Voice Referendum Polling: The Yes Vote Keeps On Falling

Current estimate of national Yes vote (two-answer basis): 54%

(Note added 12 June: an update on the Resolve poll has been added at the bottom of the article.)

---------------

Two months after my first article on Voice referendum polling, it's time for a second.  This week saw the release of a new Newspoll with the worst result for Yes of any reputable poll so far, though Yes was still ahead nationally, just, 46-43.  The poll lead to a spectacular display of unhinging on Twitter from the usual army of drips, megaphones and rusted-ons. Many of these responded by posting completely and obviously false claims about Newspoll's track record, including falsely claiming it had failed to predict Labor victories in recent elections.  I even saw one tweet that claimed Newspoll had wrongly forecast Liberal wins in all state and federal elections in the last three years.  In fact the final Newspolls in every relevant case had Labor winning (with 2PPs in chronological order of 51.5, 66, 53, 54, 54.5 and 54.5) and Labor duly won the lot (with 2PPs of 53.1, 69.7, 52.1, 54.6, 55.0 and 54.3).

One can only wonder what these people will do if, or looking more likely at the moment when, the first poll with No ahead pokes its head above the parapet.  At the current rate of progress, that day might not be far away.  The Yes vote is tanking according to polls.  While it does depend to some degree on which poll one examines and poll wording, the overall downhill trend I highlighted in early April has shown no sign of stopping in the last two months. 

Below is an updated version of my graph of polls as converted to a two-answer yes-no basis, which I am now limiting to since the announcement of the draft wording in late July 2022.  While I am not fussy about accepting question designs that I think are even arguably acceptable the list of polls I have kicked out of my dataset is growing:

* Australia Institute for reasons noted in the previous article.

* IPA Dynata (Nov-Dec 2022) among other things for pejorative wording (including claiming the Voice would advise on "all laws", and asking a leading question about racism).

* SEC Newgate for referring to the proposed change as an "update" - such wording is effectively an argument in favour by implying that a constitution that lacks an Indigenous Voice is out of date in the same way as, for instance, one that included obsolete expressions or facts.  This wording may have contributed to Newgate recording a 52-27 result in mid-April. about nine points better than the average of other polls in the field around that time on a two-answer basis and a massive outlier.  (I am not claiming this poll is intentionally skewed, just that this wording is likely to have a skewing effect.)

I've also removed a NSW-only YouGov that I mistakenly included in the previous graph, and added more accurate median dates for some older polls, and also the April sample Newspoll breakdown.

In looking at national referendum intention, I exclude polls known to be commissioned by stakeholders no matter how good their design, such as the giant Uluru Dialogue YouGov poll from March (which was only slightly more favourable for Yes than the trend of other polls at the time so would not have made much difference anyway). However I will use such polls in looking at state breakdowns.  

Here we go then down the slide:

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

It remains unclear whether the trend downhill is actually linear or accelerating over the past ten months, but here I've used the boring linear version as the percentage of variation explained no longer increases that much by doing otherwise.  According to the above, Yes is now at about 55% on a two-answer basis, and has been shedding at the rate of 1.15% per month.

But it might not be that good for Yes.  If, at the risk of cherrypicking, I limit it to this year's polls only, then Yes is at only 54% and has shed 8 points in five months, or 1.6% per month.  

That should be a rather alarming picture for the Yes side.  If Yes is really at only 54 at the moment, then given the historical tendency for referendum polls to underestimate the naying horses, and given the risk of not getting a double majority (more on that below) the Voice might not pass even if the vote were held right now.  Never mind what another four to six months of FUD-flinging from those opposed might do to it ...

Some of the drop in support might be explained by changes in question wording.  This week's Newspoll saw a change in question from:

"There is a proposal to alter the Australian Constitution to establish an Aboriginal and Torres Strait Islander Voice to Parliament. Are you personally in favour or against this proposal?”

to:

“Later this year, Australians will decide at a referendum whether to alter the Australian Constitution to recognise the First Peoples of Australia by establishing an Aboriginal and Torres Strait Islander Voice. Do you approve this proposed alteration?”

Initially uncommitted voters were then informed that voting was compulsory and asked which way they were leaning, but still allowed to be uncommitted. Frankly, I doubt this particular change would have made much if any difference (it could be argued that "recognise the First Peoples of Australia", which will be on the ballot, will drive up support). I suspect much of the change from the previous Newspoll results was down to actual changes in public opinion over two months.

(Edit added 11/6: Murray Goot has pointed out that a perhaps more meaningful change was in the array of response options, which switched from strongly in favour/partly in favour/partly against/strongly against/don't know to yes/no/don't know.  I tend to agree that this could have been a bigger factor in the results change than the question change, but if anything I think yes/no/don't know is likely to be more accurate, because "partly in favour" was a bit of a yeah-nah and could have inflated support.)

It's also the case that just because support for the Voice has been relentlessly falling for a year now, does not mean it must continue to do so.  It is risky to project such trends to continue outside the range they've operated in; the graph says what has happened to support and not necessarily what will happen.  But two months ago I was saying similar things and yet in those two months the trend has continued.  And there are reasons why it might well continue further.  Firstly the arguments for the Voice have long been out there but some of the arguments against - regarding advice to executive government and the Voice's representative structure for instance - are more novel and may still be filtering through.  Secondly the growth in the No vote that has already occurred means that voters who are sceptical but scared that voting No is racist will increasingly encounter other No voters who they don't consider racist and therefore feel licensed to vote No themselves.  

Not Just No Gaining From Undecided

There is a minor misconception that what is going on here is just undecided voters moving to No.  In fact, of the pollsters who allow undecided responses, three (Morgan, Newspoll and Resolve) aren't showing a consistent downwards trend in undecideds, and what these polls have found this year is that Yes has lost about 7% of the raw vote with No gaining about the same.  Freshwater differs; it had a much-reduced undecided vote in its most recent poll and since late last year has a swing mainly from undecided to No.  

All this doesn't mean Yes voters are flipping directly to No.  There may be a more complex churn pattern in which former Yes voters are moving to undecided and former undecided voters are moving to No.  But overall the Yes vote has been falling in raw terms across the majority of series, as well as on a two-answer basis.

State Breakdowns

There is much more state breakdown data available for various polls now (see Wikipedia compilation of it).  In the state breakdowns, what is important is where a state sample stands relative to the national result as a whole, because that can be used to estimate how the vote might go in a particular state right now.

Overall what I draw from the state breakdowns is:

* Queensland is the worst state for Yes with Yes fairly consistently running about eight points behind the nation on a two-answer basis.  This means Yes is already losing in Queensland if the polls are accurate.

* Victoria, SA and NSW are all running above the national average overall (I get by about 3, 2.5 and 1.5 points respectively).  

* The polling for Western Australia is very variable.  On average I find WA running about 2.5 points behind, but some polls have that as nothing and other polls at several points.

* There's not a lot of data for Tasmania, and while what there is appears slightly positive, I wouldn't place much trust in it.

This breakdown is actually good news for the Yes side, but the shortage of detail for Tasmania could be a big deal.  For a given national result, Yes ideally wants to run just ahead of the national vote in four or five states and take the double majority issue out of play, so losing Queensland by lots helps, all else being equal.  

Should The Voice Go Early?

The more or less linear decline in Yes support has led to a theory that the vote should be held as soon as possible before things get any worse.  There would probably be logistic issues with that but anyway I don't believe it would work.  Past experience is that campaigns concentrate voters' minds and that voting intention bounces (eg Howard 9/11) go away faster when people have a ballot box to go to.   A snap referendum might just result in a faster loss of support before the Yes campaign had time to adjust strategies.

What Is Albanese Up To?

Australian history is littered with examples of electorally dominant Prime Ministers trying to get referendums or plebiscites through and losing against Oppositions that in cases hardly existed.  John Curtin couldn't get his proposal for extra post-war powers through in 1944, Billy Hughes failed twice on conscription, and Holt and Hawke failed on the Senate-Reps nexus and simultaneous elections respectively.  Despite Coalition opposition the Voice should be easier compared to those because it only creates an advisory body.  And yet at the moment the Yes lead is modest at best and clearly shrinking.

 Despite this risks have been taken, especially on the issue of making the Voice one to executive government and not just parliament, which has been used to raise the spectre of an expensive Voice clogging the bureaucracy with incessant advice, some of which (shudder) might even be accepted. (Distastefully, some No campaigners are especially keen to reference Centrelink here, presumably as a dog-whistle to racist stereotypes about handouts.)  At the same time as taking risks, the government doesn't seem to be doing that much to arrest the slide in support for the Voice, as if it does not want to invest too much in the case of a possible failure.  

If the Voice actually fails, Albanese would not be the first PM to mistake his electoral success and polling popularity for a sign that the country had radically changed rather than a sign that it had rejected and ejected the other mob.  The Aston by-election win may have been an emboldening factor here, but what if it was largely about the Liberal Party preselecting a ring-in from the inner city and then fighting itself at state level over Moira Deeming (displeasing all sides in the process)?   

There is a difference between voters respecting leaders for taking a committed stand on an issue and voters necessarily embracing that position at a referendum.  

Something else I should mention in passing is the unfamiliarity of younger voters with referendums.  A 40 year old voter in 1988 had voted on sixteen attempts to amend the Australian constitution, but a 40 year old voter today has never voted on any.  Constitutional literacy is bound to be not what it was.  

I will continue to follow Voice polling and see if it can turn the corner or at some stage level out.  

12 June Update: We're Here Already - Resolve Poll

At the current rate of decline I was expecting the first poll with No ahead to appear sometime in July but in fact we're here already - sort-of.  Resolve has come out with a poll with No ahead 51-49 on a forced choice version but Yes ahead 42-40 with undecided voters permitted.  

The reported question (the same as the previous Resolve) wasn't entirely to my liking:  

[Edit: See 19 June update below]

It is not in fact (at least not in any policeable way) compulsory to vote yes or no; it is compulsory to attend and take a ballot paper unless one has a valid reason not to (such as religious objection or extraordinary circumstances).  Voters can vote informal though in referendums in recent decades the informal vote has been trivial (often below 2%, and even below 1% in the republic and preamble votes in 1999).  It's not obvious in which way this incorrect claim might affect responses, however.

It's also not clear to me from the online published material whether respondents got any kind of lead-in statement, as they did in Newspoll, explaining that this was a question about the upcoming expected referendum, and if so what was in it.  There is this, but it's a poor substitute for an actual script of what exactly respondents would have encountered in the course of doing the poll:  

After adding this poll to my graph I now again suspect that an accelerating decline pattern is the best interpretation (the residuals on a linear graph have an increasingly bow-shaped pattern) so here's an all-polls graph including Resolve and projecting Yes at about 52.5.  But that's probably a little pessimistic because Resolve seems to be the most negative poll for Yes and has just reported, while Essential which is the most positive has not been seen for a while (I suspect it's coming very soon).  

Where a poll offers differing forced-choice and undecided-allowed results I am prioritising forced choice in these figures (though in this case it is only the difference between 49% and 51% Yes).  I prefer Newspoll's method of trying to drive the undecided vote down as much as possible but ultimately letting voters be undecided if they must.  However I expect that forced choice will be a more accurate predictor than polls with a high undecided vote, given the history of referendum measures polling underestimating No.  While I would expect voters who fail to vote at elections for whatever reason to be more likely to be undecided than those who do vote, I would also expect non-voters to be greatly under-polled in the first place, so I expect the bulk of Resolve's 18% undecided are people who will actually vote formally in the end.  

Resolve also presented combined May-June state figures (Tas 57 from a presumably tiny sample size, Vic 56 NSW 53 SA 48 WA 49 Qld 44) - the combined national vote from these would be around 51-49.  This differs from the overall pattern in finding SA behind the national average and has not changed my assessment above that four states are probably running ahead of the average (with most doubt about Tasmania.)

[Edit: A further interesting suggestion I have seen about Resolve's question is that some respondents with low attention spans or a need for speed might be thinking the final question refers to approving or disapproving of (supposedly) being compelled to vote yes or no.]

... and Essential strikes back!

As expected Essential was not far away and it thumbed its nose at the consensus with a 60-40 Yes-No result, so what is going on here?

This is what Essential publishes as its question format (it is unclear if there are any additional instructions though with Essential being an APC member they should be publishing them if they are).  It appears to be a forced-choice question.

I am not wildly keen on this question because of the word "enshrined".  Normally when there is discussion about enshrining something it refers to giving a level (or a higher level) of protection to something that is already taken to be a thing (eg something that is an agreed right).  At the moment there is no Voice to enshrine.  The word also may imply that the Voice is akin to a right and even that it is a concept with a whiff of sanctity about it (most people do not come across this word very often).  Nonetheless whatever the wording is it has been stable.  Here is the graph again with Essential included but it now seems clear that a house effects model is needed to deal with the increasing divergence between Essential and Resolve.  As time permits I will work on this!

Overall there are a couple of sources of modelling uncertainty ignoring the house effects issue: (i) whether the pattern of decline has accelerated over time or is more or less linear (ii) whether to still include polls from late last year.  If linear and yes then the current result comes back to 55, otherwise it is about 54.  

June 19 Resolve Wording Update

An article published by the SMH now indicates a full wording for their first pass of their poll - the previous published wording appearing to be an abbreviated first pass combined with the prompt for the second question.  (A longer but also abbreviated version of the first pass actually appeared in April).  This has led to some unnecessary criticism and the moral of the story is that newspapers should publish the full wording and be crystal clear when questions are asked in stages.  Especially online versions, which do not have space constraints.

This is interesting in that the first question did allow for "probably" Yes and No answers (which do not reproduce the voting experience).  It is not explicit whether the follow-up forced-choice question did this although the articles imply a straight Yes/No.  It is also not explicit whether everyone was asked the forced-choice question (not just the initially undecided) although again the articles seem to imply this was the case.  

The first pass question is in no obvious way unfair to the Yes side (indeed it uses "enshrine" which I criticised as potentially favourable in the case of Essential) and it's a shame the wording wasn't clearer before.