DATA SOURCES AND METHODS


Medicare
Defence
Combining Medicare and Defence force data
Converting from postcode-based data
Converting from postcode-based data - data quality
Reconciling to state/territory estimates


Medicare

The ABS has evaluated a range of potential sources of administrative data for estimating interstate migration on a quarterly basis (ABS 2006a). Data supplied by the Australian Government Department of Human Services containing change of address information for people enrolled on Medicare was found to be the most effective source currently available. Since the mid 1980s, this Medicare enrolment information has been provided to the ABS as at the last day of each quarter (March, June, September, December). The de-identified data include age, sex, postcode of currently-enrolled address and postcode of enrolled address three months earlier. The data are only provided for those whose postcode of current address is different to postcode of address three months earlier. From 2012, the data will be provided based on ASGS regions.

To account for the inevitable lag between time and notification of move, the Medicare information for each quarter is used to estimate the moves that occurred in the previous quarter. This assumption of a three month lag between time of move and time of notification is consistent with the three month lag assumption that has been used in ABS’ Medicare-based interstate migration models since 1996.

The Medicare movement counts are then adjusted by incorporating 'expansion factors' (ABS 2009b). Expansion factors are calculated by the ABS to give an indication of how Medicare movement data under-enumerates the actual number of movers by age and sex, state/territory, and type of move (arrival and departure), and are applied in the models used to estimate interstate migration. These expansion factors are calculated by comparing census-based migration data with Medicare movement data for the same period. Typically, young adults are missed most in the Medicare movement data and therefore expansion factors are applied more for young adults. For moves between states, the expansion factors of the departure state and the arrival state are applied to adjust the Medicare movement counts. For inter-postcode moves within states, the average of the departure and arrival expansion factors are applied to adjust the Medicare movement counts. All expansion factor adjustments are applied by age and sex, and are applied consistently for movement counts within each state/territory.

Defence

While Medicare provides good coverage of the Australian population, it excludes Defence force personnel. To supplement the Medicare data for the calculation of interstate estimates, the ABS obtains change of location address information from the Department of Defence. From this file, the ABS has a record of quarterly moves of Defence force personnel, by age, sex, and postcode of departure and arrival. The data are considered to be reflective of the time of move, and therefore are not lagged (like the Medicare data).

Combining Medicare and Defence force data

The quarterly person Medicare movement data (lagged three months) and the Defence data are aggregated for each financial year, providing an indication of total number of arrivals to and departures from each postcode that occurred in the twelve months to 30 June each year. It is important to recognise that these annual inter-postcode migration counts are aggregated from quarterly movement snapshots, and therefore one person can be represented as moving up to four times throughout the financial year. Another important note is that because the Medicare and Defence data are snapshots as at the end of each quarter, they do not account for multiple movements of people within each quarter.

Converting from postcode-based data

While the Medicare and Defence force data are provided on a useful regional geographic basis, the Australia Post postcode, the output is most useful when prepared based on Australia’s official statistical geography. Up until 2011, the standard geography was the Australian Standard Geographical Classification (ASGC), however from 2011, the ASGC is being replaced by the ASGS.

Throughout most of Australia, the postcode geography is substantially different from the ASGC. To attempt to account for this, a postcode to ASGC correspondence has been prepared to convert the postcode-based migration counts to ASGC regions. Assuming the population of arrivals in to and departures out of postcodes is distributed in a similar way to the population used to prepare the correspondence, these correspondences can be used to convert numbers of arrivals and departures from postcode to ASGC region. For more information about the postcode to ASGC correspondence refer to the ABS Geography Portal <https://www.abs.gov.au/geography>.

Since there are two different regions associated with each move of a person (the arrival region and departure region), the application of this correspondence can be a complex procedure for some regions. A person moving from one postcode which covers X ASGC regions, to another postcode covering Y ASGC regions, could potentially be moving from one to another of X*Y pairs of ASGC regions. To account for this, all of the postcode-based Medicare movement data are effectively converted twice, once for the departure postcode, and once for the arrival postcode.

Postcode to ASGC correspondences based on the distribution of the total population were used to convert the Medicare data. Postcode to ASGC correspondences based on the distribution of persons in Defence occupations (based on the previous Census) were used to convert the Defence data.

From 2012, with the change to the supply of Medicare data directly on ASGS regions which have originally been geocoded by Medicare from a residential address, there will be no need to convert Medicare movements from postcode. This will eliminate some of the data quality issues currently associated with converting data from postcode to SLA. The ABS currently has postcode-based Medicare data dating back to 1985-86, so a historical time series of data can be converted to SLA level providing a high level of detail. The data can also be validated using the relevant Census year data, e.g. 2006 Census for 2005-06 Medicare data, using ASGC-based data, e.g. SLA. At a later date, it may also be possible that some historical ASGS coded data will be provided by the Australian Government Department of Human Services.

Converting from postcode-based data - data quality

There are several limitations to the postcode-based Medicare and Defence force data that need to be taken into account when attempting to ascertain the quality of data converted from postcode to regions such as SLAs using a postcode-based correspondence.

An underlying assumption when using correspondences to convert from one region to another confidently is that the population being converted is distributed across regions in the same way as the population used to prepare the correspondence. In converting Medicare-derived arrivals and departures data from postcode to ASGC regions, the postcode to SLA correspondence based on total population may not accurately reflect the distribution of arrivals, or departures into and out of these postcodes, for example, where there is high growth experienced with a newly formed suburb in one part of the postcode.

While some whole postcodes can be built up to form SLAs, providing high confidence in the data converted for these SLAs, most postcodes in Australia cross SLA boundaries. This creates a degree of uncertainty in relating the postcode-based data to SLAs. In some cases, postcodes encompass several SLAs, including whole SLAs and parts of SLAs. There is no over-riding pattern or relationship between postcodes and SLAs across Australia. In some capital city regions, such as Sydney and Melbourne, it is typical for several postcodes to combine to form SLAs (although most SLAs still contain several parts of postcodes). However in regions such as Brisbane, Darwin and Canberra, where SLAs coincide more with individual suburbs, there are virtually no SLAs which contain whole postcodes.

The spatial inconsistency between postcodes and ASGC regions becomes less problematic for larger ASGC regions (e.g. Statistical Regions or Statistical Divisions), because it is more likely that data from entire postcodes can be aggregated directly to these regions.

Another factor which complicates the relationship between postcode and ASGC units is the existence of 'non-mappable' postcodes, such as post office boxes and special-purpose delivery postcodes. Data coded to a post office box postcode for instance cannot be allocated to SLAs in a straightforward way. The existence of non-mappable postcodes also means that the postcode geography is not mutually exclusive and exhaustive - a point of land may be associated with more than one postcode, for example a resident of inner Sydney may consider their postcode to be either 2000 (Sydney) or 2001 (Sydney GPO). This is unlike the ASGC geography where each point of land in Australia is assigned to one and only one SLA.

These difficulties reflect the fact that postcodes are not designed to be regional entities for the collection and dissemination of data; postcodes exist to facilitate the sorting and delivery of mail by Australia Post.

When converting postcode-based migration data to SLAs, caution needs to be applied in areas where the postcode to SLA relationship is less reliable, in particular in areas where one postcode covers several SLAs, where small proportions of postcodes feed into SLAs, and other areas where there is low confidence in the postcode to SLA concordance (e.g. non-mappable postcodes). There is a general trade-off between size (of population or population flow) and accuracy/reliability. In general terms, the larger the SLA, the greater the level of confidence. However, SLAs that align with distinct localities, including small rural SLAs, can also relate well to postcodes.

To assist in assessing the confidence of data converted from postcodes, the ABS has developed an index to indicate quality of data converted from postcodes, the Concordance Confidence Index (CCI). This index can be used to identify regions where there is low confidence in the converted data, and can help identify SLAs that can be combined for output purposes. For areas where there is low confidence in the Medicare data converted from postcode to SLA, migration data from the previous census can be used to distribute arrivals and departures across SLAs to help achieve a better distribution at SLA level. Refer to Appendix 3 for more information about how the CCI has been constructed and can be applied to assess and adjust arrivals and departures data converted from postcodes.

One other known deficiency of using postcode-based data sources to estimate internal migration at the SLA level is the undercounting of moves between SLAs which occurs when they share part, or all, of the same postcode. In the most extreme case, when two or more SLAs are completely contained within a single postcode, the postcode-based Medicare and Defence data will record no movers at all between these SLAs. This effect can be significant in regional centres which contain several SLAs, as a single postcode may cover a large proportion or all of the urban area. To estimate movements missed between SLAs due to sharing parts of the same postcode, another index has been developed (the Missed Movers Index or MMI), which can be applied to adjust the Medicare and Defence force data converted from postcode to SLA. See Appendix 2 for more information on how the MMI is calculated and can be applied to adjust arrivals and departures data converted from postcodes.

Reconciling to state/territory estimates

This combined Medicare and Defence force data, aggregated from postcode to state/territory, form the basis of the ABS's quarterly estimates of interstate migration. The data, which is produced by age and sex, are published by the ABS on a quarterly basis in Australian Demographic Statistics (cat. no. 3101.0) and annually in Migration, Australia (cat. no. 3412.0). To maintain consistency with published estimates of interstate migration prepared by the ABS, the adjusted and converted Medicare and Defence force movement data at the SLA level can be constrained, or forced, to the estimates of interstate moves published quarterly by ABS, by age and sex. This is carried out using an Iterative Proportional Fitting (IPF) procedure. Corresponding adjustments for interstate moves can be applied for numbers of moves made within states/territories.