3412.0.55.001 - Information Paper: Improving Net Overseas Migration Estimation, Mar 2010  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 12/03/2010  First Issue
   Page tools: Print Print Page Print all pages in this productPrint All



1. Background
2. Current methods for estimating net overseas migration
3. Improvements to preliminary net overseas migration estimation
4. Changes to preliminary and revision timetables
5. Implementation date
6. Future directions

SUMMARY


In 2007, the Australian Bureau of Statistics (ABS) introduced new methods for the estimation of net overseas migration (NOM). With the first annual revision, released in March 2009, there was a large upward revision in the estimates of NOM and consequently the estimated resident population (ERP) in Australia for 2006-07. This was due mainly to the large revision from the preliminary to a final NOM estimate, which was much greater than previous revisions. However, it was also due to the cumulative impact of only releasing the revision to quarterly NOM estimates and ERP once a year, based on a scheduled annual revision cycle.

An opportunity to help address this issue arose in 2009 when the Australian Parliament passed the Federal Financial Relations Act 2009. The new legislation changed the date required for the Australian Statistician to make the yearly determination of the population of each state and territory (as at 31 December). It changed the due date from 'before 6 June' to 'before 31 August in the following payment year'.

In consultation with major stakeholders, this legislative change enables the ABS to provide quarterly ERP at a later date in the scheduled month of release (i.e. at the end of March, June, September and December). In turn, this enabled an additional quarter of overseas arrivals and departures data to be used allowing the methodology for preliminary NOM estimation to be improved. As a result, the ABS will implement two key changes to the methodology: changing from a 'two year ago' to a 'one year ago' propensity model, and reducing the pool of travellers using the model.

In addition, the ABS will change the current annual revision cycle for publishing final NOM to a six-monthly revision cycle. While not impacting on the difference between preliminary and final NOM estimates, this change will provide for the earlier progressive release of revised ERP figures.

The improvements outlined in this information paper are designed to capture some immediate gains to NOM estimation. However, additional investigations are being undertaken by the ABS to make further gains and build upon the improvements identified in this paper.

The purpose of this information paper is to provide an overview of the changes being undertaken to improve NOM estimation and reduce the large revisions to ERP. It will:

  • review changes being made to the methodology for estimating preliminary NOM;
  • outline changes to the preliminary and revision timetables;
  • provide implementation dates and schedules; and
  • discuss future directions the ABS is examining to make further improvements.

Estimates of both the resident population (ERP) and NOM for Australia and each of the states and territories are published quarterly in Australian Demographic Statistics (cat. no. 3101.0). The improvements outlined in this paper will be introduced from the September 2009 issue of Australian Demographic Statistics (cat. no. 3101.0), due for release on the 25 March 2010.
1. BACKGROUND

Net overseas migration is one component of population change used to estimate the Australian resident population each quarter, as are births and deaths. The ABS bases ERP on the concept of usual residence in Australia. For the purposes of NOM and therefore ERP, a person is regarded as a usual resident if they have been (or were expected to be) residing in Australia for a period of 12 months or more. As such, the ERP includes all people, regardless of nationality, citizenship or legal status, who usually live in Australia, with the exception of foreign diplomatic personnel and their families.

The quarterly peaks and troughs experienced in Australia's population growth are clearly driven by NOM, making it the major contributor to quarterly change in the ERP. In recent years, many factors have presented challenges in accurately estimating NOM, including:
  • increasing volumes of international movements across Australia's borders. For example, in 2009, there were 24.7 million border crossings, up 56% since 1999;
  • changes to the composition of international visitors and their duration of stay behaviour, such as that of international students;
  • international travel patterns of Australian residents (including duration and frequency of travel); and
  • more recently, changes in traveller behaviour due to the global financial crisis (GFC).

Each year, there are more than 20 million overseas arrivals and departures crossing Australia's borders, but the NOM is only about 1% of this figure. Accuracy in the measurement of NOM is crucial in determining accurate ERP for each of the states and territories each quarter. Currently, NOM accounts for well over half of Australia's population growth (64% for year ended 30 June 2009).
2. CURRENT METHODS FOR ESTIMATING NOM

Conceptually, the term NOM is based on an international traveller's duration of stay being in or out of Australia for 12 months or more. With the introduction in 2007 of the new methods for the estimation of NOM, the key changes were:
  • the introduction of a '12/16 month rule' for determining whether a person is a usual resident of Australia, where the 12 months do not have to be continuous and are measured over a 16 month reference period; and
  • the shift from a movements-based approach to a traveller-based approach for estimating NOM. This is done through assembling administrative data using a unique personal identifier to create individual traveller histories.

For detailed information on the methodology see Information Paper: Improved Methods for Estimating Net Overseas Migration, Australia 2006 (cat. no. 3107.0.55.003) and Information Paper: Statistical Implications of Improved Methods for Estimating Net Overseas Migration, Australia, 2007 (cat. no. 3107.0.55.005) or Chapter 6, Estimating Net Overseas Migration in Concepts, Sources and Methods, 2009 (cat. no. 3228.0.55.001).

Final NOM

It is with the final NOM estimates that the '12/16 month rule' can be fully applied. A traveller's actual duration of stay can only be calculated when data on overseas movements become available for the full 16 months following a reference period. Currently, the final NOM estimates based on the '12/16 month rule' are considered to be of high quality.

Processing the final estimation of NOM provides, for each traveller in the reference quarter, a 'migration adjustment' based on their initial category of travel.

The initial category of travel has a key role in making preliminary estimates of NOM. It is determined by a number of dimensions:
  • direction of travel - either an arrival or departure;
  • residence status - either a visitor or Australian resident (identified by the traveller themselves on the passenger card); and
  • duration of stay in Australia (or overseas) which can be either:
    • permanent traveller;
    • long-term traveller — has an initial duration of stay (or absence) of one year or more; or
    • short-term traveller — has an initial duration of stay (or absence) of less than one year.
For a visitor arriving and resident departing, it is their intended duration of stay that is recorded, whereas for a visitor departing and resident returning it is their measured duration of stay since the previous Australian border crossing that is recorded.

The initial category of travel is also determined at the time of the movement. However, all travellers are assigned to one, and only one, category of travel during a reference quarter. Further information and a list of the ten categories of travel are detailed in the Glossary.

The migration adjustment, created during final NOM processing, is the difference between the traveller's true behaviour (actual duration of stay and its effect on ERP status) as recorded in final NOM and what was deduced from their initial category of travel. In other words, an adjustment is made to correct if they should have initially been counted in or out of the population for the reference quarter. The 'migration adjustment' is not required to determine final NOM estimates, but is essential for estimating future preliminary NOM.


Preliminary NOM

Preliminary estimates of NOM are required within six months after the end of the reference quarter for the production of quarterly ERP of Australia and each of the states and territories. At that time, complete traveller histories for the 16 months following a reference quarter are not available.

To estimate preliminary NOM, the ABS developed a propensity model that uses the migration adjustments derived from final NOM during an earlier period. The migration adjustments are applied to travellers who are grouped according to their initial category of travel, age, country of citizenship and state or territory of usual/intended residence. The adjustment accounts for differences between a traveller's intended duration of stay and their actual duration of stay. The method is applied to each quarter and the preliminary estimate of annual NOM is the sum of the preliminary estimates for each quarter.

Preliminary NOM estimation is therefore modelled on patterns of traveller behaviours observed in final NOM during an earlier period. From September 2006 to June 2008 the migration adjustments used to estimate preliminary NOM were calculated from final NOM from the corresponding quarter two years earlier.
3. IMPROVEMENTS TO PRELIMINARY NOM ESTIMATION

The legislative changes in the Federal Financial Relations Act 2009 provided the opportunity for the ABS to publish ERP, for 31 December each year, at a later date. After consultation with major stakeholders in 2009, the ABS will now provide quarterly ERP at the end of each scheduled month of release (March, June, September and December). A new schedule is provided in Table 1.3 later in this paper.

This change has also made possible the use of one additional quarter of travellers movement data allowing the methodology used for preliminary NOM estimation to be improved. Using the additional one quarter of movement data (the quarter after the reference period) has enabled two key changes to the methodology:
  • changing from a 'two year ago' to a 'one year ago' propensity model and
  • reducing the pool of travellers using the propensity model.


Changing from a 'two year ago' to a 'one year ago' propensity model

Under the 12/16 rule, it can take up to 16 months after the reference quarter to determine an individual traveller's ERP status of being counted in or out of Australia's population. Since full movement histories are not available within the required time frame, preliminary NOM estimates are modelled using the migration adjustments from final NOM for an earlier period. Previously, adjustments were made based on the corresponding quarter two years earlier. The final NOM from two years earlier was used as the method needed to allow a full 16 months of data to accumulate before the final NOM could enable production of exact migration adjustments for a corresponding quarter. With the previous release schedule for ERP (prior to 2010), only 12 months of movement data were available. This was insufficient to produce exact migration adjustments for the corresponding quarter one year earlier. To be able to produce an 'exact one year ago' model would require a full 16 months of data to accumulate to be able to calculate final NOM estimates and the migration adjustments necessary for use in the propensity model. Currently it would require an additional four months of movement data post reference period.

However, by using an additional three months (one quarter) of movement data post reference period, 15 months of movement records become available for the propensity model. Analysis showed that 15 months of movement data provide enough information to produce migration adjustments for the corresponding quarter one year earlier. The analysis revealed that using the full 16 months of movement records in an 'exact one year ago' propensity model only very marginally improved results (i.e. less than 1%) when compared to using 15 months of movement data in an 'approximate one year ago' model.

The 'approximate one year ago' propensity model uses a combination of 'one year ago' and 'two year ago' propensities. First, the 15 months of movement data available are used to resolve the ERP status of as many travellers as possible (almost all travellers) for the corresponding quarter one year earlier. Second, the model uses this group of travellers to calculate 'one year ago' propensities that are then used for the majority of travellers with similar characteristics in the current reference quarter. Each quarter there is a small number of travellers whose ERP status remains indeterminate after processing the 'one year ago' propensities. For this small group, a 'two year ago' propensity is calculated and then applied to travellers with similar characteristics in the current reference quarter.

For example, if processing September quarter 2009 (July, August and September 2009), using one additional quarter of movement data (October, November and December 2009) means the ERP status can be resolved for almost all travellers in the corresponding quarter one year earlier (September quarter 2008) using the 12/16 rule. Therefore, final NOM and migration adjustments necessary for the propensity model can be calculated for almost all of these travellers, including all travellers in the first two months in the quarter (e.g. July and August 2008). July 2008 has 16 months of movement records available at November 2009, August 2008 has 16 months of movement records available at December 2009, whereas September 2008 only has 15 months of movement records available at December 2009. However, the ERP status of many travellers in the last month (e.g. September 2008) can also be resolved with only 15 months of movement records available (e.g. as at December 2009). For example, any overseas traveller who has already recorded a duration of stay in Australia for more than 12 months is considered in Australia's population as it would no longer be possible for them to be counted out of the population. Conversely, any overseas traveller who has recorded a duration of stay away from Australia for more than four months is considered out of Australia' population as it would no longer be possible for them to be counted in the population. For those travellers whose ERP status is still unresolved, a 'two year ago' propensity (e.g. from September 2007) is calculated and then used.

With the need to provide timely preliminary NOM estimates, the ABS will now use the 'approximate one year ago' model as there is very little improvement made (i.e. less than 1%) by waiting one additional month to complete the full 16 months to produce exact migration adjustments for the corresponding quarter one year earlier.


Reducing the pool of travellers using the propensity model

Many travellers' ERP status can be determined in a much shorter time frame than the full 16 months. With the availability of this additional one quarter of movement data and applying the conditions of the 12 out of 16 month rule, many of the travellers' ERP statuses can be resolved. For example, if processing the September quarter (July, August and September) using one additional quarter of movement data (October, November and December), then for the months of July and August a minimum of four months extra movement data have become available. July has four extra months of movement data by the end of November; and August has four extra months of data available by the end of December. In essence, any overseas traveller who has reached a recorded duration of stay out of Australia of four months or more is then considered out of Australia's population as it would no longer be possible for them to be in Australia for more than 12 out of 16 months. This reduces by around half, the number of travellers for which the propensity model needs to be applied to when estimating preliminary NOM.


Results of improvements to preliminary NOM estimation

For the purposes of this paper the 'one year ago' model refers to the 'approximate one year ago' propensity model that has been applied to a reduced pool of travellers.

As would be expected, the use of a 'one year ago' model is likely to be more closely aligned with capturing current changes in traveller behaviour than the 'two year ago' model. Investigations undertaken by the ABS have shown that substantial improvements are made in the estimation of preliminary NOM by the use of a 'one year ago' model compared with a 'two year ago' model.

For 2006-07, the difference between the original preliminary NOM estimation based on the 'two years ago' model and final NOM was 55,500 persons. In contrast, the difference between the new preliminary NOM estimation based on the 'one year ago' model and final NOM was 24,900 persons, a 55% improvement on the 'two year ago' model. Although there is fluctuation from quarter to quarter, a substantial improvement was made for each quarter in 2006-07 as is seen in Table 1.1.

1.1 Comparison between Preliminary NOM models —
'two year ago' model & 'one year ago' (with reduced pool of travellers) model, Australia — 2006-07

Preliminary
NOM
(2 yr ago)
(a)
Preliminary
NOM
(1 yr ago &
reduced pool)(b)
Final
NOM
(c)
Diff btw
Preliminary
(2 yr ago)
& Final
Diff btw
Preliminary
(1 yr ago &
reduced pool)
& Final
Improvement made
using
Preliminary
(1 yr ago &
reduced pool)
Ref quarter
no.
no.
no.
no.
%(d)
no.
%(d)
%(e)

Sep qtr 2006
45,619
50,821
56,940
11,321
20
6,119
11
46
Dec qtr 2006
38,358
44,701
50,618
12,260
24
5,917
12
52
Mar qtr 2007
56,963
68,869
76,071
19,108
25
7,202
9
62
Jun qtr 2007
36,430
43,498
49,195
12,765
26
5,697
12
55
2006-07
177,370
207,889
232,824
55,454
24
24,935
11
55

(a) Preliminary NOM estimates based on a propensity model using all migration adjustments from the corresponding quarter 2 years earlier.
(b) Preliminary NOM estimates based on a propensity model using the majority of migration adjustments from the corresponding quarter 1 year earlier and applied to a reduced pool of travellers.
(c) Final NOM estimates have been used in compiling Australia's official estimated resident population (ERP) for 2006-07.
(d) As a percentage of final NOM for the corresponding reference period.
(e) Improvements made as a percentage when the preliminary estimate using the '1 year ago' model (with the reduced pool of travellers) is compared to the '2 year ago' model.


Comparing estimates produced by the two models at the state and territory level showed substantial improvements for the larger states when using the 'one year ago' model. For the smaller states and territories, improvements were seen annually but some fluctuations were experienced on a quarterly basis. Much of this fluctuation can be the effect of very small numbers being calculated for the smaller states and territories. For example Tasmania, the Northern Territory and the Australian Capital Territory together represented less than 2% of Australia's total preliminary NOM estimate in 2006-07.

1.2 Comparison of NOM, Preliminary models and final - by State - 2006-07
Graph: shows an improvement using the one year ago model for each state and territory.

4. CHANGES TO PRELIMINARY AND REVISION TIMETABLES

Following the legislative changes made in 2009 and consultation with major stakeholders, the ABS will now publish quarterly ERP at the end of each scheduled month of release. This will include the quarterly release of preliminary estimates for the three components of population change: net overseas migration (NOM), net interstate migration (NIM), and natural increase (births minus deaths).

Release dates have been changed to allow for the use of additional traveller movement data in the calculation of improved preliminary NOM estimates.

Table 1.3 shows the scheduled release dates for ERP and the components of population change to be published quarterly in Australian Demographic Statistics (cat. no. 3101.0).

1.3 New scheduled release dates for Australian Demographic Statistics (cat. no. 3101.0)

Issue (Quarter)
Release Date

Sep 2009
25 Mar 2010
Dec 2009
29 Jun 2010
Mar 2010
29 Sep 2010
Jun 2010
21 Dec 2010
Sep 2010
29 Mar 2011
Dec 2010
29 Jun 2011
Mar 2011
29 Sep 2011




Six-monthly revision cycle for NOM

The quarterly fluctuations experienced in Australia's population growth and thereby ERP are currently driven by NOM. To help reduce the impact of possible large revisions to population estimates from only revising NOM estimates once a year, as was the previous practice, the ABS has consulted with major stakeholders and will change the current annual revision cycle for publishing final NOM to a six-monthly revision cycle. Table 1.4 shows the expected schedule of release for NOM estimates for the September quarter 2008 to June quarter 2012.

With the implementation of the new six-monthly revision cycle for NOM it also means the annual revision for natural increase can be released six months earlier (reverting to the previous practice) to be published in the March quarter issue released in September each year of Australian Demographic Statistics (cat. no. 3101.0) (see Table 1.4).

1.4 New release schedule, Net overseas migration and natural increase

Preliminary
Natural increase
Preliminary
NOM
Revised
Natural increase(a)
Final
NOM(b)
Reference quarter
Month of release
Month of release
Month of release
Month of release

2008
September
Mar 2009
Mar 2009
Sep 2010
Sep 2010
December
Jun 2009
Jun 2009
Sep 2010
Sep 2010
2009
March
Sep 2009
Sep 2009
Sep 2010
Mar 2011
June
Dec 2009
Dec 2009
Sep 2010
Mar 2011
September
Mar 2010
Mar 2010
Sep 2011
Sep 2011
December
Jun 2010
Jun 2010
Sep 2011
Sep 2011
2010
March
Sep 2010
Sep 2010
Sep 2011
Mar 2012
June
Dec 2010
Dec 2010
Sep 2011
Mar 2012
September
Mar 2011
Mar 2011
Sep 2012
Sep 2012
December
Jun 2011
Jun 2011
Sep 2012
Sep 2012
2011
March
Sep 2011
Sep 2011
Sep 2012
Mar 2013
June
Dec 2011
Dec 2011
Sep 2012
Mar 2013
September
Mar 2012
Mar 2012
Sep 2013
Sep 2013
December
Jun 2012
Jun 2012
Sep 2013
Sep 2013
2012
March
Sep 2012
Sep 2012
Sep 2013
Mar 2014
June
Dec 2012
Dec 2012
Sep 2013
Mar 2014

(a) The annual revision cycle for natural increase used for 2006–07 and 2007–08 had revised estimates released in the month of March, 21 months after the reference year. The new annual revision cycle for natural increase is scheduled to start from September 2010 and then each September onwards. This will be 15 months after the reference year and will be six months earlier then the previous schedule.
(b) The annual revision cycle for NOM used for 2006–07 and 2007–08 had revised estimates released in the month of March, 21 months after the reference year. A new six-monthly revision cycle is scheduled for NOM. The new quarter for release will start from September 2010 and then every six months onwards. This means the September and December quarters and therefore calendar year estimates of final NOM, will be available six months earlier than the previous schedule.

5. IMPLEMENTATION DATE

The improved methods for estimating preliminary NOM based on the 'one year ago' propensity model will be published for the first time in ABS population estimates on 25 March 2010 in Australian Demographic Statistics, September Quarter 2009 (cat. no. 3101.0).

The improved preliminary NOM methods will be applied to estimates from September quarter 2008 onwards. For September quarter 2008 to June quarter 2009, the already published preliminary NOM estimates based on the previous methods using a 'two year ago' propensity model will be replaced with the improved preliminary NOM estimates. September quarter 2009 preliminary NOM estimates will be based on the improved methods and published for the first time.

As part of the previous scheduled annual revision cycle, the September 2009 issue of Australian Demographic Statistics (cat. no. 3101.0) to be released on the 25 March 2010 will also provide revised estimates for the 2007-08 financial year for births, deaths, natural increase, NOM and ERP.
6. FUTURE DIRECTIONS

The improvements outlined in this information paper are designed to capture some immediate gains to NOM estimation. However, additional investigations are being undertaken to make further gains and build upon the improvements identified in this paper.

The ABS continues to investigate ways to improve the quality of the input data (i.e. overseas arrivals and departures data) used in the calculation of preliminary NOM. By using additional data provided by the Australian Government's Department of Immigration and Citizenship, improvements may be made to the quality of the 'initial category of travel' variable. It is on the 'initial category of travel' that the propensity model currently bases its migration adjustments to help estimate preliminary NOM. If successful, the ABS plans to implement these changes to further improve preliminary NOM estimation and publish in a future edition of Australian Demographic Statistics (cat. no. 3101.0).

With a longer time series of final NOM estimates now available, the ABS is also undertaking an examination to improve the cross-classified groupings of travellers that are used by the propensity model. Currently, groupings are made by the following variables: initial category of travel, age, country of citizenship and state or territory of usual/intended residence. The effectiveness of additional variables such as visa class will be examined and other areas of research such as the use of time series analysis may be undertaken. However, their use for improving preliminary NOM estimation will depend on the operational feasibility for the ABS to implement.
CAUTION

International migration is a volatile phenomenon caused by a wide range of demographic, social, economic and political determinants and consequences. With the improvements being implemented to estimate preliminary NOM, combined with the impact of the GFC and changes being observed in recent traveller behaviour, the ABS advises users to take care when making future assumptions based on the large upward revisions recently experienced between preliminary and final NOM estimates.


FURTHER INFORMATION

For further information relating to the improved methods for estimating net overseas migration, or the implementation of these methods, please contact:

Neil Scott
Assistant Director,
Migration Analysis and Reporting
Demography
Australian Bureau of Statistics
Locked Bag 10
BELCONNEN ACT 2616
Phone: (02) 6252 5959
Email: demography@abs.gov.au
Facsimile: (02) 6252 7494