The Changing Pattern of Results by Vote Type

With counting complete, the Australian Electoral Commission has returned the writ to the Governor-General formally declaring “The Voice” referendum defeated.

The final count has confirmed what was observed on election night, that there was a massive difference between how people voted in person on polling day compared to votes cast in the two weeks of early voting.

My professional interest in this difference is the impact the growing and variable gap between polling day and early votes has on when we know results on election night.

As I outlined in a previous post, 83.8% of votes were cast on polling day at the 1999 Republic referendum. In 2023 the figure was close to half at only 43.7%.

There has been a huge increase in pre-poll voting since its availability was first liberalised in 2010. Over the 13 years since, the number of polling day votes has declined. While pre-poll voting centres are counted and reported on election night, the larger number of votes taken per centre compared to polling places means pre-polls generally report later in the evening. At recent by-elections, all polling places have reported their results before the first pre-poll centre reported.

With pre-poll counting revealing different trends, and unreliably different trends as well, it means that close elections will take longer to call on election night.

Pre-poll and postal voting has always had a conservative lean compared to election day voting, but never have we seen a gap as wide as at the referendum.

When non-polling day votes made up less than one-in-five votes, you could factor in the last election’s postal and pre-poll trend safe in the knowledge there were not enough votes to shift a result more than a few percentage points.

With early votes now outnumbering polling day votes, an early prediction based on polling day votes can be significantly shifted. That is shown clearly by the referendum.

At the 2022 Federal election, the Labor two-party preferred vote declined 1.6 percentage points between the tally of polling day votes and the final count. That was high by past election trends.

But the shift was even greater at the referendum. The Yes% shifting down a remarkable 3.8% between the tally of polling day votes and the final result.

The table below breaks down the referendum Yes vote by vote type and compares it to the same categories for Labor’s two-party preferred vote at last year’s Federal election.

Vote By Type – 2022 Federal Election and 2023 Referendum Compared
Percent of Total Votes Percent of Vote
Vote Type 2022 2023 ALP 2PP Yes “Swing”
Polling Day Ordinary 45.1 43.7 53.7 43.7 -10.0
Pre-Poll Ordinary 33.3 35.3 50.6 35.4 -15.2
Postal 14.3 11.0 49.1 33.1 -16.0
Pre-Poll Declaration 3.6 4.3 53.3 44.7 -8.6
Absent Votes 3.2 4.4 57.4 48.9 -8.5
Other vote types 0.5 1.3 59.1 47.7 -11.4
Total .. .. 52.1 39.9 -12.2

The final column in the tables includes a calculated “swing” between last year’s Labor 2PP% and Yes% at the referendum. This shows that not only percentage vote differed by vote categoyr, but so did “swing”.

As it is standard in Australian election night analysis to predict based on swing rather than percentage vote, huge differences in swing by vote type also delay when it is safe to call an election.

While we have never seen a result by vote type gap as large as the referendum, we did see a similar phenomena play out in results at the NSW state election in March. The Labor two-party preferred percentage declined 1.4 percentage points between polling day votes and the final figure, down from 55.7% to 54.3%. The same figure had been just 0.1 percentage points in 2019.

Postal voting has always had a conservative lean. It has always included an over-representation of older voters, distant rural voters and voters from religious groups that don’t vote on Saturday. All are known to be more conservative voting than the general population. The number of postal votes did rise during Covid, but as shown in the table below, slipped back towards its usual long-term value at the referendum.

It is the different voting pattern of pre-poll votes that is most worrying for anyone calling an election. There are three to five times as many pre-poll votes as postal votes and their difference from polling day votes does not appear to be as stable.

I had always expected that as pre-poll votes increased, they would become more like polling day votes. Instead, something about the who and why of pre-poll voters has caused a growing difference between pre-poll votes and the ever diminishing group that still turn up to vote on polling day.

I can’t offer any firm answer on the cause of the vote type gap at the referendum. At some point I will break down my table by electorate to spot city-country and state differences. Maybe I’ll try and obtain roll mark-off data once it becomes available to examine the demographics of who voted when.

Without being able to offer much explanation for the voting pattern, here are some general comments on the data in the table.

  • Once again there was a decline in polling day voting, down from 45.1% of votes cast in 2022 to 43.7%. This is well down from 83.8% of votes being cast on the day at the 1999 referendum.
  • In district pre-poll voting continued the rising trend seen for the past 16 years, up from 33.3% in 2022 to 35.3% at the referendum. The 2022 figure is inflated by the inclusion of categories of early Covid votes.
  • Postal voting declined from 14.3% of votes in 2022 to 11.0% at the referendum. The long term average for postal votes used to be around 8% before a surge in 2022 related to Covid concerns. The 2023 figure appears to be a return to the normal incidence.
  • There were small increases in the two categories that largely consist of votes cast outside of home divisions, Absent votes and Pre-Poll Declaration votes. That may be an issue with timing, the referendum in October, the last Federal election in May.
  • The increase in “Other” votes is largely down to the return of Special Hospital votes, mobile polling in hospital and retirement facilities having been largely blocked in 2022 due to Covid concerns.
  • In 2022 the gap between Labor’s two-party preferred on polling day and for Pre-poll votes was -3.1 percentage points, but the same figure for Yes in 2023 was -8.3 points.
  • In 2022 the gap between polling day and postal vote Labor two-party preferred was -4.6 percentage points, at the referendum for Yes a much larger -10.6 points.
  • These differences were very noticeable in the figures on election night. While it was clear early on the night that the referendum would be defeated, the Yes percentage noticeably declined as pre-poll centres started to report their results. The Yes vote declined further in subsequent days as postal votes were counted.
  • After reaching 60.8% by the middle of the post-referendum counting week, the No% declined to 60.1% as Absent and Pre-poll Declaration votes were counted. Both categories recorded No majorities, but both recorded lower No votes than on polling day, pulling the overall figure down.
  • Absent votes have always had a left-lean compared to polling day votes. They are predominantly a category of urban voters, and they include many people who work on polling day and are outside their electorate. Like postal voting, Absent voting has a political lean reflective of the type of electors who cast these votes.
  • On election night, as well as vote totals and percentages, I had access to ‘matched’ percentages comparing Yes% to 2022 Labor 2PP% by polling place. The matched percentages allowed assumptions about the progress of the count from raw numbers to be confirmed. But this method also understated the size of the No vote early in the evening. The matched polling swing was 10%, suggesting a No result of 58%, a figure that blew out once the larger swing with pre-poll and postal votes was factored into the result.
  • Like the NSW election earlier this year, the referendum displays a marked difference between polling day swing and swing with early votes. It is another example of an election where the prediction based on polling day votes is not reflective of the final outcome. This has always been the case with postal votes and has always been accounted for in election night predictions. But that there is a difference in the ever growing category of pre-poll votes, and the sheer volume of pre-poll votes means their impact on the final outcome is greater.
  • At Federal elections this pre-poll vote issue is delaying election calls, but not preventing them on election night. The results are reported but just later in the evening.
  • There are issues in states that do not count pre-poll votes. At the 2021 WA election, the tally of votes reached only 37% on election night and it was fortunate that the result was so clear. The way in which pre-poll votes are collected in WA meant that very few could be counted on election night. I understand legislation will be introduced to improve pre-poll counting at the next WA election. I understand South Australia is also changing its legislation to permit election night counting of pre-poll votes.
  • NSW also had problems counting pre-polls at the state election in March this year. The decision to open Legislative Council pre-poll ballot boxes for all pre-poll lower house votes counted, meant only one-poll centre was counted in each district. At Victorian and Federal elections, upper house pre-poll votes are not counted on election night and I intend to propose that a similar procedure be adopted in NSW. As pre-poll ballot boxes are on secure premises and are not moved overnight as is the case with polling day ballot boxes, there is no need to carry out a verification count before the transfer of ballot boxes.
  • Victoria also has a problem with pre-poll counting that pre-poll centres cannot be separately identified in published results and election night result feeds. This makes it difficult to incorporate pre-poll counts into polling place results based on matched swing. A current parliamentary inquiry into the counting method will hopefully produce solutions.

Overall the Australian Electoral Commission has to be congratulated on being able to deliver timely results on election night. Admittedly it was a far simpler count than a Federal election, having only one ballot paper with much simpler formality rules. As can occur with counts involving only two candidates, there were some transposed results on election night, but these were all corrected with the official check count conducted once ballot boxes had been deliver back to the Returning Officer for each Division.


Overall turnout for the referendum was 89.92%, slightly up from 89.82% at last year’s House election. But it was down slightly from 90.47% at the 2022 Senate election, the figure usually used to define turnout. Commentary as late as polling day suggesting there would be a low turnout proved wide of the mark.

As the chart below shows, there was a very strong correlation between between turnout by division at the 2022 Federal election and the referendum. Turnout was high and low in the same pattern as at 2022 Federal elevction. Colours are used to show divisions that voted Yes and No.

A note on why Senate turnout is higher than for the House. This is largely due to Absent and Pre-Poll declaration voters being given the wrong House ballot paper. When declaration envelopes are processed and opened, the Senate ballot paper can be counted even if the envelope contains the wrong House ballot paper. There is also the phenomena that some voters miss the much smaller House ballot paper and only complete a Senate ballot.

The increased use of electronic roll mark-off at pre-poll and absent vote issuing desks is reducing the number of incorrect House ballot papers issued.

Informal Voting

Talk of confusion over the form of the ballot paper was also wrong. Informal voting was just 0.98%, little changed from just 0.86% at the 1999 Republic referendum. By comparison, informal voting at the 2022 House election was 5.2% and 3.4% for the Senate election.

Once again the one-box referendum ballot paper was responsible for a Federal referendum producing the lowest rate of informal voting for an Australian electoral event. Advocates of changing the ballot paper to a two-box pick-a-box format need to remember that two-box state referendum ballot papers over the past three decades have produced much higher rates of informal voting between 3% and 9%.

While actual figures are not available, all reports by scrutineers suggest that there were negligible instances of voters using ticks and crosses. Faced with a one-box ballot paper with clear instructions to write Yes or No, voters just don't respond by using ticks and crosses.

The chart below plots the rate of informal voting by division at the referendum on the vertical axis, against the rate of House informal voting by division at the 2022 Federal election on the horizontal axis.

As you would expect, the rate of referendum informal voting was high in the same divisions where it was high in 2022, but the overall level was well down.

There were only six electorates in the entire country where the rate of informal voting was above 1.5%, Calwell, McMahon, Fowler, Werriwa, Watson and Blaxland. All these electorates have well above average rates of voters whose first language is not English. All had high rates of informal voting at last year's Federal election, but the absolute rate of informal voting was much much lower at the referendum.

The mis-placed fuss over ticks and crosses was about whether the two-sides of the referendum debate had non-standard methods of voting treated equally, not about whether votes with clear intent were counted as formal, or those with unclear intent counted as informal. If only there were as much concern about allowing votes with clear first preference intent to count at House elections.

At every Federal election there are hundreds of thousands of ballot papers with a clear first preference intent rejected because of problems with a further preferences on the ballot paper, even if that further preference was never going to be distributed. Many of these votes could have their first preferences counted even without moving to optional preferential voting.

6 thoughts on “The Changing Pattern of Results by Vote Type”

  1. Thanks for the article Antony. Does the AEC, or any other electoral commission distribute counting staff by the amount each polling place (pre-poll and ordinary) received in votes (or at least the approximate figures) or are intrinsically more people allocated to on the day polling places? Although, it is likely that they cannot know about how many people voted, surely they would put around fifteen times as many people on pre-polls as opposed to the, only about a thousand vote, on the day polling places.

    Also, I would like to say that, one of the main reasons I think that many people did not put a tick or a cross in the box, is that, ticks and crosses are not how we tend to answer questions. If I asked, ‘Would you like ice-cream?’, you would not say, ‘Tick!’ I believe this, conversational example is often carried on to ‘on paper’ questions like referendums, as you do not see, ‘A Proposed Law: To alter the Constitution to recognise the First Peoples of Australia by establishing an Aboriginal and Torres Strait Islander Voice.
    Do you approve this proposed alteration?’ and think, ‘Cross.’

    COMMENT: Staffing for polling places is based on running the polling place through the day and then counting the votes in the evening. The staffing is based on the different roles required to run the polling place and there is a minimum staffing level below which the staff employed in a polling place can’t fall. So smaller polling places have proportionally more staff than larger polling places because they are there to do more than just count votes.

    Pre-poll counts employ people for the count only. In most seats all pre-poll centres are counted at a single secure premises. They have to allocate staff to each centre and there may be parallel counts being undertaken by different staff.

    Counting 15,000 votes isn’t done by throwing 15 times as many staff at the count. There comes a point where extra staff just complicate and delay the process. Better to have a smaller core of experienced counters supervising with enough staff to complete the count than have inexperienced staff engaging in parallel counts of multiple ballot boxes that have to be reconciled at the end.

    Remember, the AEC is trying to hire extra staff just to count on a Saturday night, with not great rates of pay, and on a night when the AEC is competing with other businesses seeking casual staff. The other option (often suggested by the Auditor General) is to have fewer polling places and re-direct staff to conduct the pre-poll tallies more efficiently.

    number which means Pre-poll centre counting staff

    1. Really admire your work, Antony.

      Just wanted to add some extra clarification regarding polling day staffing.

      Venues are staffed by the number of “issuing points” which ties in with the forecast number of votes expected to be taken at said venue.

      It’s something like, the total number of expected ordinary votes for the venue is divided by the number of votes that can be expected to be issued by a person / point per hour. That gives you how many ordinary issuing officers a venue will be allocated. Same for Declaration Vote Issuing Officers. Add in a ballot box guard, a queue controller, (maybe) an inquiry officer, a manager, assistant manager and that’s how many people are assigned to a venue for all the work.

      The AEC keeps a spreadsheet of “expected election day polling places” at which includes the number of expected ordinary and declaration votes and the number of assigned ordinary and declaration vote issusing officers.

        1. Hi Antony, do we know why there’s variation in turnout between different electorates or states? Is it access to polling places, demographics, something else?

          COMMENT: The turnout was up and down between states with the normal variation seen at federal elections. It was highest in Tasmania at 92.04% and South Australia 91.70%, the two states with the least movement of voters between census and elections. It was 91.36% in the ACT, 91.03% in Victoria and 90.83% in NSW. In the jurisdictions with extensive remote districts turnout was lower, 88.27% in Queensland, 87.53% in Western Australia and 71.46% in the Northern Territory. The turnout by district almost exactly matched that at last year’s Federal election.

      1. Happy New Year almost. Dealing with pre-poll/electoral day deviation may I observe as a regular party worker on both types of booths that the reasons include the different age demographic and socio-economic groupings using the booths. Pre poll voters are older, often retired, and to judge by their dress younger pre-pollers are socio-economically better off. This produces higher conservative votes numbers. If you are better off and/or self employed then you can organise your life to get it done (or as many still say on leaving “done my duty”). The few elderly you see on polling booths on the day (apart from party workers) are driven there by often grumpy children – quite a few of whom have already pre-polled. As for the greater than normal deviation between returns re Voice/general election it’s a no brainer – that vote was very very generational. I spoke to many older ALP & Greens “No” voters and quite a few younger LNP “yes” voters. Has any solid, as distinct from my anecdotal, research been done on this aspect?

Leave a Reply