Australian Bureau of Statistics

Rate the ABS website
ABS Home > Statistics > By Catalogue Number
ABS @ Facebook ABS @ Twitter ABS RSS ABS Email notification service
3116.0 - Demography Working Paper 1999/2 - Estimating Interstate Migration, 1996 - 2001, 1996 - 2001  
Latest ISSUE Released at 11:30 AM (CANBERRA TIME) 06/05/1999   
   Page tools: Print Print Page Print all pages in this productPrint All RSS Feed RSS Bookmark and Share Search this Product

Demography Working Paper 99/2

Estimating Interstate Migration
1996 - 2001



Bill Allen
Demography Section
Australian Bureau of Statistics
PO Box 10
BELCONNEN ACT 2616
bill.allen@abs.gov.au

June 1999



TABLE OF CONTENTS

Table of Contents
  1. Introduction
  2. The Method
  3. The Format of the Review
Appendix 1. Proposal for a Revised Method of Estimating Interstate Migration
Appendix 2. Summary of Comments from Round of Consultation
Appendix 3. Estimating Interstate Migration: Proposed Adjustment for Defence Personnel

1. INTRODUCTION

The ABS is required, by legislation, to provide quarterly estimates of the Australian population. As such, the estimated resident population (ERP) by age and sex is published each quarter for each of the States and Territories of Australia in Australian Demographic Statistics (ABS Cat. 3101.0).

State level ERP is calculated for the Census date (6 August) using Census counts by place of usual residence, and then adjusting for both net under-enumeration of the Census and residents temporarily overseas on census night (RTOs). The 6 August ERP is then backdated to 30 June, removing the effects of estimated births, deaths and both overseas and interstate migration for the period 30 June to 6 August. Quarterly estimates of ERP are then compiled by updating the ERP figures for the previous quarter with estimated births, deaths, overseas migration and interstate migration data for the reference quarter.

Births and deaths are calculated using birth and death registrations, while overseas migration is calculated using data collected from passenger cards, visa and passports by those travelling into and out of Australia. Data on interstate migration, however, are not directly available, but instead need to be estimated using administrative by-product data. As such, interstate migration remains the greatest individual source of potential error to State and Territory population estimates.

Various data sources for estimating interstate migration quarterly have been investigated in the past and were documented in the Demography Working Paper 96/1 "Evaluation of Administrative Data Sources for Use in Quarterly Estimation of Interstate Migration Between 1996 and 2001". This report, discussed at the 1996 Commonwealth-State Population workshop, found that despite its shortcomings Medicare data was the best administrative data source available for preparing post-censal interstate migration estimates.

With new data on interstate migration available from the 1996 Census, it was desirable to update the current models for interstate migration. As part of this process, the ABS decided to review the way in which Medicare data was used to estimate quarterly interstate migration. More specifically, the review focused on possible shortcomings of previous methods used and attempted to address these shortcomings where possible.


2. THE METHOD

Interstate migration is estimated using Medicare data on changes of address. Since 6 August 1996, interstate migration has been estimated using Medicare data on persons of all ages. For each State, sex and single year of age, the number of interstate arrivals and departures are estimated by multiplying the corresponding number of interstate arrivals and departures identified through Medicare by an 'expansion factor' to reflect that the Medicare data may not capture all interstate movements. This process can be summarised according to the following formulae:


Tt,sa
    = Tt,sa,arrivals - Tt,sa,departures
Tt,sa,i
    = Mt+1,sa,i x (Csa,i / Msa,i )
      , for males aged between 16 and 29 years inclusive
      and females aged between 18 and 24 years inclusive
    = Mt+1,sa,i
      ,otherwise

where

Tt,sa
          = Estimated net interstate movers for quarter t for state s and age-sex a.
Tt,sa,i
          = Estimated net interstate movers for quarter t for state s and age-sex a and arrival/departure status i.
Mt+1,sa,i
          = Medicare movers in quarter t+1 for state s and age-sex a and arrival/departure status i.
Csa,i
          = Census movers for state s and age-sex a and arrival/departure status i, from 6 August 1995 to 6 August 1996, adjusted for net undercount and both multiple and return movers.
Msa,i
          = Lagged Medicare movers for state s and age-sex a and arrival/departure status i from 6 August 1995 to 6 August 1996.

The component (Csa,i / Msa,i ) is often referred to as the 'expansion factor'. As noted in the above formulae, expansion factors are equal to one for all ages except for males aged between 16 and 29 inclusive and females aged between 18 and 24 inclusive.

This method is detailed further in the paper "Proposal for a Revised Method of Estimating Interstate Migration" (included as Appendix 1 in this Working Paper)

2.1 Lagging Medicare data

The model assumes an average lag of 3 months between moving address and registering the move with the Health Insurance Commission. While it is not possible to verify the assumption of a 3 month lag with the information available, a lag of 3 months would have produced a lower intercensal error for the 1991 to 1996 intercensal period than no lag.

As ABS receives Medicare data quarterly from the Health Insurance Commission within one quarter after the reference quarter, while publishing preliminary estimates two quarters after the reference quarter, it is possible to incorporate a lag of one quarter (i.e. 3 months) into the reference period. However, it would not be possible to incorporate any lags greater than one quarter given the current publication timetable and the quarterly provision of Medicare data from the Health Insurance Commission.

2.2 Smoothing

By calculating the expansion factors individually for each single year of age by sex, separately for arrivals and departures for each State and Territory, the expansion factors can be relatively volatile. Furthermore, it is reasonable to expect that consecutive ages would have similar expansion factors. As such, several adjustments are made to the expansion factors to reduce this volatility and increase the similarity in expansion factors for consecutive ages.

1. All the separate components used to calculate the expansion factors (i.e. Census movers data, Medicare movers data, Medicare multiple movers data) are smoothed across single years of age for both male and female arrivals and departures for each State and Territory using a 3 term moving average.

2. The expansion factors are smoothed using a 3 term moving average.

3. All expansion factors which are calculated as being less than one (i.e. fewer Census movers than Medicare movers) are set to one. Expansion factors less than one represent Medicare coverage of greater than 100% with movers registered through Medicare outnumbering adjusted Census movers. As such, expansion factors less than one are considered non-intuitive, instead reflecting inconsistencies between the Census and Medicare data.

These three steps generate smoothed expansion factors for all ages. Then, one additional step is applied which assign expansion factors of one (i.e. Medicare data represents actual movers exactly) for most age groups.

4. Expansion factors for males aged less than 16 or greater than 29 inclusive are set to one (assuming a coverage of 100%). Expansion factors for females aged less than 18 or greater than 24 are set to one (assuming a coverage of 100%).

2.3 Adjustment for Defence Forces

While Medicare theoretically covers all Australian usual residents as well as those non-Australian residents granted temporary registration, there are a range of Australian usual residents who do not access the Medicare system, primarily due to access to alternative health services. One such group is the military. As such, additional adjustments to compensate for defence movements not covered by Medicare are made to the estimates of interstate migration derived using the above model. These adjustments are estimated using counts by age, sex and State from the Department of Defence by attributing 70% of any change in quarterly defence numbers to interstate migration not otherwise covered by the model.

As this adjustment was introduced from 31 December 1998, an additional adjustment was made to the populations of New South Wales (- 79 persons) and the Northern Territory (+ 79 persons) to adjust for the period from 6 August 1996 to 31 December 1998.

This adjustment in explained in greater detail in Appendix 3.

3. THE FORMAT OF THE REVIEW

The review was undertaken as a five step process.

Firstly, the ABS considered alternative methods for estimating interstate migration. On the basis of the findings from Demography Working Paper 96/1 "Evaluation of Administrative Data Sources for Use in Quarterly Estimation of Interstate Migration Between 1996 and 2001" and discussions at the 1996 Commonwealth-State Population Workshop, the review focused on those methods which only sourced Medicare data. These methods were then compared, and documented in a paper "Proposal for a Revised Method of Estimating Interstate Migration" (included as Appendix 1 in this Working Paper) and a final model was recommended.

The paper was distributed to State planning authorities for comment. The State planning authorities consulted included the Department of Urban Affairs and Planning (NSW), the Department of Infrastructure, the Department of Treasury and Finance (Victoria), the Government Statistician's Office, Department of Housing, Local Government and Planning (Qld), Ministry for Planning (WA), Planning SA (SA), Treasury, the Department of Premier and Cabinet, Tasmanian Development and Resources (Tasmania). Treasury, the Department of Education, the Department of Housing and Local Government (NT), the ACT Chief Ministers Department (ACT) and the Australian Population Association (APA).

Next, the method was revised further on the basis of comments received. The comments received and the additional adjustments are detailed in Appendix 2. A final opportunity was then provided for State planning authorities to comment on the proposed method at the 1998 Australia-New Zealand Population Workshop. However, no additional refinements were required after the workshop.

The method was then implemented in December 1998 and included revisions of interstate migration back to the September quarter in 1996.

Finally, the method used to adjust the estimated interstate migration on the basis of data from the Department of Defence was revised, with a paper being distributed to the State planning authorities, with the method being implemented from 31 December 1998. An additional adjustment was also made to interstate migration for the period 1 July 1996 to 31 December 1998 to reflect this revised method. These paper detailing this review in provided in Appendix 3.

APPENDIX 1: Proposal for a Revised Method of Estimating Interstate Migration

Demography Program
Australian Bureau of Statistics
21 August 1998

Proposal for a Revised Method of Estimating Interstate Migration

1.
2.
2.1
2.1.1
2.1.2
2.1.3
2.2
2.2.1
2.2.2
2.2.3
2.2.4
2.2.5
3.
3.1
3.1.1
3.1.2
3.1.3
3.1.4
3.1.5
3.2
3.2.1
3.2.2
3.2.3
3.2.4
3.2.5
3.2.6
3.3
3.3.1
3.3.2
3.4
3.4.1
3.4.2

1. INTRODUCTION

The purpose of this paper is to present a new proposed method for estimating quarterly interstate migration.

Firstly, the paper focuses on the models used over the last two intercensal periods and then looks at some of the perceived shortcomings of these models. Next, the proposed model is described, followed by comments on how the proposed model seeks to address the perceived shortcomings of past models. This then leads on to further comments relating to the development of the proposed model.

The paper then describes the ways in which the proposed model was evaluated against other possible models. Finally, the paper comments on potential problems with the proposed model over the current intercensal period.

2. BACKGROUND

The ABS is required, by legislation, to provide quarterly estimates of the Australian population. As such, the estimated resident population (ERP) by age and sex is published each quarter for each of the States and Territories of Australia in Australian Demographic Statistics (ABS Cat. 3101.0).

State level ERP is calculated for the Census date (6 August) using Census data by place of usual residence, and then adjusting for both net under-enumeration of the Census using the Post-Enumeration Survey and residents temporarily overseas on census night (RTOs) using Incoming Passenger Cards. The 6 August ERP is then backdated to 30 June, removing the effects of estimated births, deaths and both overseas and interstate migration for the period 30 June to 6 August. Quarterly estimates of ERP are then compiled by updating the ERP figures for the previous quarter with estimated births, deaths, overseas migration and interstate migration data for the reference quarter.

Births and deaths are calculated using birth and death registrations, while overseas migration is calculated using data collected from passenger cards by those travelling into and out of Australia. Data on interstate migration, however, are not directly available, but instead need to be estimated using administrative by-product data. As such, interstate migration remains the greatest individual source of potential error to State and Territory population estimates.

Various data sources for estimating interstate migration quarterly have been investigated in the past and were documented in the Demography Working Paper 96/1 "Evaluation of Administrative Data Sources for Use in Quarterly Estimation of Interstate Migration Between 1996 and 2001". This report, discussed at the 1996 Commonwealth-State Population workshop, found that despite its shortcomings Medicare data was the best administrative data source available for preparing postcensal interstate migration estimates.

With new data on interstate migration available from the 1996 Census, it is desirable to update the current models for interstate migration. As part of this process, the ABS have decided to review the way in which Medicare data is used to estimate quarterly interstate migration. More specifically, this review has focused on possible shortcomings of previous methods used and attempting to address these shortcomings where possible.

2.1 Methods previously used to estimate interstate migration

The most recent methods used to estimate interstate migration have focused on using Medicare data on interstate movement for persons aged from 1 to 14 inclusive.

2.1.1 1986 to 1991

Medicare interstate movement data for ages 1 to 14 were used as a base to estimate interstate movement for all ages. A ratio expansion method was used by comparing Medicare interstate movers data for 1985-86 with the corresponding 1986 Census data, as described below. An expansion factor was derived for each interstate flow (e.g. NSW to NT, SA to WA, etc.), in order to allow for variation in the age profiles of the different flows.

Tts = Cs / Ms(1-14) x Mts(1-14) x Ageing Factor

where:

Tts
          = the estimated interstate movers aged 1 and over for quarter t and State s
Cs
          = the interstate movers aged 1 and over for State s from the 1986 Census
Ms(1-14)
          = the Medicare interstate movers aged 1 to 14 for State s from 1985 to 1986
Mts(1-14)
          = the Medicare interstate movers aged 1 to 14 for quarter t and State s

The age distribution for interstate movers from the 1986 Census was then applied to the estimate of total movers aged one and over (Tts), with an adjustment made to each age to allow for net under-enumeration in the Census. An adjustment was also applied to allow for the ageing of the population, with impacts on the proportion of the population which is aged between 1 and 14 inclusive.

2.1.2 1991 to 1996

The 1986 to 1991 model was revised in two ways. Firstly, 5 yearly interstate migration was used instead of 1 yearly migration. Secondly, revised interstate migration based on the 1991 Census for the 1986 to 1991 intercensal period was used instead of Census movers according to the address on Census night and the address 5 years previous.

Tts = TMs / Ms(1-14) x Mts(1-14) x Ageing Factor

where:

Tts
          = the estimated interstate movers aged 1 and over for quarter t and State s
TMs
          = the estimated interstate movers aged 1 and over for State s for 1986 to 1991
          revised on the basis of the 1991 Census
Ms(1-14)
          = the Medicare interstate movers aged 1 to 14 for State s from 1986 to 1991
Mts(1-14)
          = the Medicare interstate movers aged 1 to 14 for quarter t and State s

2.1.3 Post 30 June 1996, the current model

The current estimates of interstate migration for the 7 quarters since 30 June 1996 were produced using the method employed in the 1991 to 1996 intercensal period, but recalibrated using estimated interstate movers for 1991 to 1996, revised on the basis of the 1996 Census and corresponding Medicare data.

2.2 Possible shortcomings of past models

2.2.1 Multiple and Return Movers

It is important to recognise that the Census only measures migration by comparing the address on Census night with the address 1 year ago and 5 years ago. As such, each of these comparisons only captures one movement, when people may have moved more than once during the 1 year and 5 year periods. Medicare change of registration data on interstate movement are obtained from the Health Insurance Commission on a quarterly basis, which matches our need to update the estimated resident population quarterly.

For example, Census may identify a move from NSW to WA between 6 August 1990 and 6 August 1991, while the Medicare data might identify moves from NSW to SA and from SA to WA in successive quarters for the same person.

Furthermore, multiple movers who return to their original State or Territory may not be identified at all as an interstate move in the Census.

For example, a person identified using Medicare data as moving from NSW to WA and then back to NSW would be identified in the Census as either an intrastate move in NSW (if they moved to a different address in NSW) or no move at all (if they moved back to their original address) within the reference time period.

As such, there is a degree of inconsistency between Medicare and Census data.

2.2.2 Medicare data used in estimating interstate migration

Past methods have focused on using Medicare data only from persons aged between 1 and 14 inclusive, primarily as this section of the population has been considered to have the best coverage by Medicare. The coverage of Medicare movements for other age groups was not considered sufficiently reliable to estimate interstate migration.

However, there is a corresponding concern regarding the assumption that the relationship between the mover population aged 1 to 14 and the total mover population is both constant over time and sufficiently stable to estimate interstate migration accurately. Similarly, there is a general expectation that interstate migration might be better estimated using all the Medicare data instead of only using Medicare data for movers aged 1 to 14 inclusive.

2.2.3 Lag between move and registering with Medicare

It would be reasonable to expect that there would be a delay from the time a person moves to the time that they register their new address with the Health Insurance Commission, particularly for those persons who rarely use the Medicare system. However, estimating this lag is extremely difficult as is applying any knowledge of a lag to estimating interstate migration. For example, a lag of 6 months would cause problems because the ERP is produced 6 months after the reference period so that there would be insufficient time to collect the data and incorporate it in the quarterly ERP.

2.2.4 Coverage of Medicare

Medicare theoretically covers all Australian usual residents as well as those non-Australian residents granted temporary registration. However, as noted in previous investigations (Demography Working Paper 96/1), there are a range of Australian usual residents who do not access the Medicare system, primarily due to access to alternative health services. Such people include some Indigenous persons, defence force personnel, prisoners and persons eligible for Department of Veterans' Affairs Health Services. Furthermore, there are also those individuals who simply do not change their address with Medicare when they move, even though they continue to access the Medicare system.

As such, Medicare data on interstate movers will not cover all Australian usual residents in reality and there will be a degree of undercoverage. Previous models have adjusted for this undercoverage through applying a factor which scales data on Medicare movers up to represent the total population. In addition, manual adjustments are made to the quarterly estimates of interstate migration based on information provided by the defence forces.

2.2.5 Changes to the model over time

The models estimate interstate migration based on the relationship between the Census and Medicare over a set period of time. As such, any changes to the relationship between movers and moves as recorded in Medicare after this period will not be identified and/or incorporated into the model. This is a general problem in that the models can only be assessed and recalibrated after each Census.

In addition to assuming that these relationships are stable over time, the age-sex profiles of estimated movers are also determined using the age-sex profiles from the Census migration questions (adjusted for under-enumeration). Therefore, the age-sex distribution of movers has been assumed to be constant over time.

3. The proposed model

The proposed model differs from past models in the fundamental way that movers of each age as identified in Medicare are used to directly estimate interstate migration for that same age. As such, this technique also produces an age profile on interstate movers. Furthermore, no adjustment is made to the Medicare numbers except in the cases of males aged between 16 and 29 years inclusive and females aged between 18 and 24 years inclusive.

Tt,sa
    = Tt,sa,arrivals - Tt,sa,departures
Tt,sa,i
    = Mt+1,sa,i x (Csa,i / Msa,i )
      , for males aged between 16 and 29 years inclusive
      and females aged between 18 and 24 years inclusive
    = Mt+1,sa,i
      ,otherwise

where

Tt,sa
          = Estimated net interstate movers for quarter t for state s and age-sex a
Tt,sa,
          = Estimated net interstate movers for quarter t for state s and age-sex a and arrival/departure status i.
Mt+1,sa,i
          = Medicare movers in quarter t+1 for state s and age-sex a and arrival/departure status i.
Csa,i
          = Census movers for state s and age-sex a and arrival/departure status i, from 6 August 1990 to 6 August 1991, adjusted for net undercount and both multiple and return movers.
Msa,i
          = Lagged Medicare movers for state s and age-sex a and arrival/departure status i from 6 August 1990 to 6 August 1991.

The component (Csa,i / Msa,i ) is often referred to as the 'expansion factor'.

The investigation focused on developing a model which would address most of the perceived shortcomings of the previous models. Furthermore, it was essential that the proposed model provide the best possible estimates of interstate migration. As such, the models were evaluated by comparing estimates of the resulting intercensal discrepancy for the 1991 to 1996 intercensal period had they been adopted.

As shown in the tables below, the proposed model performs quite favourably over the 1991 to 1996 intercensal period when compared to both the model that was actually used and unadjusted Medicare data.

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period, comparison of models


NSW
no.
VIC
no.
QLD
no.
SA
no.
WA
no.
TAS
no.
NT
no.
ACT
no.
Sum of absolutes
    Raw Medicare1
-21,882
-5,708
-11,310
5,077
5,698
2,975
-260
-1,722
54,632
    Model used in 1991-1996
-15,770
-20,929
15,862
4,525
-3,339
-1,123
-5,2701
1,165
67,983
    Proposed model
-11,250
-23,630
1,835
-691
2,595
276
-534
3,533
44,344
1. 1,045 persons can be attributed to a revision to the NT population, not associated with interstate migration.

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period, comparison of models

NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
    Raw Medicare1
-0.35
-0.13
-0.34
0.34
0.32
0.63
-0.14
-0.56
2.81
    Model used in 1991-1996
-0.25
-0.46
0.48
0.31
-0.19
-0.24
-2.902
0.38
5.21
    Proposed model
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44
1 (Non-lagged) Medicare data is used to exactly represent interstate migration. This is equivalent to setting all expansion factors (Csa,i / Msa,i ) to one.
2 1,045 persons (0.6%) can be attributed to a revision to the NT population, not associated with interstate migration.

Intercensal discrepancy and the NT

Approximately 600 persons were added to the NT population estimates from NSW over the 1991 to 1996 intercensal period, based on defence force information, and are included in the estimate of intercensal discrepancy using the current (1991-1996) model. This defence adjustment resulted in approximately to an adjustment of +0.35% to the estimate of intercensal discrepancy for the NT (-3.25% + 0.35% = -2.90%).

However, no such adjustments have been included in estimating intercensal discrepancy for other models. Therefore, if these other models resulted in similar revisions to interstate migration, intercensal discrepancy could be interpreted as being 0.35% higher. For example, the proposed model would have resulted in an intercensal discrepancy of 0.06% (-0.29% + 0.35%).

3.1 Addressing the shortcomings

3.1.1 Multiple and Return Movers

Multiple and return movers were both included in the determination of the expansion factors. As the quarterly Medicare data would include individuals who had moved before (return movers and multiple movers), it is important that the expansion factor attempts to cater for all these groups. As such, the Census figures were increased to allow for multiple and return movers:

Adjusted Census = Census x (1+ proportion of movers in Medicare which were multiple movers)

The proportion of multiple movers in Medicare was estimated by matching individual changes of registration by postcode, sex and date of birth, using the September 1995 to September 1996 quarterly Medicare change of registration files. Similarly, the period September 1990 to September 1991 was used to estimate interstate migration for the 1991 to 1996 intercensal period.

The impact on adjusting for multiple movers on the accuracy of the model is shown in the table below:

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period,
with and without multiple and return movers


NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
With
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44
Without
-0.17
-0.50
0.01
-0.02
0.10
0.14
-0.54
1.26
2.74

3.1.2 Medicare data used in estimating interstate migration

The suggested model uses Medicare data for movers of all ages, thereby using more information to predict interstate migration. In addition, each cohort of Medicare movers is used to estimate the actual movers for that same cohort, so there is no reliance on the relationship between movers in different age groups.

However, an additional degree of volatility has been introduced by calculating the expansion factors individually for each single year of age by sex, separately for arrivals and departures for each State and Territory. Furthermore, it would be reasonable to expect that consecutive ages would have similar expansion factors. Several adjustments are made to the expansion factors to reduce this volatility and increase the similarity in expansion factors for consecutive ages.

1. All the separate components used to calculate the expansion factors (i.e. Census movers data, Medicare movers data, Medicare multiple movers data) are smoothed across single years of age for both male and female arrivals and departures for each State and Territory using a 3 term moving average.

2. All expansion factors which are greater than two are capped and set to two. An expansion factor of greater than two represents less than 50% coverage of Medicare and is not considered likely to represent a long-term average over the next inter-censal period. It is also important that this is carried out before the next step so that these 'outlier' expansion factors do not unduly impact on the expansion factors for consecutive ages.

3. The 'capped' expansion factors are smoothed using a 3 term moving average.

4. All expansion factors which are calculated as being less than one (i.e. less Census movers than Medicare movers) are set to one. Expansion factors less than one represent Medicare coverage of greater than 100% with movers registered through Medicare outnumbering adjusted Census movers. As such, expansion factors less than one are considered non-intuitive, instead reflecting inconsistencies between the Census and Medicare numbers.

These four steps generate smoothed expansion factors for all ages. Then, two additional steps are applied which assign expansion factors of one (i.e. Medicare data represents actual movers exactly) for most age groups.

5. Expansion factors for males aged less than 16 or between 29 and 54 inclusive are set to one (assuming a coverage of 100%). Expansion factors for females aged less than 18 or between 24 and 54 inclusive are set to one (assuming a coverage of 100%).

6. Expansion factors for ages greater than 54 are set to one (assuming a coverage of 100%). (In practice, Steps 5 and 6 are carried out simultaneously).

Steps 4 through to 6 have been represented pictorially below, showing how the expansion factors (XF) vary by age.


The fifth and sixth steps were introduced for two main reasons. Firstly, the estimated intercensal discrepancy for 1991-96 is significantly lower if Steps 5 and 6 are included (as shown in the table below). Secondly, the estimated interstate migration since the 1996 Census appeared more reasonable when both Steps 5 and 6 were included (also shown below).

The expansion factors between 16 and 29 for males and 18 and 24 for females were not set to one because it was expected that the data on Medicare movers would significantly under-represent these ages, based on the observed relationship between (adjusted) Census movers and Medicare movers. This was further supported by the fact that the proposed model produced more desirable estimates of intercensal discrepancy.

It should be noted that Step 5 has minimal impact on the expansion factors set to one for ages less than 54 where the observed relationship between Medicare and Census data suggest close to 100% coverage.

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period,
with and without setting expansion factors outside the pre-set range to one


NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
steps 1-4
-0.04
-0.92
0.35
-0.13
-0.36
-0.24
-2.15
5.71
9.90
steps 1-51
-0.18
-0.67
0.25
-0.22
0.17
-0.16
-0.13
2.01
3.80
steps 1-6
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44
Raw Medicare2
-0.35
-0.13
-0.34
0.34
0.32
0.63
-0.14
-0.56
2.81
1 Expansion factors for ages greater than 54 smoothed using a 7 term moving average.
2 (Non-lagged) Medicare data is used to exactly represent interstate migration. This is equivalent to setting all expansion factors (Csa,i / Msa,i) to one.


Estimated Net Interstate Migration from June 30, 1996 to March 31, 1998

NSW
no.
VIC
no.
QLD
no.
SA
no.
WA
no.
TAS
no.
NT
no.
ACT
no.
Preliminary estimates
-22,883
-7,150
41,519
-8,335
6,610
-7,684
2,623
-4,700
steps 1-4
-21,333
-1,798
30,865
-5,968
9,860
-6,353
250
-5,523
steps 1-51
-21,337
-2,542
31,505
-6,054
9,937
-6,387
631
-5,753
steps 1-6
-21,439
-3,899
33,988
-6,784
10,002
-6,580
714
-6,002
Non-lagged Medicare
-23,452
-9,595
41,326
-7,313
9,371
-6,031
1,162
-5,468
Lagged Medicare
-22,702
-6,719
37,758
-6,421
9,605
-6,232
356
-5,542
1 Expansion factors for ages greater than 54 smoothed using a 7 term moving average.

Thus, while the observed relationship between Census and Medicare data might suggest that there is a degree of undercoverage of Medicare data for the older ages, the above results suggest that total movements are better estimated through the inclusion of Step 6 (i.e. through setting expansion factors for ages greater than 54 to one). One possible interpretation of this is that the estimated expansion factors for Medicare movers of 54 years or more are less reliable and the relationship is better described using expansion factors of one.

3.1.3 Lag between move and registering with Medicare

The recommended model assumes an average lag of 3 months between moving address and registering the move with the Health Insurance Commission. While it is not possible to verify the assumption of a 3 month lag with the information available, this model did perform better than the corresponding model which assumed no lag.

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period,
assuming different registration lags

Assumed lag
NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
0 months
-0.18
-0.54
0.06
0.07
0.04
0.44
-0.94
1.25
3.52
3 months
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44

As ABS receives Medicare data quarterly from the Health Insurance Commission within one quarter after the reference quarter, while publishing preliminary estimates two quarters after the reference quarter, it is possible to incorporate a lag of one quarter (i.e. 3 months) into the reference period. However, it would not be possible to incorporate any lags greater than one quarter given the current publication timetable.

3.1.4 Coverage of Medicare

As with previous models, undercoverage of movers by Medicare is incorporated into the model through the use of expansion factors which scale up Medicare data to reflect the higher levels of Census movers (adjusted for net undercount and multiple movers). With the proposed model, an expansion factor greater than one simply reflects less than 100% coverage by Medicare.

It is tempting to want to adjust for the undercoverage directly where possible. In particular, one option could be to separately estimate the interstate moves of military personnel directly using data obtained from Defence. However, Census figures also include military personnel so it is not possible to generate reliable expansion factors for non-military and make separate adjustments for the military. As such, changes to the military population for each State and Territory will continue to be monitored as part of the process of the estimating quarterly interstate migration. In addition, the ABS plans to have further talks with Defence to investigate whether better information on interstate moves by military personnel is available.

3.1.5 Changes to the model over time

Unfortunately the only times it is possible to recalibrate the interstate migration model are just after each Census. While it is acknowledged that the models use an assumed relationship between the Census and Medicare over a fixed period of time (6 August 1995 to 6 August 1996), there are no additional data available to recalibrate the models should these relationships differ over time.

However, there is one significant difference between the proposed model and previous models. In the past, Medicare has been used to estimate total movers (by sex) and then an age distribution according to movers from the most recent Census was applied to get total movers by age (and sex). The proposed model, however, is able to directly estimate the number of interstate movers for each single year of age, so the age distribution of movers will be able to change over the intercensal period.

3.2 Additional issues considered in developing the proposed interstate migration model

3.2.1 Using the Census 1 year question rather than the 5 year question

The 1 year ago question was used in developing the expansion factors, primarily because multiple and return movers become a much greater issue over longer time periods. Multiple and return movers over the five year period would need to be estimated from Medicare, resulting in a very large adjustment to Census figures. As such, it was felt that such an estimate would impact too greatly on the final expansion factors, thus introducing too much potential error into the model.

Use of the 1 year question does introduce potential problems by attempting to estimate the expansion factors over a shorter (and thus potentially more volatile and less representative) time period. The smoothing algorithms used in generating the expansion factors will help address these problems.

3.2.2 Allocation of 'not stated' address 1 year ago

As mentioned previously, interstate migration is estimated through the Census by comparing the address on Census night and the address 1 year previous to Census night. This leads to problems where no response is given to the question on the address 1 year ago (noting that place of enumeration is imputed for non-response to address on Census night).

For the purposes of this investigation, persons who did not state their address 1 year ago were allocated as interstate movers according to the observed proportions of interstate movers for those respondents who did indicate their address 1 year ago. It is possible to suggest theories that the 'not stated' respondents should be allocated otherwise (e.g. non-movers are more likely not to respond as they consider the question irrelevant or movers are more likely not to respond because they cannot remember exactly when they moved). However, there is no direct information suggesting an allocation method preferable to the proportional allocation used. Allocating all 'not stated ' respondents as movers results in poor intercensal discrepancies (as shown in the table below), whereas allocating them as non-movers produces similar results to the proposed allocation, except for the NT which benefits significantly from the proportional allocation.

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period,
using different allocation methods for address 1 year ago not stated

    Allocation of not stated
NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
    All non-mover
-0.19
-0.49
0.02
0.00
0.14
0.16
-0.59
1.10
2.69
    Proportion as per stated
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44
    All movers1
-0.57
-0.64
0.81
-0.16
0.77
-0.52
1.09
-0.36
4.92
1 Nearly all expansion factors for males aged 16 to 29 and females 18 to 24 end up being capped at 2.

3.2.3 Adjustment of Census counts for net undercount

Census figures were also adjusted for net undercount, as estimated using the 1996 Post-enumeration Survey. As such, Census figures were fully adjusted as follows:

Adjusted Census = Census x (1+ proportion of movers in Medicare which were multiple movers) x (1 + net undercount)

3.2.4 Adjustment for 6 August Census date

While the Census uses addresses on 6 August 1995 and 6 August 1996 to estimate interstate migration, Medicare data reflects interstate migration occurring at any time in a given quarter. As such, data from the relevant September quarters were proportionally allocated to the Census reference period according to the proportion of days in the quarter falling into the reference period.

3.2.5 Adjustment for ageing due to 3 month Medicare registration lag

In creating the expansion factors, it is important to consider lags for both Census and Medicare:
- on average, there will be a 6 month lag from the time of moving to the time it is recorded in the Census 1 year movement question
- on average, there is assumed to be a 3 month lag from the time of moving to the time it is registered in Medicare

As such, it is necessary to reverse age Census movers by 3 months before calculating the expansion factors, as determined at the time of registration.

Then, the expansion factors are applied to quarterly Medicare registrations to provide an age profile of movers at the time of registration. Therefore, it is necessary to reverse age the Medicare movers from the quarterly Medicare file to allow for the 3 month lag between the move and the registration. This further adjustment provides the age profile of estimated movers at the time of move.

3.2.6 Other Territories

Some adjustments needed to be made for Jervis Bay Territory in estimating intercensal discrepancy for the ACT, primarily as Jervis Bay was considered part of the ACT until September 1993. No attempt is made to estimate interstate migration for either the Cocos (Keeling) Islands or the Christmas Islands, due to the small numbers involved.

3.3 Assessing Model Quality

Models were assessed using two main criteria:

1. On the basis of the resulting intercensal discrepancy for 30 June 1996, assuming that the model had been applied over the 1991 to 1996 intercensal period. Intercensal discrepancy for the 1991 to 1996 intercensal period is the difference between the 1996 Census based ERP and the 1991 Census based ERP updated with components of population change up to 30 June 1996. For the purposes of this investigation, intercensal discrepancy has been calculated before any intercensal revisions are made to interstate migration.

2. The model made intuitive sense.

The second criteria was included to avoid problems over choosing non-intuitive models which coincidentally produced lower intercensal discrepancies. Similarly, it was important that any relationships assumed by the model finally proposed would be considered sustainable over the next intercensal period to 6 August 2001.

In addition, it was desirable, but not essential, that all States and Territories shared the same model. This preference stemmed from a concern that models developed for individual states might be effective in reproducing interstate migration for the 1991 to 1996 intercensal period, without adequately describing the underlying relationships between Medicare data and total movers. As a result, such models would not translate well in estimating interstate migration post 30 June 1996.

A large range of models other than those listed previously in this paper (in the section "Addressing the shortcomings") were also tested, primarily relating to different combinations of adjustments and different ways of smoothing the expansion factors. Rather than provide details on all these models, this paper has only looked to present the main models investigated .

In addition to these two criteria, models which showed considerable potential were assessed using a number of different criteria which have been detailed below.

3.3.1 Intercensal discrepancy from the 1986 to 1991 intercensal period

It is important that the final model proposed is applicable to intercensal periods other than 1991 to 1996, as the method will be eventually applied to the 1996 to 2001 intercensal period. As such, the model was also assessed in terms of the resulting intercensal discrepancy if the model had been applied over the 1986 to 1991 intercensal period.

Estimated Intercensal Discrepancy for the 1986-1991 and the 1991-1996 Intercensal Periods

NSW
%
VIC
%
QLD
%
SA
%
WA1
%
TAS
%
NT
%
ACT
%
Sum of absolutes
1986-1991Proposed
-0.07
0.28
-0.03
0.50
1.15
-0.35
2.21
-0.58
5.16
Current
0.04
0.16
0.34
0.72
1.83
-1.35
-4.05
1.45
9.94
1991-1996Proposed
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44
Current
-0.25
-0.46
0.48
0.31
-0.19
-0.24
-2.902
0.38
5.21
1 The high intercensal discrepancies for WA are strongly influenced by an over-estimation of approximately 9,000 persons in 1986 form the Post Enumeration Survey.
2 1,045 persons (0.6%) can be attributed to a revision to the NT population, not associated with interstate migration.

3.3.2 Estimated revisions to interstate migration after 30 June 1996

Finally, it is essential that the proposed model makes intuitive sense when estimating interstate migration post 30 June 1996. As such, interstate migration was estimated using the proposed model and compared with the current estimates. To further aid in the evaluation, both lagged Medicare data and non-lagged Medicare were also referenced.

Estimated interstate migration from 30 June 1996 to 31 March 1998

Net
Proposed
Current
Lagged Medicare data1
Non-lagged Medicare data
Proposed revision to Current
NSW
-21437
-22883
-22702
-23452
1446
Vic
-3899
-7150
-6719
-9595
3251
QLD
33988
41519
37755
41326
-7531
SA
-6784
-8335
-6421
-7313
1551
WA
10002
6610
9505
9371
3392
Tas
-6580
-7684
-6232
-6031
1104
NT2
714
2623
356
1162
-1909
ACT
-6002
-4700
-5542
-5468
-1302
Arrivals
NSW
162058
160906
151748
152394
1152
Vic
117282
111797
109718
108921
5485
QLD
171194
188022
165375
168981
-16828
SA
49623
49010
48201
48068
613
WA
60882
59248
56700
56592
1634
Tas
19159
17845
18217
18389
1314
NT2
31047
36515
27329
28110
-5468
ACT
30523
31482
28165
28664
-959
Departures
NSW
183495
183789
174450
175846
-294
Vic
121181
118947
116437
118516
2234
QLD
137207
146503
127620
127655
-9297
SA
56406
57345
54622
55381
-938
WA
50879
52638
47195
47221
-1758
Tas
25738
25529
24449
24420
210
NT
30333
33892
26973
26948
-3559
ACT
36525
36182
33707
34132
343

1 The proposed model for estimating interstate migration assumes an average lag of 3 months between moving and registering with Medicare.
2 Preliminary investigations suggest up to 230 persons might be added to the NT from other States or Territories, through manual adjustments due to interstate movements involving defence force personnel.

In comparing the proposed and current estimates of post 30 June 1996 interstate migration, it is important to consider three primary issues:

1. The methodology used to generate the current estimates

2. Seasonality in the interstate migration.

The methodology used to generate the current estimates

The current methodology uses the same methodology used for the 1991 to 1996 intercensal period, as described in 2.1.2, which is influenced by the respective intercensal discrepancy for each State and Territory:

1. As part of the process of revising interstate migration for the 1991 and 1996 intercensal period, an attempt is made to reduce intercensal discrepancy for the respective States and Territories. For example, the NT had a high (negative) intercensal discrepancy, so interstate migration for the NT was adjusted to reflect more arrivals and less departures.

2. Then, these revised interstate migration figures were compared to Medicare data (for 1 to 14 year olds) over the same five year period to generate expansion factors (revised migration for all ages / Medicare movers aged 1 to 14). Following the NT example, these new expansion factors reflected the greater number of arrivals and reduced number of departures because of the high (negative) intercensal discrepancy over the last intercensal period.

3. For each quarter post 30 June 1996, these expansion factors were applied to Medicare data for movers aged 1 to 14 to get estimates of total movers (all ages). Again following the NT example, these expansion factors resulted in higher net migration.

As such, in revising interstate migration, it is possible that the expansion factors have over-compensated for high intercensal discrepancies and thus impacted on the current estimates. For example, the current model for NT may be generating too many arrivals and/or too few departures in attempting to compensate for the large (negative) intercensal discrepancy.

Seasonality

In comparing the proposed (which uses lagged Medicare data) and current (which uses non-lagged Medicare data) estimates of interstate migration since 30 June 1996, it is important to note that only 7 quarters of interstate migration are estimated. As such, it is possible that differences between the proposed and current estimates will be influenced by seasonality.

For example, the September quarter of Medicare data traditionally has higher net interstate migration in the NT. However, the lagged Medicare data only accesses one September quarter of interstate migration, while non-lagged Medicare data access two September quarters. As a result, the lagged Medicare data indicates fewer movers over the 7 quarters than the non-lagged data and this is reflected in the lower proposed estimate of interstate migration. Therefore, it would be reasonable to expect a reduction in the proposed revision when comparing over a full eight quarters (which will remove the impact of seasonality).

For example, if we compare the lagged and non-lagged Medicare data in the NT for just the first 4 reference quarters, the difference between the lagged and non-lagged Medicare is less than 40 people (912 persons using lagged Medicare and 879 using non-lagged Medicare).


    NT
    Reference Quarter
Lagged Medicare
Non-Lagged Medicare
    1996
    Sept
Dec
163
Sept
536
    1996
    Dec
March
185
Dec
163
    1997
    March
June
-5
March
185
    1997
    June
Sept
569
June
-5
    1997
    Sept
Dec
-143
Sept
569
    1997
    Dec
March
-143
Dec
-143
    1998
    March
June
-270
March
-143
    NT Total
356
1162


Use of lagged Medicare data

The use of lagged Medicare data in the proposed model further complicates post-censal comparisons of interstate migration. As is shown in the table above, only the current model references Medicare data for the September 1996 quarter, while only the proposed model uses June 1998 Medicare data. As such, significant proportions of the proposed revisions can be attributed to the use of different Medicare data:


State
Proposed Revision to Net
Estimated Movers
Difference between lagged Medicare movers and non-lagged Medicare movers
Lagged Medicare movers - June 1998 quarter
Non-lagged Medicare movers - Sept 1996 quarter




NSW
1,446
750
-2,836
-3,586
VIC
3,251
2,876
8
-2,868
QLD
-7,531
-3,571
3,642
7,213
SA
1,551
892
-553
-1,445
WA
3,392
134
1,628
1,494
TAS
1,104
-201
-963
-762
NT
-1,909
-806
-270
536
ACT
-1,302
-74
-656
-582



Thus, in Queensland for example, there are 3,571 fewer Medicare interstate movers just due to using a one quarter lag for Medicare data (compared with a revision of -7,531 interstate movers). Similarly, much of the proposed revision in the Victoria ERP (+3,251) can be attributed to an increase of 2,876 Medicare movers.

The exclusion of September 1996 Medicare data from the calculation of post 30 June 1996 interstate migration is intuitively preferable as it would be expected that most people who had registered a change of address with Medicare in the September 1996 quarter would have indicated their new address in the 1996 Census.

Quarter by quarter estimates of interstate migration have been provided in the appendix for each State and Territory.

3.4 Problems for the future

While the proposed model attempts to address most of the identified concerns regarding previous models, it would be naive to suggest that this model has no shortcomings. Rather, no appropriate strategies have been identified which will sufficiently address these problems.

3.4.1 Reliance on Medicare data

The interstate model relies on the ABS receiving quarterly Medicare data from the Health Insurance Commission. Furthermore, the model assumes that the relationships between Medicare movers and Census movers remains constant over the intercensal period. This is not ideal, but no mechanism other than the Census has been identified which allows for regular recalibrating of the model.

In particular, changes to Medicare arrangements (e.g. eligibility, changes to benefits, changes to procedures, etc) as they impact on the relationship between Medicare movers and Census movers, are a particular weak point of any model using Medicare data. For example, many people in their late teens and early twenties tend not to register their changes of address with Medicare, which is reflected in higher expansion factors. Should Medicare arrangements change which cause for a greater proportion of these movers to register, then the model would predict too many movers.

Similarly, the Health Insurance Commission is currently trialing a Medicare 'autoteller' in Palmerston in the Northern Territory. Should this trial prove successful and the Health Insurance Commission adopt this practice on a wide scale, there could be significant implications on the accuracy of the model.

3.4.2 Changes in the Military Personnel Arrangements

As mentioned previously, the movements of military personnel are implicitly included in the expansion factors, as these persons reflect a deficit in the coverage of Medicare data. However, once again the model is reliant on a consistency in the relationships between Census and Medicare over time.

For example, the current method assumes that for every one Medicare mover, there will be somewhere between one and two Census movers. This assumption has obvious problems in the case of large scale interstate transfers of military personnel, where it would be unlikely that there would be a corresponding interstate movement from non-military personnel. As such, it will remain essential to monitor changes in military numbers for each State and Territory through liaison with Defence.

Similarly, potential outsourcing of some military functions may have an impact on the undercoverage of Medicare with contractors being covered under Medicare. However, the current method will not adjust the expansion factors to reflect this change in coverage.


APPENDIX
Comparison of quarterly interstate migration estimates for proposed and current methods,
September quarter 1996 to March quarter 1998

Net
Year Quarter
Proposed
Current
Lagged Medicare data1
Non-lagged Medicare data
Proposed revision to Current
NSW1996Sept
-3213
-4435
-3340
-3586
1222
1996Dec
-3492
-3448
-3623
-3340
-44
1997March
-2837
-2894
-3060
-3623
57
1997June
-1805
-2833
-2060
-3060
1028
1997Sept
-3460
-1250
-3654
-2060
-2210
1997Dec
-3987
-3805
-4129
-3654
-182
1998March
-2642
-4218
-2836
-4129
1576
NSW Total
-21437
-22883
-22702
-23452
1446
Vic1996Sept
-2203
-2331
-2472
-2868
128
1996Dec
-939
-2464
-1402
-2472
1525
1997March
-346
-1181
-756
-1402
835
1997June
-1187
-331
-1574
-756
-856
1997Sept
-232
-1442
-552
-1574
1210
1997Dec
530
254
29
-552
276
1998March
477
345
8
29
132
Vic Total
-3899
-7150
-6719
-9595
3251
QLD1996Sept
6339
6952
6759
7213
-613
1996Dec
5566
7352
6137
6759
-1786
1997March
3770
5699
4407
6137
-1929
1997June
4527
3428
5118
4407
1099
1997Sept
5445
4680
5891
5118
765
1997Dec
5229
6395
5801
5891
-1166
1998March
3112
7013
3642
5801
-3901
QLD Total
33988
41519
37755
41326
-7531
SA1996Sept
-1322
-1746
-1264
-1445
424
1996Dec
-1185
-1565
-1126
-1264
380
1997March
-1020
-1368
-957
-1126
348
1997June
-832
-506
-784
-957
-326
1997Sept
-749
-654
-722
-784
-95
1997Dec
-1084
-871
-1015
-722
-213
1998March
-592
-1625
-553
-1015
1033
SA Total
-6784
-8335
-6421
-7313
1551
WA1996Sept
1819
1297
1728
1494
522
1996Dec
1370
1869
1266
1728
-499
1997March
2130
4
1969
1266
2126
1997June
909
2096
856
1969
-1187
1997Sept
980
434
982
856
546
1997Dec
1128
504
1076
982
624
1998March
1666
406
1628
1076
1260
WA Total
10002
6610
9505
9371
3392
Tas1996Sept
-775
-659
-750
-762
-116
1996Dec
-1022
-1186
-959
-750
164
1997March
-829
-1189
-779
-959
360
1997June
-1096
-679
-1031
-779
-417
1997Sept
-869
-1277
-842
-1031
408
1997Dec
-961
-1448
-908
-842
487
1998March
-1029
-1246
-963
-908
217
Tas Total
-6580
-7684
-6232
-6031
1104
NT1996Sept
237
1205
163
536
-968
1996Dec
235
436
185
163
-201
1997March
21
354
-5
185
-333
1997June
679
-195
569
-5
874
1997Sept
-58
671
-143
569
-729
1997Dec
-121
-160
-143
-143
39
1998March
-279
312
-270
-143
-591
NT Total
714
2623
356
1162
-1909
ACT1996Sept
-881
-283
-824
-582
-598
1996Dec
-533
-994
-478
-824
461
1997March
-889
575
-819
-478
-1464
1997June
-1196
-980
-1094
-819
-216
1997Sept
-1057
-1162
-960
-1094
105
1997Dec
-733
-869
-711
-960
136
1998March
-714
-987
-656
-711
273
ACT Total
-6002
-4700
-5542
-5468
-1302
Arrivals
Year Quarter
Proposed
Current
Lagged Medicare data1
Non-lagged Medicare data
Proposed revision to Current
NSW1996Sept
20584
21874
19298
21845
-1290
1996Dec
25712
19516
24081
19298
6196
1997March
23763
27056
22219
24081
-3293
1997June
23468
23138
21969
22219
330
1997Sept
20077
23230
18790
21969
-3153
1997Dec
25801
19157
24192
18790
6644
1998March
22651
26935
21199
24192
-4284
NSW Total
162058
160906
151748
152394
1152
Vic1996Sept
14438
15193
13506
14826
-755
1996Dec
18551
13088
17342
13506
5463
1997March
17063
18674
15952
17342
-1611
1997June
15916
15974
14850
15952
-59
1997Sept
15028
14526
14114
14850
502
1997Dec
19556
14137
18331
14114
5419
1998March
16730
20205
15623
18331
-3475
Vic Total
117282
111797
109718
108921
5485
QLD1996Sept
23810
27413
23004
25259
-3603
1996Dec
27597
24943
26632
23004
2654
1997March
23886
30480
23062
26632
-6594
1997June
23619
25006
22844
23062
-1387
1997Sept
22697
24886
21937
22844
-2189
1997Dec
27148
23996
26243
21937
3152
1998March
22437
31298
21653
26243
-8861
QLD Total
171194
188022
165375
168981
-16828
SA1996Sept
6050
6601
5872
6594
-551
1996Dec
8112
5774
7887
5872
2338
1997March
7195
8433
6979
7887
-1238
1997June
7098
7114
6887
6979
-16
1997Sept
6381
7070
6204
6887
-689
1997Dec
7861
6293
7645
6204
1568
1998March
6925
7725
6727
7645
-800
SA Total
49623
49010
48201
48068
613
WA1996Sept
8162
8647
7611
8092
-485
1996Dec
9461
8085
8802
7611
1376
1997March
9333
9232
8647
8802
101
1997June
8140
9184
7545
8647
-1044
1997Sept
7474
7683
6993
7545
-209
1997Dec
9538
6861
8902
6993
2677
1998March
8773
9556
8200
8902
-783
WA Total
60882
59248
56700
56592
1634
Tas1996Sept
2522
2684
2398
2720
-162
1996Dec
3184
2094
3027
2398
1090
1997March
2782
3115
2648
3027
-333
1997June
2598
2733
2471
2648
-135
1997Sept
2276
2414
2168
2471
-138
1997Dec
3117
1843
2957
2168
1274
1998March
2681
2962
2548
2957
-281
Tas Total
19159
17845
18217
18389
1314
NT1996Sept
4056
5568
3550
4173
-1512
1996Dec
5107
4722
4517
3550
385
1997March
4182
5995
3676
4517
-1813
1997June
4695
4281
4131
3676
414
1997Sept
4126
5115
3601
4131
-989
1997Dec
5027
4436
4462
3601
591
1998March
3853
6398
3392
4462
-2545
NT Total
31047
36515
27329
28110
-5468
ACT1996Sept
3759
4339
3468
4217
-580
1996Dec
5434
3519
5007
3468
1915
1997March
4406
6478
4063
5007
-2072
1997June
3952
4272
3652
4063
-320
1997Sept
3616
3785
3359
3652
-169
1997Dec
5311
3554
4898
3359
1757
1998March
4045
5535
3718
4898
-1490
ACT Total
30523
31482
28165
28664
-959
Departures
Year Quarter
Proposed
Current
Lagged Medicare data1
Non-lagged Medicare data
Proposed revision to Current
NSW1996Sept
23798
26309
22638
25431
-2511
1996Dec
29204
22964
27704
22638
6240
1997March
26601
29950
25279
27704
-3349
1997June
25273
25971
24029
25279
-698
1997Sept
23538
24480
22444
24029
-942
1997Dec
29788
22962
28321
22444
6826
1998March
25293
31153
24035
28321
-5860
NSW Total
183495
183789
174450
175846
-294
Vic1996Sept
16641
17524
15978
17694
-883
1996Dec
19490
15552
18744
15978
3938
1997March
17409
19855
16708
18744
-2446
1997June
17102
16305
16424
16708
797
1997Sept
15260
15968
14666
16424
-708
1997Dec
19026
13883
18302
14666
5143
1998March
16253
19860
15615
18302
-3607
Vic Total
121181
118947
116437
118516
2234
QLD1996Sept
17472
20461
16245
18046
-2989
1996Dec
22031
17591
20495
16245
4440
1997March
20116
24781
18655
20495
-4665
1997June
19092
21578
17726
18655
-2486
1997Sept
17252
20206
16046
17726
-2954
1997Dec
21919
17601
20442
16046
4318
1998March
19325
24285
18011
20442
-4960
QLD Total
137207
146503
127620
127655
-9297
SA1996Sept
7373
8347
7136
8039
-974
1996Dec
9297
7339
9013
7136
1958
1997March
8214
9801
7936
9013
-1587
1997June
7930
7620
7671
7936
310
1997Sept
7130
7724
6926
7671
-594
1997Dec
8945
7164
8660
6926
1781
1998March
7517
9350
7280
8660
-1833
SA Total
56406
57345
54622
55381
-938
WA1996Sept
6343
7350
5883
6598
-1007
1996Dec
8091
6216
7536
5883
1875
1997March
7203
9228
6678
7536
-2025
1997June
7231
7088
6689
6678
143
1997Sept
6494
7249
6011
6689
-755
1997Dec
8410
6357
7826
6011
2053
1998March
7107
9150
6572
7826
-2043
WA Total
50879
52638
47195
47221
-1758
Tas1996Sept
3297
3343
3148
3482
-46
1996Dec
4205
3280
3986
3148
925
1997March
3610
4304
3427
3986
-694
1997June
3693
3412
3502
3427
281
1997Sept
3145
3691
3010
3502
-546
1997Dec
4078
3291
3865
3010
787
1998March
3710
4208
3511
3865
-498
Tas Total
25738
25529
24449
24420
210
NT1996Sept
3818
4363
3387
3637
-545
1996Dec
4873
4286
4332
3387
587
1997March
4162
5641
3681
4332
-1479
1997June
4016
4476
3562
3681
-460
1997Sept
4184
4444
3744
3562
-260
1997Dec
5148
4596
4605
3744
552
1998March
4132
6086
3662
4605
-1954
NT Total
30333
33892
26973
26948
-3559
ACT1996Sept
4641
4622
4292
4799
19
1996Dec
5966
4513
5485
4292
1453
1997March
5295
5903
4882
5485
-608
1997June
5148
5252
4746
4882
-104
1997Sept
4673
4947
4319
4746
-274
1997Dec
6044
4423
5609
4319
1621
1998March
4758
6522
4374
5609
-1764
ACT Total
36525
36182
33707
34132
343
1 The proposed model for estimating interstate migration assumes an average lag of 3 months between moving and registering with Medicare. ie Dec 1996 - June 1998 Medicare data is used to estimate Sept 1996 - March 1998 migration.



APPENDIX 2: Summary of Comments from Round of Consultation

Following the round of consultation with the relevant State planning authorities, the various comments were collated. The key issues raised were:
  • the capping of expansion factors,
  • the proposed lag of 3 months in sourcing the Medicare data,
  • the impact of the backdating adjustment from 6 August 1996 to 30 June 1996, and
  • the use of Defence data, in relation to both the data sourced and the way it is used.

A2.1 The capping of expansion factors

Two comments were made regarding the capping of expansion factors, with the first comment being made by a number of planning agencies:
  • Questioning of the upper limit of 2.0 for expansion factors for ales aged from 16 to 29 inclusive and females aged 18 to 24 inclusive.
  • Questioning of the setting of expansion factors for ages greater than 55 to one

Upper limit of 2.0 for expansion factors

The capping of expansion factors was originally undertaken to avoid outliers over influencing the estimates and resulted in a reduced intercensal discrepancy in 1991-96 estimates (see table below).

Estimated Intercensal Discrepancy for 1991-96, with different capping regimes

NSW
VIC
QLD
SA
WA
TAS
NT
ACT
AUS
    Cap of 3.0
-0.13
-0.52
0.03
-0.08
0.01
0.07
-1.39
1.92
4.15
    Cap of 2.5
-0.13
-0.51
0.03
-0.08
0.03
0.07
-1.19
1.62
3.66
    Cap of 2.0 except NT
-0.17
-0.51
0.07
-0.04
0.16
0.08
-1.41
1.18
3.61
    Cap of 2.0
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44

However, while there would have been 43 expansion factors greater than 2.0 without capping for the 1991 to 1996 period (State by age by sex by arrival/departure), there would only be 10 expansion factors greater than 2.0 for the post June 1996 period. As the relationship between Medicare data and 1996 Census based estimates of interstate movements for males aged 16 to 29 and females aged 18 to 24 in 1995-96 was only minimally affected by outliers, the proposal was revised to remove the caps of 2.0.

Setting expansion factors for other ages to 1.0

The decision to set expansion factors for the ages to 1.0 (except males aged 16 to 29 and females aged 18 to 24) was made primarily on the basis that it significantly reduced intercensal discrepancy for the 1991 to 1996 intercensal period (see below), particularly as it related to both Victoria and the ACT which both had greater intercensal discrepancy under the proposed method compared to the other States and Territories.

Estimated Net Interstate Migration from June 30, 1996 to March 31, 1998

NSW
no.
VIC
no.
QLD
no.
SA
no.
WA
no.
TAS
no.
NT
no.
ACT
no.
Sum of
absolutes
    No cap of 1.0
-0.04
-0.92
0.35
-0.13
-0.36
-0.24
-2.15
5.71
9.90
    No cap of 1.0 for ages >= 55
-0.18
-0.67
0.25
-0.22
0.17
-0.16
-0.13
2.01
3.80
    Cap of 1.0
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44

It is also important to note that the setting of expansion factors to one for the other ages will not necessarily result in a reduction in net interstate migration for a State or Territory. For example, in Queensland, setting expansion factors for ages greater than or equal to 55 to one had the effect of raising all arrival expansion factors to one (except for two female ages), as they would have had expansion factors less than one. Conversely, all departure expansion factors were lowered to one (except for five ages), as they had expansion factors greater than one. Therefore, as suggested in 3.1.2 of the proposal, the removal of the cap of one for ages greater than or equal to 55 would result in a further downward revision to the post 30 June 1996 net interstate migration for Queensland by approximately 2,500 persons.

As such, it was decided to set expansion factors to one except for males aged 16 to 29 and females aged 18 to 24.

A2.2 The proposed lag of 3 months in sourcing the Medicare data

The main effect of implementing a 3 month lag is to move quarterly interstate migration back one quarter - Medicare movements identified in a given quarter are used to estimate interstate migration for the previous quarter. Two comments were received with regards to the implementation of the proposed 3 month lag:
  • NT Treasury argued against the implementation of a lag, and
  • The Victorian Department of Infrastructure questioned whether some other time period was more appropriate.

The implementation of a lag

The average number of Medicare services per capita nationally in 1996-97 was 10.7 and in the NT was 5.7. This suggests an average use of Medicare services of almost 1 a month for Australia and 1 every two months in NT. As such the argument was put forward that a 0 month lag was a closer approximation to reality than a 3 month lag. Even in the case of the NT, it was suggested that the area of NT most affected by interstate migration (Darwin) could well attract more than 1 service every two months.

While this argument sounds plausible, only 28.6% nationally (28.3% in the NT) of Medicare Services are not bulk billed and bulk bills are extremely unlikely to identify a change of address. Furthermore, cash claims and EFTPOS transactions (12.3% nationally) will not identify changes of addresses. As a result, the number of times an individual is likely to use Medicare and be able to register a change of address is significantly smaller.

Medicare Statistics 1996-97


    State
% Bulk billed
Average services per year
Average months between visits
    NSW
74.7%
11.6
4.1
    VIC
70.2%
10.8
3.7
    QLD
71.0%
10.5
3.9
    SA
66.8%
10.1
3.6
    WA
72.2%
9.5
4.5
    TAS
58.0%
9.7
2.9
    NT
71.7%
5.7
7.4
    ACT
60.2%
8.9
3.4
    AUS
71.4%
10.7
3.9
Source: Medicare Statistics, Department of Health and Family Services

However, estimating the average duration between visits in much more complicated. For example, the average number of months between visits could be expected to reduce after adjusting for the fact that all members of a family often use the same Medicare card. Conversely, it would also be reasonable to expect that all Medicare services are not received evenly throughout the year, but rather in clusters (for individuals and families), thus increasing the duration between visits. As such, it is difficult to use these data to accurately estimate the most appropriate lag.

The assumption in using lagged data is that, on average, there will be a 3 month lag between the time an individual moves interstate and when they register this change of address through the Medicare system (non-bulk billed). In addition this seems to be supported by the analysis which indicates better results are obtained using a lagged model (see below).

Estimated Intercensal Discrepancy for 1991-1996 Intercensal Period,
assuming different registration lags


Assumed lag
NSW
%
VIC
%
QLD
%
SA
%
WA
%
TAS
%
NT
%
ACT
%
Sum of absolutes
0 months
-0.18
-0.54
0.06
0.07
0.04
0.44
-0.94
1.25
3.52
3 months
-0.18
-0.52
0.05
-0.05
0.15
0.06
-0.29
1.14
2.44

The lower intercensal discrepancy in the NT from using a 3 month lag can be attributed to high net migration Medicare figures (approximately 500 persons) to the NT in the September 1996 Quarter. In using the lagged model, these Medicare data are not referenced post 30 June 1996. Instead, this quarter is to estimate interstate migration in the June 1996 quarter and plays a significant role in decreasing the estimated intercensal discrepancy in the 1991 to 1996 intercensal period. This appears more intuitive, with the 0 month lag suggesting that people would move to the NT in the September quarter, at the end of the dry season.

Thus, while Medicare address data suggest that address information is regularly updated, the data are not clear. Furthermore, it seems intuitive that an average lag of three months between moving and registering is more probable than no lag at all. As such, the decision was to proceed with using lagged Medicare data.

The choice of a 3 month lag

Given that the current ABS publication schedule and the fact that the ABS receives Medicare data quarterly from the HIC, only lags of 3 months or 0 months can be accommodated. The 3 month lag was chosen in preference to the 0 month lag as it was thought to be more intuitive and produced a lower intercensal discrepancy for the 1991-96 intercensal period (see previous table).

Despite the difficulty in implementing longer lags, lags of both 6 months and 12 months were also investigated during preliminary investigations, with intercensal discrepancy generally minimised with lags of 3 and 6 months.

A2.3 The impact of the backdating adjustment from 6 August 1996 to 30 June 1996

The estimated resident population for 30 June 1996 was originally calculated by adjusting the 6 August 1996 (Census date) estimated resident population for births, deaths, interstate migration and international migration during the intervening time. As part of this process, interstate migration for the period between 30 June 1996 and 6 August 1996 was estimated using the previous interstate migration model (as detailed in 2.1.2 in the proposal).

As such, it is necessary to ensure that revisions made to post-30 June 1996 interstate migration are consistent with the method used to backdate the 6 August 1996 estimated resident population back to 30 June 1996. This is achieved by only introducing the proposed method from 6 August 1996 instead of 30 June 1996 (see below).



A2.4 The use of Defence data, in relation to both the data sourced and the way it is used

NT Treasury expressed concern over both the method used to adjust the estimated interstate migration on the basis of data provided by the Department of Defence, and the quality of the data provided by the Department of Defence. While it was not possible to investigate these concerns in sufficient detail and still implement the proposed model by December 1999, ABS provided a commitment to investigate NT Treasury's concerns with the intention that a satisfactory outcome would be achieved before the release of the December 1998 population estimates (due June 1999).

A2.5 Summary of changes made to the proposed model

On the basis of the comments received, the following adjustments were made to the proposed model for estimating interstate migration:

1. The cap of 2.0 for expansion factors was removed.
2. An adjustment was introduced to allow for the fact that the population estimates had been backdated from 6 August 1996 to 30 June 1996 using the 1991 to 1996 model.

In addition, a commitment was made by ABS to further investigate the method used to adjust the estimated interstate migration on the basis of data obtained from the Department of Defence. The results of this investigation are detailed further in Appendix 3.

APPENDIX 3: Estimating Interstate Migration: Proposed Adjustment for Defence Personnel

Following the circulation of the ABS discussion paper Proposal for a Revised Method of Estimating Interstate Migration (August 1998) and comments received, particularly from NT Treasury, the ABS has undertaken further analysis of defence data.

In summary, the ABS proposes:
  1. For the period 30 June 1997 to 31 December 1998, to increase the NT ERP by an additional 79, and decrease the NSW ERP population by 79.
  2. For the period after 31 December 1998, to use 'new' DOD data with the assumption that 70% of interstate movements of defence personnel are not covered by the existing method.

Four sources of data have been investigated:
  1. Department of Defence (DOD) data
  2. Defence Corporate Support Centre (DCSC) data for Northern Territory (NT), previously referred to as Northern Command (NORCOM) data
  3. Medicare data from the Health Insurance Commission
  4. Census of Population and Housing data on the defence industry and defence occupations.

In addition, consultation with staff within DOD and DCSC has helped to shed light on the strengths and weaknesses of the respective defence data.

The main features of our investigation are as follows:

1. The interstate migration model assumes for every one Medicare mover there is effectively between one and three actual movers. This ratio or expansion factor recognises that the movement of a family or household may be captured only partially by Medicare enrolments. The expansion factors are complicated by the fact that military personnel may be covered by the military's own 'on-base' health provision.

2. DOD data by 'place of posting' provides an internally consistent time series back to June 1981 for all States/Territories. This data suggests that NT is the only State/Territory to have increased its number of defence personnel from 1996 to 1998 (Figure 1).


Figure 1.
        Note: Where defence data is unavailable for some quarters, the latest available figure has been used. For example, navy data for the June 1992 quarter has been used for all subsequent quarters.

3. The ABS has been advised by DOD that 'place of administration' is a better indicator of usual residence than 'place of posting'. From December 1998, the ABS will receive DOD data by 'place of administration', although data by 'place of posting' will be still be sought to facilitate the transition to the former.

4. DCSC data and DOD data (especially on the 'new' basis) are consistent with each at the aggregate level. However, there are inconsistencies between these two data sources at the broad service level that warrant further investigation (Tables 1 and 2).


Table 1.




Table 2.

The DCSC data suggests a loss of some 400 personnel from the NT compared with the DOD data which suggests a small gain (Table 3). The differences could be due to timing (eg. paydays v end of month) and conceptual differences (eg. with respect to duration of residence, treatment of personnel in transit, and treatment of personnel on short-term exercises). The DCSC data may not be any superior to that provided by the DOD. Nevertheless, the DOD has offered to investigate the differences at the unit record level if December 1998 DCSC unit record data is supplied to it.


Table 3.

5. Comparison of DOD, Medicare and population data (Census or ERP) by postcode has not helped to quantify how well Medicare data covers defence personnel, mainly because these data sources do not share a common reference date nor consistent postcodes.

6. The existing interstate migration model theoretically allows for defence personnel in the calculation of the expansion factors. However, it is appreciated that the model may not necessarily cover large-scale movements of defence personnel given the provision of non-Medicare services (i.e. 'on-base' medical treatment) for defence personnel.

'New' DOD data allows some analysis of defence personnel by marital status and their number of dependents. Based on the December 1998 DOD data, approximately 60% of defence personnel are either partnered or have dependents, and they may be listed on the same Medicare card as their partners and/or dependents who are not covered by 'on-base' medical treatment. Therefore, the Medicare data may not fully cover the remaining 40% of defence personnel. The validity of this proportion is affected by a number of factors:

a) Some defence personnel will be individually registered with Medicare. While most defence personnel are expected to opt for the military's health cover, some may remain with or choose Medicare (eg. when on leave) which will lower the proportion of 40%.

b) Families of defence personnel may not register with Medicare. Investigations support the assumption that families of defence personnel are not covered by the military's health care, so it is assumed they are as likely to register with Medicare as any other family.

c) Defence personnel may not be equally likely to transfer interstate. If defence personnel with partners and/or dependents are less likely to migrate than unpartnered personnel without dependents, then this would tend to raise the proportion. There is some evidence of this (eg. ABS pers. comm. with NORCOM).

d) When defence personnel migrate, their families may not accompany them. It is possible that a number of defence personnel will migrate without their families (so the partner and dependents remain where they are already established), which would again raise the proportion.

e) Changes in defence numbers also include enlistment and discharges. As such, not all changes in numbers of defence personnel can be attributed to interstate migration, which would tend to lower the proportion.

The ABS considers the 40% figure, or a ratio of 0.4, as the lower bound of the proportion of defence personnel who will not be incorporated in the Medicare data. At the other extreme is a ratio of 1.0, where no defence personnel are included in the Medicare data. Given the factors mentioned above, the ABS considers an average of the theoretically lower (0.4) and upper (1.0) limits gives a realistic ratio (0.7).

7. There is some State variation in the proportion of defence personnel without partners and/or dependents (Table 4). Excluding the two States with the lowest number of Defence personnel, Tasmania and South Australia, the proportion is in the range 0.36-0.43. The ABS will monitor the family characteristics of the defence personnel, but at this stage recommend using a single ratio of 0.7 across all States.


Table 4.




8. It is proposed that a ratio of 0.7 be applied to the change in each State's defence personnel, by age and sex, from December 1998 onwards using 'new' DOD data. The advantage of this method are that it allows a quarterly adjustment to be made for each and every State, although a further adjustment may be needed to ensure that net interstate migration sums to zero. The method relies on the continued receipt of timely defence force numbers from DOD.

9. Historical defence data does not give the ABS sufficient confidence in a statewide adjustment for defence force movements prior to 1999, as the 'old' DOD data is coded by 'place of posting' and is incomplete for navy (there is no navy data between June 1992 and December 1998). The proposed adjustment will only be applied to NT offset by New South Wales, based on advice from the DOD that the largest movement of army personnel to NT were from NSW in the period June 1997 to December 1998.

10. The effect of the proposed adjustment on the population of NT is summarised in Table 5. As the proposed method of adjustment would have given a higher NT population at 30 June 1997 (+65 people), this adjustment may be carried forward to the subsequent period (30 June 1997 to 31 December 1998) to give a total adjustment of +134. As +55 of this adjustment has already been incorporated in the current method, the additional adjustment of the NT population is equivalent to +79.


Table 5.


11. The ABS will continue to source both DOD and DCSC defence data to monitor the interstate movement of military personnel. The ABS has requested data on migration flows of defence personnel, which may assist in a subsequent review of the method of adjusting estimates of interstate migration of defence personnel.

In summary, the ABS proposes:
  1. For the period 30 June 1997 to 31 December 1998, to increase the NT ERP by an additional 79, and decrease the NSW ERP population by 79.
  2. For the period after 31 December 1998, to use 'new' DOD data with the assumption that 70% of interstate movements of defence personnel are not covered by the existing method.


Demography Program
ABS
May 1999

Bookmark and Share. Opens in a new window

Commonwealth of Australia 2014

Unless otherwise noted, content on this website is licensed under a Creative Commons Attribution 2.5 Australia Licence together with any terms, conditions and exclusions as set out in the website Copyright notice. For permission to do anything beyond the scope of this licence and copyright terms contact us.