MOSES: Modelling and Simulation for e-Social
Science
Mark Birkin, Haibo Chen, Martin Clarke, Pete Dew, Justin Keen, Phil Rees, Jie Xu
University of Leeds
Abstract
This poster describes an e-Social Science research programme at the University of Leeds with a
specific focus on Modelling and Simulation (MOSES – Modelling and Simulation for e-Social
Science). The cornerstone of the programme is the creation of a dynamic simulation model of the UK
population, represented as a series of richly disaggregated individuals and households. We aim to use
the power of e-Science Technologies to deliver a complete representation of the population which
draws on attributes from a diverse portfolio of databases. The simulation model will be applied to
address research questions in three social science domains, relating to healthcare policy and practice,
transport & environmental sustainability, and the business impacts of socio-demographic change.
1.
Background
Social modelling and simulation has obvious
appeal to game players of all ages: examples
like ‘The Sims’™ as well as ‘SimCity’™
spring immediately to mind. On the other
hand, simulation modelling has had a rather
more chequered history as both an academic
and applied method within the social sciences,
ever since Lee’s (1973) damning early critique.
And yet the value of scenario-based
approaches to policy development, evaluation,
planning and research remains widely
recognised (e.g. Masser et al, 1992).
In this research, we argue that the renaissance
in social simulation, and specifically in urban
and regional modelling is long overdue. The
idea of developing real urban simulations
which are calibrated using widespread social
and behavioural data has appeal from a number
of perspectives. In the first place, the exercise
is academically and intellectually challenging.
There are many social scientists who would
deny vehemently that it is possible to represent
cities in such an analytical fashion and to
derive meaningful outputs from this process.
To demonstrate a capability to reproduce
social behaviours and patterns within cities is
therefore an objective in its own right.
Secondly, the analogue of social simulations to
a wind tunnel has value. We can legitimately
ask whether such models might not be used in
real planning environments, to test and predict
the outcomes of different policy interventions.
Thirdly, there is a different kind of analogy to
flight simulation, in which pilots are trained to
fly aircraft within exceptionally realistic VR
environments which can provide learning
opportunities with no risk to expensive
equipment or human life. Could the same
opportunities be used in the urban planning
context, to learn about the impact of alternative
strategies without the need to experience
negative outcomes from poor decisions?
2.
Aims and Objectives
Within this context, the MOSES project has a
number of aims. The overarching goal is to
create an activity in which the capabilities of
Grid Computing are mobilised to develop tools
for social modelling and simulation whose
power and flexibility surpasses existing and
previous research outputs. Furthermore, we
seek to demonstrate the applicability of gridenabled modelling and simulation tools within
a variety of substantive research and policy
environments; to provide a generic framework
through which grid-enabled modelling and
simulation might be exploited within any
problem domain; and to encourage the creation
of a community of social scientists and policy
users with a shared interest in modelling and
simulation for e-social science problems.
In order to promote these aims, our main
building block will be a richly disaggregate
synthetic model of the UK population. It is
our objective to develop this model with both a
baseline and a short-to-medium term
forecasting component. The model will be
deployed in selected application domains,
comprising health, business and transport, to
demonstrate policy impacts and the valueadded through simulation.
From these
examples, we hope to generalise the
application of these techniques to more varied
domains.
3.
Relevance of e-Science
3.1
Rationale
The concept of e-Science and Grid Computing
is crucially important to the reinvigoration of
urban simulation for a number of reasons.
Firstly, and most obviously, the programme
demands sharing of data, for example between
the core model and the various application
domains. Furthermore, this sharing of data
may only be possible with strict attention to
problems of security and confidentiality: for
example, if patient records are being accessed
for the purposes of service delivery planning.
Many of the simulations may be complex and
computationally onerous, especially if
forecasts are to be derived as some kind of best
guess or ‘average’ from a large universe of
possibilities. The academic possibilities will
only be realised fully through the pursuit of
diverse multi-disciplinary collaborations, while
under the proposed model planning
departments themselves might increasingly
adopt
the
characteristics
of
virtual
organisations. Finally, it is evident that part of
the appeal of simulation games lies in their
excellent interfaces and visual representation
of outcomes. It is possible that academic
implementations of e-social science might
equally well benefit from the application of the
latest visualisation technologies.
3.2
Proof of Concept
The application of Grid technology to spatial
decision
support
systems
has
been
demonstrated within the context of a healthcare planning scenario through the Hydra
project (Birkin et al, 2005). Hydra assumes a
scenario in which health care services targeted
at a particular demographic group are made
available through a dispersed network of
providers. The technology is designed to
support a wide range of contemporary
problems such as growth of care in the
community services for the elderly, and
increased local provision of services like
cancer screening. The Hydra demonstrator
incorporates a service-based grid architecture
which provides secure access to a variety of
capabilities, including a (virtual) data service,
modelling and optimisation, mapping and
collaborative services, delivered through an
easy-to-use portal.
The Hydra portal is
illustrated in Figure 2.
Figure 1. The Hydra Portal
4.
Methodology
The demographic simulation model will be
developed in a four stage process –
representation,
behavioural
modelling,
forecasting and application testing.
4.1
Stage 1. Representation.
The objective is to generate a complete
synthetic representation of people and
households in the UK. The building blocks
within this process will be the Sample of
Anonymised Records (SARs) from the 2001
Census of Population and Households.
Repeated sampling from the SARs will be
used as a means to recreate small area
populations in accordance with known census
distributions (compare Williamson et al,
1998). As their description implies, the SARs
are fully anonymised and there is no
possibility that real individuals or households
may be identified through this process. Thus
the baseline population for the model will be a
synthetic
but
completely
realistic
representation. This recreation process is
commonly referred to as ‘microsimulation’.
4.2
Stage 2. Behavioural modelling.
The second stage of the project will be
concerned with the addition of likely activity
patterns for our synthetic population. This will
include travel to work patterns, migration,
retailing, leisure and education. A variety of
secondary databases from government and
commercial providers will be used to inform
this process.
4.3
Stage 3. Forecasting.
The population will be projected forward to
the year 2031 using a combination of static and
dynamic ageing. ‘Static ageing’ is a process in
which the core database is resampled in order
to match a change in the underlying population
distribution.
For example, suppose that
government projections show an expected
growth in the young and affluent communities
of metropolitan Leeds. The population would
we resampled with increased selection
probabilities for the young and affluent target
group. ‘Dynamic ageing’ is a method in which
individual processes of ageing, household
formation, labour market migration, and so
forth, are modelled explicitly for individual
members of the population.
Thus an
individual aged 25 in 2001 will be aged 35 in
2011, to use a straightforward example. The
dynamic ageing method is more resource
intensive than static ageing, but potentially
more effective.
Figure 2. Four stage modelling process
2001 Census
SAR
Microdata
2001 Census
Area Statistics
Tables
2001 Census
Commuting
Data
Representation
Model
Retail & Other
Activity Data
Behavioural
Model
2001 Census
UK Microdata (1)
Residential
Attributes
1
.
.
.
.
.
.
H
.
.
.
.
.
.
…………
…………
…………
…………
…………
…………
…………
1
.
.
.
.
.
.
P
.
.
.
.
.
.
Immigration
Emigration
Domain
behaviour
data
Domain Applications
5.1
Business
In this application area we propose building a
model that sits on top of the individual and
household microsimulation model and
simulates the effects of a number of critical
personal financial service events and scenarios
to examine their potential impact both at a
national but also, importantly, at a local level.
The events we propose exploring relate to the
increased level of personal indebtedness in the
UK. Latest Bank of England estimates suggest
that personal indebtedness have reached £1
trillion, equivalent to annual GDP. Several key
factors come in to play looking towards the
future :
• The pensions timebomb: as individuals
recognise that their pension is unlikely
to fund their current lifestyle they will
look to liquidate assets (mainly
property) to top up the shortfall in
pension payments
• Relating to this the increased use of
Equity Release Products to generate
annuity incomes
• The reduction of inter-generational
transmission of wealth
• Potential deflation in house prices
• Potential rise in interest rates
• Increase in household formation (more
smaller households)
1
.
.
.
.
.
.
H
.
.
.
.
.
.
UK Microdata 2006 (3)
Residential
Attributes
UK Microdata (4)
Domain
Attributes
………………
………………
………………
………………
………………
………………
………………
Domain
infrastructure
data
Application
Model
Forecasting
Model
4.4
5.
Domain
Attributes
………………
………………
………………
………………
………………
………………
………………
Residential
Attributes
1
.
.
.
.
.
.
H
.
.
.
.
.
.
1
.
.
.
.
.
.
P
.
.
.
.
.
.
Activity
Attributes
Domain
Attributes
…………………………
…………………………
…………………………
…………………………
…………………………
…………………………
……...
1
.
.
.
.
.
.
H
.
.
.
.
.
.
Stage 4. Application testing.
At this stage we would look to model
processes relating to the specific application
domains. This will require data relating to
both infrastructure and behaviour: for example,
the provision of hospital beds and patient
referral patterns in the healthcare sector; or
road networks and traffic counts for transport
analysis. The extent to which applications can
be generalised across domains remains an open
Process summary
The components of the simulation model are
summarised in diagrammatic form at Figure 2.
Next, we consider some of the issues which
might be considered within our chosen
application domains.
Deaths
Births
1
.
.
.
.
.
.
P
.
.
.
.
.
.
4.5
2001 Census
UK Microdata (2)
Residential
Attributes
1
.
.
.
.
.
.
P
.
.
.
.
.
.
2001 Census
Workplace Data
question for further research and investigation
within the project.
We propose building a simulation model that
would explore the interdependencies of these
potential events over the next decade. We
believe that the impacts of a fall in house
prices/ increase in interest rates will have
substantively different impacts in different
regions/localities of the UK that the simulation
model should be able to detect and predict.
5.2
Transport
Many regional development agencies have
ambitious plans for expansion within their
local economies. For example, the recently
published business plan for the Northern Way
anticipates substantial increases in the
throughput of business and leisure trips
through northern airports, together with a
growth in the region’s share of ship arrivals
and container freight. At the same time, the
plan aspires to reduce congestion in the interurban strategic road network to below the
national average by 2010. This challenging
objective will only be achievable through some
combination of a reduction in intra-regional
business and leisure trips (for example, in
relation to increased home-working), a
redistribution of trips towards uncongested
routes, changing modes of transport (for
example, increased rail traffic), or investments
in the transport infrastructure, e.g. new roads
or improved junctions. The articulation of
scenarios relating to changing demographics
and business activity, together with economic
forecasts in line with the ambitions of the
Northern Way, and specific ‘what if’ changes
to the local infrastructure, would be an
example of a suitable challenge for the
MOSES simulation technology.
5.3
Health care
One of the fundamental problems in health
care modelling and analysis is that services are
provided and monitored by organisations
which are vertically-oriented, but that use
profiles for individuals are not constrained by
the same boundaries. For example, care for
the elderly is provided through a rich
combination of primary and secondary health
care together with social services, local
voluntary organisations, and informal support
within a family or neighbourhood. However it
is extremely difficult to get a view of service
use at the level of individual patients, even
across the two major services of health and
social care because of the disassociation of the
provider organisations. This is important, not
least because service provision and utilisation
are inextricable linked: for example, if
hospital beds are limited within a particular
area, then more intensive use of social services
is one likely outcome. Therefore one objective
within this part of the project would be to
explore the capabilities of Grid technology to
integrate data from diverse sources, including
health and social care, to provided a balanced
picture of service use. Another interesting
question is the way in which ‘social networks’
might support more formal regimes of health
and social care.
For example, can one
demonstrate that communities with strong
social networks are persistently less dependent
on formal care?
The representation of
individual members of the population within
this project provides an ideal platform for
investigations of this type, in which links
within a social network could be simulated in a
similar way to behaviours or other activity
patterns.
6.
Conclusion – Towards
generic social science applications
We are now in a position to add an
applications layer to our description of the
modelling process. In order to build problemfocused applications on top of the
microsimulation model, two types of inputs are
required – data about individual characteristics
and behaviours (morbidity, propensity to own
a personal pension, preferred mode of
transport) and information relating to
infrastructure and service provision (hospital
treatment rates, house price data, trip cost by
mode).
There seems no reason why a general model of
this type might not be applied to a wide range
of problems. For example, a user with an
interest in crime patterns might access data on
propensities to commit crimes (or the
likelihood of falling victim to crime) together
with intelligence relating to the crimes
reported to various local police forces. This
could lead to a model which allows the
effectiveness of crime prevention to be
benchmarked.
References
Birkin M., Dew P., McFarland, O., Hodrien, J.
(2005)
Hydra: A prototype grid-enabled
decision-support system, Proceedings of the
First International Conference on e-Social
Science, National Centre for e-Social Science,
Manchester.
Lee, D.B. (1973) Requiem for Large Scale
Urban Modelling, Journal of the American
Institute of Planners, 39, 163-178.
Masser I., Suiden O., Wegener M. (1992) The
geography of Europe’s futures, Belhaven,
London.
Williamson, P., Birkin, M., Rees, P. (1998),
The estimation of population microdata by
using data from small area statistics and
samples of anonymised records, Environment
and Planning A, 30, 785-816