How Might the Referendum Results Come In

Where to find referendum results? Google points a lot of people looking for election result here to my blog on election night. Currently I’m busy on television so can’t be of much assistance. If you are live results, check out the ABC’s Referendum results site for everything you need to know after 6pm eastern time.

When results for ‘The Voice’ referendum report on Saturday night, they will arrive in a fearsome rush.

In preparation for Saturday night’s coverage, I’ve been digging back through some of my working documents for the 1999 Republic referendum. I also checked out the television coverage through ABC archives.

My memory was that results poured in, but checking the archives proved the rate of reporting was faster than I remembered. By 7:30pm in the 1999 count, half the national vote had reported and NSW had reached 68% counted and Victoria 65%. That is around twice the first preference votes that would have been reported by that time at a general election.

The results will report just as quickly on Saturday, though they may not reach the same percentage counted by 7:30. As I pointed out in a previous post, 80% of votes were cast on election day in 1999 where the equivalent figure in 2023 will be under 50%.

But fewer votes in polling places means the votes will be quicker to count, so the votes reported may reach 50% very quickly. From there it may slow down for a while given how long it takes to count some of the giant pre-poll centres. It has been a feature of recent by-election counts that there is a pause between the point when all polling places have reported their results, and when pre-poll counts are completed and reported.

But what will the early figures look like on election night? For that I again went back to my 1999 figures.

The Counting Process

A key point to remember about election night is that picking the winner is not a process of analysing the behaviour of voters. What you are doing is modelling the trend caused by the order in which polling places report.

In the days before polling place results were published on election night, the method was to correct for the early count trend observed at previous elections. Since a change in counting procedures in 1987, small polling places in country areas have been the first to report. As small rural booths historically favour the Coalition, simple tally counts produce a trend where the Labor percentage vote starts low and rises. Picking election winners used to be about picking when this trend levelled off and approached the final level of vote.

Picking winners was revolutionised during my early years of covering election by two major changes. The first occurred in 1990 when it became possible to match up polling place results with the same result by polling place at the previous election. The second was the introduction of indicative preference counts at the 1993 Federal election.

With accurate polling place figures, you can calculate swing by polling place. The standard deviation of swing between polling places is a third to a half of the standard deviation of two-party preferred percentages. More importantly, swing is generally not correlated with polling places size, unlike two-party preferred results. If you operate on swing as a predictor rather than simple percentage, you have a better predictive measure that has a smaller standard deviation and less statistical bias. (I gave a speech on the technique two years ago which you can find here.)

On a side note, a problem that has emerged in recent years is that swings now sometimes correlate with booth size, though at the opposite end of the count. Pre-poll centres reporting late in the count have greater weight on the final result because of their size, and have proven to sometimes have swings that differ from votes on polling day. That was a problem at the NSW election when the different swings with pre-poll and postal votes caused a late shift in the swing predicted from the smaller polling places that reported first on election night.

For The Voice referendum there will be polling place results available. For the Republic count in 1999 they were not. Nor was there a history of republic voting by polling place on which to base predictions. As I explained in my post on the 1999 Referendum results, you could calculate ‘swing’ between the 1998 federal election and the 1999 referendum, but it produces a massive range of ‘swings’ from -30% to +20%. You never see swing ranges like that at a Federal election, so using polling place swing calculations might give you a predicted percentage just as bouncy as sticking to simple cumulative percentages.

So with only simple cumulative percentages in 1999, what did I do to work out what was going on?

In What Order do Polling Places Report?

The factors that determine the speed at which polling places report their results are –

• How many votes are taken. Staff at a polling place with fewer ballot papers can count and report results quicker than at a polling place taking more votes.
• How many candidates there are. The more candidates on the ballot paper, the longer it takes to sort the ballot papers before counting begins.
• The number of votes cast for candidates other than the top two. The more votes for other candidates, the longer it takes to conduct an indicative preference count.

Only the first factor matters at a referendum for a single proposition. Every ballot paper in every electorate is the same – a one box ballot paper that resolves into a two-horse race between candidates called Yes and No. How long it takes a polling place to report its results depends overwhelmingly on how many votes there are to be counted.

Which means early figures are heavily weighted towards small country polling places. As mentioned already, small rural polling places have a distinct conservative lean. That was expected in the 1999 Republic count, occurred in the actual count, and is again expected to be evident in early figures for the Voice referendum. The Yes % in all states will start low and trend up towards its final level.

As a working guide for the 1999 Republic referendum, I took all two-party preferred results by polling place at the 1998 Federal election, sorted them by state into ascending number of votes taken, and then calculated cumulative votes and percentages for each state.

This created a trend line of the worst case scenario where all polling places reported in strict booth size order. I then kept a running tally of Yes and No percentages by percentage vote counted through the night.

So, how did this method perform?

How the Republic Count Unfolded in NSW

Fortunately I kept my 1999 referendum tally sheets. The rush of figures meant I had gaps in scorekeeping on the night but there are still enough points to plot the result.

In the chart below, the grey line is the 1998 cumulative trend line for NSW, the blue line the track of NSW Yes % that actually unfolded. I have added an extra point for each line representing the final declared result.

The trend line predicted a low Yes % that would rise and level out. The actual results also started low and rose, but did not rise as steeply and then levelled out more quickly.

The reason that the actual result levelled out more quickly is that the assumption that booths report strictly in size order was a worse case scenario. The way results are phoned through and data entered means you can get backlogs in the entry of small polling places that overlap with larger polling places reporting from other electorates.

Secondly, a key feature of the 1999 results, as outlined in a previous post, was that the Yes % performed far worse at polling places in capital city Labor seats than Labor had performed at the 1998 election. The trend line uplift in Labor vote was not repeated at the Republic referendum because the Yes vote did not perform as well in safe Labor voting polling places. Yes did better in many Liberal voting polling places, but not enough to overcome the poor Yes performance in Labor voting seats.

How Will Results Unfold for ‘The Voice’?

As in 1999, the Yes% in every state will start low and rise. The question is, at what point will it stop rising and give a close approximation of the final outcome?

If the Yes% in a state stops rising at a low percentage, say 44% with 20% counted, you can be confident of a clear majority for No in that state. If the rise reaches 49% or 51% with 20% counted, a prediction will have to wait for more counting as the votes to come could still change the outcome. That’s especially the case with so many pre-poll votes to come, and postal votes in the days after the election.

Any state where Yes has a narrow lead at the end of election night counting could have its lead overturned by postal votes.

One big advantage available for coverage of ‘The Voice’ count is access to polling place results. On early figures, it is possible to use a comparison with 2022 Federal election results to stabilise the early counts.

It may be possible to make early calls on this 2022 to 2023 comparison of results. I am reticent to call this calculation ‘swing’ because the ‘swing’ figure will be nonsense if talked about on its own. But if used as a method of cancelling out early statistical bias in the referendum returns, it will be useful.

On the ABC’s election coverage tonight, only Yes/No votes and simple percentages will be broadcast. The same numbers will be published on the ABC’s election website. The website will include full electoral level results.

ABC general election coverages use projected percentages, but at no point will projected Yes and No percentages be published as part of the ABC’s coverage. I will have access to numbers that provide projected Yes and No percentages which will aid predictions, but projected Yes/No shares will not be broadcast. It is impossible to know beforehand how steady and reliable the projected percentages will be so they are not being published.

Covering the referendum will be like a US election, all about vote numbers and progressive percentages, not the projected vote shares we normally use for Australian elections.

How Tonight’s Count Could Unfold

The 1999 referendum count highlights how quickly results will pour in. Opinion polls indicate a clear majority for No at ‘The Voice’ referendum. It all points to there being an early call for each state tonight.

The ABC coverage starts at 5:30pm eastern summer time with results from 6pm. As at general elections, the broadcast will be live into the states in later time zones, so from 2:30pm in Western Australia and equivalent times in other jurisdictions.

The first result expected within five minutes will be for Norfolk Island which is one hour ahead of the east coast. For electoral purposes, Norfolk Island is in the southern ACT electorate of Bean. Results will pour in from around 6:15pm and a result may be clear in each state within an hour of the count beginning.

Counting starts at 6pm local time in each state. In eastern daylight time that means counting starts at

• 6:00pm – New South Wales, Victoria, Tasmania, ACT
• 6:30pm – South Australia
• 7:00pm – Queensland
• 7:30pm – Northern Territory
• 9:00pm – Western Australia

If the opinion polls are correct, it is possible that most states will have a clear result in the first hour after the state’s count begins. It is highly likely the referendum will be decided before counting starts in Western Australia. If the results in southern states are very clear, we may even have a result before the first results report from Queensland.

Tune in tonight and see. And there will be full results published down to electorate level at the ABC news website, including the nifty hexagon map of results shown below that will colour in during the night.

Declaration Under ABC social media standards, I would like to make it clear this post was written by me based on my own personal experience covering elections and referendums. The opinions stated here are my own, I have not sought approval for the content of this post and nor am I speaking on behalf of the ABC.