Voice Referendum Polling: How Low Can Yes Go?

Two-answer estimate Yes trails 42.3-57.7 (as of last data September 8)

Yes still behind in every state.  Fourth state now behind national average

(Update added at bottom for Resolve September 11)

 

Key to colours: Green  - Newspoll, Magenta - Resolve, Yellow - Essential, Dark blue - JWS, Light blue - Freshwater, Black - Morgan, Red - Redbridge.  

Time for another Voice roundup following a flurry of polls in the past week.  In the three weeks since the previous edition there's been another chance for the trend line to do something, anything, other than simple accelerating decline, and again this hasn't happened.  

This week's offerings have been the first Pyxis Newspoll at 38-53-9 (yes-no-undecided), a Redbridge forced choice at 39-61, a Freshwater poll at 35-50-15 and 41-59 forced choice, and Essential at 42-48-10.  The field dates for Freshwater were Sep 2-5.  

The poll to most recently leave the field was Freshwater on 5 Sep, so there doesn't seem to be anything since Tuesday in the mix.  It's possible the trend has changed as a result of this week's campaigning, though every time I have said that over the last five months it has then failed to do so.

I should note that the raw-result trendline above actually slightly exaggerates how fast Yes is falling in my overall estimate.  The reason for that is that in the last few months most of the polling has come from Resolve and Newspoll (which I estimate as having moderate pro-No house effects compared with the polling average) and Redbridge (which I estimate as having a larger leaning to No) with relatively little from Essential (which has a strong pro-Yes house effect) and one poll from Freshwater (which seems not to have a significant house effect so far).  After I adjust the recent polls for these apparent house effects the rate of decline is a little less steep, but the decline is still accelerating.  With house effect adjustments included I estimate the polls as having Yes on 42.3 as of the apparent most recent in-field date.  If the trend has continued and the polls are right Yes has probably dropped most of another point in the five days since, and would still be on course for a sub-40 result (say 36.5).    

Way too many Yes supporters on social media are responding to the ongoing decline in polling by engaging in blatant poll denying and spreading disinformation about polls.  Others, however, are just struggling to process how badly Yes is polling because it's not consistent with how they feel things are going on the ground.  I've seen a few tweets reporting feedback from handing out to be about 60-40 Yes, which given the inner-city skew on such issues, and given that No supporters might not make their disagreement obvious in such settings, actually seems more to confirm the national polls than refute them.  Living as I do in an inner-city part of one of the country's most left-leaning divisions (Clark) with Yes signs locally abundant and barely a sign here that the No campaign even exists it's easy for me to sympathise with the feel of unreality.  But there is nothing historically unusual about what the polls are saying and there have been a lot of problems with the Yes campaign.  

State picture

I now have the following state estimates based on aggregated polling; the figure in brackets is the difference from the national estimate

Victoria 45.7 (+3.4)

Tasmania 44.6 (+2.3)

New South Wales 42.9 (+0.6)

South Australia 42.1 (-0.2)

Western Australia 40.3 (-2.0)

Queensland 37.0 (-5.3)

A bad breakdown for Yes in South Australia from Essential has now pushed that state slightly below the national average, the first time I have had three states below the national average (albeit very slightly).  It's academic for now because of the national picture.

Other polling

Among polls I have not included in my aggregate, I mention the following:

1. SEC Newgate:  I had previously rejected this poll from my aggregate for using the term "update".  Fortunately it has now changed its question to:

Before the end of 2023, the Federal Government plans to hold a referendum to alter the constitution to

recognise the First Peoples of Australia by establishing an Aboriginal and Torres Strait Islander Voice.

Do you approve this proposed alteration of the constitution? 

Following this change, support for the Voice dropped from 56-44 (43-34-23) in June to 46-54 in August (a large drop which may well have been largely caused by the wording change).  However I've since found another issue with this poll; it describes its sample as "Australian residents" and doesn't seem based on the disclosure statement to have made any attempt to deter non-voters (c. 14% of voting age population and with a lower average age) from answering the Voice question.   

The poll attracted attention for a purported finding that Yes voters (8.3/10) were far more likely to vote than No voters (5.4/10), an obviously unlikely finding for a compulsory election.  It's been a long time working out what happened here, because the initial disclosure statement omitted the question, but the question was:

How likely are you to vote in the planned referendum to alter the constitution to recognise the First Peoples of Australia by establishing an Aboriginal and Torres Strait Islander Voice? 

The problem with this question is that it can very easily be read as asking not how likely you are to vote in the referendum at all, but as asking how likely you are to vote in favour of altering the constitution, ie to vote Yes.  As a result it is not a valid turnout predictor and the difference between Yes and No in it would most likely be caused by some No supporters interpreting it that way. 

2.  Australia Institute (SA):  One of the reasons I was not surprised by this week's Newspoll was that an Australia Institute poll of South Australia had come out with Yes ahead only 43-39.  Given that Australia Institute polling has so far skewed to Yes by several points that was consistent with Yes being actually well behind in South Australia, and on that basis I would have been surprised if the Newspoll wasn't about where it was.  (TAI are excluded from my graphing for skewed wording and lack of neutrality.)

3.  Insighfully (Tas): A poll of Tasmania by Insightfully for the IPA had a 42-53-5 result.  While it wouldn't surprise me greatly if the end Tasmanian result was about that, I'm ignoring this one both for being a lobby group poll and for some peculiar properties.  The poll had divisional breakdowns that featured only a 3% spread between the best and worst divisions in the state.  Tasmania is very polarised between the northern electorates (federally often marginal and conservative at state level) and the southern electorates (federally strongly left).  There is no way the break will be anything like this small.  The poll also had implausibly high No votes in the 18-34 age group relative to older demographics (the poll is described as a phone poll but it is unclear to me if it was a robopoll).

4. No campaign polling:  There have been reports of robopolling with skewed wording asking questions about race.  Regular pollsters have been blamed for this but these polls use neither the methods nor the wordings of the established poll, and appear to be the work of No campaign groups.

5. Age opt-in: An unscientific Age reader poll showing a 72% Yes vote was passed off in some circles as an actual poll and in others even as early voting figures when voting hasn't started yet. The Age is read by lefties and would still be read by lefties even if its copy was entirely sourced from the Herald-Sun.  Film at 11.  

Are there unusual reasons for caution?

My polling denial register notes reasons why a lot of the anti-polling copium doing the rounds on social media is nonsense.  Voting intention can change, but as a snapshot of the current standings of the campaign are there any special reasons for caution, and if it turns out that the final polls for once seriously underestimate Yes, does anything stick out in advance as a potential cause?  I've been thinking about this but I really cannot find a lot.  

The first thing I would note here is that the Voice polls market has consisted almost completely of online panel polls.  Six of the seven pollsters I am aggregating use online panel polls, the seventh (Morgan SMS) having not polled since May.  Filling out online surveys for sweatshop level financial rewards is a fairly unusual activity and there's potential that someday people who do this might break oddly in their views in a way that their demographic data don't predict.  However, it has never so far happened, nor is there any specific reason to expect it in this case.  In general, differences between polling houses with the same polling mode are far more significant than differences between polling modes.  It's also notable that Brexit 2016 was a major success for online panel polling, which was relatively accurate while phone polling was getting it wrong.  So I don't see any rational basis for concern here, but it would be interesting to see results from other methods.  

Several of the polls are more or less replicating the wording on the ballot paper by including "Do you approve this proposed alteration?" (Redbridge has "Do you approve of [..]") This may read somewhat oddly compared to polling questions respondents would be used to, and it may well be that some voters will just vote Yes or No without actually reading the question on the ballot paper.  This could make a slight difference compared to a wording that quoted the ballot paper then asked respondents how they would vote, but I doubt it would be significant.

Various claims have been made about likely turnout issues.  Having debunked the most startling example above, I don't find claims about No voters being softer convincing, as there have been plenty of polls finding evidence otherwise.  If anything, potential Yes voters are more likely to be younger and hence I'd expect more likely to fail to vote as a result of enrolment issues.  

Why Is No Winning?

The following are, in my view, several reasons why No is currently winning that go beyond the fact that any contested referendum will see scare campaigns and disinformation from opponents.  For the most part these are my interpretations that I don't always claim there to be polling evidence on, but I offer them in case they are of interest.

1. The history: referendums proposed by Labor governments, held in mid-term and lacking bipartisan support all have poor track records individually, and this one is all three, which may be enough to cause the referendum's defeat alone without any of the other factors.

2. The timing: while the Albanese Government is still polling fairly well at the moment, one of the most striking features of issues polling this year has been the sheer degree of pessimism among voters about the direction of the country and their own finances.  This is likely to fuel any by-election backlash vote; it is easy for people to want to send a message to the government that this is not where its focus should be.  

3. Conceptual tension: the Yes side needs to convince people that the Voice will not be an ineffectual waste of money and will drive change.  But the more the case is made that its advice will lead to real outcomes, the more that plays into concerns that its advice could lead to other real and more controversial outcomes (Treaty, rampant advice to bureaucracy and so on).  After all no-one can really know what exactly the Voice will advise and whether that advice will be accepted.  One of the most common criticisms of the Yes campaign has been the lack of any clear message.  The problem is it's hard to spread a clear message when you can't actually say what your proposal will accomplish in real terms.  

4. Blunders on detail: the Government was concerned about repeating the 1999 referendum experience where being tied to a specific model caused the defeat of the republic.  But in that case the detail had to be specified in the amendment and there were irresolvable differences between republican camps on how to deliver the Republic.  Enough supporters of both direct and indirect election were strongly opposed to the other model that neither could win. In this case the lack of a formal exposure-draft type model with a government commitment behind it has made it easy for No supporters to generate doubts about the lack of detail.  However the existence of draft proposals that some Indigenous players do not like has made it easy to make negative claims about the detail too.  The government could have usefully spent more time on this.  

5. Format: Giving the Voice a chapter of its own in the Constitution has fuelled No claims that the High Court will regard the Voice as extra special and read intentions into it, even though constitutional experts have mostly dismissed this.  This was unnecessary.  

6. Campaign role models: The Yes campaign has had plenty of support from celebrities, corporates and politicians - all of which plays into a conspiracy narrative about elites and perception of the Voice as a politicians' solution.  Too little has been seen from everyday non-Indigenous Australians who support the Voice (and no, this doesn't count actors in ads.)  There is not that much sense of community ownership.  

7. Soft racism: The argument about the Voice as a method to help closing outcome gaps rests on the idea that gaps in health and other outcomes (which governments have not known how to address) are a result of historic dispossession and ongoing discrimination. So if Indigenous Australians are given more influence on policy there will be a major improvement.   But lots of voters who don't express overtly racist/paranoid views probably still don't buy this and instead buy into a racialised version of the same idea as that underlining the "dole bludger" myth - that if people are in bad circumstances it must be partly their fault, and that enshrining the Voice is not going to fix it.

8. Scepticism among older voters: Many older voters can vaguely remember (with an emphasis on "vaguely") what happened to ATSIC and earlier such bodies.  Some are likely to have negative views about the prospects and governance of Indigenous organisations in general, and hence to sympathise with the idea that the Voice should have been legislated first to see if it would work and maybe entrenched later.  

9. Lack of discipline and respect:  Too many Yes supporters online have harmed the cause by abusing No supporters.  What I am calling "rotten-egg Yes supporters" (a minority of course) do not only regard anyone who supports No (unless they are Indigenous) as racist. They also apply the "racist" tag to people who publish facts that they think are inconvenient to the Yes campaign - even when these people are likely to be Yes voters!  The Yes campaign has done almost nothing about this but has continued to try to claim a moral and intellectual high ground that is just not justified by the abusive and disinforming online behaviour of its worst supporters.  Early in the campaign I thought there was more such bad behaviour from the Trumpy/conspiracist No types but I can no longer see a difference.   

10. The "divided by race" argument:  Affirmative action is commonplace and isn't racism, and usually only a few culture-warriors and racists complain about it.  That is, until everyone gets forced to vote on an example because government again outsources failure to fix a problem to the voters.  Redbridge's latest had "It divides us" as the leading objection among No voters - this is not just an excuse; I think the Voice has run up against strong sentiment that the Constitution shouldn't make special cases.

Perhaps this can be turned around in the time remaining, somehow (Yes's greater ground game and resources are the main ways this is argued.).  That would, however,  be unusual.  It would be better to not be this far behind in the first place.  

Update Resolve 11 Sep

A new Resolve poll has been released today with a 43-57 forced choice result and 35-49-15 prior to forcing.  This is actually just a little bit better for Yes than I would have expected.  The impact on my aggregate is that it stays at 42.3 but becomes that number as of September 8 rather than September 5; as noted by Jim Reed this poll shows that the much-vaunted John Farnham "You're The Voice" ad has not stemmed the bleeding in the polls.

After including the state data I have the following state aggregates: Victoria 46.1 (+3.8), Tasmania 45.5 (+3.2), NSW 43.1 (+0.8), SA 41.3 (-1), WA 39.8 (-2.5), Queensland 36.6 (-5.7).  Regarding Tasmania, all the recent data comes from Resolve and there have been large differences between different polls with Resolve having Tasmania about +10 to the national average, Morgan +5 and YouGov/YouGov Newspoll a bit below 0.  So Tasmania should be treated with enormous caution.  

Sep 12 ACM

There have been reports of another round of the Canberra Times/ACM reader survey (Yes 34 No 61 undecided 5) but this is not a genuine poll; mostly an opt-in with a small panel component.  As I noted back in July:

The data have been reweighted, but only in a primitive fashion, reportedly by age, gender and metro/regional location.  But this is not enough to eliminate any skews caused by the reader base of the publications in question, or by whether or not their circulation within regional areas is politically representative.  (There was also a survey panel used, but it accounted for only about 10% of respondents.)  It's also prone to the same risk as robo-polling: relatively few young voters would be captured and those who were might be unrepresentative of their cohorts in a way reweighting would not repair (especially not if insufficiently fine-scale)

The two-answer shift in this survey since early July (about 5%) is actually less than the shift in aggregated polling in the same time (around 8%), for what it's worth.  

Sep 13 Painted Dog

I previously also dismissed WA-based pollster Painted Dog from serious consideration for insufficient methods details, use of signup promotions alongside poll reports and conspicuous Yesdom but I note in passing that its latest has the Voice in WA at 39-61 forced choice, down from 57-43 (which for some reason the West Australian now reports as 58-42) in June, from a rather large sample of 1285.