts this by presenting unweighted voting intentions that show no evidence of such selection as the results are very different. But then the question "If an election were to be held next weekend, which of the following parties would receive your first preference?" presents results that almost exactly match the YouGov poll.  So it looks more like the sample has been scaled to match YouGov rather than selected.)

So the claim that the LNP is ahead 56-44 in this poll reduces to the pollster applying a ludicrously small and potentially otherwise flawed preference sample to a sample selected and/or scaled to match a poll released over three months ago!  (And indeed, even the pollster doesn't think the lead is that large, describing the election as "heading to a marginal result according to our latest polling" on the basis of various other results that seem to point to a much closer contest, though I'm unconvinced that all of them do.  For instance there's a narrow 44-41 lead for the LNP on a follow up question on preferred result that includes a "hung parliament" option, but by far the biggest voting group choosing that option are "Nationalists" who would mostly specifically want an LNP minority government. )

And still there is more!  The Courier-Mail also failed to disclose that the Australian Institute for Progress has links with the LNP.  This information was readily available after the AiP unsuccessfully but understandably sought declaratory relief concerning the impact of new Queensland electoral laws on its activities, including polling.  The court judgement noted:

The Australian Institute for Progress (“AIP”) is a “think tank” based in Queensland. Most of the AIP Board has a connection with the Liberal National Party. However, many of its members do not. The AIP says that it “stands ready to cooperate with any political parties, or its members or supporters who are prepared to support good policy”. It undertakes research and advocacy on both federal and state issues. According to its Executive Director, as the AIP is “ideologically centre-right”, its criticisms tend to favour parties of the right of politics.

(Directors of the AIP include former LNP Premier Campbell Newman and former LNP MP Jane Prentice, among others.)

The Courier-Mail presented this as rigorous polling, "exhaustive polling" even, without asking any of the many questions that clearly needed asking re this poll.  A complete debacle of poll reporting even by the Courier-Mail's parlous standards and it's clearly time for one of these:

Porcupine Fish Award for Ultra-Fishy Reporting Of A Very Fishy Poll (image credit)

YouGov (Currumbin, Redlands and Mansfield)

A YouGov poll of 200 voters in each of Currumbin (LNP 3.3%, 1.2% at by-election), Redlands (ALP 3.1%) and Mansfield (ALP 1.6%) has been reported but details are vague.  The Gold Coast Bulletin reports results said to be for Currumbin only, while mistakenly believing Mansfield to be in Victoria. (There is a town by that name in Victoria, but the electorate is in Queensland).  The Currumbin result is said to be 50.8% to Labor, which isn't implausible given that the LNP had a swing against it at the by-election, and the new MP has had no real time to build a personal following.  However I understand there were not in fact seat-specific breakdowns and William Bowe (who reports the poll as showing a 2% swing to Labor overall) notes that the sample size of this poll (commissioned by the Australian Conservation Foundation) was a trivial 200 votes per seat.  If we treat this as similar to a single-seat poll then, given the reliability issues with seat polling, it adds very little evidence to a widespread existing belief that Labor could perform fairly strongly in Brisbane and the Gold Coast.  However, I understand this poll has used similar selection processes to YouGov's federal polling so let's see how it goes.  

Omnipoll (Ipswich, Keppel, Mackay, Thuringowa)

Omnipoll is the old Newspoll, which ceased doing Newspoll in 2015 and has been relatively rarely seen since.  It has showed up with a sample of 2000 voters across four seats outside Brisbane - four seats where One Nation did well in 2017.  This is part of a poll conducted for a coal industry source that includes a very large number of questions related to coal issues.  Sadly, even the full poll report does not provide verbatim wording for the issue questions or the order they were asked in, so I suggest ignoring all reported findings of the issue questions.

These four seat polls report primary swings away from Labor in three of the four seats (an average of five points per seat overall) but also report massive swings to the LNP and away from One Nation, which supposedly loses seven tenths of its 2017 vote.  This would be quite remarkable given that One Nation's Senate vote in Queensland was so stable at the 2019 Senate election.  It would also be rather boring in one sense, because it would eliminate otherwise fascinating three-cornered contests in both Keppel and Thuringowa.  Thuringowa is especially interesting if One Nation's vote holds up, because One Nation would have probably just won that seat in 2017 had the LNP preferenced it.  Keppel is also interesting because One Nation finished second, but only just, so a swing directly from Labor to the LNP could in theory knock One Nation out and hand the seat back to Labor.  

However, the massive swings found in these seat polls would see the LNP win Thuringowa and Keppel easily, with Mackay unclear between the major parties.  It's just all a bit hard to credit though: One Nation on 5% in Ipswich???  The relative scarcity of recent polling by Omnipoll to assess their form against is also a problem here.  It would be quite a coup for them if it turned out to be true.

Lonergan (Maiwar, McConnell, South Brisbane)

The Greens have released seat polling by Lonergan that has them:

* Winning Maiwar off primaries of LNP 37, Green 36, Labor 17, 2CP 56-44

* Ahead in South Brisbane off primaries of Greens 36, Labor 30, LNP 21.  This would be a win if the LNP preferenced the Greens, but touch and go if they preferenced Labor.

* Supposedly winning McConnel off primaries of LNP 31, Green 30, Labor 29.  However margin of error issues alone mean it is unclear from this sample which of the three parties would finish third.  If it is Labor, the Greens win, if it is the Greens, Labor wins, and if it is the LNP then it comes down to their how-to-vote card and whether their voters follow it.

The biggest issue is that this is internal party polling, and deliberately released party polling tends to overestimate support for the party releasing it.  That isn't always the case (eg the Greens' Essential sample of Maiwar in 2017 had them very narrowly losing,  but they won) but in the Greens' case especially, there is a long history of it.  Greens internals by Lonergan were inaccurate in the federal seats of Higgins and Batman in 2016, though a Lonergan seat poll of the 2018 Batman federal by-election that wasn't for the Greens was accurate.  

I am preparing some detailed seat modelling for the Queensland election (though not on the same scale as 2017) but am hoping before I release it that we will get some statewide polling from an established pollster!


  1. The AIP polling was originally reported in the Courier-Mail, including in the print edition, as being "exhaustive polling *by YouGov*". But clearly that wasn't the case.

  2. Speaking of preferences, would it be possible to figure out how much a fairly organized/resourced minor's (e.g. Greens, UAP) preferencing recommendations are worth?
    e.g. if One Nation switches from recommending preferences against the incumbent major in each seat to recommending preferences to the LNP, how much might that shift their voters' preference flow by?

    1. In 2017 the difference between seats where One Nation preferenced the LNP and those where they preferenced Labor suggests that about 11% of their voters were following the cards. For the Greens the figure is generally in my experience somewhat less. For the UAP there's insufficient data but I doubt their primary vote will be significant in most places anyway.

    2. (So for instance if the flow from One Nation to LNP in a seat is 66% with One Nation preferencing the LNP, it would be 55% to LNP if One Nation preferenced Labor instead.)