Net interstate migration review

Australia’s official population estimates are calculated by adding and subtracting components of population change from a five-yearly Census base. Net interstate migration (NIM), the number of interstate migrant arrivals minus departures, is a key component of these estimates for the states and territories. Accurate estimates of NIM are critical for high quality population estimates.

There is no direct measure of interstate migration, making it more difficult to measure than the other components of overseas migration and natural increase. The ABS has used change of address data from Medicare (Australia’s universal health insurance scheme) to model interstate migration since the 1980s, incorporating Census-based expansion factors since 1996. As defence force personnel are generally not included in Medicare, Department of Defence movement data has been used over this time to supplement the Medicare data. 

After each Census, the ABS reviews the interstate migration model and calculates new expansion factors which are then used until the next Census. The ABS also periodically assesses other administrative data sources for the purposes of measuring interstate migration.

The Census also provides an opportunity to assess the quality of population estimates. One measure of this quality is the difference between the estimate updated from the previous Census (using data from Medicare and other sources), and the new Census-based estimate, referred to as the intercensal difference. The 2021 Census revealed larger than usual intercensal differences for some jurisdictions, with the Australian Capital Territory (5.1%) and Tasmania (4.9%) having the largest differences. For more information see Methodology used in rebased population estimates, June 2021.

The COVID-19 pandemic posed unique challenges to the estimation of NIM, with changes in Medicare address-updating behaviour in 2021 and 2022 due to Australia’s COVID-19 vaccination program. 

To address these issues and improve the accuracy of state and territory population estimates going forward, the ABS has introduced a mid-cycle (between Censuses) review of NIM to reduce error accumulating over a full intercensal cycle. The review comprised three main components:

• addressing COVID-19 impacts in the source Medicare data used to estimate interstate migration
• evaluating the existing expansion factor methodology, including aspects such as capping and age ranges
• exploring newly available linked administrative data from the Person Level Integrated Data Asset (PLIDA) to inform changes in interstate migration patterns between Censuses.

The review confirmed that Medicare is still the best single data source for estimating quarterly interstate migration due to its population coverage and availability, and identified improvements to the way Census-based expansion factors are calculated and applied. It also found that updating these factors annually using change of address data from the Australian Taxation Office (ATO) Client Register (from now on referred to as tax data) in PLIDA enhances their accuracy between Censuses, particularly for more mobile age groups.

Following this review, the ABS has:

• adjusted the source Medicare data to reduce the impact of COVID-19 vaccination-induced address updates over 2021 and 2022
• introduced enhanced 2021 Census-based Medicare adjustment factors, to replace the previously used expansion factors
• introduced a new annual adjustment to these factors incorporating PLIDA tax data
• applied these new factors to revise interstate arrival and departure estimates for the September 2021 to September 2024 quarters, and will use these factors for the December 2024 quarter onwards.

Medicare change of address data has been used since the 1980s as the main input to measure Australia’s interstate migration, and is supplied quarterly to the ABS by Services Australia. The review confirmed that Medicare remains the best available data source for measuring quarterly interstate movement due to its coverage of the overall population and the timeliness of data supply, and that it remains fit for purpose for the estimation of NIM. 

Australian residents are generally not required to report to government agencies when they move. Instead, people update address details with government agencies when convenient. For example, a person may not report their address change to Medicare until they visit a doctor. To account for the delay between someone moving and updating their address, Medicare moves for a given quarter are used to estimate NIM for the previous quarter (i.e. are lagged by three months).

Australia’s response to the COVID-19 pandemic had an unprecedented impact on this data and the lags associated with it. The COVID-19 vaccination program required nearly every Australian to interact with the Medicare system over 2021 and 2022. While there has always been a lag between people moving and updating their address with Medicare, this one-off event led to a large number of interstate movers reporting their change of address at the time of vaccination, rather than when they actually moved. This resulted in an artificial increase in moves on the Medicare system over this time. The ABS initially dealt with this by not applying expansion factors for the affected quarters. 

Further analysis found that these increases in moves impacted states and territories in different ways in different quarters, reflecting the timing of vaccination pushes in each state. The higher-than-usual Medicare arrivals for each state correlated with higher increases in dose 1 vaccination rates for those states, as published by the Department of Health at the end of each quarter during the vaccination uptake period. 

The Medicare data was significantly impacted for the June 2021 to March 2022 quarters, which relate to the internal migration estimates for the March to December 2021 quarters (after factoring in the three-month lag between Medicare moves and the estimation period). These quarters straddle the 2021 rebasing point.

Rebasing is the process of constructing a new base population figure for 30 June of the Census year and updating population estimates for the five years between Censuses to incorporate information from the most recent Census. Some states had their vaccination push prior to the 2021 Census, while others had theirs after. For these latter states, the excess moves registered at the time of vaccination had likely been accounted for in the Census rebasing process and may have subsequently been double counted.

To account for those excess moves, reduction factors for arrivals into each state for each of the impacted quarters were derived after assessing levels and trends of interstate moves before and after the vaccination period, and dose 1 vaccination rates, by state and quarter. The factors were determined using a combination of methods including linear regression and other modelling techniques. 

The arrivals into each state were adjusted by applying these reduction factors to all moves into those states, which consequently reduced departures from other states. Larger factors were used for states with higher increases in vaccination rates that quarter. The result is a smoother time series of quarterly Medicare-derived interstate arrivals and departures, and consequently the net interstate migration series for each state and territory.

As the internal migration estimates for the March and June 2021 quarters were finalised as part of 2021 rebasing, only the September and December 2021 quarters were revised based on this treatment as part of the review. 

The Medicare system covers Australian citizens, permanent residents and certain temporary visa holders. There are people who are included in estimated resident population (ERP) but are not covered by Medicare, such as people on international student visas or temporary work visas.

Most people report change of address to Medicare in a reasonable timeframe, usually because they interact with the health care system. Some people, particularly younger adults, do not report changes of address with Medicare or do so long after they move. This means that the Medicare data can underestimate interstate migration for certain age groups.

To account for these issues, expansion factors are calculated to calibrate the Medicare change of address data with the interstate migration patterns observed every five years in the Census, based on the address of usual residence one year ago question. In calculating these factors, several data treatments are applied (e.g. defence adjustment, multiple mover factor, age-specific smoothing and capping) and the age range for which the factors are used is determined. Once calculated, these factors are applied by state, age, sex and move type (arrival or departure) for the 20 quarters up until the next Census. For further information, see 2021 Census update of the Net Interstate Migration model. 

As part of this review, the ABS reviewed the expansion factor methodology and implemented some improvements.

These changes give us an improved set of base Medicare adjustment factors, which can:

• account for moves not reported and those not eligible for Medicare
• adjust moves based on observed under/over-estimates from previous Censuses.

However, because these factors are based on the five-yearly Census, they cannot react to changing:

• trends in Medicare address reporting within an intercensal period
• patterns of those not eligible for Medicare (i.e. recent migrants).

Migration and Medicare address reporting patterns may change between Censuses, diminishing the effectiveness of Census-based adjustment factors. Any error resulting from this accumulates over the 20 quarters between Censuses and contributes to intercensal difference.

As such, a method has been developed to adjust the factors between Censuses to reflect changes in the moves being missed by Medicare and identified in tax change of address data from PLIDA. These adjusted factors are then used to revise NIM, better aligning it with the other components of population change which are revised regularly between Censuses. 

The Person Level Integrated Data Asset (PLIDA) is a series of linked administrative datasets combining information on health, education, government payments, income and taxation, employment, and population demographics over time. Datasets are linked together using the Person Linkage Spine (the Spine), which contains the combined population from the following three datasets: 

• Medicare Consumer Directory (MCD) – containing all persons with an active Medicare enrolment
• ATO Client Register – containing demographic data about individuals who require a tax file number to interact with government, business, financial, educational and other community institutions. This dataset is referred to as tax throughout this paper.
• DOMINO Centrelink Administrative Data (DOMINO) – containing a snapshot of the characteristics of recipients of government payments such as the Age Pension and JobSeeker.

These three datasets contain residential address information and are updated on an annual basis in PLIDA. As such, the address information can be compared at different points in time to measure interstate address changes.

To assess the suitability of PLIDA address data for the NIM review, several rules and treatments were applied.

PLIDA presented an opportunity to explore alternative address data through an (anonymised) person-level lens. De-identified individual records were examined to determine whether people moved in one, two or all three of the available address data sources over a specified period, to identify the specific moves picked up in tax or DOMINO but missed in MCD. However as different people may interact with different data sources at different times, taking all interstate moves from each of the three sources over an annual period produced an unrealistic number of moves, far exceeding expectations and alternative data sources like the Census.

Instead, patterns and trends in aggregate data i.e. total interstate moves by state, age and sex across the PLIDA data sources were explored. This revealed that the aggregate tax data provided a useful indication of changes in the moves being missed by Medicare over the 2016–2021 intercensal period and should continue to do so in future. Further information on this analysis and findings follows below.

The MCD data in PLIDA is comparable to the Medicare data currently used for NIM. The review confirmed that MCD had the best coverage of interstate moves for the overall population of the three address data sources in PLIDA, and that it continues to be fit for purpose in the production of NIM. 

MCD was used to evaluate the suitability of tax data and DOMINO to supplement NIM. Introducing new data to the estimation of interstate migration may create inconsistencies around scope and lags in address updating, so the potential benefits of using tax and/or DOMINO have been weighed against these risks.

The DOMINO address updates were found to have similar coverage to MCD, as many people eligible for government payments are also eligible for Medicare. The inclusion of DOMINO address information added little improvement to NIM. 

The tax data captures address updates from many residents who pay income tax, including those who aren’t eligible for Medicare or are less likely to update their Medicare address. This includes recent overseas migrants and younger residents (aged 20 to 39 years) who are in the workforce. These characteristics made tax the best available data source to inform changes to adjustment factors between Censuses.

The additional benefit of the tax data over DOMINO is seen looking at the composition of elements (i.e. people) on the Spine. Of the 39,011,000 elements (reflecting all people resident in Australia since 2006):

• 34,406,000 (88.2%) are found on the MCD
• an additional 108,000 (0.3%) are found on DOMINO only
• an additional 4,456,000 (11.4%) are found on tax only.

Most of this 11.4% are overseas migrants, and therefore generally not eligible for Medicare or government payments. Changes in the volume and geographic distribution of this population between Censuses would not be picked up in the current NIM model.

The tax data captured changes in interstate migration levels and trends between the 2016 and 2021 Censuses that the Medicare data alone (supplemented with defence force moves and 2016 Census-based expansion factors) did not. For example, the following graph compares net migration for ages 20 to 39 in Tasmania and the ACT according to Medicare and tax. These were the states with the largest percentage intercensal differences in 2021, with population growth under-estimated between the 2016 and 2021 Censuses. 

The strong seasonality in tax address updates (associated with lodgement of tax returns) means that it can only provide an annual indication of interstate movements. It has strong coverage for the mobile population group where Medicare has higher undercoverage, but limited coverage for people under the age of 18 who are less likely to have a tax file number. As such, the tax data is not suitable to replace Medicare in the quarterly calculation of NIM but can be used to supplement it.

The additional information that the tax data provides between Censuses has now been incorporated into the estimation of interstate migration via annual updates to the base Medicare adjustment factors. The base Medicare adjustments factors are those described earlier in this paper, calculated using Census and Medicare migration data and incorporating changes to caps, widening of the age range and treatment for ages 15 to 19.

To make these updates, annual tax-Medicare ratios were created by dividing the annual interstate moves captured in the tax data by those captured in the Medicare data provided by Services Australia. These ratios are calculated by state/territory, sex, age, and move type (arrival or departure), with the input data smoothed (by taking a three-term moving average across single years of age) to reduce volatility. 

Comparing these ratios over time indicates where moves being missed by Medicare relative to tax change over time, and the base Medicare adjustment factors are updated accordingly. 

To make this comparison, a baseline period of tax and Medicare data was selected to best match the period of Medicare data feeding into the 2021 base adjustment factors. Due to the one-off COVID-19 vaccination impacts in the 2020–21 Medicare data, these factors were calculated using an average of the 2016–17 to 2019–20 Medicare data. For consistency, the baseline tax-Medicare ratios used for preparing updated Medicare adjustment factors are the average of the 2016–17 to 2019–20 ratios.

Corresponding ratios were calculated for 2021–22, 2022–23 and 2023–24, and divided by the baseline ratios. These quotients, representing the change in the ratios over time, were then multiplied by the base Medicare adjustment factors to create tax-updated adjustment factors for each year. These updated factors can be higher or lower than the base factors. The Appendix provides an example of this calculation. 

As the tax data provides the most benefit for capturing additional moves for ages 20 to 39, the base Medicare adjustment factors are only updated for these ages. These are the most mobile ages and where moves are more likely to be missed by Medicare. For all other ages the base factors are carried forward each quarter.

This method was trialled by preparing a test series of interstate migration estimates for the previous (2016–2021) intercensal period, recalculating corresponding population estimates and assessing their intercensal differences. For most state and territories, this method reduced or had little impact on these differences. The largest improvements were gained in the ACT and Tasmania, which had the largest percentage intercensal differences in 2021.

While the 2016–2021 analysis showed that applying the full effect of the tax adjustment worked well, this was based on more mature tax data that was several years old and included late tax returns. For the 2021–2026 period, the updated Medicare adjustment factors for each year (referred to as annual adjustment factors) use the average of the base factors and the tax-updated factors, halving the impact of the tax adjustment. This more conservative approach reflects the less-mature tax data being used for estimating more recent interstate migration, and exercises caution in introducing new data sources. This method will be assessed following the 2026 Census and the full effect of the tax adjustment may be applied in future. 

The introduction of the annual tax adjustment to update the Medicare adjustment factors potentially enables the ABS to introduce an ongoing revision process to the interstate migration series, similar to that for births, deaths and overseas migration. However, this mid-cycle review is the only committed revision to interstate migration during the 2021 to 2026 intercensal period. The ABS will review tax adjustments for 2024–25 and 2025–26 subject to available data and resources. Additional data sources will also be considered as part of future reviews.

The implications of the revision to interstate migration and population estimates below the state/territory level will likewise be determined in future. 

The effectiveness of the review and adjustments made will be assessed following the 2026 Census.