Safety or change? The 2023 Australian voice referendum

ABSTRACT

The 2023 Australian Indigenous Voice to Parliament referendum joined the 37 other referendums which have been defeated since federation, out of a total of 45 referendums. In contrast to the 1967 referendum on Indigenous Australians, which attracted a record majority of 90.8 percent, the Voice referendum gained just 39.9 percent support. This paper uses a large, nationally representative survey to explain voting in the Voice referendum and to compare it with the 1999 republic referendum. The results show that the absence of bipartisanship on the Voice resulted in voters prioritizing the potential risks of constitutional change over the prospect of better outcomes for Indigenous people. In the absence of bipartisanship, voters opted for safety over change – as they did in the 1999 republic referendum and in other unsuccessful referendums.

KEYWORDS:

• Referendums
• indigenous politics
• voting behaviour

The 2023 Aboriginal and Torres Strait Islander Voice was the 45th constitutional referendum held since federation in 1901.¹ The referendum proposal (hereinafter referred to as the Voice) failed to meet the dual thresholds of winning a majority of the national vote and a majority of the six states. The Voice therefore joined 37 other proposals which have been defeated, or 82 percent of all the proposals brought before voters. Of the eight successful proposals, all had bipartisan support and only one – the 1946 social services referendum – was held during a period of Labor government.

Placed in historical perspective, the fate of the Voice proposal is not surprising given the absence of bipartisanship. More surprising is that its failure contrasted with another constitutional change relating to Indigenous Australians – the 1967 referendum to include Indigenous Australians in the census count and to allow the Commonwealth government to make laws on behalf of Indigenous people – which was carried by a record majority of 90.8 percent of voters and produced an overwhelming majority in every state (Bennett 1985). At the outset, the 2023 Voice referendum was also expected to attract strong public support and opinion polls conducted in 2022 and early 2023 suggested that around seven out of every 10 voters supported constitutional change. In the event, support declined in published opinion polls throughout the remainder of 2023 and the Voice attracted support from just 39.9 percent of voters, the eighth lowest yes vote of the 37 other failed referendum proposals.²

Why did the Voice proposal fail so decisively? And what does the result tell us about patterns of referendum voting behaviour in Australia? To address these questions we use a large, nationally representative survey conducted immediately after the 14 October 2023 referendum, and a similar national survey conducted after the 1999 republic referendum. We reach three main conclusions. First, we find that the Voice proposal failed because of a lack of bipartisanship, despite widespread public support for better outcomes for Indigenous Australians. Second, a majority of voters considered the Voice to be risky would not deliver on its promised outcomes. Third, both the Voice and the republic referendums show that unless voters are convinced of the merits of a proposal, they will choose safety and the avoidance of risk over change and the prospect of a better future.

The paper proceeds as follows. The next section outlines the background to the Voice referendum while the second section examines voting behaviour in referendums compared to general elections and advances four main hypotheses. The third section outlines the data and the measures. The fourth section presents the results, separated into social background, information and campaign effects, the role of parties and leaders, and attitudes towards the benefits and risks of the constitutional change. The conclusion places the Voice referendum and the absence of bipartisanship in historical perspective.

The voice referendum

The origins of the 2023 Voice referendum lie in the establishment of the Referendum Council in 2015. Established with bipartisan support, the Referendum Council was charged with advising the government on the necessary steps to recognize Indigenous peoples in the constitution. The council organized a series of regional consultations with Indigenous communities which resulted in the May 2017 Uluru Statement from the Heart. The Uluru Statement³ identified the need for a permanent Voice to Parliament to advise on the special needs of Indigenous people, as well as a commission which would lead ultimately to a treaty between the government and Indigenous people (Appleby and Davis 2018; Davis and Williams 2021). The Voice proposal was rejected by the Liberal government as being unlikely to be supported in a referendum.

In contrast to the Liberal government’s response, Labor had a longstanding commitment to implementing the Voice and in 2022 stated that ‘if elected, we will move quickly on a referendum to enshrine a Voice to Parliament in our Constitution’ (Australian Labor Party 2022, 2). This commitment figured prominently in Labor’s election campaign and on election night, Anthony Albanese began his victory speech by committing the new Labor government to hold a referendum to implement the Uluru Statement in full.

Opinion polling following the 2017 Uluru Statement suggested that the proposal attracted majority support, with those in favour outnumbering opponents by at least two to one, albeit with significant minorities expressing no opinion. However, these early polls were conducted in advance of public debate around what institutional changes the Voice would require and prior to the Liberal/National Coalition indicating they would oppose the yes vote. Instead, the polls reflected general support for Indigenous reconciliation. Public support remained consistently high until late 2022, when the National Party announced on 28 November 2022 that it would oppose the Voice, arguing instead for ‘empowering local Indigenous communities, giving them the power at a local level, not creating another layer of bureaucracy here in Canberra’ (Grattan 2022). Thereafter public support began to wane, notably after the Liberal Party’s announcement on 4 April 2023 that it too would campaign against the Voice.

Figure 1 shows the trends in public support for the Voice during 2023.⁴ In the first three months of 2023, opinion polls showed that the average yes vote was 51 percent; in the three months after the Liberal announcement that they would oppose the Voice, the average yes vote was 45 percent. Thereafter the yes vote declined consistently and from early July onwards, the no vote exceeded the yes vote. At the same time, the proportion of respondents who said they did not know how they would vote declined, averaging about one in 10 of the electorate in the two months prior to the referendum. While this a major change in public opinion on a single issue over a relatively short period of time, it is in line with opinion change in the run-up to the 1999 republic referendum, when support for the change declined from 58 percent to 42 percent in the space of eight months (McAllister 2001: Figure 2). The consistent decline in the yes vote leading up to the referendum vote is also similar to what has been found internationally (LeDuc 2002a: Table 9.2).⁵

Figure 1. Voting Intention on the Voice, February-October 2023. Question wordings vary. Only polls which included a ‘don’t know’ option are included. Source https://simonjackman.github.io/poll_averaging_voice_2023/poll_averaging.html.

Figure 2. Generations and referendum voting. The generations are defined as follows: Great (born before 1928), Silent (1928-1945), Boomers (1946-1965), Gen X (1966-1980), Millennials (1981-1994), Gen Z (1995 or after). Sources 1999 and 2023 ACRS.

Referendum voting

Referendums are similar to general elections insofar as voters are presented with arguments and asked to make a choice at the ballot box. There the similarity mostly ends. In a referendum, voters are offered a ballot paper offering a binary choice about change to the political rules, while in a general election voters are presented with a ballot paper which offers a broad spectrum of choices between parties and candidates. As a consequence of these different choice sets, the cues provided to voters in referendums and general elections, both in the lead-up to the vote and at the vote itself, are very different (LeDuc 2007). While campaign effects are similar between referendums and general elections, the weight that voters place on the issue, and on partisanship and leaders, depends heavily on the context of the referendum, as well as on the cognitive abilities of individual voters (Garry 2014; Laycock 2013).

The decision voters must make in a referendum is whether or not to support a legal change to the constitution. To vote yes for a constitutional change, voters are opting for the prospect of a better future that will ultimately provide lasting benefit to themselves and their society. Casting a no vote implies a vote for safety, with voters seeing the proposed change as a risk to the political system or to their way of life. Each of these two options brings into play separate emotions. Change implies hope, and an expectation that better outcomes will be forthcoming in the future. In contrast, voting for safety implies an aversion to risk and opting for the certainty of what is known (Beach 2018). Prioritizing safety over change also raises the fear that something tangible might be lost if a proposal is passed, such as a well-paid job or a comfortable standard of living. Not surprisingly, most referendum campaigns have used change and hope for a better future as arguments in favour of a yes vote, while safety and the avoidance of risk are used to argue the case for a no vote (Black et al. 2023).

When weighing up whether to vote yes or no in a referendum, what cues do voters rely on to make their choice, given the institutional differences between a referendum and a general election vote? In this paper, we examine four main explanations covering social background, information and campaign effects, the role of parties and leaders, and attitudes towards the referendum issue itself, expressed as safety versus change. For each of these four explanations, we advance a hypothesis to test their importance in shaping vote choice in the 2023 Voice referendum, with comparisons to the 1999 republic referendum. These explanations necessarily overlap but by estimating a series of multivariate models we also show which of the four explanations (if any) mattered most to the referendum outcomes in 1999 and 2023.

Comparing voting in the Voice referendum with the 1999 republic referendum allows us to draw some general conclusions about referendum voting behaviour in Australia. The two referendums shared three main characteristics. First, both proposals lacked bipartisan support, and in each referendum the Liberal Party was more internally divided than Labor. Second, in each case the public was broadly supportive of the principle underlying each proposal – the creation of a republic in 1999 and reconciliation with Indigenous Australians in 2023 – but divided over how to operationalise it. Third, as we explain later, the questions put in front of voters lacked the crucial political clarity required to elicit broad public support. In both cases, voters were not convinced that the change would have positive benefits and consequently opted for safety and the status quo.

Hypotheses

Since voting behaviour has traditionally been based around social cleavages, we might expect that voters’ social backgrounds will play an important role in their voting decision. Social cleavages are most likely to shape a referendum vote when the referendum question aligns with current political debates, such as economic management, or with issues of identity. Voters with higher socioeconomic status are more economically secure and therefore in a better position to tolerate risk. Those with lower socioeconomic status are likely to be more concerned about the threat to their jobs or living standards. These were significant factors during the 2016 Brexit vote in the UK (Hobolt 2016; Steenbergen and Siczek 2017). In addition, if the referendum question is a moral issue that aligns with a liberal-conservative dimension, such as the 2018 Australian same sex marriage plebiscite (Gravelle and Carson 2019; McAllister and Snagovsky 2018) or the 2018 Irish abortion referendum (Elkink et al. 2017; 2020), generational effects are likely to be a major driver of the vote. In the context of the Voice, this leads to two related hypotheses:

H1a Voters with higher socioeconomic status will be more likely to support the Voice compared to voters with lower socioeconomic status.

H1b Voters who are millennials and gen Z will be more likely to support the Voice compared to voters from older generations.

A second potential influence on the referendum vote is information. In the absence of strong party cues, voters will seek political information in lieu of partisanship to guide their decision. This is especially important when the decision is technical or legal, which is often the case in a referendum. The flow of information is likely to increase in a more intensive referendum campaign, and here voters will rely more heavily on their pre-existing attitudes and opinions than on their partisanship to make a decision. For example, the information flow during the (unsuccessful) 2011 British referendum to change the electoral system – a complex, technical question – proved crucial to the outcome (Vowles 2013), as was information about the European Union during the various (mostly successful) ratification referendums held in the various member countries (Beach 2018; Hobolt 2005).

The information flow is mediated by the campaign, so how the debates over the referendum question affect voters can be crucial (de Vreese and Semetko 2004). When the referendum question is unfamiliar to voters, the campaign represents the only means by which voters can gain a better insight into what is being proposed (LeDuc 2002b). As well as conveying direct information about the proposed change, the campaign can also highlight the positions of parties and leaders from which voters can derive cues. The extensive use of voting initiatives on US state ballots has generated research to show that campaigns are crucial in shaping voters’ choices (for a review, see Lupia and Matsusaka 2004). The campaign also highlights the salience of the issue; in the case of high salience issues voters engage with the issue itself, while with low salience issues, voters are more likely to rely on other cues (Hug 2002). The impact of information and the campaign on the referendum outcome is expressed in the second pair of hypotheses:

H2a Voters who pay more attention to the campaign will be more likely to support the Voice compared to voters who pay less attention to the campaign.

H2b Voters who have more political information will be more likely to support the Voice compared to voters who have less political information.

The positions of the various parties and leaders on the referendum question provide heuristic shortcuts for voters, especially when the question is not anchored in citizens’ belief structures. There is considerable research to show that parties, and the actions of party and non-party leaders, can have a major influence on a referendum vote (Pammett and LeDuc 2001). The personalization of politics has provided an added dimension to party and leader cues, and there is evidence from specific referendums, such as the 1999 Australian republic referendum (Bean 2002) and the 2020 Italian electoral representation referendum (Bergman and Passarelli 2023), to show that feelings towards leaders have a greater impact on the vote than feelings towards parties. If the leader is a person who voters like, then voters are more likely to follow their view on the referendum question (Hobolt 2005). This leads to the third hypothesis:

H3 Feelings towards leaders will have a greater influence on the Voice vote than feelings towards parties.

Finally, pre-existing attitudes and opinions will play a role in shaping a voter’s choice in the referendum. Two views have dominated Australian attitudes towards Indigenous people. One is reconciliation and the need to recognize Indigenous Australians in the constitution. Such a change is relatively uncontroversial and widely supported by the mass public (Levy and McAllister 2022). Related to reconciliation is the need to design policies aimed at reducing the socioeconomic disadvantages faced by Indigenous people compared to other Australians. This view sees change as delivering better lived experiences for Indigenous people and presents positive outcomes for the Voice. The second view concerns the widespread belief that Indigenous people receive special benefits from government that are not available to other Australians, and that this undermines the principle of equality.⁶ This reflects the view that any change to the constitution brings with it major risks. These two contrasting views are expressed in two final hypotheses:

H4a Voters who are more concerned with the positive benefits of change are more likely to vote yes compared to voters who are less concerned with the positive benefits of change.

H4b Voters who are more concerned with the risks of change are more likely to vote no compared to voters who are less concerned with the risks of change.

Data, measurement and method

Data. The 2023 Australian Constitutional Referendum Study (ACRS) survey was conducted as part of the ANUpoll survey series. The survey was fielded online among a nationally representative sample of 4,219 adults between 17 and 29 October 2023 with a response rate of 74.6 percent from the Life in Australia probability-based panel. The October 2023 survey can also be linked at the individual respondent level to the January, April, and August 2023 ANUpolls, permitting a detailed tracking of how the referendum vote changed over the campaign period. The survey is weighted to represent the characteristics of the national electorate, including the actual yes vote. The 1999 republic ACRS survey was conducted immediately after the 6 November 1999 referendum and collected a nationally representative sample of 3,341 respondents from the electoral register, with a response rate of 58.9 percent (see McAllister 2001: Appendix). Estimates from both the 1999 and 2023 surveys are restricted to those who reported casting a valid vote in the referendum.⁷

Measurement. The Voice survey replicated many items from the 1999 republic ACRS survey to ensure overtime comparability. In both surveys, the main dependent variable is referendum vote, scored yes versus no. The detailed definitions of the variables, question wordings, and means, are shown in the Appendix.

Method. Since the dependent variable is dichotomous, we use logistic regression procedures.

Results

Social Background. If a referendum issue aligns with current patterns of party competition, it is more likely that voting will be anchored in social cleavages and opinion change will be less (LeDuc 2002a). Since a voter who is more economically prosperous should be better able to tolerate the risk involved in a constitutional change, we would expect them to be more likely to vote yes. Similarly, more education should produce more engagement and understanding of the referendum issue and be more likely to result in a yes vote (H1a). In recent elections, generation has emerged as being at least as important as economic status in shaping a person’s vote. We might expect, therefore, that those who vote for change in a referendum will be more likely to be millennial or gen Z members compared to other voters (H1b).

To test the first hypothesis, Table 1 shows the proportion voting yes and no in the 2023 referendum, with a comparison to the vote in the 1999 republic referendum. The patterns are broadly consistent across the two elections, almost a quarter of a century apart. In both referendums, the yes vote is drawn from the better educated, those with higher incomes and those living in major cities. In both referendums, for example, yes voters were more than twice as likely to be drawn from tertiary educated voters as voters with no educational qualifications. We therefore confirm H1a.

Table 1. Social background and the vote.

	1999 vote		2022 vote
	Yes	No	Yes	No
Indigenous	–	–	3	2
Gender
Female	51	52	53	46*
Male	49	48	47	54*
Educational qualifications
Tertiary qualification	25	9*	41	20*
No qualifications	42	54*	22	31*
Family income
Bottom quintile	10	13*	15	22*
Top quintile	25	17*	27	19*
Ethnicity
Born English speaking country	9	12*	11	11
Born non-English speaking country	13	7*	12	11
City resident	57	42*	71	60*
(N)	(1,543)	(1,700)	(1,261)	(1,872)

Figures are the percent of yes and no voters with a particular characteristic. * Difference between yes and no voters statistically significant at p<.01. Sources 1999 and 2023 ACRS.

There are also two differences in social background between the two referendums which may reflect the nature of the two different referendum questions. First, women were more likely to vote yes in 2023 compared to men while in 1999 there was no significant gender difference. Second, while there were no birthplace differences in 2023, in 1999 non-English speaking migrants were more likely to vote yes for the republic compared to those born in an English-speaking country, mainly the UK, which retains the British monarchy.

To test the hypothesis (H1b) that generational differences were important drivers of the vote, Figure 2 shows the yes vote in 1999 and 2023 separately for six generations. In 1999 the difference in the yes vote between the oldest (great) generation and the youngest (gen X) is 25 percentage points. In 2023 the difference between the oldest (silent) generation and the youngest is 34 percentage points. With the exception of gen X in 1999, in each referendum a new generation is more likely to vote for the proposition than the generation that preceded it. We therefore confirm H1b.⁸ We can interpret these generational patterns in various ways, but one conclusion is that newer generations are more likely to be comfortable with change since it aligns with their liberal values, while older generations opt for safety and resist change, unless they are convinced that the change will ultimately benefit the country.

Information and the Campaign. How much attention did voters pay to the referendum campaign and how much did they know about the referendum question? And did the 2023 Voice differ on the levels of information and the progress of the campaign compared to the 1999 republic referendum? To answer these questions, Table 2 shows how the campaign and the levels of information affected the vote in the two referendums. Overall, the Voice was clearly a lower salience referendum compared to the 1999 republic. Around one in four said they discussed the Voice referendum compared to one in three who said they discussed the republic. The proportion who cared about the outcome of the election was also lower in 2023 than in 1999. The dynamics of the two referendum campaigns also differed significantly, with more saying they knew enough about the issue to make a decision in 2023 than in 1999, and fewer in 2023 than in 1999 saying they thought of changing their vote. Voters in 2023 also decided on their vote closer to election day than in 1999.

Table 2. Campaign and information effects.

	1999 vote		2023 vote
	Yes	No	Yes	No
Political interest (percent ‘good deal’)
Interest in election campaign	44	32*	46	31*
Discussed the election	36	31*	29	24*
Cared about outcome	73	72	76	62*
Voting decision (percent yes)
Vote if not compulsory	90	83*	88	73*
Knew enough about issue	52	53	66	57*
Thought of changing vote	33	30*	24	25
Decided vote long time ago	46	39*	37	25*
(N)	(1,543)	(1,700)	(1,261)	(1,872)

Figures are the percent of yes and no voters with a particular view. * Difference between yes and no voters statistically significant at p<.01. Sources 1999 and 2023 ACRS.

How did information and the campaign correlate with the vote in the two referendums? We hypothesize that the more attention a voter paid to the campaign the more likely they were to vote yes (H2a) and the more information they possessed the more likely they would be to vote yes (H2b). These hypotheses are tested in Table 2. There are significant differences between yes and no voters in both referendums. For example, in both 1999 and 2023 significantly more yes voters than no voters said they had a good deal of interest in the campaign. The only exception is in 1999 when there is no significant difference between yes and no voters caring about the referendum outcome. Based on the other large differences between yes and no voters, we therefore confirm H2a.

In terms of how the voting decision was influenced by the level of information, Table 2 shows that in both referendums, yes voters were generally more committed to their view than no voters. Significantly more yes voters than no voters said they would vote if the election was voluntary rather than compulsory and had arrived at their voting decision a long time ago. In 2023 more yes voters than no voters said they knew enough about the issue while there was no significant difference between the two groups in 1999. We therefore find partial support for H2b. Overall, information mattered more to the 2023 referendum outcome compared to 1999, and the commitment of yes voters to their view in 2023 was considerably greater than in 1999.

Parties and Leaders. The positions of the major parties and their leaders on a referendum question represent a key influence on the outcome. If the political elite adopts a consensual position on a referendum question, particularly a question which is technical or complex in nature, it will be difficult for opponents of change to make a convincing case against it (de Vreese 2007; Kriesi 2006). Not least, opponents of change will have difficulty in publicizing their arguments in the mass media given their lower political profile compared to an established and united political elite. We also expect that the public’s closeness to leaders will be more important than their closeness to parties in influencing the vote given the personalization of politics (H3).

To evaluate the importance of bipartisanship on the referendum outcome, Table 3 estimates the yes vote in 1999 and in 2023 by general election vote. In the case of 2023, we are also able to make use of the panel component of the survey to estimate how the party distribution of the vote changed in the months leading up to the referendum. Partisanship was clearly important in shaping the vote in both referendums, but more so in 2023 than in 1999. In 1999, 35 percent of Coalition supporters voted for the republic, compared to 60 percent of Labor supporters. The panel component of the 2023 survey shows that by the time the referendum was held in October, just 8 percent of Coalition voters supported the change compared to 62 percent of Labor supporters. Indeed, the proportion of Coalition voters supporting the yes side declined by 30 percent over the four survey waves, compared to a decline of 16 percent among Labor supporters. In contrast to 1999, when the Liberal Party was divided over the republic, in 2023 the party was broadly united in its opposition to the Voice, except for a small number of individuals.⁹

Table 3. Yes Vote by Federal Election Vote

	2023
Vote	1999	January	April	August	October	(Change)
Coalition	35	39	35	24	8	(−30)
Labor	60	78	80	70	62	(−16)
Green/Democrat	59	94	90	81	78	(−16)
Other/undecided	46	53	47	34	22	(−31)
(N)	(3,122)	(2,277)	(3,674)	(3,673)	(3,644)

Election vote in 1999 is recalled vote in the 1998 election, and in 2023 voting intention if an election were held tomorrow. The 2023 estimates are based on those who completed the October 2023 survey and voted in the referendum.Sources January, April, August 2023 ANUpolls; 1999 and 2023 ACRS.

While party attachment was obviously important in determining the outcome of the 2023 Voice referendum, how did it compare with the influence of the party leaders? To answer this question, we use the likeability thermometer scales for party and leader which are available in both the 1999 and 2023 surveys. Table 4 shows the mean score for each major party and its leader on the zero to 10 scales for the yes and no votes in the two referendums. The estimates are presented separately for Coalition and Labor voters. To evaluate the relative importance of party as opposed to leader, Table 4 shows the difference in the mean score in parentheses between yes and no voters separately for the parties and leaders. These differences show whether party or leader was more important in shaping the referendum vote. For example, in 1999, John Howard rated 6.6 on the scale among Coalition yes voters compared to 7.6 among Coalition no voters. The difference of −1.0 suggests that Howard was a stronger driver of the vote than the Liberal Party, which produces a difference between yes and no voters of −0.6.

Table 4. The Pull of the Parties and Leaders

	1999				2023
	Howard	Beazley	Liberal	Labor	Dutton	Albanese	Liberal	Labor
	(Coalition)
Yes	6.6	4.9	7.2	4.3	4.5	4.9	6.3	4.5
No	7.6	3.8	7.8	3.2	6.2	2.7	7.1	3.0
(Yes-No)	(−1.0)	(1.1)	(−0.6)	(1.1)	(−1.7)	(2.4)	(−0.8)	(1.5)
	(Labor)
Yes	2.7	7.8	3.0	8.1	1.6	7.7	2.4	7.7
No	3.5	6.8	3.4	7.6	3.4	6.1	3.6	6.8
(Yes-No)	(−0.8)	(1.0)	(−0.4)	(0.5)	(−1.8)	(1.6)	(−1.2)	(0.9)

Estimates are means from zero to 10 thermometer scales. In 1999 vote is 1998 federal election vote, in 2023 vote if an election were held tomorrow. Coalition N = 1,393 (1999), 1,145 (2023). Labor N = 1,272 (1999), 1,110 (2023). Sources 1999 and 2023 ACRS.

The patterns of the party and leader vote across the two referendums show that leaders were consistently more important than party in shaping the vote. In 2023, the difference between the ratings of Peter Dutton and the ratings of the Liberal Party are −1.7 and −0.8, respectively, suggesting that feelings about Dutton were a more important driver of the vote than feelings about the Liberal Party. The patterns are very similar in 1999, with one exception: Kim Beazley’s rating was the same (1.1) as the rating for the Labor Party. Otherwise, the results in Table 4 suggest that in both referendums, feelings about leaders were more important to voters in shaping their vote than feelings about the major political parties. We therefore confirm H3.

Referendum Benefits and Risks. Voters’ attitudes will matter more to their referendum vote if the question is one they are familiar with and aligns with their pre-existing opinions. In the case of attitudes on Indigenous issues, public debate is rare and political elites have endeavored to maintain bipartisanship. Both parties have endorsed a policy agenda aimed at reducing the socioeconomic differences between Indigenous and non-Indigenous Australians. There are three prominent exceptions to this broad bipartisanship: the 1992 Mabo decision and the economic implications of land rights (Goot and Rowse 2007, 89–91; Patapan 2000, 141–145); the apology to the Stolen Generation; and the 2007–2012 Northern Territory Intervention (Macoun 2011). In contrast to 2023, there had been extensive public discussion about the advantages and disadvantages of the proposed change to the republic well in advance of the 1999 referendum so in many respects, voters were better informed about the issue (McAllister 2001).

A plethora of arguments were advanced to support the yes and no cases during the Voice referendum campaign. For the yes case, the formal booklet distributed by the Australian Electoral Commission listed the importance of recognition for Indigenous people, listening to their advice on matters affecting them, and achieving better results in narrowing the social and economic gap between Indigenous and non-Indigenous Australians. For the no case, the booklet listed the risk involved to the current system of government, the lack of detail about how the Voice would operate, and the risk of division by creating different classes of citizens. Very similar arguments contrasting change and hope as opposed to safety and risk were advanced during the 1999 republic referendum campaign (Tranter 2003).

Both the 1999 and 2023 referendum surveys asked the respondents what they considered were the important considerations in their vote, separated between factors that they regarded as benefits for the future and potential risks. For the Voice, Table 5 shows that the benefits endorsed by a majority of voters ranged from listening to Indigenous views (which 84 percent regarded as ‘very’ or ‘fairly’ important) to connecting with Indigenous history (56 percent). The risks were seen as less important than the benefits, though 65 percent said the risk that the Voice would divide the country was important to their vote. While the question about benefits versus risks was asked in a different form in the 1999 survey, the same pattern emerges (Table 6), with benefits for the future attracting more support than the possible risks.

Table 5. Benefits and risks on the voice.

	Very important	Fairly important	Not very important	Not at all important	Total
Benefits
Listen to Indigenous views	48	36	10	6	100
Better Indigenous outcomes	41	28	17	14	100
Connect with Indigenous history	31	25	24	20	100
Risks
Concerned about dividing country	45	20	14	21	100
Change too risky	32	21	19	28	100

N = 3,134. See Appendix Table B for question wording. Source 2023 ACRS.

Table 6. Benefits and risk on the republic.

	Strongly agree	Agree	Neither	Disagree	Strongly disagree	Total
Benefits
Australia more independent	17	41	22	15	5	100
Australia’s standing improved	12	27	37	19	5	100
Government more democratic	5	18	43	26	8	100
Risks
Political stability endangered	7	17	35	31	10	100
Future reform difficult	8	14	28	32	18	100
Weaken federalism	5	12	39	34	10	100

N = 3,243. See Appendix Table C for question wording. Source 1999 ACRS.

How far did the possible benefits as opposed to concern about the risks influence the vote in the two referendums? Our expectation is that voters motivated mainly by the benefits of the proposal will vote yes, while those seeing greater risks in the proposal will vote no. Figure 3 addresses this question by combining the change and risk items into multiple item scales, scored from zero (least support) to 10 (highest support) and compares the means for yes and no voters in the 1999 and 2023 referendums. The factor analysis results for the two sets of questions are shown in Appendix Tables B and C and show that the two underlying dimensions – benefit and risk, respectively – are statistically appropriate.¹⁰

Figure 3. . Referendum benefits and risks. Figures are means on zero to 10 scales based on the factors in Appendix Tables A and B. Sources 1999 and 2023 ACRS.

The results in Figure 3 show that the benefits/risks calculation was significantly greater between yes and no voters in the Voice referendum compared to the republic. In 2023, for example, yes voters saw the benefits of the change as 8.6 on the scale, compared to 2.6 for no voters – a significant difference of 6.0 on the zero to 10 scale. By contrast, the same difference among yes voters for the republic was just 3.3. There are similar, though less substantial, differences among no voters in the two referendums. Overall, while many voters saw benefits in both referendum proposals, their votes in the two referendums appear to have been driven more by their concerns for the risks involved rather than by the positive benefits the changes might deliver. We therefore confirm both H4a and H4b but find more support for the latter than the former.

Evaluating the Explanations. It remains to evaluate the impact of these various explanations on the referendum vote, taking all other factors into account. To make these estimates, we calculate a logistic regression equation for each of the two surveys predicting the yes versus no vote. These estimates are shown in Table 7 which show the parameter estimates (B) and the odds ratios for each of the variables.

Table 7. Predicting the yes versus no referendum vote.

	B (Odds Ratios)
	1999	2023
Social Background
ATSI	–	.13 (1.14)
Female	.14 (1.15)	-.14 (.87)
Generation (Baby Boomers)
Great	-.78 (.46) **	–
Silent	-.18 (.83)	-.52 (.59)
Gen X	.05 (1.05)	.09 (1.09)
Millennials	–	.06 (1.06)
Gen Z	–	.30 (1.35)
University education	.63 (1.88) **	.45 (1.57) **
High family income	.13 (1.14)	.18 (1.20)
Born English speaking country	.17 (1.18)	.57 (1.77) *
Born non-English speaking country	.40 (1.49) *	.64 (1.90) **
City resident	.26 (1.30) **	.57 (1.77) **
Campaign
Interest in election campaign	.33 (1.39) **	.14 (1.15)
Discussed the election	.09 (1.09)	-.38 (.69) **
Cared about outcome	-.64 (.53) **	.37 (1.45) *
Voting Decision
Definitely vote if not compulsory	.40 (1.49) *	.55 (1.73) **
Knew enough about issue	-.13 (.88)	-.09 (.91)
Thought of changing vote	-.19 (.83)	.29 (1.34)
Decided vote long time ago	-.05 (.96)	-.08 (.93)
Parties and Leaders
Kim Beazley/Anthony Albanese	.13 (1.14) **	.30 (1.34) **
Labor	.01 (1.00)	-.08 (.92)
John Howard/Peter Dutton	-.10 (.90) **	-.10 (.91) **
Liberal	-.01 (.99)	-.16 (.85) **
Attitudes to Referendum Issue
Benefits	.47 (1.60) **	.64 (1.90) **
Risks	-.59 (.56) **	-.64 (.53) **
Constant	-.44	−2.61
Nagelkerke R-sq	.56	.83
(N)	(3,243)	(3,134)

* p<.05, ** p<.01.

Logistic regression equations showing parameter estimates and (in parentheses) odds ratios predicting the yes vs the no vote in the republic and Voice referendums. See Appendix Table A for details of variables and scoring. Sources 1999 and 2023 ACRS.

In both referendums by far the most important predictor of the vote was how far the voter balanced the possible benefits and risks inherent in each proposal. The risks of the change were a more significant driver of the no vote than the possible future benefits was to the yes vote. Next in importance in both referendums is the impact of feelings about the parties and the leaders, with leaders being the stronger predictor of the two. This confirms the findings reported earlier, this time controlling for a wide range of potentially confounding factors, and highlights the importance of leader cues in influencing voters’ choices in a referendum. In particular, views about Albanese were a strong predictor of the vote, reflecting his personal commitment to the Voice. In 1999, the influence of views about Kim Beazley, the Labor leader, were less than half as important as that of Albanese in 2023, though Beazley was also a strong supporter of the referendum proposal.

While attitudes towards the benefits and risks of the referendum issues and views of leaders and parties were by far the most important predictors of the vote, there are also some common patterns between the two referendums. In terms of social background, university educated voters and those living in cities were more likely to support change, more so in 1999 than in 2023. Surprisingly, given the patterns in Figure 2, generation is unimportant once other factors are taken into account, except for the Greatest generation in 1999, who had experienced the Second World War and were significantly less likely to vote for a republic. There are few commonalities between the impact of the campaign and the voting decision between the two referendums, suggesting that the two very different referendum issues shaped the views of voters in distinct ways.

The relationship between the benefit/risk calculation, leader cues and party loyalties in shaping the referendum yes votes in 1999 and 2023 is illustrated in Figure 4. Since all of the variables share the same zero to 10 scale they show graphically how important the benefit/risk calculation dominated voters’ decisions in each referendum. National leadership also mattered, but Anthony Albanese’s personal commitment to the Voice is distinctive, certainly in comparison to his Labor counterpart in 1999, Kim Beazley, himself a strong supporter of the republic. By contrast, voters were much less likely to follow the cues provided by their party loyalties, with only a Liberal loyalty in 2023 being of importance, which reflects the formal position the party took against the Voice.

Figure 4. . Leaders, Parties, Benefits and Risks and the Yes Vote. Figures are odds ratios from Table 7 showing 95 percent confidence intervals. Sources 1999 and 2023 ACRS.

Overall, what mattered most in the 2023 Voice referendum vote were concerns about the risks of change, followed by the arguments put forward to Albanese for the yes case, and the position of the Liberal Party (and to a lesser extent Peter Dutton) on the no side. Ultimately, the yes campaign was unable to convince a sufficient number of voters that the Voice was not a traditional party political issue, and that the proposal would not confer special privileges on one group. While the goal of reconciliation is broadly supported across the population, the desire to ensure equality of treatment by the political institutions proved to be a stronger driver of the vote. For many voters, equality of treatment was seen as more important than giving special rights to one group, however deserving.

Conclusion

The two aims of this paper have been to better understand why the 2023 Voice referendum failed and to compare voting patterns in 2023 with those of the previous referendum in 1999. On the first question, the failure of the Voice referendum to attract a majority of voters can be explained by voters’ concerns about the risks involved in a constitutional change and to the role of the parties and leaders during the campaign. The absence of bipartisan agreement on the proposal led voters to rely on their pre-existing opinions about Indigenous issues and on party political cues in deciding how to vote. As Australian and international research shows, bipartisanship is critical to a successful referendum outcome, and coalition formation among political elites is a key determinant of the outcome (Kriesi 2006; LeDuc 2002b; McAllister 2001). In the absence of bipartisanship, the government was unable to convince enough voters that the goal of Indigenous reconciliation should take precedence over maintaining equality of treatment in the constitution.

Would bipartisan agreement on the Voice have changed the outcome of the 2023 referendum? If we assume that the same proportions of party supporters in January 2023 would have voted yes in October 2023 vote, then the proposal would have passed, with 60 percent voting yes and 40 percent voting no.¹¹ When confronted with a complex referendum question, voters rely on the cues provided by their political leaders. If there is elite disagreement, then the balance of risk versus change will shift towards the former and the referendum will fail.

Comparing the 1999 survey results with those for 2023 enables us to address the second question and to identify some common patterns in referendum voting in Australia. The overarching finding is how the absence of bipartisanship ensured the failure of both the republic and the Voice proposals, but particularly in latter case. In the absence of bipartisanship, voters saw the risks involved in the change to be much greater than the potential benefits and voted accordingly. During the 1999 republic campaign, with a majority supporting the principle of a republic, the question was redefined from a choice between a monarchy or a republic, to whether the republic should have an elected or appointed president (Higley and McAllister 2002).¹² In 2023, with most voters favouring some form of reconciliation with Indigenous Australians, the core question was similar to 1999, namely whether the model proposed in the referendum could deliver an agreed outcome. In both referendums, a majority of voters thought the model that was on offer would not work.

Ultimately, changing the rules of the political game through a referendum requires agreement between the major players. In the absence of that agreement, voters are likely to see too many risks in the change and will instead opt for safety.

Acknowledgements

An earlier version of the paper was presented to the Australian Society for Quantitative Political Science, Griffith University, 7–8 December 2023. Our thanks to two anonymous referees from this journal for their constructive comments on an earlier draft of the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data is available from the Australian Data Archive ( doi:10.26193/13NPGQ

 Google Scholar

).

Additional information

Notes on contributors

Ian McAllister

Ian McAllister is Distinguished Professor of Political Science at The Australian National University, Canberra.

Nicholas Biddle

Nicholas Biddle is Professor of Economics and Public Policy at The Australian National University.

Notes

1 There have also been three plebiscites, two (in 1916 and 1917, respectively) on conscription, both of which failed to win a majority, and one (in 1977) on changing the national anthem. In addition, there was one postal survey (in 2017) on same sex marriage.

2 The lowest yes vote was for the Rights and Freedoms proposal in 1988, which attracted a yes vote of 30.8 percent.

3 The Statement from the Heart is usually abbreviated to the Uluru Statement, named after the place where it was drawn up, Uluru (formerly Ayer’s Rock) in the Northern Territory.

4 The opinion results are from https://en.wikipedia.org/wiki/Opinion_polling_for_the_2023_Australian_Indigenous_Voice_referendum. Only polls are included which had a ‘don’t know’ opinion. See also https://simonjackman.github.io/poll_averaging_voice_2023/poll_averaging.html

5 LeDuc (2002a: Table 9.2) finds that based on 23 referendum proposals, the mean change from the start of the campaign to the outcome is a decline in the yes vote of 13 percentage points.

6 This view was central to the rapid electoral rise of Paul Hanson’s One Nation Party in the late 1990s (Gibson, McAllister, and Swenson 2002).

7 Of the total sample of 4,219 respondents in 2023, 16 percent were not eligible to vote, while 8 percent either didn't vote or cast an informal vote, and 2 percent refused to answer the question. The resulting sample is composed of 3,134 respondents who voted either yes or no in the referendum. The 1999 sample was drawn from the electoral register and of the total sample of 3,431respondents, 2 percent cast an informal vote or didn’t vote, and 4 percent refused to answer the question, resulting in a sample of 3,243 who voted either yes or no in the referendum.

8 The mean age of yes voters in 1999 was 44.5 years and no voters 49.4. In 2023, the mean age of yes voters was 45.3 years, and no voters 52.5 years.

9 In 2023 three Liberal House of Representatives members, Bridget Archer, Andrew Bragg and Julian Lesser, indicated publicly they would vote yes. Lesser was shadow Attorney-General and resigned from the shadow ministry.

10 The scales were constructed by first coding missing values on any one item to the mean and then combining the items identified in Appendix Tables B and C. Since the resulting scale had no natural metric it was transformed into a zero to 10 scale using the calculation (score-lowest values/range)*10.

11 This estimate is based on the party breakdown of the referendum vote for January 2023 in Table 3.

12 A similar change occurred in the 1986 Irish divorce referendum when voters were persuaded that divorce was not a traditional cleavage issue and the proposal passed (Darcy and Laver 1990; LeDuc 2002a).

Appendices

Appendix Table A: variables, scoring, means.

Table

		Means
Variable	Scoring	1999	2023
Vote	1 = yes, 0 = no	.48	.40
Indigenous	1 = yes, 0 = no	–	.02
Female	1 = yes, 0 = no	.52	.49
Great	1 = yes, 0 = no	.10	–
Silent	1 = yes, 0 = no	.24	.05
Baby Boomers	1 = yes, 0 = no	.42	.34
Gen X	1 = yes, 0 = no	.24	.22
Millennials	1 = yes, 0 = no	–	.23
Gen Z	1 = yes, 0 = no	–	.16
University education	1 = yes, 0 = no	.17	.29
High family income	1 = yes, 0 = no	.21	.22
Born ES country	1 = yes, 0 = no	.10	.11
Born NES country	1 = yes, 0 = no	.10	.11
City resident	1 = yes, 0 = no		.65
Interest in campaign	From 1 (none) to 4 (good deal)	3.13	3.02
Discussed the election	From 1 (not at all) to 4 (a lot)	3.08	2.97
Cared about outcome	From 1(none) to 3 (good deal)	2.70	2.62
Vote if not compulsory	1 = definitely, probably, 0 = no	.86	.79
Knew enough about issue	1 = yes, 0 = no	.53	.60
Thought of changing vote	1 = yes, 0 = no	.31	.24
Decided long time ago	From 1 (polling day) to 6 (long time ago)	4.58	4.11
Kim Beazley/Anthony Albanese	From 0 (dislike) to 10 (like)	5.66	4.83
Labor Party	From 0 (dislike) to 10 (like)	5.57	4.99
John Howard/Peter Dutton	From 0 (dislike) to 10 (like)	5.00	3.69
Liberal Party	From 0 (dislike) to 10 (like)	5.26	4.28
Referendum benefits	From 0 to 10	5.48	6.57
Referendum risks	From 0 to 10	4.25	5.76
(N)		(3,243)	(3,134)

Sources 1999 and 2023 ACRS.

Appendix Table B: Public Opinion on the Voice

Table

	Factor loadings
	1	2
Benefits
Listen to Indigenous views	.87	.09
Better Indigenous outcomes	.82	-.31
Connect with Indigenous history	.78	-.34
Risks
Concerned about dividing country	-.09	.91
Change too risky	-.21	.87
Eigenvalues	2.65	1.23
Percent variance explained	53	25

N = 3,134.Varimax rotated factor loadings with unities in the main diagonal. No other factor had an eigenvalue greater than one. Percent combines ‘very important’ and ‘important’. Cronbach’s alpha for factor 1 is 0.80 and for factor 2 0.79. The question was: ‘Which of the following, if any, were important factors in how you voted in the Constitutional Referendum about the Voice? I listened to the views of Aboriginal and Torres Strait Islander people. I thought it would help deliver better outcomes for First Nations communities. It would connect Australia’s founding document with 65,000 years of First Nations history. The change was too risky. I was concerned about dividing our country’. Source 2023 ACRS.

Appendix Table C: Public opinion on the republic.

Table

	Factor loadings
	1	2
Risks
Further reform difficult	.86	-.17
Political stability endangered	.84	-.10
Federal system would be weakened	.82	-.04
Benefits
Australia more independent country	-.15	.83
Australia's standing improved	-.31	.80
Government more democratic	.08	.78
Eigenvalues	2.71	1.51
Percent variance explained	45	25

N = 3,243.Varimax rotated factor loadings with unities in the main diagonal. No other factor had an eigenvalue greater than one. Percent combines ‘strongly agree’ and ‘agree’. Cronbach’s alpha for factor 1 is 0.81 and for factor 2 0.75. The question was: ‘Please indicate how strongly you agree or disagree with each of the following statements about what would happen under the proposed change to a Republic. Australia's record of political stability would be endangered. Further constitutional reform would become more difficult. The states in our federal system would be weakened. Australia would become a more independent country. Australia's standing in the world would be improved. Australia's government would be more democratic’.

Source 1999 ACRS.