Dynamic microsimulation with spatial interactions

1
Dynamic microsimulationwith spatial interactions

• B.M.Wu, M.H.Birkin and P.H.Rees
• School of Geography
• University of Leeds

2
Outline
Introduction Modelling objectives Model
description Initial analysis Model
improvment Conclusion and future work
3
Introduction

• Moses
• Modelling and Simulation of e-Social Science
• Modelling objectives
• To develop a complete representation of the UK
  population at a fine spatial scale
• To produce rich, detailed and robust forecasts
  of the future population of the UK
• To investigate scenarios which relate
  demographics to service provision - emphasis on
  policy applications within the health and
  transport policy sectors

4
Some large scale MSMs

• DYNASIM (Orcutt 1986)
• CORSIM (Caldwell, 1998), DYNACAN (Morrison,
  2003) and SVERIGE (Rephann, 1999)
• APPSIM (Harding, 2007)
• EUROMOD (Sutherland, 2007)

5
Modelling Description(1)

• Dynamic representation of key demographic events
  /transactions in a geographically identified
  population
• Macrosimulation and microsimulation models (MSM)
  are alternative ways of realising the processes
  (van Imhoff and Post, 1998)
• We use a spatial MSM of the population and its
  dynamics, but the structure parallels the macro
  multi-state cohort-component (MSCC) projection
  model
• An MSM depends on good data on the important
  transitions experienced by individuals
• We experimented with an Agent Based Model(ABM)
  for a sub-population, students, where empirical
  data on migration has often proved problematic

6
Model Description (2)

• Individual-based representations, forecasts and
  scenarios
• What does this mean?
• Leeds population720,000 UK 60 million
• Each individual has about 60 individual
  variables
• 20 household variables area variables
• Various probabilities/rates eg localised single
  year of age based mortality rates for Leeds
• Distinctive behaviours from various population
  groups in different demographic processes
• Interdependency of household and individual
  variables in different demographic processes

7
Demographic processes in the MSM

• 6 modularised processes
• simple processes
• multi-stage processes
• Household formation and dissolution

8
Initial Results (1)
An example of standard age-sex representations of
Leeds population
9
Initial Results (2)

10
Improving Migration Model

• We combine two approaches
• A person-specific general model, using
  probabilities of migration derived from the BHPS
  applied to cloned individuals in households
  derived from the 2001 Census SAR
• Location specific information about migration
  intensities in small areas (2001 Census SMS),
  which are used to modify the results of the
  person-specific model
• The model has a two stage procedure
• Migrant generation procedure
• Migrant distribution procedure

11
Migrant generation procedure

• Assess migration probabilities from an analysis
  of BHPS data, 2000-2004 for
• a) households
• b) groups
• c) individuals
• Major drivers of migration identified using a
  stepwise chi-squared estimation procedure
• Households age of head, household size, housing
  type
• Individuals age, household size, marital status
• Groups merged with individuals (small numbers)
• National rates are locally adjusted by age using
  the Census Special Migration Statistics (SMS)

12
Migrant distribution procedure

• The process is explored through a number of
  simplifying assumptions (later to be relaxed)
• Net migration balance of zero between emigration
  from the city region and immigration to the city
  region
• No new housing
• No change in individual or household
  characteristics
• Only considers complete household moves
• Vacancy chain model of household migration

13
Migrant distribution procedure

• The problem can be described as follows
• Estimate migration rates by location, age,
  household size and housing type this process
  creates a stock of vacant housing
• For each migrant, by location and household type
  (age, size) find a destination location by
  location and house type
• Calibrate this process using data on known moves
  (by distance from the census SMS) and known
  assignments of household type to house type
  (BHPS)

14

Simulation Database
Update Location and Dwelling Characteristics
1
5
Migrant generation model
2
2
Aggregate To Migrant Population
Aggregate To Vacant Dwellings
Migration distribution procedure (Birkin and
Clarke 1987 Wu et al, 2008)
Spatial Interaction Model
3
Compute dwelling preference for each migrant
4
15
Migration Results
16
Characteristics of student migrants

• Students are highly mobile during their studies
  in the universities
• Mostly only move around the area close to the
  universities where
• they study, not in the suburban areas.
• More importantly, most of them will leave the
  city once they finish their study, instead of
  settling down and growing old in the area
• Due to the replenishment of the student
  population each year, the population of the wards
  in which university student stay tends to
• remain younger than that in other wards.

17
ABM

• An alternative approach that models individuals
  as agents through their interactions with each
  other and the environment that they live in.
• It is very flexible to introduce heterogeneous
  agents with distinctive behaviours through their
  built-in rules
• It is useful in modelling features in the model
  where knowledge and theory is lacking (Billari et
  al. , 2002).

18
Student Migrants experimenting with ABM

• We recognise the following groups
• First year undergraduates
• Other undergraduates
• Master students
• Doctoral students
• We apply the following rules
• Each group is allowed set years to stay in the
  area
• Students prefer to stay with their fellow
  students
• Students stay close to their university of study,
  subject to housing availability
• They dont do marriage and fertility

19
Comparison of Results Pure MSM

Observed
Predicted
20
Comparison of Results MSM with ABM

Observed
Predicted
21
Comparison of Results Observed, MSM and ABM
Observed MSM
ABM
22
Potential usage of the model
Limiting long-term illness in Leeds 2031
23
Conclusions and Future Work

• We have built the foundations of an ambitious
  hybrid model which combines MSM, SIM and ABM
  features
• Next steps
• Genesis (Generative e-Social Science)
• One Result alignment - towards validation - by
  matching the assumptions used in ONS projections
• Two Learn from the model and improve various
  sub-models according to the recent population
  trends etc. until satisfied reality is being
  reproduced.
• Three Explore the potential of usage of ABM in
  conjunction with MSM, eg interaction between
  individuals/environment, individual behaviours,
  impact of personal history etc.

24
References

• Billari, F., Ongaro, F., Prskawetz, A. (2002).
  Agent-based computational demography Using
  simulation to improve our understanding of
  demographic behaviour, in F. Billari A.
  Prskawetz (Eds.), (pp. 118). London/Heidelberg
  Springer/Physica.
• Birkin M. and Clarke M. (1987) Comprehensive
  models and efficient accounting frameworks for
  urban and regional systems. In Griffith D., and
  Haining R. (Eds) Transformations through space
  and time, Martinus Nijhoff, The Hague, 169-195.
• Caldwell, S. Clarke, G. and Keister, L. (1998)
  Modelling regional changes in US household income
  and wealth a research agenda. Environment and
  Planning C Government and Policy 16 707722.
• Champion T., Fotheringham S., Rees P., Bramley G.
  and others (2002) Development of a Migration
  Model. Office of the Deputy Prime Minister,
  London. Online at http//www.odpm.gov.uk/stellent
  /groups/odpm_housing/documents/page/odpm_house_601
  865.pdf
• Harding, A(2007)APPSIM The Australian Dynamic
  Population and Policy Microsimulation Model, the
  1st General Conference of the International
  Microsimulation Association, Vienna, Austria.
• ...

25

• Morrison, R.J. (2003) Making Pensions out of
  Nothing at All, The International microsimulation
  Conference on Population, Ageing and Health
  Modelling our Future.
• Orcutt, G., J. Merz and H. Quinke, eds. (1986)
  Microanalytic simulation models to support social
  and financial policy, North Holland Amsterdam.
• Rephann, T. J. (1999) The education module for
  SVERIGE Documentation V 1.0, available at
  http//www.equotient.net/papers/educate.pdf
• Sutherland, Holly (2007) EUROMOD - the
  tax-benefit microsimulation model for the
  European Union, in Anil Gupta , Ann Harding
  Modelling our Future population ageing health
  and aged care , Elsevier Science BV, chapter 10,
  477-482, 2007
• van Imhoff E. and Post W. (1998) Microsimulation
  methods for population projection. Population An
  English Selection, 10 97-138.
• Wu, B.M. Birkin, M.H.and Rees, P.H. (2008)A
  spatial microsimulation model with student
  agents, Computers, Environment and Urban Systems
  32, 440453