Australian election data

Elections tend to create fascinating data sets. They are spatial in nature, comparable over time — and more importantly they are consequential for all Australians.

Australia’s compulsory voting system is a remarkable feature of our Federation. Every three-ish years we all turn out at over 7,000 polling booths at our local schools, churches, and community centres to cast a ballot and pick up an obligatory election day sausage. The byproduct is a fascinating longitudinal and spatial data set.

One word of warning: I use the terms electorate, division, and seat interchangeably throughout this chapter.

Packages

library(tidyverse)
library(sf)
library(strayr)
library(readaec)
library(eechidna)

The Australian Electoral Commission publishes detailed results for every federal election on their tally room at results.aec.gov.au. For elections from 2007 onwards this includes first preference votes, two-party preferred, two-candidate preferred, booth-level results, polling place coordinates, Senate counts, enrolment, and turnout — all updated live on election night.

The catch is that each election has its own URL structure built around an internal event ID, and column names shift between years without warning. The readaec package wraps the AEC’s CSV downloads in a consistent, tidy interface. One function call returns a clean data frame. Results are cached locally so you’re not hitting the AEC’s servers on every call.

For census analysis, the eechidna package takes a different approach: it bundles House of Representatives results from 2001 to 2022 as ready-to-use R data frames, with ABS demographic variables already apportioned to electoral boundaries. readaec and eechidna are complementary — readaec for live, current election results; eechidna for the census join.

The strayr package provides electoral boundary shapefiles from the ABS.

What’s available

list_elections()
  year event_id       date               type has_downloads
1 2001    10822 2001-11-10            general         FALSE
2 2004    12246 2004-10-09            general         FALSE
3 2007    13745 2007-11-24            general          TRUE
4 2010    15508 2010-08-21            general          TRUE
5 2013    17496 2013-09-07            general          TRUE
6 2016    20499 2016-07-02 double_dissolution          TRUE
7 2019    24310 2019-05-18            general          TRUE
8 2022    27966 2022-05-21            general          TRUE
9 2025    31496 2025-05-03            general          TRUE

readaec covers all federal elections from 2007 to 2025. The AEC doesn’t provide the structured CSV downloads that readaec depends on for elections before 2007.

2025 results

Let’s start with the most recent election. Who won each seat?

members_2025 <- get_members_elected(2025)
head(members_2025, 10)
# A tibble: 10 × 9
   divisionid divisionnm stateab candidateid givennm   surname   partynm partyab
        <dbl> <chr>      <chr>         <dbl> <chr>     <chr>     <chr>   <chr>  
 1        179 Adelaide   SA            41210 Steve     GEORGANAS Austra… ALP    
 2        197 Aston      VIC           40809 Mary      DOYLE     Austra… ALP    
 3        198 Ballarat   VIC           40839 Catherine KING      Austra… ALP    
 4        103 Banks      NSW           41350 Zhi       SOON      Austra… ALP    
 5        180 Barker     SA            41016 Tony      PASIN     Liberal LP     
 6        104 Barton     NSW           41339 Ash       AMBIHAIP… Austra… ALP    
 7        192 Bass       TAS           41546 Jess      TEESDALE  Austra… ALP    
 8        318 Bean       ACT           40676 David     SMITH     Austra… ALP    
 9        200 Bendigo    VIC           40869 Lisa      CHESTERS  Austra… ALP    
10        105 Bennelong  NSW           41329 Jerome    LAXALE    Austra… ALP    
# ℹ 1 more variable: year <dbl>

A quick tally by party:

members_2025 |>
  count(partyab, sort = TRUE)
# A tibble: 8 × 2
  partyab     n
  <chr>   <int>
1 ALP        94
2 LP         18
3 LNP        16
4 IND        10
5 NP          9
6 GRN         1
7 KAP         1
8 XEN         1

Margins

Which seats were the most marginal?

get_tpp(2025) |>
  mutate(margin = abs(alp_pct - 50)) |>
  arrange(margin) |>
  select(division, state, alp_pct, lnp_pct, margin) |>
  head(15)
# A tibble: 15 × 5
   division   state alp_pct lnp_pct margin
   <chr>      <chr>   <dbl>   <dbl>  <dbl>
 1 Longman    QLD      49.9    50.1  0.110
 2 Bullwinkel WA       50.5    49.5  0.510
 3 Wentworth  NSW      50.6    49.4  0.560
 4 Menzies    VIC      51.1    48.9  1.08 
 5 Petrie     QLD      51.2    48.8  1.17 
 6 Solomon    NT       51.3    48.7  1.31 
 7 Bendigo    VIC      51.4    48.6  1.40 
 8 Berowra    NSW      48.4    51.6  1.63 
 9 Forde      QLD      51.8    48.2  1.77 
10 La Trobe   VIC      47.9    52.1  2.06 
11 Curtin     WA       47.8    52.2  2.16 
12 Forrest    WA       47.8    52.2  2.24 
13 Kooyong    VIC      47.6    52.4  2.37 
14 Banks      NSW      52.4    47.6  2.39 
15 Bowman     QLD      47.6    52.4  2.43

And the safest seats — won by the largest margins:

get_tpp(2025) |>
  mutate(margin = abs(alp_pct - 50)) |>
  arrange(desc(margin)) |>
  select(division, state, alp_pct, lnp_pct, margin) |>
  head(10)
# A tibble: 10 × 5
   division  state alp_pct lnp_pct margin
   <chr>     <chr>   <dbl>   <dbl>  <dbl>
 1 Wills     VIC      80.9    19.1   30.9
 2 Grayndler NSW      80.2    19.8   30.2
 3 Cooper    VIC      78.5    21.5   28.5
 4 Sydney    NSW      78.1    21.9   28.1
 5 Canberra  ACT      76.4    23.6   26.4
 6 Melbourne VIC      74.0    26.0   24.0
 7 Maranoa   QLD      26.0    74.0   24.0
 8 Watson    NSW      72.9    27.1   22.9
 9 Fenner    ACT      72.1    27.9   22.1
10 Fraser    VIC      72.0    28.0   22.0

Election maps

Geography matters enormously in Australian politics. Electorates vary from Durack in Western Australia (1.63 million square kilometres) to Grayndler in inner Sydney (32 square kilometres). The AEC carves up the population by state and territory, using a formula to allocate seats, then draws boundaries to achieve roughly equal populations per electorate.

At time of writing, the seat allocation is:

State/Territory Seats
New South Wales 47
Victoria 39
Queensland 30
Western Australia 15
South Australia 10
Tasmania 5
Australian Capital Territory 3
Northern Territory 2
Total 151

Let’s map the 2025 TPP result across all 151 divisions.

ced2021 <- strayr::read_absmap("ced2021")
tpp_2025 <- get_tpp(2025) |>
  mutate(division = str_to_title(division))

map_data <- ced2021 |>
  rename(division = ced_name_2021) |>
  left_join(tpp_2025, by = "division")

ggplot(map_data) +
  geom_sf(aes(fill = alp_pct), colour = "white", linewidth = 0.1) +
  scale_fill_gradient2(
    low = "#80b1d3", mid = "white", high = "#E4281B",
    midpoint = 50, name = "ALP TPP %"
  ) +
  labs(
    title   = "2025 federal election: ALP two-party preferred",
    caption = "Source: AEC via readaec; boundaries: ABS 2021"
  ) +
  theme_void() +
  theme(plot.title = element_text(face = "bold", size = 12))

Booth-level data

The AEC publishes polling place locations and vote counts at the booth level. With over 7,000 booths nationally, this is where the spatial story gets genuinely granular — even within a single electorate, booths can lean very differently.

booths <- get_polling_places(2025)
head(booths, 10)
# A tibble: 10 × 16
   state divisionid divisionnm pollingplaceid pollingplacetypeid pollingplacenm 
   <chr>      <dbl> <chr>               <dbl>              <dbl> <chr>          
 1 ACT          318 Bean                93925                  5 Belconnen BEAN…
 2 ACT          318 Bean                11877                  1 Bonython       
 3 ACT          318 Bean                11452                  1 Calwell        
 4 ACT          318 Bean                 8761                  1 Chapman        
 5 ACT          318 Bean                 8763                  1 Chisholm       
 6 ACT          318 Bean                93916                  1 City (Bean)    
 7 ACT          318 Bean                93922                  5 City BEAN PPVC 
 8 ACT          318 Bean               124111                  1 City North (Be…
 9 ACT          318 Bean               124103                  5 City North BEA…
10 ACT          318 Bean                31298                  1 Conder         
# ℹ 10 more variables: premisesnm <chr>, premisesaddress1 <chr>,
#   premisesaddress2 <chr>, premisesaddress3 <lgl>, premisessuburb <chr>,
#   premisesstateab <chr>, premisespostcode <chr>, latitude <dbl>,
#   longitude <dbl>, year <dbl>
ggplot() +
  geom_sf(data = ced2021, fill = "grey95", colour = "grey80", linewidth = 0.1) +
  geom_point(
    data = booths,
    aes(x = longitude, y = latitude),
    colour = "purple", size = 0.8, alpha = 0.3
  ) +
  xlim(112, 157) + ylim(-44, -11) +
  labs(
    title   = "Polling booths in Australia — 2025 federal election",
    caption = "Source: AEC via readaec"
  ) +
  theme_void() +
  theme(plot.title = element_text(size = 12))

We can also pull TPP results at the booth level. Figuring out where votes come from within an electorate is fundamental to campaign strategy — even small seats have pockets of strongly opposing sentiment.

tpp_booths <- get_tpp_by_booth(2025)
head(tpp_booths)
# A tibble: 6 × 12
  stateab divisionid divisionnm pollingplaceid pollingplace       
  <chr>        <dbl> <chr>               <dbl> <chr>              
1 ACT            318 Bean                93925 Belconnen BEAN PPVC
2 ACT            318 Bean                11877 Bonython           
3 ACT            318 Bean                11452 Calwell            
4 ACT            318 Bean                 8761 Chapman            
5 ACT            318 Bean                 8763 Chisholm           
6 ACT            318 Bean                93916 City (Bean)        
# ℹ 7 more variables: `australian labor party votes` <dbl>,
#   `australian labor party percentage` <dbl>,
#   `liberal/national coalition votes` <dbl>,
#   `liberal/national coalition percentage` <dbl>, totalvotes <dbl>,
#   swing <dbl>, year <dbl>

Census analysis

eechidna does something genuinely difficult: it apportions ABS census variables to Commonwealth Electoral Division boundaries. Census data is collected at the mesh block level, but electoral boundaries don’t align with census boundaries — the eechidna team has done the spatial concordance work so you don’t have to. The result is a data frame with ~80 demographic variables, one row per electorate, ready to join directly with election results.

data(tpp22)
data(abs2021)

election2022 <- left_join(abs2021, tpp22, by = "DivisionNm")

A simple but telling question: does education level predict LNP support?

ggplot(election2022, aes(x = BachelorAbv, y = LNP_Percent)) +
  geom_point(alpha = 0.6) +
  geom_smooth(method = "lm", se = TRUE, colour = "#80b1d3") +
  labs(
    title   = "Education vs LNP two-party preferred vote — 2022",
    x       = "% with bachelor degree or above",
    y       = "LNP TPP %",
    caption = "Source: eechidna (AEC + ABS 2021)"
  ) +
  theme_minimal()

The negative relationship is striking — and quite recent. A university-educated vote was historically a Liberal stronghold; the teal wave of 2022 flipped that relationship in inner-city and coastal seats.

A lean regression with a handful of theoretically interesting variables:

model <- lm(
  LNP_Percent ~ BachelorAbv + MedianPersonalIncome + MedianAge + NoReligion + Indigenous,
  data = election2022
)

summary(model)

Call:
lm(formula = LNP_Percent ~ BachelorAbv + MedianPersonalIncome + 
    MedianAge + NoReligion + Indigenous, data = election2022)

Residuals:
     Min       1Q   Median       3Q      Max 
-22.3464  -6.5855  -0.2546   6.4105  19.0103 

Coefficients:
                      Estimate Std. Error t value Pr(>|t|)    
(Intercept)          -24.57466    8.85838  -2.774  0.00626 ** 
BachelorAbv           -0.43279    0.10439  -4.146 5.74e-05 ***
MedianPersonalIncome   0.04856    0.01122   4.329 2.78e-05 ***
MedianAge              1.81639    0.19911   9.123 5.66e-16 ***
NoReligion            -0.36076    0.08669  -4.161 5.40e-05 ***
Indigenous             0.40329    0.17452   2.311  0.02225 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 8.02 on 145 degrees of freedom
Multiple R-squared:  0.4916,    Adjusted R-squared:  0.4741 
F-statistic: 28.04 on 5 and 145 DF,  p-value: < 2.2e-16
hist(model$residuals, col = "bisque", freq = FALSE,
     main = "Residuals", xlab = "")
lines(density(model$residuals), col = "red")

One caution: this is an ecological regression — we’re modelling electorate-level aggregates, not individual votes. Interpreting the coefficients as individual-level effects would be the ecological fallacy. It tells us about the geographic pattern of voting, not about individuals.

Informal votes

Over 700,000 people — around 5% of all votes cast — vote informally each election. Of these, over half have ‘no clear first preference’, meaning their vote did not contribute to the tally of any candidate.

Informal votes are genuinely fascinating. Not only are there 8 official categories (the AEC publishes a detailed breakdown here), but the rate of informal voting varies tremendously by electorate.

Broadly, informal votes fall into two buckets. If we want to get particular about it:

  1. Protest votes — a person that thinks they are voting against:
    • the democratic system,
    • their local selection of candidates on the ballot, or
    • the two most likely candidates for PM.
  2. Stuff-ups — people who:
    • filled in the form wrong but a clear preference was still made, or
    • stuffed up the form entirely and it didn’t contribute towards the tally for any candidate.

The AEC works tirelessly to reduce stuff-ups on ballot papers through clear instructions and ballot design, but there isn’t much of a solution for protest votes.

The geographic distribution of informal voting is worth exploring in its own right. It is not randomly distributed — the variation tells a story about disengagement and political alienation that the formal results don’t capture.