|Page tools: Print Page Print All|
Despite these limitations, Medicare data is the most effective source of internal migration currently available, based on timeliness and population scope. Address change data from the Defence force data and from the Census data are used to supplement the Medicare data, to address some of the known limitations.
Defence force adjustments
Australian defence force personnel have access to alternative health services and so may not use Medicare's services. To account for this, 70% of interstate movements by defence force personnel (calculated by age, sex and state/territory of arrival and departure) are added to the Medicare data. This data is provided to the ABS by the Department of Defence quarterly. It is not known what proportion of defence personnel opt to use Medicare instead of Defence's health system. The 70% factor is an estimate based on the assumption that single people are most likely to exclusively use the Defence health service whereas personnel with a partner or dependents are likely to be listed on the same Medicare card as their family members and so captured in the Medicare address change data. In 2016, 71.5% of defence personnel who moved interstate had neither a partner nor children.
The defence adjustment has a small total impact on net interstate migration, accounting for less than 3% of all movements. This impact varies markedly across states and territories, from 6.5% of movements to and from the Northern Territory in 2015/16, to 0.6% of movements to and from Tasmania.
The Australian Census of Population and Housing includes a question on address of usual residence one year ago and address five years ago, so that alternative interstate migration estimates can be calculated. These estimates are complementary to the Medicare-based estimates, rather than being superior. Census only provides a one-year and five-year snapshot of interstate migration, whereas quarterly estimates are required for the purposes of calculating the population. Census is limited in its ability to capture multiple interstate movements by the same person within the one or five year period. Although the scope of Census covers the whole population, non-response is still a factor - either where no Census form was received from an individual, or the specific question was missed.
Extent of non-response in Census data
Of all Census records for usual residents, 8.6% had no response stated for their address one year ago. There was little variation in response rate between sexes or between states, other than Tasmania which had a non-response rate of 14.2% for this item. The item response rate also varied by age, with people over age 70 more likely not to answer. Most of non-responses came from imputed records. This means either that no form was collected but a Census record was created (with basic demographic information imputed), or that age or sex was not stated on a form received. Census records for which the person's sex or age or both was imputed accounted for 61% of records with address one year ago not-stated, compared to only 6.0% of all records. Imputation rates also varied by state, between 5.0% for Tasmania and 12.0% for the Northern Territory.
, where a = age/ sex/ state (e.g. 20 year-old males in New South Wales)
2) Item non-responses are pro-rated across states by age and sex. For example, there were 3,800 males aged 20 living in New South Wales whose address five years ago was not stated. Of those who did answer that Census question, 97% lived in New South Wales five years ago, 0.6% lived in Victoria, 1.1% lived in Queensland, etc. The 3,800 not stated responses were distributed according to these proportions.
3) People aged 0-4 on Census night have no response for 5 year-ago question. For 0-1 year-olds these are estimated based on the one year ago data for 1 year olds. For 2-4 year-olds an estimate is produced based on the data for five year olds, using the relationship between 5 year olds and younger ages in the Medicare data.
The data resulting from this adjustment process is what we call the 'Census-based' interstate migration estimates.
According to these adjusted Census results, 340,000 people lived in a different state on Census night than where they lived one year earlier. Slightly more males than females moved interstate despite there being more females in the population. Interstate movers also had a more pronounced age distribution than the general population. People aged 18 to 40 made up over half (54%) of all interstate movers, compared to only one-third (32%) of the total population. These trends broadly align with published preliminary estimates of net interstate migration.
Calculating expansion factors
The interstate migration estimates used in calculating the quarterly ERP incorporate each of the above data sources. The model includes an 'expansion factor' calculated from the Census-based estimates to account for under-reporting of address change to Medicare, as follows:
Expansion factors are calculated for each age/sex/state/move type (ie arrival or departure) combination, and applied to certain age groups as:
Multiple mover factor
To calculate the multiple mover factor, anonymous records from the quarterly Medicare data were matched by age, date of birth, enrolment type and postcode to estimate which movement records constituted multiple interstate movements by the same person. This percentage was applied to the Census data by single year of age/sex/state.
The multiple mover calculation was last done in 2006, when it was found that people who moved interstate more than once accounted for 7.0% of all interstate movements. In 2016 this percentage was found to be 6.5%.
Census estimates include defence force personnel. Approximately 88% of the defence adjustments applied to Medicare were conceptually covered by Census, and were removed from the Census estimate to allow for a direct comparison with the Medicare data. The remaining 12% were conceptually not able to be captured by Census and therefore were removed from the defence adjustments before subtracting from Census.
This included all movements for people who moved interstate but returned to their original state, and the intermediate movements of a multiple interstate move (e.g. from New South Wales to the Australian Capital Territory to Victoria).
Smoothing and capping
Because the expansion factors are based on comparison of only one year of data (2015-16), they are potentially volatile and are limited in their ability to represent a longer period. The relationship between Census and Medicare for a given age/sex/state observed to particularly fluctuate as actual migration behaviour changes. Smoothing and capping are both treatments which help to 'future-proof' the expansion factors.
All inputs, as well as the expansion factors themselves, were smoothed by taking a three-term moving average across single years of age. For example, the smoothed figure for age 24 is the average of the figures for ages 23, 24 and 25. This reduces the impact on future estimates of random noise within the historical data. This smoothing also addresses the theoretical inconsistency that Census data gives age at the end of the period, rather than age at move. In previous reviews, adjusting the age of the Census data was not found to have a significant impact on the data. Smoothing was not applied where this would change the real pattern of the data - i.e., defence was only smoothed starting at age 19, because smoothing ages 17 and 18 distorted the real pattern.
Expansion factors were capped at 2, as has been the practice in the past. Capping the expansion factors at 2.0 limits the influence of any one age group, whose behaviour may change over time.
Expansion factors are calculated for all ages, but are only applied to certain ages. When this model was originally designed in 1996, only a small number of consecutive ages were under-represented in the Medicare data, and the extent of undercoverage was relatively small. As can be seen in the graph Interstate moves, ratio of Census to Medicare-based, by age, Australia above, both the number of ages under-represented and the extent of this under-coverage, have increased over the last 20 years. In 2016, the possibility of applying all factors greater than 1.0 (rather than limiting to an age range) was considered, however the results were too variable, especially for smaller states.
The original intention of the model was to apply an expansion factor to all ages that experienced under-reporting in the Medicare change of address data - with an upper age limit of 55 years, as data becomes more volatile in older ages. That is, any age (below 55) for which the expansion factor was greater than 1.0. When this model was originally designed (in 1996), only a relatively small number of consecutive ages were under-represented, and only by a relatively small amount.
In 2016, over half the ages had expansion factors greater than 1.0, not necessarily consecutively (see the graph Interstate moves, ratio of Census to Medicare-based, by age, Australia above). To reduce this variability and 'future-proof' the model as much as possible, the age range to which the expansion factors are applied was limited to only addressing the main bulk of the under-reporting, at the Australia level. Note that expansion factors are calculated and applied at the state level - it is only the age range which is determined at the Australia level. This approach produced less variability when applying different Census' expansion factors to the same Medicare data, and also performs well in minimising intercensal difference.
The resulting age range to which the expansion factors were ultimately applied was 17-35 for males, and 17-30 for females. These age ranges cover 91% of under-reporting according to the 2016 Census.
Revising interstate migration
The new, 2016 Census-based expansion factors have now been applied to all data from September 2011 onwards, and will continue to be used in the preliminary migration model until after the 2021 Census. Prior to this revision, preliminary interstate migration (that is, from September 2011 onwards) was based on the expansion factors calculated from the 2011 Census. Data up to June 2016 is now considered final. Data from September 2016 onwards will be revised following the 2021 Census.
The expansion factors are available from the Downloads tab in the Interstate Migration Expansion Factors datacube.
Prior to this revision, preliminary interstate migration (that is, from September 2011 onwards) was based on the expansion factors calculated from the 2011 Census. The newly-calculated 2016 Census-based expansion factors will be applied from September 2016 - so the last 5 quarters of published data are being revised due to the change. These new factors will continue to be used in the preliminary model until after the 2021 Census. Data from September 2011 to June 2016 has been revised according to the method below, and is now 'final'.
In previous years, the new expansion factors have been applied only to estimates forward from the most recent Census. The revision to the previous intercensal period was done differently.
Difference to previous revision method
The method of revision is reviewed and adjusted each Census cycle in order to produce the best final estimates. One indicator of the accuracy of the interstate migration estimates is intercensal difference. Intercensal difference is the difference between preliminary population estimates based on the 2011 Census (updated using births, deaths, overseas and interstate migration data), and the 'rebased' population estimate based on the 2016 Census. If the final interstate migration estimates result in a smaller intercensal difference, this may indicate an improvement in the estimate. Intercensal difference cannot be wholly attributed to interstate migration - other components as well as the 2011 base population and 2016 population estimate may contribute.
In the past, the difference between the Census-based estimate and the Medicare-based estimate was highly correlated with preliminary intercensal difference (R2=0.82 in 1996). The revision to net interstate migration (NIM) was made by adjusting the Medicare estimate for each state by the amount of the intercensal difference. This adjustment consistently resulted in a NIM estimate that was closer to the Census estimate than the preliminary estimate had been. Over time the relationship between Census, Medicare and Intercensal difference weakened such that the previously observed correlation no longer exists (R2=0.05 in 2016). Neither the newly calculated Census-based estimates nor the previously published Medicare-based estimates are observed to be systematically 'more accurate' than the other for every state in 2016.
The estimation method used in recent revision cycles was therefore not appropriate to the new data. It was decided that a more appropriate treatment would be to revise the 2011-2016 data using the same model as the preliminary estimates, but updated with the 2016 expansion factors. This method treats all states comparably, draws on the strengths of both Census and Medicare and aligns well, overall, with known trends over the five-year period. It also produced more plausible and consistent results than other methods considered, when tested against other Census years.
The graphs and tables below show the differences between preliminary and final NIM estimates.
These documents will be presented in a new window.
3101.0.55.003 - Information Paper: 2016 Census Update of the Net Interstate Migration Model, 2011-2016
Latest ISSUE Released at 11:30 AM (CANBERRA TIME) 19/12/2019 First Issue