Blog/

Analysis

James Kanagasooriam

Chief Research Officer

Bi_Focal #9: Australia & the Voice referendum: The noes have it

October 11, 2023

On 14 October, Australian voters will cast their ballots in a historic referendum, known as ‘The Voice’. We forecast a "No" result.

Introduction

This weekend, on 14 October, Australian voters will cast their ballots in a historic referendum, known as ‘The Voice’. The “Yes” camp proposes to amend the constitution to create an advisory body to the Australian government on Indigenous matters. This referendum is the 45th since Australia’s founding in 1901, and only eight out of the previous 44 have passed.

Figure 1: Results from historical Australian referenda (1900s-present)

Why “No” will win

Our research suggests that No will win in a landslide: 61% No vs 39% Yes. For those of you who don’t want to wade through 25 charts, one map and three tables we think No is going to win for six key reasons: 

  1. The “Voice” is a hard to understand concept that even after months of campaigning a significant proportion of Australians don’t really understand. And even when they have understood they are being told it is simultaneously both an important and modest change.
  2. “Yes” side only has 1 cogent argument to win – that it will improve the lives of Indigenous Australians. The “No” side has deployed three separate arguments that only partially overlap – and therefore offer much wider electoral coverage. These three reasons are (i) now is not the time for a referendum due to the cost of living crisis; (ii) the Voice undermines the principle of equality; and more controversially (iii) the concept of “First Australians” is itself flawed. 
  3. The failure of Yes to present the Voice as a bi-partisan campaign. The Coalition vote is registering 82% against. Elite Liberal party stakeholders such as ex state and Federal leaders have not swung the vote towards yes. Engaging more with Liberal grassroots could have shifted the dial.
  4. The look and feel of Yes is hugely far away from the median voter. Albanese has traded a centrist, mixed rural/urban coalition for one that more closely resembles the politics of his own seat of Grayndler where voters are disproportionately concerned with Indigenous affairs, racism and the environment vs the rest of Australia that is far more concerned with economics. Celebrity endorsements, corporate sponsors have given the whole yes campaign a strong Clinton 2016, Remain 2016 feel. 
  5. Labor has forgotten quite how far apart its rural voters in QLD and WA are vs NSW and VC voters. Almost no centre-left party in the English speaking world would hold these types of voters - and the yes side were relying on partisanship kicking in on the Labor side that ran up against demographic and attitudinal trends in these states.
  6. The bias towards the status quo – almost no referendum proposal has succeeded without winning over all states and generally greater than 70% support. The only “contested” referendum to pass was the 1946 social services referendum with 54.4% support.

What hasn’t led to the result…

  1. Misinformation. We specifically polled on this question. Of the 37% of Australians that thought that No had a better campaign than yes (29% the other way), only 7% believed that Yes had a bad campaign because of campaign misinformation from No. Instead most Australians thought that the lack of clarity (30%) and tone of the yes campaign (18%) mattered more. 
  2. Albanese poor personal ratings / Dutton strong ratings. The PM’s ratings have held up remarkably throughout the campaign. Whilst Labor commitment for the vote has been less than full throated - the declining yes vote, hasn’t been matched by a similar collapse in PM Albanese numbers This could come later though in the aftermath. So far we don’t see evidence that Dutton has personally carried the No vote. That said his signalling against Yes has proved critical. We will release voting intention estimates using the model in later issues, and will perform a deep dive on Dutton and the Coalition in general. 

How we got here

The last vote, a write-in mass postal ballot on same-sex marriage, passed with 62% of the vote. We are projecting that the ‘No’ campaign against the amendment will win by a similar margin, if not more. Our poll shows a structural split of  the Labor party between its older, rural and traditionalist voters and its younger, urban, university-educated base. This is a pattern of votes that could have been lifted straight from the UK’s EU referendum – but with the Yes side doing materially worse than Remain. It seems that Prime Minister Albanese has traded in a winning centrist, ideologically heterodox, widespread coalition in the federal elections last year for a vote that largely tracks density, age and mildly along education lines – whilst losing out median Aussies that don’t fit either a ‘Yes’ or ‘No’ archetype.  

Why should you care? Well the straight answer is that Australia’s is the latest socio-cultural issue which has split another Western democracy. That’s not how the referendum was envisaged back when it was announced and yes to the change was polling at 60% (Figure 2). Instead our estimates point to a significant loss, and the potential for a realignment in Australian politics that looks quite a lot like the US and UK realignment. Indigenous affairs poll low on salience/relevance (see Figure 3) – so the realignment may be a slower burn, and the Liberal leader Peter Dutton (read Conservative for UK readers!) may not be the person to enact such change.

Figure 2: Opinion polling for 2023 Australian Indigenous Voice referendum 

Figure 3: Top three issues facing Australia today (% including in top three)

 

To produce these estimates, the team at Focaldata conducted a 4,608-respondent poll across Australia. Fieldwork was carried out between 18 September and 2 October. We fielded our mega survey as six state representative polls, merged them, and then modelled individual vote intention to come to a view on whether the Referendum will pass. We’ve done it this way for two reasons. First, in order to collect a sample big enough to model how every CED (think constituency seat, UK readers!) will vote, and provide an even more robust projection than a nationally weighted poll. Secondly, to be able to dig into the detail of why the ‘No’ vote is as strong as it is, by breaking down the political make-up of the No coalition into a patchwork of different subgroups and motivations. 

Our argument is that the ‘Yes’ campaign’s arguments, associated with the progressive/Green wing of Australian politics, have failed to convince the average Aussie voter. In fact, a staggering 67% of ‘No’ voters think none of the pro arguments were convincing, compared to the 7% of Yes voters who thought the same for the ‘No’ campaign. This asymmetry has important implications. If voters are more convinced by the logic of the ‘No’ campaign, ‘Yes’ votes are fragile and we could yet see more conversion to the ‘No’ camp over the next few days. Indeed, when we asked undecided voters how they would vote if forced, 65% said no. The Australian government is hurtling towards a big defeat.  

Australia votes

Let’s cut straight to it. Our prediction as mentioned up top is 61%-39% for the No vote. This is at the upper end of public polls – particularly with recent ones showing a smaller No lead. This is based on reallocating undecided voters, who are then presented with a binary choice Yes/No. However, we anticipate it could still get worse for the Yes side. Their vote is softer: there are roughly 2x more voters who are currently voting Yes who could switch to No, than the other way round. More worrying still for the Yes side - if you take just voters with a ‘hard’ voting intention (i.e. very likely to vote and likely to vote) No lead by a crushing margin of 70/30. More realistically if you take the ‘flippers’ (Yes voters who are open to voting No, and vice-versa) the polling moves to 63% No, 37% Yes. 

Figure 4: Headline vote intention 

This headline vote conceals considerable variance by demographics and geography. The sharpest split is by age (Figure 5). This age curve will look familiar to observers of the UK (see Brexit!) politics, except the ‘flipping age’ is much younger, and the average position is more firmly against. The fact the Yes campaign has not convinced a majority of 35-44 year olds is a telltale sign that it has failed to win over the median voter. The gender split (at Figure 6) is worth highlighting, though not because it mirrors Brexit voting patterns (there was no significant gender split in EU referendum voting). Rather, it reflects a male/female drift in voting behaviour we have been seeing across Western democracies, with women peeling away from right-leaning parties and issues in favour of left-leaning ones. 

   
Figure 5: Vote Intention by Age
Figure 6: Vote Intention by Gender

Most remarkably, and again in contrast to Brexit, the ‘graduate’ effect on vote intention is mild (Figure 7). Whereas graduates were split 3:1 for Remain in 2016, graduates here are split 48:52 – an indication that the Yes coalition is very ideologically narrow indeed. 

 

Figure 7: Vote Intention by Educational Attainment  

Let’s turn to geography. Taking our nationally representative poll, we’ve modelled down to the level of the Australian constituency using a technique called MRP. This provides us with estimates for Yes/No voters for every CED (Commonwealth Electoral Division) in the country.  The result is stark. As Figure 8 shows, the vast majority (85%) of seats are expected to swing against the government. 

Figure 8: Percentage projected to vote Yes, by CED 

What’s most interesting is how these seats are distributed, cutting in many ways across political dividing lines. The 22 seats projected to vote Yes are all metropolitan – high-density, high degree-educated and skew young. A majority (>50%) for Labor across most of these seats in 2022. High proportions also voted Green (~25%), far higher than their national vote share (12%). Looking at these seats, the ‘Yes’ vote comes in a single shade: bits of the country which are very similar to each other. 

Table 1: Top 10 seats, projected Yes 

By contrast, ‘No’ seats are highly variegated. They include suburban districts – Sydney’s Wentworth, Melbourne Fraser – but also, more surprisingly, Cunningham, Lilley and Bendigo. These otherwise safe Labor seats nevertheless share a common feature: they are relatively low-density. In fact, the ‘Yes’ vote seems to track population density most strongly, over and above other dimensions like education (Figure 9). 

Table 2: Top 10 Seats, projected No 

Figure 9: Percentage projected to vote Yes vs logged density

Map 1: Percentage projected to vote Yes, by CED 

Swing from 2017

We might well expect that the pattern of results in this referendum would be similar to the pattern of the 2017 referendum. The graph below shows this, by Commonwealth Electoral Division: points further to the right voted the most Yes to same-sex marriage, while points towards the top are projected in our model to be the strongest Yes in the coming vote. The diagonal black line shows points where there would be no change.

Figure 10: Flipping points: projected percentage voting Yes in 2023 referendum vs 2017 same-sex marriage vote, by CED and % born in Australia

The cloud of points is towards the right - this is because the 2017 referendum proposition was much more popular than this one is now. What’s more, there is indeed a strong relationship between the two. The most ‘Yes’ areas in 2017, such as central Sydney, Melbourne and Brisbane, are also forecast to be the most ‘Yes’ now, while the most ‘No’ areas before, such as  Maranoa and Kennedy, are also expected to be the most ‘No’. The upwards diagonal shape of the cloud shows that this relationship holds across almost all districts.

However, this pattern does not hold everywhere. There are exceptions – the districts off towards the upper left. These are districts where the Voice is almost, or even more, popular than same-sex marriage. And there is a clear pattern to these: they are the areas with the most foreign-born residents. The shading of the points shows the % of the population who were born in Australia, and the lighter blue increases off towards the upper left.

Explaining the map 

While demographics explain part of this story, they do not explain why exactly the ‘No’ majority line has fallen where it has – on age, density or even party splits – and aggressively against the government. The fact the vast majority of electoral districts sit beneath the line in Figure 10 suggests many persuadable voters have been passed over. 

The dividing lines of the vote come into sharper relief when we examine vote intention by 2022 Party Vote (Figure 11). Albanese’s government has only just managed to persuade a majority of his 2022 voters (33% of the electorate) to approve the proposal, and the only decisively convinced faction are Green voters (12%). The rest of the electorate (55%) are firmly against it. Even more troubling for Albanese, our MRP indicates a majority of Labor voters in Queensland and Western Australia – his ‘hero voters’ – opted for No (Table 3). 

Figure 11: Vote intention by 2022 federal elections first vote 

Table 3: Projected Yes vote by state, by 2022 federal elections first vote 

The sense that the Yes campaign has elided the median voter is reinforced by looking at voters’ political self-placement (Figure 13). Only the self-reported ‘left-leaning’ are overwhelmingly Yes. ‘Centrists’ (a catch-all term which likely includes the politically disengaged, too) have swung against the Government. 

Figure 12: Political self-placement histogram

Figure 13: Vote intention by political self-placement

These plots suggest a gap between the political positioning of the Yes campaign and the mood of the majority of the voters. To validate this, we asked voters directly about their reasons for holding their view. Voters in the Yes camp opted, overall, for the option suggesting the law would benefit disadvantaged Indigenous Australians (Figure 14). 71% picked this option among their top three, and a large proportion as their top reason (39%). The next two reasons – focused on ‘healing the nation’ (20%) and ‘apologizing for Australia’s colonial past’ (12%) – were only selected by small minorities as their top option. The Yes vote comes in mostly in a single shade of what I would call “sacred progressivism”. 

Yet this particular argument only came through strongly for a specific group of voters – the Greens, followed by Labor’s younger, more educated and urban voters. Other Yes voters, including those who voted for the Coalition in 2022, picked a more varied set of reasons from our list. It doesn’t seem like the ‘disadvantaged Indigenous Australians’ argument landed convincingly for them. Instead, the fact the second most popular argument – around ‘healing the nation’ – has shown to be demonstrably false throughout the campaign might explain why a bitter, angry campaign has seen so many right-leaning voters defect from undecideds or Yes, to No. Other aspects of “sacred progressivism” are also reflected in quite how different voter priorities are between Yes and No voters (Figure 21).  

Figure 14: Top three reasons for voting Yes

Figure 15: Top Reason for voting Yes, by party 

Turning to ‘No’ voters, there is more heterogeneity (Figure 15). No single argument convinced more than a third of them. But the different voter factions were aligned on a trifecta of arguments – against the idea of a two-nation constitution defined on race (30% top 1 reason), against the idea of First People’s as a concept (18%), and against the necessity of the referendum in the first place during a cost of living crisis (18%).  

Figure 16: Top three reasons for voting No 

Figure 17: Top three reasons for voting No, by 2022 first vote 

Crucially, we also asked voters what they thought the most persuasive arguments for the opposite camp were. Having been shown the same set of options, a convincing 67% of No voters thought none of the pro arguments were convincing on any level, in contrast to 7% of Yes Voters. There could be several reasons for this. The fact that, as the polling trajectory has shown, many current No voters may have been Yes voters only a few months ago could reflect a journey of persuasion that has resulted in a much more robust set of convictions. Or, more fundamentally, the dominant Yes argument may be too wrapped up in the values and associations of a progressive wing of Australian politics to convert the required number of these ‘centrists’ to the cause.  

Figure 18: Proportion saying ‘None of these’ arguments are convincing, by vote intention

We put this theory of an ‘ideological gap’ to the test. We asked a series of questions focused on key issues – like housing, equality and geopolitics – to map the ideological terrain of Australian politics. We then ‘summarised’ the two underlying dimensions that best define this political space. The result is an electorate scattered along two key dimensions, a cultural values axis, and a Europe-facing / Asia-facing alignment axis. The plot, along with 2022 voters’ average position on these axes, are outlined below (Figure 19). 

Figure 19: 2022 Party and Voice referendum in political values space 

A number of things immediately stand out. Self-reported ‘left-leaning’ voters – who make up over 40% of the Yes coalition – are, unsurprisingly, positioned deep in top-left quadrant territory. Crucially however, they are at a considerable distance away from the centre of Australian politics, where the vast majority of voters are. When we overlay the positions of Yes, No and Don’t know voters, this spatial distance becomes more obvious. Yes voters hold deeply progressive, or what we might call postmaterialist, values. They hold these even more than Green voters. No voters sit among a wider group of right-leaning voters. They are closest to the Coalition’s 2022 voters, and outflanked far to the right by Pauline Hanson’s One Nation. 

The implication is clear: the values to which the Yes campaign appealed are values held strongly by a vocal minority, themselves far removed from the centre ground of Australian politics. They are certainly far removed from the voters yet to be convinced (DKs) – who lean left, but are yet to make the leap to a Yes vote. In the end the Yes campaign looked more like the Greens than No looked like One Nation. The role that Jacinta Price played in this crucial difference will be one analysed in-depth post vote. 

 

By way of illustrating this lack of congruence between the ‘Yes’ campaign and its potential voters, we also plotted the top arguments from both sides of the campaign on the values map (Figure 20). It is a telling sign that the top ranked argument for the ‘Yes’ vote – focused on disadvantaged Indigenous people – is more progressive than its wider Yes coalition. By contrast, the arguments by the No campaign – as well as No voters themselves – are all clustered further in towards the median voter. This suggests a deeper coverage among these voters, and the wider resonance of the No arguments among its proponent base.

Figure 20: Top reasons, 2022 party and Voice referendum VI in political values space 

Figure 21: Top concerns of Australians vs yes/no split

Where Australia goes next  

Looking forward, we expect Albanese will suffer from some kind of voting intention discount in the months to come. More than twice the number of voters think worse (12%) of him than better (5%) as a consequence of the Referendum. The opposite is true of Liberal leader Peter Dutton. However, at this stage Australian polls still show a strong preference of voters for Albanese over Dutton – and the ALP still showing a strong voting intention lead (we will be doing more on this in the months to come). Sporting and corporate organisations are I’m sure reflecting on how hard they rowed into this Referendum - largely on the Yes side, when both the axes of division were sharply demographically clear, and in the end the win for No looked so comprehensive. Proper scenario planning, research and historic analysis would have shown the risks of this it poses to consumer brands and to potential customer backlash. I’m sceptical that corporate Australia undertook the necessary research before wading in.

 
Figure 22: Albanese Approval
Figure 23: Dutton Approval

Another sign that our polling may not be overestimating the final result is that when we asked respondents what they thought the No vote would be in their local neighbourhood (Figure 24). The average figure they gave was 61%; but with the 57-65% No vote estimates varying only slightly by type of respondent and age-cohort except for the very young. This suggests that some of the sampling flaws of online polls may be partially ameliorated by asking people to estimate how their community is going to vote. Certainly in prior referendums – marriage equality in Ireland, and Brexit in the UK – that has proved an interesting alternative method of forecasting. 

Figure 24: Wisdom of Crowds. Percentage of people in local neighbourhood that will vote no, by demographic bucket 

Figure 25: Impact of a No vote in the referendum (% net improve)

Postscript: Off only a base size of c.250 respondents we are not publicising our Indigenous yes/no vote share break - as we believe that this finding would drown out other findings based on more data. With a sample size this small the margin of error is large, and sampling frame even more difficult. BUT - it does look like 30% of Indigenous voters opted for No, and 70% yes. Rather extraordinary to think that when this Referendum was announced yes was over 70% for all of Australia. 

How we might be wrong?

Our poll has a higher no vote than many of the recent polls (except Newspoll) conducted – few of which have them over 61%. This is our estimate. Thoughtful polling and modelling should be  transparent over the limitations of our approach and where we might have sources of error:

  1. Our poll was conducted over two and a half weeks from 18 September to 2 October. The vote is on 14 October. If there is a last 2-week swing to Yes, we will miss it. However, the political science literature is fairly clear on last minute campaign polls often being less predictive than those taken at the beginning.
  2. We have asked respondents how they voted at the 2022 election and have not adjusted for false recall. This could well have the effect of upweighting Liberal/National voters beyond their correct level if people have forgotten they have voted for the losing side (as is the case for the 2017 Same-sex referendum). Our raw data of all respondents is 56% no, 44% yes – but was not fully representative of Federal 2022 election (too many Labor voters) and gender (too many females) in particular so was adjusted on this basis to be nationally representative 
  3. Our NSW and Victoria subsamples (1,270, and 1,068 in size each and representative) required significant upweighting of Coalition 2022 voters (average weight of 1.3x on average respectively) to be representative at a state level; raising the prospect that we were not able to capture the time poor high income liberals were more likely to vote yes. The actual CED estimates may yet reveal a more variegated map with more liberal areas in Melbourne and Sydney conurbations opting more strongly for yes
  4. Our Independents sub-sample surprised us with their strong no preference. Given the profile and demographics of the Teal independents (a substantial proportion of independents) of last year this surprised us. If there was one aspect of our data collection that challenged our priors, and continues to do so – it was this. We could well be structurally underestimating the yes vote across the Teal seats – qualitative assessment certainly suggests so. It’s also worth noting that YouGov’s Teal estimations were the worst of its 2022 model 
  5. Regularisation is a constraint of MRP – and one of the potential errors of our seat and vote forecasts is that we have underplayed how high the no vote could be in Queensland and Western Australia and yes vote in inner city seats. Our estimates were higher for No in Australia in aggregate, they were not higher for WA and QLD than public polls. 
  6. We may well be overestimating the No vote in population centres with high numbers of Indigenous people - there has been public scepticism - in particular from the Australian PM on our estimates. MRP is designed to get the overall seat count and national story correct though - and we remain confident in that.  

Sample

Focaldata online panel polling of 4,608 respondents across Australia. Fieldwork was carried out between 18 September and 2 October.

MRP model specification

  • Random effects: Age, Gender,  Education,  Income, birth place, Commonwealth Electoral District, 2022 Federal vote primary vote = 380,520 post strat sheet
  • Fixed effects: State + 2022 Federal vote simplified

Questions about this memo to Focaldata Chief Research Officer James Kanagasooriam at james@focaldata.com or Oliver Sheinwald at oliver@focaldata.com 

About Focaldata

We're on a mission to power the world’s understanding of what people think and do.

We research public opinion for data-driven organisations. With Focaldata, decision makers and researchers get the most rigorous data and analysis at 5x the speed — reducing the time to decision while delivering uncommonly actionable insights. We are a team of engineers, data scientists, product specialists and researchers building outstanding technology and next-generation services. Our services are global. We are non-political and non-ideological. 

Find our more: www.focaldata.com

Sign up to receive the next issue of Bi_Focal

Book a demo

Stay connected

Subscribe to get the Bi_Focal newsletter delivered directly to your inbox.

Subscribe
Button Text