Multiregional and bi-regional population projection methods: a brief overview

1

• Multiregional and bi-regional population
  projection methods a brief overview
• Prepared for
• Summer School on Projection Methods for Ethnicity
  and Immigration Status
• Leeds University, July 2009
• Tom Wilson

2
Multiregional and Bi-regional Population
Projection Methods

• What is a multiregional model?
• Formal multiregional demography ... is concerned
  with the mathematical description of the
  evolution of human populations over time and
  space Rogers, 1995 p 1
• A multiregional model is one which explicitly
  includes migration between regions

3
Multiregional and Bi-regional Population
Projection Methods

• Why use a multiregional model ?
• (rather than a simple net migration / net
  migration rate model)
• Conceptual reasons
• No such thing as a net migrant, just people
  moving from one place to another
• Net migration rates do not relate to a correct
  PAR
• Origin-destination migration rates model
  migration as a function of origin population size
• Practical reasons
• Net migration models can give negative
  populations
• They can result in rogue cohorts thus
  implausible age profiles
• If you are dealing with many regions it is
  difficult to ensure that migration within the
  system sums to zero.
• Difficult to smooth net migration rate profiles
  because they are highly variable in shape there
  are no models/theory to draw upon (much easier to
  do this for directional migration age profiles)

4
Multiregional and Bi-regional Population
Projection Methods

• Birth of multiregional demography in a nutshell
• Began in 1960s with work of Andrei Rogers
• He was interested in generalising the uniregional
  Leslie matrix and the uniregional life table to
  the multiregional case
• Many other researchers contributed to this
  emerging stream of demographic work in the 1960s
  1970s
• Phil Rees and Alan Wilson approached
  multiregional demography from a population
  accounting perspective
• Much theoretical development and empirical work
  done at IIASA in 1970s and 80s

5
Multiregional and Bi-regional Population
Projection Methods

• Population accounts (movement perspective)
• K Population M internal migration E
  emigration
• I immigration D deaths
• Ø ignored R accounting residual

6
Multiregional and Bi-regional Population
Projection Methods

• Notation
• a a1 subscript denotes the period-cohort aged
  a at time t and aged a1 at time t1

7
Multiregional and Bi-regional Population
Projection Methods

• Movement accounts multiregional model
• Accounting equation
• Projection equation
• Either solve this iteratively, or . . .

8
Multiregional and Bi-regional Population
Projection Methods

• Movement accounts multiregional model
• Matrix M
• Matrix projection equation
• See Rogers (1995) Multiregional Demography

9
Multiregional and Bi-regional Population
Projection Methods

• Population accounts (transition perspective)
• Keisj Pop existing in i at time t and surviving
  in j at time t1

10
Multiregional and Bi-regional Population
Projection Methods

• Transition accounts multiregional model
• Accounting equation
• Projection equations (term by term)
• etc.
• See Rees Wilson (1977) Spatial Population
  Analysis

11
Multiregional and Bi-regional Population
Projection Methods

• Practical challenges with the implementation of
  multiregional models
• Models can contain a large number of elements
• No. of migr. rates
• no. of regions
• x no. of regions -1
• x no. of sexes
• x no. of age groups
• x no. of time periods
• Requires large data to be purchased
• Suppression of small values in cells
• Lots of random noise in data
• Difficult to smooth age profiles with vast
  amounts of noise
• Computing difficulties
• Issue of setting assumptions

12
Multiregional and Bi-regional Population
Projection Methods

• What can be done?
• Aggregation
• of age groups, spatial units, time periods etc.
• e.g. switch from single year age groups to 5
  year age groups
• Decomposition
• Break up multiregional systems into a no. of
  relatively independent systems
• e.g. model O-D flows between areas within a
  region/state and a simple distribution for
  other migrations
• Parameterisation
• Replace age profiles with a mathematical model
  described by a small no. of parameters
• Migration pool models
• Out-migration from each region goes into a pool
  it is then distributed back to regions using
  shares of the pool

13
Multiregional and Bi-regional Population
Projection Methods

• What can be done? (2)
• Bi-regional models (a form of spatial
  aggregation)
• Model migration between a region and the rest of
  the system
• Use uniregional net migration models? NO!

14
Multiregional and Bi-regional Population
Projection Methods

• Decomposition
• Fully multiregional model used for flows within a
  broad region or state
• Pool model used for flows to distant areas
• Out-migr to distant areas out-migr rate x PAR
• Migration to area j out-migr to distant areas x
  share going to j
• e.g. migration between regions within Australian
  states modelled using a fully multiregional
  model migration between regions in different
  states modelled using shares of total
  out-migration

15
Multiregional and Bi-regional Population
Projection Methods

• Migration pool model
• Step 1
• Out-migr from each regions is product of
  out-migration rate and population at risk
• Step 2
• All out-migration placed in a common 'pool'
• Step 3
• Pool of migrants distributed to regions on basis
  of shares

16
Multiregional and Bi-regional Population
Projection Methods

• Parameterised model migration schedules
• Developed by Rogers, Castro colleagues
• Migration intensity
• childhood component
• labour force component
• retirement component
• constant component

17
Multiregional and Bi-regional Population
Projection Methods

• Bi-regional models
• Internal system (excl. rest of the world) is
  divided into just two regions. Actually series of
  bi-regional models knitted together.
• Migration between each region rest of
  system
• modelled in turn
• etc.
• Minor adjustment required to ensure that
  projected internal out-migration and in-migration
  to/from all regions is the same.

18
Multiregional and Bi-regional Population
Projection Methods

• Do bi-regional models give similar results to a
  fully multiregional model?
• Series of empirical tests with fully
  multiregional model several other models,
  including the bi-regional (BR) model

19
Multiregional and Bi-regional Population
Projection Methods

• Do bi-regional models give similar results to a
  fully multiregional model? (2)
• Similarity of projected age profiles

20
Multiregional and Bi-regional Population
Projection Methods

• Strengths of the bi-regional model
• Maintains the important conceptual strength of
  the fully multiregional model of handling place
  to place migration flows (rather than net
  migration)
• Considerably reduced data requirements (and )
• Less randomness in age profiles less
  adjustment/smoothing needed (and less staff time)
• Considerably reduced no. of assumptions that need
  to be made
• ? important in persuading govt ( others) to use
  such methods
• Empirical results not far off those produced by
  the fully multiregional model
• Less projections data to check

21
Multiregional and Bi-regional Population
Projection Methods
Population projections for NSW Model New South
Wales Demographic Simulation System
(NEWDSS) Projections calculated at 3 levels
(1) NSW as a whole Rest of Aus
(2) Regions of NSW
(3) Statistical Local Areas (SLAs)
Local Government Areas (LGAs)
Strategy regions / subregions
22
Multiregional and Bi-regional Population
Projection Methods

• Regional geography

23
Multiregional and Bi-regional Population
Projection Methods

• NEWDSS
• State Rest of Australia level
• movement accounts-based bi-regional model
• single years of age 0-120
• Regional level
• movement accounts-based tri-regional model
• (i) region (ii) rest of NSW (iii) rest of
  Australia
• single years of age 0-120
• Constrained to State level
• Statistical Local Area level
• transition accounts-based bi-regional model
• (i) Statistical Local Area (ii) rest of
  Australia
• 5 year age groups to 85
• Constrained to regional level

24
Multiregional and Bi-regional Population
Projection Methods
NEWDSS input and output Data assembled,
checked adjusted in Excel workbook ? Input
files prepared as csv files ? Projection
program (fortran 95) ? Output csv files ? Excel
workbook used to visualise outputs
25
Multiregional and Bi-regional Population
Projection Methods
Age patterns of interstate migration Bi-regional
model reduces no. of age profiles required just
in- and out-migration profiles needed (not whole
set of O-D profiles) Much easier to smooth as
graphs demonstrate NSW male out-migration
NSW male in-migration
26
Multiregional and Bi-regional Population
Projection Methods

• The changing age profile of Sydney
• Bulge in age groups in
• 20s 30s due to migration
• bulk of net migration gains
• concentrated at these ages

27
Multiregional and Bi-regional Population
Projection Methods
Projected population of the Mid-North
Coast 2006 297,000 2036 387,000
28
Multiregional and Bi-regional Population
Projection Methods
Projected population of the Central West 2006
179,000 2036 183,000
29
Multiregional and Bi-regional Population
Projection Methods

• Bi-regional Statistical Local Area model gives
  plausible age profiles
• Leichhardt council Hunter's Hill
• The previous net migration rate model produced
  several 'rogue' cohorts

30
Multiregional and Bi-regional Population
Projection Methods

• Projected pop'n change 2006-21, NSW Statistical
  Local Areas
• text

31
Multiregional and Bi-regional Population
Projection Methods

• Concluding remarks
• Multiregional bi-regional models are
  conceptually much better than net migration
  models because they incorporate place to place
  migration flows
• The bi-regional model is a reduced form of the
  fully multiregional model which retains
  directional migration whilst reducing data
  requirements and input assumptions
• Much work remains to be done on
  assumption-setting, esp. in having regionally
  consistent assumptions

32
Multiregional and Bi-regional Population
Projection Methods

• The end
• Any questions?