Spatial Microsimulation Models: challenges of dealing with demographics at the small area level

1
Spatial Microsimulation Models challenges of
dealing with demographics at the small area level

• Mark Birkin and Martin Clarke

2
Microsimulation a potted history

• Origins of microsimulation can be traced back to
  the work of Guy Orcutt (1959, 1962)
• Principal characteristic representation of a
  (economic) system at the micro-level
  individual, household, firm.
• Individuals are then processed in some
  probabilistic way (e.g. Monte Carlo simulation)
  or rules applied
• and results aggregated to produce outputs
• During the 1960s the technique was predominantly
  used by economists to examine impacts of policies
  e.g. Change in taxation or housing benefits

3
1970s

• Orcutt moved to the Urban Institute and developed
  DYNASIM (Dynamic Simulation of Income) and TRIM
  (Transfer of Income Model)
• Kain and Apgar (1977) Housing Model
• These economic models had no spatial
  representation but did model household dynamics
• Aided and abetted in the US by the availability
  of micro-data

4
Spatial Microsimulation

• Often traced back to Hagerstrands paper What
  about people in Regional Science in 1970
• However, microsimulation is not about people
  but a methodology for efficient representation of
  a complex system and an effective way of
  modelling change to this system
• First reference to spatial microsimulation is
  Wilson and Pownall (1975) which started a rich
  tradition of microsimulation modelling at Leeds
  which continues to the present day

5
Model Applications

• Economists finance, taxation, pensions, benefits
• Leeds
• Health Care
• Education
• Retail
• Income/Housing

6
Different types of Microsimulation Models

• Static Microsimulation used to look at impacts
  of policy changes e.g. taxation
• - relatively straightforward in most cases
• Dynamic Microsimulation modelling household
  dynamics.
• - intrinsically difficult from a modelling
  perspective
• - interdependence of events
• e.g. labour market,
  housing market and migration
• marriage and
  migration
• competition
  between individuals for housing, jobs and
  marriage partners
• Attempts made to link macro- and micro-models to
  model markets

7
Creating an initial population

• In an ideal world we would access relevant real
  world micro-data with suitable specified
  attributes
• In some cases it is possible to merge different
  micro-data sets to create a new combined one
• In the 70s and 80s use of Iterative Proportional
  Fitting to create a synthetic population
  consistent with known aggregate distributions
• More recently we have seen the use of genetic
  algorithms to create populations at the small
  area by sampling from National databases such as
  the Sample of Anonymised Records. Computationally
  very time consuming!

8
Computational Issues

• Earliest attempts at spatial microsimulation were
  seriously compromised by lack of computing power.
  Sample data of say 5000 households were about the
  limits of what was possible for a 10 year dynamic
  model looking at household dynamics, housing and
  labour markets
• Nowadays this is not so much a problem but as we
  shall see increasingly ambitious models pose
  their own computational problems

9
Changing Times

• In the 1970s the household composition of the UK
  was relatively straightforward most people
  lived in nuclear households and typically lived
  near where they worked
• In the 2000s household composition is much more
  complicated
• - Houses of multiple
  occupation
• - Married but living apart
• - long distance commuting
• - increased student population
• - people in institutions e.g.
  care homes
• These changes present significant challenges to
  both the specification of initial populations and
  modelling household dynamics

10
Spatial Representation a paradox

• Models of 1970s and 80s would typically use
  census wards 33 in Leeds with a population of
  c25,000
• - using Monte Carlo sampling
  provided fairly robust results at this level
• 2000s use Output Areas 2400 in Leeds with a
  population of c300
• - using Monte Carlo sampling
  does not provide robust results at this level

11
A Microsimulation Research Agenda

• Development of methods/ optimisation
• Dynamic modelling
• Agent-based simulation
• Applications
• Validation

12
A Microsimulation Research Agenda

• Development of methods/ optimisation
• Dynamic modelling
• Agent-based simulation
• Applications
• Validation

13
Methods
Harland, Heppenstall, Birkin, Smith (2009)
14
Methods
15
Dynamics
2006
2001
2031
2016
16
Dynamics
17
Agent-based models
18
Agents
19
Agents
20
Applications
21
Validation
Predicted

Observed
22
Conclusion

• P. Williamson (1992) Community Health Care
  Policies for the Elderly A Microsimulation
  Approach.
• J. Jin (2009) A Small Area Microsimulation Model
  for Water Demand.
• J. Williams (1993) A simulation model of the
  diffusion of HIV and AIDS in the United Kingdom
• C. Duley (1989) A model for updating
  census-based household and population information
  for inter-censal years
• B. Wu (2010???) A hybrid geographical decision
  support system for strategic planning in UK