How Referendum Results Relate to Levels of Party Support

The Voice Referendum is being put by and overwhelmingly backed by the Albanese Labor government. It is opposed by the National Party and is opposed by large parts of the Liberal Party including Opposition Leader Peter Dutton. It is largely supported by the Greens and ‘teal’ Independents, and opposed by Pauline Hanson’s One Nation.

It is a pattern of party support that suggests referendum night results might follow traditional party divides. This is despite the referendum being held away from a general election, away from a campaign with how-to-votes advocating a vote for or against the government as well as for or against the referendum.

To examine the role of partisanship in referendum voting patterns, I look back at two very different referendums and the relationship between Labor/Coalition election voting by electoral division, and Yes/No referendum results.

The first is the Simultaneous Elections referendum held in conjunction with the 1984 Federal election. With both Labor and Coalition how-to-votes having clear Yes and No referendum recommendations, there was an extremely strong relationship between two-party preferred results by division and Yes/No referendum results.

But the second case, the 1999 republic referendum, was a very different campaign and produced results with a much weaker link between party voting and Yes/No results. Held separately from a general election, the republic referendum was not combined with the partisanship inducing vote for or against the government. The Republic referendum is also unique in being the only referendum put by a Prime Minister who advocated a No vote.

Like The Voice referendum, the Republic was backed by Labor and generally opposed by the Coalition, though with some significant Liberal supporters of a republic. The result produced a confusing mosaic of results where safe Liberal seats voted Yes and safe Labor seats voted no.

As I outline in this post, you can explain more about the pattern of Republic referendum results by looking at the social status of electorates rather then the level of Labor or Coalition support at the previous year’s Federal election.

So are these high social status Republic supporting electorates the ‘elites’ campaigned against in 1999 and so often mentioned again in The Voice campaign? The majority of voters have more interest in getting by day to day than worrying whether ‘The Voice’ will improve the position of First Nations Australians. Does railing against ‘elites’ tap into resentment against those with more time and money to worry about such issues?

Having covered the 1999 Republic campaign and written on the results at the time, I see strong similarities with the current Voice referendum. And on Saturday night I expect to see a very similar pattern of results, with Yes results strongest in high social status metropolitan seats irrespective of whether they are Labor, Liberal, Green or Independent held.

A quick technical point before the post. Most of this post is written comparing Labor two-party preferred percentages by electoral division to Yes percentages at referendums. It could have been written comparing Coalition 2PP% to No% and produce the same findings. You just have to choose one of the two methods of measure, and focussing on the smaller number of Yes voting divisions is easier to measure and explain.

1984 Simultaneous Elections Referendum

The Hawke government held two referendums in conjunction with the 1984 Federal election. One was to allow the States and Commonwealth to voluntarily exchange powers, the second another attempt at introducing simultaneous elections for the House and Senate.

Both referendums were defeated. The exchange of powers question achieved only 47.1% national support and failed in all states. The Simultaneous elections referendum won the national vote but only achieved majority support in the two largest states, NSW and Victoria.

Referendums to require simultaneous elections were put four times between 1974 and 1988. As governments have voluntarily chosen to hold simultaneous election at every Federal election since 1974, Australians have collectively forgotten why it was such an issue in the 1970s and 1980s.

The decision of Robert Menzies to call an early House election in 1963 put House and Senate terms out of step for a decade. In the 14 years between 1963 and 1977, Australians voted at 13 federal electoral events – four separate House elections, three separate half-Senate elections, three dates when referendums were held and three joint elections for the House and Senate.

Annoyance at so many elections, plus voters treating separate half-senate elections as giant by-elections, saw a push to require House and Senate elections be held on the same day. The fixed six-year terms of Senators was proposed to become two terms of the House. The 1988 version of the referendum differed in being for four-year variable terms and electing the whole Senate at each election.

The results for the four simultaneous election referendums were –

• 1974 – Defeated, 48.3% national Yes support, majority in only NSW
• 1977 – Defeated despite 62.2% national Yes support, majority support in only three states, NSW, Victoria and South Australia
• 1984 – Defeated despite 50.6% national support, majority support in NSW and Victoria
• 1988 – Defeated, 32.9% national Yes support

Labor’s proposals in 1974, 1984 and 1988 were opposed by the Coalition. In 1977 the Coalition under Malcolm Fraser put the referendum and was supported for a Yes vote by Labor, but the proposal was again defeated despite winning a clear 62.2% of the national vote. The referendum was opposed by Queensland Premier Joh Bjelke-Petersen and WA Premier Charles Court, their vigorous campaigning defeating the referendum in the three outer states. Voters were persuaded that the power of the smaller states would be weakened by tying Senate terms to the House.

Only eight of 44 federal constitutional referendums have passed, so it is worth mentioning the three that did pass on the same day in 1977. One was to change the method of filling Senate casual vacancies. The second allowed Territory electors to vote at referendums, and the third introduced a retirement age for judges. From a field of four songs, Advance Australia Fair was also elected as the ‘national song’ with a majority of the two-song preferred vote. (That is not a joke.)

Having come close to success in 1977, the simultaneous elections referendum was tried again by the Hawke government in 1984, in conjunction with that year’s early federal election. Labor’s how-to-vote material for the election advocated a Yes for both referendums. The Coalition switched back to opposing the change and recommended a No vote for both referendums.

Unsurprisingly given the positions of Labor and the Coalition, the breakdown of the referendum result by electoral division closely matched two-party preferred support for Labor and the Coalition. The 1984 referendum is a case study in the pattern of support at referendums when party position is a key determinant of whether electors vote Yes or No.

The chart below plots Labor’s two-party preferred percentage against the Yes % in each division.

The above graph is exactly what you would expect at a referendum fought at the same time as the partisan contest for government. The chart shows clearly that the higher the Labor vote, the higher the Yes vote. The points are clustered around the chart's diagonal, which is the line where the Labor 2PP% and Yes% are equal.

The chart is divided into four quadrants representing majority ALP/Coalition and Yes/No combinations. The bottom left quadrant has 62 points representing divisions that recorded a majority Coalition 2PP% and majority No%. There are 71 points in the top right quadrant representing divisions with a Labor 2PP% majority and a Yes% majority. That most points lie in these two quadrants is what you would expect with a strong relationship between two-party vote and Yes/No percentage.

Only 15 divisions had conflicted results. There are four points in the top left quadrant representing Coalition voting divisions that also voted Yes. There are 11 points in bottom right quadrant, Labor voting seats with a No majority for the referendum.

Corresponding to results in each state, Yes results are mostly below the diagonal line for seats in the No voting states, Queensland, Western Australia, South Australia and Tasmania.

The points are not evenly distributed around and along the diagonal. There are more points above the diagonal in Coalition majority seats at the left of the chart, and more points below in Labor majority seats to the right. This reflects the trend line of the data with the Yes% by division rising at a slower rate than the Labor 2PP%.

The linear regression for the trend line gives an r-squared of 0.91, indicating a very strong relationship between Labor 2PP% and Yes %. From the equation, the slope of the line is 0.81 and the constant 8.4%. This means that for every 10% rise in Labor 2PP%, the Yes% rose only 8.1%.

Or to put it another way, every 10% rise in Labor 2PP% saw the gap between the Labor vote and Yes% increase by 1.9 percentage points. The relationship is clearer in the chart below which plots the gap value against Labor 2PP% by division. In this case I have also included the trend line. (The vertical scale of the chart is narrower than it should be, but I have chosen the vertical axis values to match the same data I plot for the 1999 Referendum later in this post.)

Why the drift away? If you hover over one of the NSW or Victorian points, you can see that each state's points have the same slower rise in Yes% than occurs with the Labor 2PP% rise. So state differences don't appear to explain the drift. There was an unusually high informal vote at the House election in 1984 but I don't think it explains the drift. Perhaps it's a sign of a "just vote No" attitude to voting at referendums.

But whatever the reason for the slower growth of Yes %, it does not detract from the clear and very strong relationship between Labor 2PP% and Yes% by division at the 1984 election and referendum.

But what about at a stand alone referendum where voting patterns aren't defined by the contest for government? And a referendum where the parties have looser links with the groups running Yes and No campaigns?

We have such an example in the 1999 referendum. And it shows a considerably weaker relationship between party vote and Yes/No voting by division.

1999 Republic Referendum

The Republic Referendum is unique in being the only referendum put by a Prime Minister who advocated voting No. As Opposition Leader in 1996, John Howard had promised to hold a Constitutional Convention to draft a proposed Republic model. If a model was agreed to, he also promised to hold a referendum to vote on amending the Constitution to become a republic.

Howard arranged and supported a second referendum held the same day to insert a new preamble into the Constitution. It was easily defeated with 39.3% support against 45.1% for the Republic.

There was no election held the same day as the referendum. How-to-vote material was handed out by the Yes and No campaigners, some of whom were party activists. But the campaign groups were broader and less clearly aligned with normal party politics. There were very few Labor Party activists opposed to the referendum, but a significant minority of Liberal Party members supported a Yes vote.

Without the focus of a general election the same day, the results diverged wildly from the party link that drove the 1984 referendum results.

The chart below plots the Republic Yes% for each division against the Labor 2PP% in the same division at the 1998 Federal election. As is clear from the chart, the relationship between two-party preferred percentage and Yes/No % was radically different in 1999 compared to 1984.

The chart has the same diagonal line where Labor 2PP% and Yes% are equal, the same quadrant structure as the 1984 result chart, but displays a completely different pattern of results. The off-centre straight line is the poorly fitting trend line discussed below.

The clear relationship between Labor vote and Yes vote in 1984 is replaced by a scatter of points across the chart. There is no clustering around the diagonal line, points instead spread far and wide in all four quadrants. The points are also dispersed from the trend line because the linear regression discussed below reveals a very weak relationship between Labor 2PP% and Yes%.

Looking at the quadrants, there are 64 points in the bottom left quadrant indicating divisions that voted for the Coalition in 1998 and voted No to the Republic. There are only 24 points in the top right quadrant representing the 1998 Labor voting divisions that voted Yes to the Republic.

Where in 1984 there were only 15 conflicted divisions, in 1999 there were 59. There were 17 points in the top left quadrant representing 1998 Coalition voting seats that voted Yes to the Republic. And in the bottom right quadrant there are 42 points representing 1998 Labor voting divisions that voted No to the Republic. (The Labor seat of Newcastle also voted Yes but is excluded from the analysis for reasons explained at the bottom of this post.)

And where in 1984 the conflicted divisions were clustered around the centre of the graph, that is near the diagonal and trend lines, in 1999 the conflicted divisions are all over the chart from safe Coalition seats to safe Labor seats. This is another sign of the lack of relationship between two-party vote and Yes/No results.

Performing a linear regression on the data, the r-squared measure is a low 0.12, the line slope 0.31 and constant value of 29%. The F-Statistic says a relationship exists, but the flat slope and high constant suggest that Labor/Coalition support is a poor explainer of Yes/No support by electoral division. There are other factors explaining the pattern of support.

The lack of party/referendum alignment is even is even clearer if you plot the gap between 1998 Labor 2PP% by division and the Republic Yes %. For every 10% increase in Labor 2PP%, the gap between the Labor 2PP% and the Yes% increases by 6.9%. That is a much higher divergence rate than 1.9% in 1984.

Some of the gaps are enormous. In the ultra safe Liberal seat of Bradfield on Sydney's north shore, the Labor 2PP% in 1998 was 26.8%, the Yes% in 1999 55.6%, the gap in favour of Yes 28.8 percentage points. In the ultra-safe Labor seat of Maribyrnong in Melbourne's west, the Labor 2PP% in 1998 was 72.1%, the Yes% in 1999 36.4%, a gap in No's favour of 35.7 percentage points.

With partisanship being a poor predictor of support for the Republic, what was driving the pattern of support for the republic by electoral division?

Referendum Voting and the 'Elites'

Even a passing knowledge of Australian political geography gives you a good idea of what explains the pattern of support for the Republic.

The Liberal voting seats that voted Yes to the Republic in 1999 are amongst the most affluent in the country. Seats like Wentworth (Sydney), Kooyong (Melbourne), Ryan (Brisbane) and Curtin (Perth) are electorates that the residents of each city will tell you are ranked highly for social status.

Or as it might be described in a term that has become a regular epithet in referendum campaigns, they are the seats where the 'elites' live.

(Incidentally, only five of the 17 Liberal seats that voted Yes to the Republic are still held by the Liberal Party. Six are now held by 'teal' Independents, five by Labor and one by the Greens. That's the subject of another post before Saturday.)

On the other side of the ledger, the Labor seats that voted Yes include two types of electorates. Some are inner city Labor seats with above average proportions of university educated and professionally employed residents. Others are very safe Labor seats further out in the suburbs with such large Labor margins that even a large gap between Labor 2PP and Yes support left the seat with a Yes majority.

More important to examine than an electorate's binary choice of voting Yes or No is the gap between Yes% and Labor 2PP%. (I could equally use Coalition 2PP% and No% but am sticking to one measure for consistency. It is easier to analyse why change was achieved or failed rather to explain why a status quo was maintained.)

The 17 Liberal seats that voted Yes had a Yes% on average 15 percentage points higher than than Labor 2PP% at the 1998 election. The 24 Yes voting Labor seats had a Yes% on average six percentage points lower than the Labor 2PP%, but this is an average that hides a much greater range. Nine electorates had a gap beyond -10%, reaching -24.4% in the ultra-safe Labor seat of Fowler, which recorded a narrow Yes majority. The already mentioned more affluent inner-city Labor seats had a closer match between support for the Republic and Labor 2PP%.

But how do you measure social status? There are a range of education, occupation and income measures you can use. In this post I am going to use the proportion of each electorate's adult population with university qualifications.

The chart plots 1996 electorate census data on university education against an electorate's Yes% on the Republic. I have drawn a trend line on this graph and it produces an upward sloping curve that approximates the plotted results. The r-squared measure on this is a strong 0.57 and the slope 1.05, so on average a 10% increase in a division's % University educated corresponded to a 10% increase in Yes %. The constant term of 27.8% suggests there is a fair bit of variance still to explain, mainly because there are a wide rage of results to the left of the graph where the bulk of electorates with low rates of university education are clustered.

Let me make it clear that I am not saying that University educated voters were all voting for the republic. That would be ecological fallacy, trying to infer behaviour of individuals from relationships derived using aggregate data.

I have simply used the percentage of University educated residents as one of several possible variables representing the social status of an electorate. I could have used percentage professionally employed, percentage leaving school by age 15, or income statistics. All produce a similar relationship. I am using percentage university educated as a proxy for electorate social status, and it seems to explain more about the republic result than two-party preferred voting patterns at the 1998 election.

No wonder the No case in 1999, and the No case in 2023, have used the word 'elites' so often in campaigning. At a time when cost of living pressures are biting, it's not hard to stoke a bit of resentment against the 'elites' on an issue that most voters see as having not much to do with their day to day lives.

Footnote - As mentioned earlier, the electorate of Newcastle has been excluded from the 1999 analysis. Newcastle voted 51.0% Yes at the referendum but there was no Labor two-party preferred percentage for the 1998 election. A candidate died during the 1998 campaign and a supplementary election had to be held a month after the general election. The Liberal Party did not nominate a candidate for the supplementary election.