Table of Contents

Go Back

Spatial urban dynamics*

A vision on the future of urban dynamics:
Forrester revisited

Peter Sanders
Delft University of technology
Faculty of technology, policy and management
PO Box 5015, 2600 GA Delft, the Netherlands
Phone: (+31)646023307
E-mail: p.sanders@tbm.tudelft.nl

Frank Sanders
Delft University of technology
Faculty of civil engineering
PO Box 5048, 2600 GA Delft, the Netherlands
Phone: (+31)15271780
E-mail: f.m.sanders@citg.tudelft.nl

*
This research was conducted during an internship at Rand Europe.

Abstract
The development of a dynamic spatial model of an urban area is described in this
work. The system dynamics method is used to create a model that copes with the
criticism on the original Urban Dynamics model (Forrester 1969) by disaggregating
the urban area into 16 zones. It was found that the trajectory of the behavior (growth,
overshoot and stagnation) in the Forrester model is observed in each zone, but the
overall behavior of the city shows a relatively small overshoot. The zonal division
creates opportunities to explain and understand the dynamic behavior within the city
in a more satisfactory manner. Finally, it is concluded that the system dynamics
method remains very useful for creating insight in urban management for urban
planners and students, despite the fact system dynamics has never become an
established method in the field of urban planning (Alfeld 1995).

1 Introduction
1.1 Urban development
Urban development is the growth process of urban areas. An urban area is
traditionally defined as a system of interacting industries (figure 1), housing and
people (Forrester 1969). Under favorable conditions the interplay between the parts of
a new area cause it to develop. As the area develops and its land area fills, the
processes of aging cause stagnation. As the urban area moves from the growth phase
to the equilibrium phase, the population mix and the economic activity change. Unless
there is continuing renewal, the filling of the land converts the area from one marked
by innovation and growth to one characterized by aging housing and declining
industry.

In this research, an urban area is defined as a system of interacting business structures,

housing and people with an explicit formulation for the spatial component.

1
In figure 1, the system observed in this study is displayed. The behavior of the system
is endogenously driven. The conditions within the system control the dynamics of the
system and these conditions are part of various feedback structures. The urban system
incorporates two markets:
1. Labor market
2. Housing market
+ Population -
1 2
- +
Job Housing
availability availability

+ 3 - + 4 -

Business
Houses
structures
-
5 6
-
Land
availability

Figure 1: Urban system area is a system of interacting sectors (Alfeld and Graham 1976)

Figure 1 shows that an expanding population can strain the area’s resources. As the
population increases the job availability (loop 1) and housing availability (loop 2) will
decrease. The lower job and housing availability correspond with a lower
attractiveness of the urban area due to perception of high rents, severe crowding,
unemployment and low pay. The diminished attractiveness leads to less immigration
and more out migration.
A high housing availability (4) leads to lower rents and therefore suppresses new
construction of housing. Similarly in loop 3, a shortage of jobs implies an excess of
potential employees, who are probably available on short notice and for low wages.
This stimulates the construction of new business structures and expanding of existing
structures.
The land availability modulates the construction of both housing and industry in a
non-linear way (loop 5,6). When almost no land is occupied, an increase in housing
and business structures stimulates further development because an area becomes more
attractive, due better transportation facilities and more attractive potential for market
growth. In the period that the area starts to approach full land occupancy, the
construction of housing and business structures decline. The parcels that are
developed last are probably fragmented and situated at the least desirable locations.

1.2 The problem of urban management

The problems in urban areas identified by Forrester (Forrester 1969) are still present.
The municipality of Rotterdam (second largest city in the Netherlands) published a
report (Rotterdam 2003), which identifies problems as urban decay, filtering down of
housing and a large fraction of underemployed population in the city. In the report, a
systematic approach is advocated.

The phenomenon that Forrester calls “counterintuitive behavior of social systems” (

Forrester 1969, Forrester 1971) is as insightful as three decades ago. Counterintuitive

2
behavior or policy resistance (Sterman 2000) means that actions taken to alter the
state of the system feed back (unanticipated effects) into the system. Unanticipated
effects often arise in complex systems, like urban centers, because cause and effect
are distant in time and space.
As explained in figure 1, a city is a complex system consisting of an interlocking
structure of feedback loops. The system dynamics literature contains a large body of
evidence stating that people are incapable of assessing the effects of their actions in a
complex system (Paich and Sterman 1993). Urban planners encounter the same
problem in their work. This leads to the conclusion that a simulation model can assist
them in building understanding in the behavior of an urban system.
An example of policy resistance in urban management in the Netherlands is the policy
whereby the underemployed population was subsidized so they could live in worker
housing (in a neighborhood in Den Haag). The worker housing filtered down very fast
because labor population left this neighborhood and the abandoned worker housing
(economic obsolete) could only be filled with underemployed population. The urban
planners underestimated or neglected the effect of social mixture on the migration.
The result of good intentions to relieve the city of the pressure on underemployed
housing market is a neighborhood with a poor image and prospects. The current
mixture of the population and the characteristics of the neighborhood create the
challenge of reviving this neighborhood.

The objective in this research is to build understanding in how urban areas develop as
a result of the interactions between people, housing and business structures. This is
vital in order to develop a good policy to alter the behavior of a system.

2 From urban dynamics to spatial urban dynamics

2.1 Traditional urban dynamics
Jay W. Forrester’s well-known publication Urban Dynamics in 1969 introduced a
new perspective on analyzing urban problems, forming a bridge between engineering
and the social sciences. Today, this perspective is known as System Dynamics. The
use of the system dynamics methodology to analyze socio-economic processes in
order to simulate urban development is called urban dynamics.

The traditional urban dynamics models define an urban area as a system of interacting
industries, housing and people. The area:
Could be the political boundary of a city but usually will differ. The area treated here (in
urban dynamics) would be only a part of our larger cities. The appropriate area is small
enough so that cultural, economic and educational interchange is possible between its
component populations. It could be a suburban area or the core area of a city but probably
not an area containing both (Forrester 1969).
Figure 2 illustrates this definition. A suburban area would be considered part of the
environment or an endogenously driven system of its own.

3
 Environment

City

Figure 2: The urban area in its limitless environment

The urban area is represented as a social system set in an environment with which it
communicates. People can move in and out of the area. The flow to and from the city
depends on the relative attractiveness compared to its surrounding environment.
Conditions in the surrounding environment are taken as a reference (which can
change) the attractiveness rises or falls with respect to that moving reference. The
attractiveness of the area, compared to its environment, depends on the conditions and
the activities within it. The environment is taken as a limitless source. It can absorb or
supply people as long as the area is more attractive respectively less attractive. These
three concepts:
1. specific area;
2. relative attractiveness;
3. limitless environment;
are important in understanding the model of urban processes.

The boundary of the urban system

The boundary is chosen so that it includes the concepts that interact to produce the
behavior of interest. In the classical urban dynamics model, this behavior is the
growth, overshoot and stagnation of an urban area. Urban growth, overshoot and
stagnation do not appear to require changes in the world environment as a cause. The
behavior of a city is much more directly dependent on its own economic merit and its
changing internal mix of industry, housing and population. It is assumed that an urban
area in the urban dynamics models is small enough in the world setting not to affect
the outside environment. The urban area is therefore taken as a living system that
communicates with an environment it does not influence.

Relative attractiveness
Using the environment as a reference point means that conditions within the urban
model are generated relative to the environment. The model shows how the area
becomes more or less attractive than the surrounding country and other cities and
thereby causes the movement of industry and population to and from the area. Only
differences between the area and environment are significant. The model does not
deal with changing technology and rising national standard of living as it assumes
these developments to be equal for the urban area and its environment.

2.2 Reviewing urban dynamics

The publication of Urban Dynamics in 1969 documents one of the most insightful
applications of the system dynamics (SD) methodology. However, the book generated

4
intense controversy (Alfeld 1995). The fundamental criticism on the methodology of
urban dynamics breaks down in three parts:
1. the boundary problem;
2. the problem of the limitless environment;
3. the use of data

The boundary problem

The problem of choosing the system boundary in a way that its environment does not
substantially influence the system and vice versa is called the boundary problem. The
main criticism deals with the perceived exclusion of interactions between the city and
its suburbs (Graham 1974; Gray, Pessel, and Varaiya 1972; Rochberg 1972).
In the eyes of critics there are a number of observations that invalid the assumption
that its environment does not substantially influence the urban area (Garn and Wilson
1972). First, the urban dynamics model does not include commuting across the
boundary; people who work in this city reside there. Second, the model rules out the
possible effects of actions taken to improve the situation in the city on the larger
society, by the choice of the closed boundary.
In other words, critics challenge the principle that the system does not influence its
environment in significant way. They question the assumption that the dynamics of
cities are endogenously driven.

The problem of the limitless environment

The unlimited environment means that people are available from the outside for
migration into the area whenever the area appears more attractive than the point from
which people may come from (and vice versa). In this formulation, the limitless
environment is actually the rest of the universe.
Babcock (Babcock 1972) and Gray (Gray, Pessel, and Varaiya 1972) observe that the
majority of people in the US live in cities. Thus, the environment should be viewed as
a collection of cities. This naturally leads to the criticism that Forrester’s model
encourages local (citywide) optimization rather than global (national) optimization.
The concept of the limitless environment omits the spatial dimension. The
development of the city Amsterdam in the Netherlands will be hardly influenced by
its relative attractiveness compared to a town in China. However it will be influenced
by the attractiveness of the satellite city of Almere.

The use of data

The urban dynamics model was formulated without recording empirical data.
Therefore the model specifies a hypothetical city. Garn and Wilson explain this point
as followed (Garn and Wilson 1972):
It is one thing to make the methodological point that the latter functions (parameters
and multiplier functions) should be nonlinear. It is quite another to know what the
critical functions are, as well as their range and shape. Again, there is little empirical
support provided for the assumptions.
This point of criticism is well known for SD applications in general. The standard
response of SD practitioners is that it is better to include an estimation of a variable
than omitting this variable, in other words denying its influence (Forrester 1969;
Forrester 1980). The value of the mental database is emphasized in the urban
dynamics model (Forrester 1969).
The use of the data invalids the results of the Forrester model (Gray, Pessel, and
Varaiya 1972). Forrester draws general conclusions from his specific model.

5
Alternative data should be presented to valid the policy recommendations (Jaeckel
1972).

System dynamics practitioners reacted on the criticism in Readings in Urban

Dynamics volume I (Mass 1974) and II (Schroeder, Sweeney, and Alfeld 1975).

2.3 Building on the insights of urban dynamics and its critics

2.3.1 The reaction on the criticism
The most fundamental criticism concerns the boundary choice of the urban dynamics
model. In a reaction to this criticism Walter Schroeder (Schroeder 1975) formulates
an alteration of the Forrester model to include the relationships between the city and
its suburbs, similar extensions are made by Kadanoff (Kadanoff and Weinblatt 1972),
Laudet and Fournier (Laudet and Fournier 1979; Fournier 1986).
The research suggested by Kadanoff is comprehensive, however the formulation for
the interaction between the different sub sectors is primitive. Burdekin (Burdekin
1979) divides the city in 16 zones for which the construction rates of different housing
types and business structures depend on a combination of citywide and zonal
characteristics. The development of the population is not disaggregated but specified
like in the Forrester model.

Figure 3 illustrates the development of urban dynamics research described in the

section above.

6
 Environment
Environment

City City

Suburb

1) The original model 2) The Schroeder model, the Kadanoff

Forrester and Weinblatt model
1969 Laudet and Fournier model
1975,1972,1979, 1986

Environment Environment

Zone 1 Zone.. Zone 1 Zone..

Zone.. Zone 16 Zone.. Zone 16

3) Burdekin 4) Sanders
1979 2004

Figure 3: The development of the spatial dimension in Urban Dynamics

All the research mentioned in figure 3 utilizes the attractiveness principle to regulate
the migration flows (between the city, the environment and zones). Therefore it is
important to understand how this principle works. The basic assumption is that there
is a normal fraction of residents that migrate into the city (zone) or out of the city
(zone). These normal fraction rates are modulated by the attractiveness of the city
compared to its environment (which has attractiveness ‘1’) or other zones.
Alternatives for this formulation are logit models or gravity models (Wilson 1974)
where the available residents will be allocated again and again every time step.

Schroeder 1974 and Burdekin 1979

The model of Schroeder will be discussed briefly in this section. In figure 4, the
simple feedback relation model is formulated.

7
 city conditions
affects
suburb conditions

City Suburb

suburb conditions
affects
city conditions

Figure 4: A city-suburb feedback relationship

Schroeder wants to answer two questions. First, if any feedback relationship between
the city and the suburb explain the urban life cycle of growth, maturation and decline?
Second, if the feedback between the city and suburb changes the policy
recommendations of the Forrester model.
The approach adopted by Schroeder is to develop two (Forrester) urban dynamics
models, which are identical. One model is representing the city and the second model
represents the suburb. The total migration into the city is determined by the total
attractiveness of the city (central city plus the suburbs) relative to its environment.
The total migration into the city is distributed over the central city and suburb based
on their relative attractiveness. The migration between the central city and suburb
depends on the residential conditions. In short, people work and live in one area or
they work in one area and live in the other area. However it is assumed when people
arrive in one area they work there too. To illustrate this point, if people arrive in the
central city they work there too and the only possibility is that they start living in the
suburb.
Schroeder’s model is a simple extension of the Forrester model to tackle the model
boundary problem (figure 3). This extension is subject to the same criticism as the
Forrester model. The major assumption is that the behavior of the two areas is
endogenously driven and that the influence of the environment does not significantly
contribute to the behavior of the city and the suburbs.
The division of the metropolitan area in a city and suburbs introduces the spatial
dimension. However, this is done in a correct but unclear way. The visual complexity
of the model is increased tremendously by the introduction of suburbs in Schroeder’s
model.

Burdekin (Burdekin 1979) splits the urban area in 16 zones, in order to model the
mixture of different housing types in zones of a city. It is assumed that the residents in
the urban area are distributed over the zones proportional to the housing types in the
various zones. The migration into and out of the city is based on the citywide
conditions but the population is not disaggregated over the different zones (figure 3).

Sanders 2004
The concept adopted in this research is to divide the urban area into 16 zones. This
represents the reality in a more accurate way, because it includes the zones in a city as
individual endogenously driven systems that all communicate with their environment
(figure 3).

8
This is an improvement to the previous versions of urban dynamics models, in which
the suburbs are aggregated (into a suburb ring). This aggregation does not justify the
settlement patterns observed in the reality and the aggregation omits the possibility
that the zones in a city have different characteristics (mixture of population, housing
and business) and therefore different functions in a city. For this reason it is better to
formulate different zones in a city.

A second reason to disaggregate city in different zones is the possibility to introduce

competition between the zones. Furthermore, the disaggregation allows a better
representation of the behavior of individuals in an urban system, because individuals
relocate over relatively short distances (within the city). The relocations of people
within the city are very important in the development of neighborhoods in the city.
The social environment of a person allows him to observe opportunities within the
city (zones) sooner than in the surrounding environment. In other words, if an area has
high-perceived attractiveness, this will observed sooner by individuals that live closer
to this area.
The statement above is supported by the evidence presented in Blijie, de Bok and
Sanders (Blijie, de Bok, and Sanders 2003). The majority of the relocations of
households are over a very short distance. 90 % of all relocations in the Netherlands
are within a distance band of 40.5 km (with an average of only 13.5 km). Depending
on the reason for the relocation the distance varies. If the reason to move is job related
the distance is on average longer (50.8 km).
The inter-zonal migration can influence the total migration to and out of the city. A
difference between the attractiveness of two zones will be balanced by the internal
migration and consequently make the total attractiveness of the city relative to its
environment different from what it otherwise would have been. The latter leads to
different migration to and out of the city.

The difference between the Burdekin model and spatial urban dynamics model is that
the population is disaggregated based on the zones in the city. This allows the model
to include zonal factors as social mix, housing availability and job accessibility, which
are important determinants in movement behavior of individuals.

2.4 Spatial Dynamics

2.4.1 Spatial dynamics in biological systems
The spatial dimension has not received a great deal of attention in system dynamics
modeling. An intensive literature review showed that there are only a number of
articles dealing with this subject. A summary of the relevant literature is given in
Modeling dynamic biological systems (Hannon and Ruth 1997) and in Modeling the
environment: An introduction to system dynamics modeling of environmental systems
(Ford 1999).

Hannon and Ruth (1997) recognize the importance of the spatial dimension in the
modeling of biological systems. A mobility framework is specified in which migration
from one state to another depends only on the current status of the donor state. They
use this formulation to introduce a spatial dimension to the classical predator-prey
model, making it possible for the prey to migrate to another region. The migration of
the prey is referred to as mobility. The spatial aspect is modeled by defining four
subdivisions of the landscape as laid out in figure 5.

9
 1 2

3 4

Figure 5: The spatial aspect in the classical predator prey model.

The specification in this model only allows the predators to migrate to an adjacent
cell. This means that only migration from cell 1 to cell 2 or 3 is possible. The prey
migrates routinely regardless of their population in the starting or the receiving cells,
and predators migrate to a new cell when they begin to starve in their current cell.

The examples described in the book are very simple models. Despite the simplicity of
the model the specification already leads to a great deal of visual complexity. A
second observation is that the interaction between cells is only with adjacent cells. For
many biological problems, like the migration of toxic fluids across a landscape, this is
fine. However if one looks at the migration of people this implicit assumption is
incorrect.

It can be concluded that there is analogy between the biological examples and the
application of urban dynamics. However, a different formulation for the spatial
dimension is necessary.

The obvious way to link urban dynamics with the insights in this section is to define
an endogenously driven system for each cell. This endogenously driven system
determines the internal conditions for the total area. These internal conditions (for
example job opportunities, housing availability or availability of accommodation for
business structures) represent the relative attractiveness of each area. Based on this
relative attractiveness, people and firms will relocate (change between different
areas).
The major difference with the spatial dynamics described in biological systems is that
migration is possible between areas that are not flanking (figure 6). The reader should
keep in mind that time scale of the dynamics of interest is different for biological
systems and urban systems. During the migration of people in an urban area they have
to travel through adjacent cells, however this short-term dynamic behavior is not
necessary to capture in the long run dynamics.

10
 Traditional spatial dynamics in SD literature New approach for spatial dynamics

1 2 3 1 2 3

4 5 6 4 5 6

7 8 9 7 8 9

Figure 6: The difference between the traditional and new approach for spatial dynamics

In the traditional approach the migration out of area 1 must be to an adjacent area 2 or
area 4. The total number of possible flows is 18 divided by 2, because the relative
attractiveness of two areas leads to one flow. If area 1 is more attractive than area 2
only a flow between area 2 and area 1 is specified and no flow between area 1 and
area 2.
The new approach for spatial dynamics makes it possible to specify the migration
flows between all 9 areas. A migration flow is the total number of people who leave
zone 1 to settle in for example zone 5. The total number of possible flows is 72
(=9*8) divided by 2. The challenge is how to define the flows between the regions. A
good suggestion is to use matrix based calculations, which are possible through the
array functions in the SD software packages.

2.4.2 The formulation of an alternative hypothesis

The alternative hypothesis proposed in this paper challenges two elements in
traditional urban dynamics research. First, the poor representation of the spatial
dimension in the mentioned urban dynamics models is challenged. The author feels
that the influence of the spatial dimension is underestimated in these models. For
example the distance between a persons job and home is very important in the
decision to commute or to migrate.
Second, to view urban areas as endogenously driven systems is a good assumption.
The urban area is divided into 16 zones. This alteration makes it possible to judge the
influence of the different zones on each other and on the total behavior of the system.
This introduces competition between the different zones in a city.
Furthermore this will create different attractiveness levels for the different zones.
These attractiveness levels control the migration flows within the city (between the
zones) and from and to the city. The behavior is not necessarily different from the
behavior observed in the Forrester model. However, the introduction of a more
sophisticated spatial representation will lead to better understanding of the urban
system and can present evidence on the value of the traditional criticism. The users
will perceive the model as a more accurate representation of the reality, because it
matches their firsthand knowledge about the system (the state of different
neighborhoods in a city). This hints at a possible failure of the Forrester model. The

11
level of aggregation may be to high, because users of the model do not recognize the
problems experienced by them in various zones in the city in the overall behavior.
The level of aggregation should be zonal. Evidence for this statement is given in the
next table that includes the migration for the city of Rotterdam (the second largest city
in the Netherlands). The migration consists of two types, inter-zonal and migration to
and out of Rotterdam in the year 2002.
Table 1: The migration for Rotterdam in the year 2002 (COS 2004)

# of migrants # of migrants to # migrants to the # migrants out

The 15 zones
from other zones other zones Rotterdam of the Rotterdam
Stadscentrum 2103 2547 2527 2100
Delfshaven 6538 8144 5467 4886
Overschie 908 893 470 682
Noord 3954 4768 3448 3123
Hillegersberg-Schiebroek 3176 1919 1626 1635
Kralingen-Crooswijk 3939 4325 2962 2850
Prins Alexander 4244 3266 2889 3319
Feijenoord 5758 6245 3803 3117
IJsselmonde 4439 2908 2544 2788
Charlois 6019 6439 4861 4455
Pernis 76 85 63 115
Hoogvliet 1881 1518 1510 1602
Hoek van Holland 116 86 312 276
Habor 11 19 24 21
Total 43162 43162 32506 30969
The information in the table illustrates that the inter-zonal migration is considerable in
relation to the cross-city boundary migration.

The main purpose of modeling is problem solving. In light of double-loop learning

models, Argyris and Schön (Argryris and Schön 1978), problem solving can take
different forms. At one level, the main challenge is to convince urban planners and
politicians that a problem exists and that improvement are possible.
The model should include behavior that the users observe in reality. As the users of
the model gain trust in the insights, one could use aggregated models to explain the
urban system to third parties. As the problem is acknowledged, the problem is often
dealt with at another level where more detailed and advanced policy analysis may be
appropriate.
The model described in this paper belongs to the first category, because it focuses on
building of understanding and facilitating learning of the urban system in general. An
application of this model to a specific city would belong to the second category.
The complexity of the urban system makes simulation absolutely necessary in the
development of effective policies.

The developments in the SD modeling software make it possible to reduce the visual
complexity of the model compared to previous research (through arrays, figure 3) and
therefore test the alterations in the model structure and the model boundary.

The hypothesis:
The behavior of the original Urban Dynamics (growth, overshoot and stagnation) will change
when one includes the spatial dimension in an explicit way and additionally will result in
better understanding for policymakers.

12
3 The spatial urban dynamics model
The dynamic hypothesis in this research builds on the model of Forrester. Therefore it
is important to show the relationship with the Forrester model. The core of the model
is quite similar to Forrester. The added structures are:
1. Explicit spatial dimension (figure 7)
The city area is divided in 16 equal squares of side 3.125 miles. The 16 squares add
up to the 100,000 acres in the spatial urban dynamics model like in the Burdekin
model (Burdekin 1979).
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16

Figure 7: The zones of the city

2. Disaggregation
The different types of housing, industry levels (including an additional group of
industry, service industry) and people are identified within each zone. Each zone has
its own stock of housing, industry and residents.

3.1 The spatial urban dynamics model

In figure 1 (section 1) the interactions between the important stocks within the urban
system can be observed. The figure shows that people need a house and a job and the
housing and industry sector compete for the available land. In this section the
interactions between the three major sectors are explained, beginning with the housing
sectors. The industry sector and the migration of individuals are described thereafter.

3.1.1 The housing sector

In order to explain the spatial urban dynamics model it is crucial to show the relations
between the important stocks (housing, industry and people). In the next three
sections it is illustrated how the different stocks are connected with each other.

PHC Premium WHO Worker

housing construction i housing construction i

PH Premium WH Worker UH
housing i Underemployed
PHO Premium housing housing i WHO Worker housing SHD Slum housing
housing i
obsolescence i obsolescence i demolition i

Figure 8: The housing sector

In figure 8, the process of housing filtering down is specified. Housing filter down
denotes the process whereby aging and obsolescence of housing units invite
occupancy by successively less affluent residents (Alfeld and Graham 1976). The
physical deterioration (as repairs are never as good as the original) reduces the
desirability and profitability of a housing unit. Another mechanism of importance is
obsolescence. As the living standards rise and technological progress continues,

13
occupants start looking for new facilities in their housing units. As the process of
aging and obsolescence continues the maintenance costs continue to rise. At some
point in time, the housing unit is no longer viable as an economic entity. The
premium-housing unit filters down to the worker-housing sector.
The reader may argue that the opposite process of aging and obsolescence can be
observed in reality (retrofitting). In the current model this is possible when there is
high demand for premium housing, a lack of space and there is a negative utility of
living with lower economic groups. The obsolescence of premium housing can
become very low in this case.

The next question is how the flows of housing construction and obsolescence are
influenced by the state of the city. The construction rate and the rate of obsolescence
are specified by the product of a set of multipliers, which consists of the various
forces influencing the rates and the current level of housing units. This multiplier
combines the influence of 8 forces:
• The demand for housing in the total city. As housing becomes scare, this provides
incentives to increase the construction of housing.
• The social mixture of a particular zone. This accounts for the fact that the
construction rate is dependent of the residential mixture in the zone.
• The blend of the housing types in a zone. As the fraction of premium housing
increases the construction rate of premium-housing rate will increase relative to
the construction of worker housing.
• The land availability in a zone. As the land in a zones fills the construction rate
will increase because of the increased utility. The construction rate will diminish
as the land becomes fully occupied.
• The tax level in the city. This factor illustrates the city’s attractiveness compared
to its surrounding based on the tax level. As the tax level in the city rises, the
construction rate declines.
• The business confidence in the city. The establishment of new enterprises tends to
encourage construction of new housing for the people associated with the
industrial expansion.
• The momentum of recent construction (rate of change) in housing itself tends to
carry forward and cause still more housing construction in a zone.
• The attractiveness of the zone based on the accessibility of the zone to jobs. As it
becomes easier to reach jobs in or from a zone, the construction rate in this zone
will grow.

This last factor reflects the zone’s favorability for future households based on the
access of the zone to jobs. If a zone has relative (to other zones) good access to
employment the construction rate should be higher than it otherwise would have been.
The zone’s accessibility to jobs is measured as the number of jobs divided by the
distance (to the power 2) to those jobs. The zone’s job accessibility is compared to the
average job accessibility of all zones to indicate its attractiveness for construction
(equation 1). The numerator in equation (1) specifies how easy it is to reach the jobs
in all zones from zone 1 (or any other zone).
The area’s relative job accessibility:
N  MJ (k)  N  MJ(k) 
∑  
 TDFC (i, k )  ∑  
 TDFC (i, k )  (1)
i =1   i ,k =1  
N = The number of zones in the city

14
 MJ = Manager jobs in zone k
TDFC = Trip distribution function construction i
The trip distribution function is the distance between zone i and zone k to the power
2, to include the deterrent effect of distance. The distance is taken as the length of the
straight line between the midpoints of the zones. A more detailed description of the
trip distribution function is given in Wilson (Wilson 1974).

The construction of housing in the zones is based on a combination of zonal factors

and citywide factors. The term ‘citywide’ means that an influence is the same on all
the zones in the city and consequently is not specified on a zonal level. The citywide
factors are the impact of tax, the business confidence and the influence of the demand
for housing. The five other mentioned forces are influences on a zonal level.
The construction rate for housing is constrained by the available labor for
construction. The assumption hereby is that the construction rate is only constrained
by citywide available labor, meaning that construction workers are able to travel
throughout the complete city to do their work. This concept is also used in the
industry sector and the worker-housing sector.

The factors controlling the construction of housing have the inverse effect on the
filtration of housing. As the urgency of new housing construction increases, the
obsolescence rates declines and the average life of housing unit’s increase.

The worker housing and underemployed housing sector

The premium housing and the worker-housing sector are specified in the same way.
However the premium-housing sector is influenced by managerial-professional sector
and the factors related to this sector. For example, the pressure of the location
attractiveness of a zone for premium housing construction is dependent of the
accessibility to managerial jobs and the worker housing construction is controlled by
the accessibility to labor jobs.
In the underemployed housing sector, the slum housing demolition multiplier controls
the slum housing demolition. The latter is partly dependent of the accessibility of a
zone to labor jobs. This means that the underemployed provide pressure to decrease
the slum demolition rate in a zone if the accessibility to labor jobs is above average. It
can be questioned if the underemployed can exercise this pressure.

3.1.2 The industry sector

In the model, two types of industry are defined: the basic industry and the service
industry. The basic industry exists of new enterprises, mature business and declining
business (figure 9).

The basic industry sector

NEC New enterprise

construction i

NE New MB Mature DI Declining

enterprise i business i MBD Mature industry i DID Declining
NED New
enterprise decline i business decline i industry demolition i

15
 Figure 9: The basic industry sector

The basic industry sector is very similar to housing sector in figure 8. The
construction of new enterprises changes through 6 forces:
• The fraction of managers in the city. As the fraction of managers grows in
proportion to the managerial jobs the likelihood of establishing new enterprises
increases, because the managerial population has more time and incentives to start
new enterprises.
• The availability of land in the city. This concept is the same as discussed in the
housing sector section, except that it deals with the citywide land availability.
• The availability of labor in the city. The construction of new enterprises is
depressed if there is a shortage of labor in the city and encouraged if there is an
excess of labor.
• The influence of the tax level on the new enterprise formation. As the tax level is
high relative to the city’s surrounding, the formation of new enterprises is
discouraged.
• The influence of business confidence on the formation of enterprises. If the recent
growth rate for new enterprises has been high, this will encourage further growth
and vice versa.
• The mixture of the basic industry types. The new enterprise has weighting factor
of one, meaning that a unit of new enterprises is most likely to form another new
unit. The mature business has a weighting factor of 0.5 and the declining industry
0.3.
The creation rate of new enterprises is controlled by citywide conditions relative to
the environment of the city like in the Forrester model. However the basic industry
has a prior claim to certain zones (zone 1,5,6,9,10 and 13). This means that the growth
of industry is only allowed in these zones. This concept implies that it is an external
decision that determines in which zones basic industry is allowed. The designation of
zones to industry could also be used as a policy variable.
The filtration of basic industry is influenced by the factors above in an inversed way,
like housing demolition in the housing sector.

The service industry

Figure 10 displays the stock and flow diagram for the service industry.
SI Service
SIC Service industry industry i SID Service industry
construction i demolition i

Figure 10: The service industry

The service industry is second type of industry added to the model and consists of one
stock for reasons of simplicity. The service industry is location sensitive. Location
sensitive implies that construction of service units in a zone is dependent of the
accessibility of that zone to housing units. In other words, the formation of service
industry units will increase as the zone has more access to households (see equation
1). Besides this influence, these 7 pressures control the construction of service
industry:
• The fraction of managers in the city. As the fraction of managers grows in
proportion to the managerial jobs the likelihood of establishing new service units

16
 increases, because the managerial population has more time and incentive to start
new enterprises.
• The availability of labor in the city. The construction of new enterprises is
depressed if there is a shortage of labor in the city and encouraged if there is an
excess of labor.
• The influence of the tax level on the service industry formation. As the tax level is
high relative to the city’s surrounding tax, the formation of new enterprises is
discouraged.
• The influence of recent developments on the formation of service industry. If the
recent growth rate for service industry has been high, this will encourage further
growth and vice versa.
• The availability of land in the zone. This concept is the same as discussed in the
housing sector section.
• The influence of the density of housing units in a zone. The idea is that a normal
service level should be maintained in a zone. The number of service units in
proportion to the number of housing units should be around 1/150. If this ratio is
smaller than 1/150 this provides extra incentives to construct service units.
• The influence of the citywide density of people. The normal service level in the
city is one service unit for every 900 persons. If the current citywide service level
is lower than the normal service level, this encourages the formation of service
industry.
The construction and filtration of service industry is controlled by a blend of citywide
and zonal pressures.

3.1.3 Attractiveness principle and migration of people

The migration in Urban Dynamics is directed by the attractiveness principle. The
moment the attractiveness within the area is higher than the surrounding environment
the migration to this area increases and migration out of the area decreases. In figure
11, the three different groups of persons in the model are displayed and how they
relate to each other.

MPB
LB Labor births i Managerial-professional
UB Underemployed births i
births i
LA Labor arrivals i MA Manager
LTU Labor to arrivals i
underemployed i
U
Underemployed L Labor i MP Managerial-
UA Underemployed i UTL Underemployed LTM Labor to professional i DID Declining
arrivals i to labor i manager i industry demolition i
UD Underemployed LD Labor MD Manager
departures i departures i departures i

UNZM Underemployed LNZM Labor net MNZM Manager net

net zone movement i zone movement i zone movement i

U
Underemployed MP Managerial-
L Labor k
k professional k

Figure 11: The migration of persons

17
The first flows into the stocks of persons are the births of the three groups of persons.
The flow is actually the net difference between the birth rate and the death rate in the
city and is proportional with the distribution of the persons over the zones.

The arrivals and departures of residents into or out of the city are controlled by the
attractiveness of the city relative to its environment. As the attractiveness of the city
exceeds the attractiveness of the environment the arrivals will grow and the departures
will reduce. The supposition is that persons will migrate to and out of the city based
on the citywide attractiveness and consequently the flows are proportional with the
distribution of the residents over the zones. The attractiveness is specified the same as
in the original Forrester model (see appendix A or (Forrester 1969), p 137).

The flow of underemployed to labor represents the upward economic mobility. The
upward economic mobility is influenced by the social and economic conditions in the
city. As the scarcity of labor increases the upward mobility of underemployed will
rise. The upward mobility is more difficult as the labor group decreases relative to the
underemployed. It covers the influence of dominant blocks of underemployed in a
particular zone where they do not intermingle with other economic groups. The effect
of education in upgrading economic skills is made a function of the tax per capita.
The flow of labor to managers is specified in a similar way. On the other hand, the
flow of labor to underemployed grows as the proportion of labor to labor-jobs rises
and is suppressed as this proportion falls.

The specification of the departure and arrival of the different groups of persons are
controlled by the concept ‘attractiveness’. The attractiveness is a product of different
components of attractiveness: the upward mobility of underemployed, housing
availability, economic opportunities and the public expenditure per capita.
The equations for the arrivals and departures from and to the environment can be
combined to specify the migration flows between the zones within the urban area as
suggested by Kadanoff and Weinblatt (Kadanoff and Weinblatt 1972). The
underemployed arrivals are specified as:
UA Underemployed arrivals i = (U+L)*UAN*AMMP (2)
U = Underemployed i (person)
L = Labor i (person)
UAN = Underemployed arrival normal (fraction/year)
AMMP = Arrival migration multiplier perceived (dimensionless)

And the equation for underemployed departures is:

UD Underemployed departures i = U*UDN*UDM (3)
U = Underemployed i (person)
UDN = Underemployed departures normal (fraction/year)
UDM = Underemployed departures multiplier (dimensionless)

This leads to the following equation for the migration (relocation of households) for
underemployed population from zone 1 to zone 2:
(U1 A1 ) × (U 2 × AP2 ) (4)
Total _ U
Zone 1 includes underemployed population U1 and A1 is defined as the attractiveness
of zone 1 for underemployed population U1. Zone 2 has a perceived attractiveness
AP2 for the underemployed in zone 1 and represents a proportion of the total citywide
underemployed group U, given by U2 divided by the total underemployed group. It is

18
assumed that the migration is governed by the attractiveness in the same fashion as in
equation (2) and (3). The probability that an underemployed person will migrate from
zone 1 to zone 2 is proportional to the number of underemployed in that zone relative
to the number of underemployed (citywide). The migration rate from 1 to 2 is a
constant times the equation (4).

The next important question is how to formulate the attractiveness factor controlling
the inter-zonal migration flows. Van Ommeren (Ommeren 2000) presents evidence
suggesting that the two major forces in choosing a new residence location are: job
opportunities and housing opportunities. In other words, a person that decides to move
is basically tide to a job and to a housing elastic.
The accessibility of a zone to jobs is specified as in equation (1) and reflects the force
of the job opportunities. The housing opportunities are included as a ratio of the
number of houses and households of the corresponding level.
In the model, a third important force is included, namely the social mixture of a zone.
The social mixture is important because managers and professional workers
preferably do not choose a residence location (zone) in which a high fraction of
underemployed persons is living.
The influence of tax level used in the Forrester model to partly determine the
attractiveness for cross-city boundary migration has no influence in the inter-zonal
migration because it is the same for all zones.

In summary, there are three migration flows. First, the migration flow to and out of
the city is based on the citywide attractiveness but proportional with the density of the
three economic groups in the zones. Second, there is the upward economic migration.
This flow is controlled by a combination of zonal and citywide conditions. The last
type of migration is the inter-zonal migration, which is naturally managed by the
zonal attractiveness.

Implications of the new structure

The two markets mentioned in section 1 are disaggregated into zones in different
ways. In the labor market, the zones in which the industry construction is allowed
provide the demand for labor. The supply of labor is partly allocated by the access of
people to jobs of their level of competence.
The housing market is now divided into zones. This means there is housing market in
each zone. The market for service industry structures is basically the same as the
housing market with no restriction on the location. The basic industry structure
location on the other hand is pre-determined by external decision. Therefore no
influence of the available labor on the location choice of basic industry is assumed.

4 The behavior of the model

The citywide initial conditions of the model are the same as in the Forrester model.
The zonal initial conditions mainly distributed in zone 5 and 9:
• Premium housing: 2150 (zones 5 and 9), 50 in remaining zones
• Worker housing: 10150 (zones 5 and 9), 50 in remaining zones
• Underemployed housing: 480 (zones 5 and 9), 10 in remaining zones
• New enterprise: 100 (zones 5 and 9), 0.01 in remaining zones
• Mature business: 500 (zones 5 and 9), 0.001 in remaining zones
• Declining industry: 50 (zones 5 and 9), 0.001 in remaining zones

19
• Service industry: 85 (zones 5 and 9), 1 in remaining zones
• Managerial-Professional: 1676 (zones 5 and 9), 39 in remaining zones proportional with the
premium housing density
• Labor: 6766 (zones 5 and 9), 33 in remaining zones proportional with the worker housing density
• Underemployed: 523 (zones 5 and 9), 11 in remaining zones proportional with the underemployed
housing density

In order to present the results in a clear manner it should be noted that the results of
zone 1-8 are symmetrical to the results of zone 9-16. The reason is the initial values
for the stocks in the zones. The geographically location of zone 1-8 in relation to zone
9 is the similar to the geographically location of zone 9-16 in relation to zone 5 (see
figure 8). The moment one changes the initial values of the stocks in the zones the
symmetry of the zones would be lost and the results of all the 16 zones should be
represented. In this paper, the output of zone 1-8 is presented in the graphs.

The equilibrium that is reached by the model is not sensitive to the initialization of the
model. The equilibrium totals of the population, housing and industry units are
constrained by the available land. However with a different initialization the
distribution of the population, housing and industry units over the zones and the
mixture in the zones will change.

4.1 The base behavior

The aggregate behavior
The equilibrium totals of the population and housing are equal and the total number of
industry units is lower relative to the results of the Forrester model (figure 12). The
latter is due to the mixture the industry sector. The service industry has the same
characteristics (higher number of jobs per industry unit) as the new enterprises in the
basic industry. This means in the current model a lower number of industry units offer
the same amount of job opportunities in the city.
Population Housing Industry
8M 1M 45,000
800,000 3 3
2 1 2
1 1 1 2 31 2 12
3 2 2 12 12 12 750,000 2 12 12 12 12 12 12 1 33,750
1 2 2
2 1 12 12 2 2 2 2 2
1 3 1
4M 1 1 1 1 1
400,000 500,000 1 22,500 2
1 1
3
2 12
1 250,000 11,250 1 2
1
0 3 23
3 312 2 12
0 12
12 1
12 2 12
0 25 50 75 100 125 150 175 200 225 250 0 1 2 0 1 2 12

Time (year) 0 25 50 75 100 125 150 175 200 225 250 0 25 50 75 100 125 150 175 200 225 250
Time (year) Time (year)
P Population : Sanders 2004 1 1 1 1 1 1 1 1 1 1
"P Population Forrerster (1969)" : Reference 2 2 2 2 2 2 2 HUT Housing units total urban area : Sanders 2004 1 1 1 1 1 1 PUT Productive units total urban area : Sanders 2004 1 1 1 1 1 1
P Population Rotterdam : Reference 3 3 3 3 3 3 3 3 3 "H Housing Forrerster (1969)" : Reference 2 2 2 2 2 2 2 "I Industry Forrerster (1969)" : Reference 2 2 2 2 2 2 2

Figure 12: The base behavior in comparison with Forrester (1969)

The behavior type (growth, overshoot and stagnation) for population can be observed
in the City of Rotterdam (line 3 in the first graph of figure 12). The overshoot of the
spatial urban dynamics model is very small compared to the Forrester model and the
city of Rotterdam (after WO II, there is second growth period because of new
housing-types and new available land). However until WO II the growth pattern of the
city of Rotterdam seems to be more consistent with spatial urban dynamics model
than with original model. The match between real data and the model should not be
emphasized and no conclusions should be drawn from this information; it is given as
an illustration of the pattern of development of real city.
The reason that the overshoot is smaller than in the Forrester model is that it takes
place in every zone and on different points in time (see figure 17-19). For example,
the land in zone 5 fills up and an overshoot can be observed in zone 5. This overshoot

20
is compensated by the growth that starts zone 6 and 1 (figure 17). Basically, the
limited area in a zone does not impact the development of the total city very much
because it is balanced by the growth in areas where land is still available.

The behavior of this complex model is not easy to analyze and explain. The reader
must keep in mind that only the flows can change the stocks (housing, population and
business structures). If growth takes place in for example the premium-housing sector
this means that the construction rate is higher than the filtration rate. The causes for
this situation can be found by examining the equations or by examining the behavior
of the variables after simulation. This strengthens the author in his belief that an
interactive learning environment (Davidsen 2000) would be an efficient tool in
explaining the observed behavior in this model.
In an interactive learning environment (ILE), the user can be directed to the important
structures to understand the behavior in an automated fashion. The attention of the
users is led to the causes of the dynamic behavior. The interactive learning
environment forms an environment in which the users of the ILE can execute policy
testing and experiments in a controlled manner.

The zonal behavior

In zone 5 and 9 most of the early growth takes place because the construction of
housing depends on zonal conditions that are higher in these zones than in the rest of
the zones. The growth of the basic industry depends on the citywide conditions,
causing the housing construction to spread outwards to other zones as zone 5 and 9
fail to provide the necessary labor and housing. The expansion and the growth of
housing, due to the industry expansion, introduce delays in the total development
process.
Premium housing [i] The number of premium
25,000 housing in zone 5 and 9 peak
around year 30 (figure 13).
18,750 2
After that the inner-city zones
4 peak every 10-15 years,
3
8 beginning with zone 6 and 10.
12,500 34 7 8 3
4
7 8
2
3 4
In the zones where basic
2
5
7
8 7 industry is present the
6,250 1
6 2 premium housing almost
1
disappears (except of zone 2
0 4
1 2 34 6 78 2 3 5 56 1 56 1 56 1 and 14). The assumption that
0 25 50 75 100 125 150 175 200 225 250
Time (year) the location attractiveness
multiplier is less strong than
Figure 13: The base behavior for premium housing for worker housing and
underemployed housing,
causes the premium housing to distribute mainly in outer zones (without basic
industry). This assumption implies that people (managers) who live in premium
housing are willing (and have means) to travel longer distances to their job than
people in the other housing types.

21
 Worker housing [i] Worker housing gets
35,000 established in all zones
5
2 because of the stronger
26,250
6 7 location attractiveness
8
1
4
2 1 1
multiplier and the filtration of
3
34
6
7 8
2
3 4
56
7 8
2
3 4
premium housing.
17,500 5 1 5

6
The reader should note that
5
the overshoot and stagnation
8,750
1
behavior in the Urban
7
8 Dynamics model can be
0
1 234 6 78 2 34 observed in every zone (figure
0 25 50 75 100 125 150 175 200 225 250
Time (year) 13-15).

Figure 14: The base behavior for worker housing The development of the
underemployed housing is
more stable (figure 15). The volatile conduct as observed in the premium housing and
worker-housing sector is more dampened. This is rooted in the filtration process, as
the filtration from premium housing to worker housing is expressed as the number of
premium housing multiplied with a fraction. The variation in the level of worker
housing becomes less volatile because the flow of premium housing to worker
housing dampens the variation effect (in this specification).
Underemployed housing[i] The distribution of
35,000 underemployed housing over
5 5 the zones is originated in the
6
26,250
competition for available
1 5
1 6
1
56
1
land. In the non-industry
2
6
34
7
8
2 7 2 zones, the service industry
17,500 7 3 3
8
4 8 4
and the two other housing
types (worker and premium
8,750
housing) are established and
5 2
4 sustained sooner.
3
0 78
1 234 6 78 1 2 34
0 25 50 75 100 125 150 175 200 225 250
The basic industry is not very
Time (year)
interesting to discuss in detail.
Figure 15: The base behavior for underemployed housing The growth is basically
specified in the same way as
in the Forrester model, because no location attractiveness multiplier is included and
the growth is influenced by citywide factors.

22
 The service industry on the
SI Service industry i[i]
600
other hand shows more
interesting conduct (figure 16.
7
In equilibrium condition the
450 5

2
7 2 8
2
3 distribution of the service
78
2
8
3
4
4
units over the zones is
300
5
3
4
6 roughly proportional with the
6 5
4 6 1
5
5
6
5
6
zones location attractiveness,
150
1
3
1 1
as the total number of
households is evenly
1
0
1 23 4 67 8 23 4
78
distributed over the zones.
0 25 50 75 100 125
Time (year)
150 175 200 225 250
The second mechanism that
contributes to the
Figure 16: The base behavior of the service industry development of the service
industry is that zones 1,5 6, 9,
10 and 13 are already filled with other functions (basic industry and housing) and
therefore no land is available in the zone.

The different classes of population are distributed over the different zones by the
attractiveness principle. As one zone becomes more attractiveness more residents will
migrate to this zone. The attractiveness of a zone is a combination of the housing
availability, social mixture and the accessibility to jobs.
The dispersal of residents over the zones in the equilibrium state is proportional to the
distribution of the matching housing sectors (premium housing for managers, etc),
corrected for the zones accessibility to jobs and social mixture. The behavior has the
same trajectory as the development of the housing types, because the jobs develop in
a stable manner.

In figure 17-19, the reader can observe that the distribution of the persons is different
from the distribution in the housing sector (figure 13-15) for the reason explained
above.
MP Manager[i] L Labor i[i]
20,000 70,000
1
2

15,000 52,500 6
2 7
6
10,000 35,000 6 1 56 1
78 3 7 2 6
7 5 1 5
4 23 5
8 23 8 5 3 2
4 4 4 8 7 2
3 4 7
5,000 1
17,500 8
5 2 3 3
4 8
4
7
8
0 1 5 0
234 56 1 34
1 2 34 6 78 56 1 56 1 1 2 34 6781 234 78
0 25 50 75 100 125 150 175 200 225 250 0 25 50 75 100 125 150 175 200 225 250
Time (year) Time (year)

Figure 17: The base behavior of managers Figure 18: The base behavior of labor

23
 U Underemployed i[i]
The reader can judge the effect of the
60,000 job accessibility on the movement
5
6
behavior, if he compares the order of
45,000 5
56
56
the zones in figure 17-19 with the
1 1
1 1 order of the zones in figure 13-15. For
30,000 6
2
2
example in figure 17, the order of the
15,000
7
7 7
2
zones is 7,2,3,8,4,2,1,6 and finally 5.
3
5
2
4 8
3
4 8
3
4
In figure 13 the order is 7,8,3,4,2,1,6
0
1 2 34 6 78 1 2 34 78 34
8
and 5, meaning that zone 8 has the
0 25 50 75 100 125
Time (year)
150 175 200 225 250 second highest number of premium
houses. If the managers were
Figure 19: The base behavior of underemployed distributed proportionally with the
premium-housing zone 8 would have
the second highest number of managers, which is not the case. Zone 8 has less access
to manager jobs and is consequently less attractive for settlement.

Notes on the overall behavior

The moment that the growth of basic industry is equal for all the zones where it is
allowed (zone 1,5,6,9,10,13), the overall growth of basic industry is explained by the
citywide conditions (like in the Forrester model). It should be noted that it takes time
before zone 1,6,10 and 13 match the growth of 5 and 9, because of the higher initial
values in zone 5 and 9.
The total behavior of the model is quite insensitive to changes in a single table
function or parameter. The cause is the assumption of the fixed area that limits both
the total numbers of industry and housing. However the industry and housing sector
are tied together by the need of people to have both a house and a job. The fact that
one zone can compensate for another zone causes the net cross boundary movement
(citywide arrivals and departures) not to be very different from the original Forrester
model.

Policy implications
The natural tendency of the city toward imbalance in which housing dominates
industry might be corrected by urban policies that encourage industry and that
discourage the construction of excess of worker housing. This policy is called
‘Encouraging industry’ in the Urban Dynamics. This policy is selected because it
must be considered in the context the particular area to be revived (Forrester 1969).
The insights of the explicit spatial dimension could be very useful as Forrester
suggests using this policy on a local basis.

The policy consists of encouraging industry (increasing the new enterprise

construction rate form 5 to 7% per year) and discouraging the construction of excess
worker housing (slum-housing demolition 5 % per year) in all zones. The results of
the Forrester model and the spatial urban dynamics are summarized in table 2. T = -5
year is the value of the model in equilibrium 5 years before the policy is applied (T =
0).

24
 Table 2: The changes caused by the policy described above. (T = year)
Urban dynamics Spatial urban dynamics
Variable Symbol T = -5 T = 50 Change(%) T = -5 T = 50 Change(%)
New enterprise NE 4900 8000 63% 3400 7200 111%
Mature business MB 7800 12000 64% 5400 10500 94%
Declining industry DI 16500 22200 35% 11400 15400 35%
Service industry SI ----------- ----------- ----------------- 5300 5300 0%
Premium housing PH 110900 152800 38% 113400 161000 42%
Worker housing WH 335600 450600 34% 311600 425500 37%
Underemployed housing UH 310100 175300 -43% 314100 173600 -45%
Managerial-professional MP 71100 108700 53% 73000 109600 50%
Labor L 392600 600000 53% 381000 587200 54%
Underemployed U 377300 335900 -11% 378900 331400 -13%
Underemployed to labor net UTLN 5500 9200 67% 5600 9700 73%

The table shows that the effect of the policy on overall behavior of the city does not
change very much in comparison to the Forrester model. It should be noted that the
model has not reached its equilibrium state yet (at T=50) after applying the policy. T
= 50 is included in order to make comparison with the Forrester model possible.
The most significant change is the 111% change in new enterprises. In the spatial
urban dynamics model there is a relatively small portion of new basic industry due to
the addition of the service industry sector. In this policy, the land that becomes
available due to the slum housing demolition is used to construct new basic industry
(and not for service industry). The absolute growth of the total basic industry sector
(new enterprises, mature business and declining industry) is not very different from
the Forrester model (13.000 units versus 12.900 units in the spatial urban dynamics
model). In conclusion, on an aggregate level the spatial urban dynamics model
supports the policy recommendation of the Forrester model.

The benefit of the explicit spatial dimension lies in the fact that the reader can obtain
qualitative insights. The model provides results for the various zones. In figure 20, the
results of the policy for the housing sector in zone 1-8 are given. The policy is applied
after year 250 because the spatial urban dynamics model is in equilibrium then. The
graphs show how the policy changes the equilibrium state.
Underemployed housing[i] Worker housing [i]
25,000 35,000
5 1
1 5 6
6 1
6 5 56 1 5 6 1
18,750 2 28,750
3
4 1 2 2 2
2
7 1 7 7 7
3 4 8 34 8 3 4 3 4
12,500 22,500 7 8 8
8 5 6 5 6 56 1 2
1
2 1 78 2 3 7 8 1 2 3 45 6 7 8 1 23 4 6 3
4
3 23 4 4 5
4 7 8 2 78
34
6,250 16,250

0 10,000
240 250 260 270 280 290 300 310 320 330 340 350 240 250 260 270 280 290 300 310 320 330 340 350
Time (year) Time (year)

25
 The results tell the following story.
Premium housing [i]
18,000 First, the reader should remember that
7 3 4 78 34 7
8 3 4 7
8 3 4

the basic industry is only allowed in

8
3 2
4 2 2 2
13,500
3
4
78 2zones (1,5,6,9,10 and 13). These zones
9,000
2 are characterized by mixture of
industry and housing biased towards
4,500 underemployed housing. Naturally, the
industry growth will occur in these
0
1 56 1
zones. The unfavorable balance (old,
5 6 1 5 6 1 56 1 5 6 1
240 250 260 270 280 290 300 310 320 330 340 350
Time (year) low-income housing and basic
industry) in the industry zones is
Figure 20: The effects of policy 1 on the housing sector corrected by this policy, creating a
better balance between the housing
types in the zones (figure 20). This improving balance is the opposite process of urban
decay.

The policy at a zonal level

In the above description, the policy is applied to all the zones in the city. It is
thinkable that in real life a policy is not applied to the whole city but only to the part
with problematic behavior. The spatial urban dynamics model allows testing the
implications of this fact. If the policy is only applied to zone 5 and 6, the fraction of
underemployed relative to the total population in those zones decreases. The fraction
of the underemployed in the other zones will increase (figure 21).
Fraction underemployed in zone i Fraction underemployed in zone i
0.8 0.8

8 3 4 8 34
34
0.65 0.65 78 7 7
5 1 1 1 1 56 34
6 1 7 8
1 1
2 2 2 2
2 7 7 7 1 3
2 7 2
0.5 3 8 3 8 3 0.5 4
3 8 7
7 4 4
3 8 4 4 2
8 5
3 45 3 6
8
4 6 4 12
0.35 0.35 56 12 5 6 1
2
5 6
2
1 5
56 5 6 5 6 5

0.2 0.2
240 250 260 270 280 290 300 310 320 330 340 350 240 250 260 270 280 290 300 310 320 330 340 350
Time (year) Time (year)

Figure 21: The fraction of underemployed in zone i Figure 22: The fraction of underemployed in zone i

The total system is slightly better off as the decrease in zone 5 and 6 is not fully
compensated by the increase in the other zones (the fraction of underemployed in the
total city is 53.33% versus 52.40% after applying the policy).
If the policy is applied to the zones where basic industry is allowed (zone
1,5,6,9,10,13 and 2,14) interesting results can be observed (figure 22). The zones
where the policy is not applied compensate (and become problematic) for the
diminished fraction of underemployed in the zone where the policy is applied. In the
total urban area the fraction of underemployed falls from 53.33% to 48.54% after the
policy becomes operative.
The results described above illustrate the additional insights that the spatial urban
dynamics model can provide relative to the Forrester model. Although the policy
leads to a better situation for the total urban area, under particular circumstances the
problematic behavior reoccurs in other zones within the city (figure 22). This may
cause the following situation when the insights of the Forrester model are used on a
zonal level. The policy is applied to a number of zones with success; the problematic

26
behavior resurfaces in different zones in the city. The attention of the city shifts to
these zones and the problematic behavior could reemerge in the original zones.

5 Discussion
In the previous section it is outlined that the citywide totals of population, housing
and business structures in steady state are close to the original Forrester model
(Forrester 1969). The trajectory of reaching this equilibrium condition is more stable.
There is only a small overshoot in the citywide behavior, because of the division in 16
zones. The overshoot is visible in each zone on different points in time, but is
compensated in the total growth of the city by the growth in other zones.

The division into the 16 zones allows for the reader to develop further insight and
additional policies can be tested with this model. The internal dynamics of cities can
be illustrated in a comprehensive manner without the model becoming too widely and
losing its transparency. The current model shows the reader the different roles that the
zones represent in the city. Zone 5 is a mixture of mainly basic industry and
underemployed housing. This can be seen as the core of an old industrial town. The
author feels that the division into zones has more appeal to the users of the model
because it can illustrate why a policy is not effective on the citywide level when
applied to one zone. Moreover, personal knowledge and interest is likely to be on a
zonal level, so the model can facilitate effective learning.
The latter claim is strengthened by the possibility to show problematic behavior on a
zonal level. Urban planners can learn how to manage the urban system by observing
the interactions of people, housing, industry and the different zones. Policies trying to
counter problems in one district in a town (as still can be observed in urban
management) have little effect because of the interactions with other zones. A
systematic approach of policies applied in a set of districts appears to be necessary.

The author believes that the inter-zonal interactions are undervalued in comparison to
cross-city boundary interactions in the original Forrester model. This claim is
supported by evidence presented in section 1 (table 1) and by the policy insights
presented in this work.

The model is a representation of a generic theory of urban development and

consequently does not focus on numerical data. The conclusions of the model are
generic too. In order to draw conclusions for a specific city, one must customize the
model to this city.

6 Conclusions
The result for the citywide behavior (S-shaped growth and stagnation) in the model
presented in section 3 is slightly different from the Forrester model. However, on a
zonal level the conduct (growth, overshoot and stagnation) of the Forrester model is
still present and the overshoot is even larger. The poor long-term equilibrium
conditions of the Forrester model can also be observed in the spatial urban dynamics
model.
The additional level of detail in the spatial component of the model facilitates the
building of deeper understanding of the complexity of the interactions in a city,
because the user can learn about the different effect of policies when applied on a

27
zonal level or on a citywide level (section 4). Regarding, the use of the policy of
encouraging industry on a zonal level the following is concluded. The problematic
situation one wants to alter in a particular zone can resurface in other zones within the
city if the ‘encouraging industry’ policy is used.
In conclusion, the policies recommendations of Forrester are still valid in a
disaggregated model when applied to the complete city. However when the policies
are applied to only part of the city, poor conditions can develop in the other zones in
the city. A zonal applied policy could be interpreted as successful if the side effects
are not judged as structural consequences of this policy.

Possible further developments of this model can be divided in two groups. First,
changes in the model that challenges the theory on which the model is built. A good
suggestion is to make the fixed area constraint flexible. The industry growth could be
made dependent of the demand for goods (in contrast to the current model in which
the available land and labor are the constraints for the growth). The urban spread
could be modeled in a more satisfactory manner. A generic approach is suggested by
Despotakis (Despotakis and Giaoutzi 1996).
The second group of developments deals with the question how to explain the results
to users. The author proposes the development of system dynamics based learning
environment as advocated by Pål Davidsen (Davidsen 2000). This is very interesting
for educational purposes emphasizing both the management of complex systems and
the principles of urban planning (Erkut 1997). An example could be to visualize the
growth of the city in a spatial dimension.

The model, like every model, has shortcomings. An explicit representation of the
transportation sector would help to build confidence in the theory of this model. The
current model is capable of representing a city with its surrounding suburbs. However,
this feature is not used to its full potential.
As pointed out by Peter Allen (Allen 1997) it should be noted that this model is
deterministic. Allen acknowledges the value of system dynamics models in urban
planning. His models are also built form differential equations with a more prominent
role for stochastic functions and variation.
A third remark is the absence of empirical data. This devaluates the results in the eyes
of certain groups of users (with a strong appreciation for models that match numerical
data). A solution is the application of the model to a real city.

The system dynamics methodology proves to be flexible and appropriate in analyzing

urban systems, despite the fact system dynamics has never become an established
method in the field of urban planning (Alfeld 1995). The flexibility is made clear in
the development of the current model and the omissions of the Forrester model in the
eyes of its critics have been included to a large extend.

Note
The reader can contact the author at peter.sanders@ifi.uib.no for the complete list of
model equations or check the supplementary file in the proceedings of the 22nd
International of the System dynamics society for the model in Vensim format.

28
References
Alfeld, Louis E. 1995. Urban dynamics- The first fifty years. System Dynamics
Review 11 (3):199-217.

Alfeld, Louis E., and Alan K. Graham. 1976. Introduction to Urban Dynamics.
Cambridge MA: Productivity Press.

Allen, Peter M. 1997. Cities and Regions as Self-Organizing Systems : Models of

Complexity. Vol. 1, Environmental Problems and Social Dynamics series.
Amsterdam: Gordon and Breach Science Publishers.

Argryris, C., and D. Schön. 1978. Organizational learning: A theory of action

perspective: Addison-Wesley.

Babcock, D. L. 1972. Assumptions in Forrester's Urban Dynamics Model and their

Implications. I. E. E. E. Transactions on System Man and Cybernetics (April).

Blijie, H.P, M.A. de Bok, and F. M. Sanders. 2003. De invloed van transport op het
woonlocatiekeuzegedrag van huishoudens: Introductie en bepaling van een
zoekgebied voor verhuizende huishoudens. Paper read at Transportation
Planning Research Colloquium, at Antwerpen.

Burdekin, R. 1979. A dynamic spatial urban model: a generalization of Forrester's

Urban dynamics model. Urban systems 4:93-120.

COS. 2004. Cijfers over Rotterdam [Internet]. www.cos.nl 2004 [cited 2004].
Available from www.cos.nl.

Davidsen, Pål I. 2000. Symposium Issue: System Dynamics and Interactive Learning
Environments, Part 1 - Issues in the Design and Use of System-Dynamics-
Based Interactive Learning Environments. Simulation & gaming 31 (2):170.

Despotakis, Vassilios K. , and Maria Giaoutzi. 1996. Spatial Modeling of Urban

Dynamics. Paper read at 1996 International System Dynamics Conference, at
Cambridge, Massachusetts.

Erkut, Gulden. 1997. The Use of Systems Thinking and System Dynamics in Urban
Planning and Education. Paper read at 15th International System Dynamics
Conference: "Systems Approach to Learning and Education into the 21st
Century", at Istanbul, Turkey.

Ford, Andrew. 1999. Modeling the Environment: An introduction to System Dynamics

Modeling of Envrionmental Systems. Washington: Island Press.

Forrester, Jay W. 1969. Urban Dynamics.

Forrester, Jay Wright. 1971. Counterintuitive Behavior of Social Systems.

Technology Review 73 (3):52-68.

———. 1980. Information Sources for Modeling the National Economy. Journal of the
American Statistical Association 75 (371):555-574.

29
Fournier, Serge. 1986. Dynamics of the urban growth of an average french city: the
CARPE model. In Advances in urban advances modelling.

Garn, H. A., and R. H. Wilson. 1972. A Look at the Urban Dynamics. I. E. E. E.

Transactions on System Man and Cybernetics (April).

Graham, Alan K. 1974. Understanding Urban Dynamics: An Analysis of Garn's

Interpretation. In Readings in Urban Dynamics, edited by N. J. Mass.
Cambridge MA: Productivity Press.

Gray, J. N., D. Pessel, and P. P. Varaiya. 1972. A Critic of Forrester's Model of an

Urban Area. I. E. E. E. Transactions on System Man and Cybernetics (April).

Hannon, Bruce M. , and Matthias Ruth. 1997. Modeling Dynamic Biological Systems.
New York: Springer-Verlag.

Jaeckel, M. T. 1972. Forrester's Urban Dynamics: a Sociologist's Inductive Critique. I.

E. E. E. Transactions on System Man and Cybernetics (April).

Kadanoff, L. P., and H. Weinblatt. 1972. Public Policy Conclusions from Urban
Growth Models. I. E. E. E. Transactions on System Man and Cybernetics
(April).

Laudet, Claire, and Serge Fournier. 1979. La simulation de politiques urbaines.

These de Doctorat, U.E.R. de Sciences Economiques, Universite Paris X -
Nanterre, Paris.

Mass, Nathaniel J., ed. 1974. Readings in Urban Dynamics. 2 vols. Vol. 1.
Cambridge MA: Productivity Press. Original edition, Wright-Allen Press and
MIT Press.

Ommeren, J. van. 2000. Commuting and relocation of jobs and residences.

Alderschot: Ashgate Publishing Ltd, UK.

Paich, Mark, and John D. Sterman. 1993. Boom, Bust, and Failures to Learn in
Experimental Markets. Management Science 39 (12):1439-1458.

Rochberg, Richard. 1972. Some questionable assumptions in the model and method
of Urban Dynamics. In Urban dynamics: Extensions and reflections, edited by
K. Chen. San Francisco: San Francisco Press Inc.

Rotterdam. 2003. Rotterdam zet door- Op weg naar een stad in balans. Rotterdam.

Schroeder, Walter W., III. 1975. Urban Dynamics and the Suburbs. In Readings in
Urban Dynamics, edited by W. W. Schroeder, III, R. E. Sweeney and L. E.
Alfeld. Cambridge MA: Productivity Press.

Schroeder, Walter W., III, Robert E. Sweeney, and Louis E. Alfeld, eds. 1975.
Readings in Urban Dynamics. 2 vols. Vol. 2. Cambridge MA: Productivity
Press. Original edition, Wright-Allen Press and MIT Press.

Sterman, John D. 2000. Business Dynamics: Systems thinking and modeling for a
complex world: McGraw-Hill companies.

30
Wilson, A.G. 1974. Urban &regional planning models in geography & planning.
London: John Wiley & Sons.

Appendix A
In this appendix, a detailed description of the stock and flow diagrams and equations
is given. The section looks at the attractiveness principle and the migration of persons.

The attractiveness principle and migration of persons

U Underemployed i
UA Underemployed (12)
arrivals i
(13)

UAN Underemlpoyed
<L Labor i> UD Underemployed
arrivals normal UDN Underemployed
departures i departures normal
(14)

UDM Underemployed
AMMP departure multiplier AMF Attractiveness
Attractiveness for for migration factor
migration multiplier CAMMP Change
perceived in AMMP
UAMM Underemployed UHPM Underemployed
AMM Attractiveness
arrivals mobility housing program
for migration multiplier
multiplier multiplier
AMMPT Attractiveness for
migration multiplier UJM
UHM
perception time Underemployed/job
Underemployed/housing
multiplier
multiplier PEM Public
expenditure
multiplier

Figure 23: Migration out and to an area (selection of the underemployed view of the model)

The population is divided in three groups: unskilled workers (U), skilled workers (L)
and management and professional workers (MP). The attractiveness principle works
in the same way for the three groups. For the simplicity, the other flows influencing
the U Underemployed namely UB Underemployed births, social migration to and
from labor are omitted in figure 23 (the model includes these flows).
U Underemployed i = INTEGRAL (UA-UD) (12)
UA = Underemployed arrivals i (person/year)
UD = Underemployed departures i (person/year)

The UA underemployed arrivals are determined by a normal arrival rate multiplied

with the underemployed and the attractiveness of the area as perceived by the
environment.
UA Underemployed arrivals i = (U+L)*UAN*AMMP (13)
U = Underemployed (person)
L = Labor (person)
UAN = Underemployed arrival normal (fraction/year)
AMMP = Arrival migration multiplier perceived (dimensionless)

The UD underemployed departures are expressed by a normal departure rate

multiplied with the underemployed and the attractiveness of the area compared to the
environment as perceived by the persons in the city.
UD Underemployed departures i = U*UDN*UDM (14)
U = Underemployed i (person)
UDN = Underemployed departures normal (fraction/year)
UDM = Underemployed departures multiplier (dimensionless)

31
UDM underemployed departures multiplier in equation (14) is the inverse of the
attractiveness multiplier that controls the UA underemployed arrivals.

The attractiveness factor for internal migration (AIMLM, (16) in figure 24) is the
product of the housing availability (LAHM, (17)), the social mixture (LAUM, (18))
and the accessibility to jobs (LLAM, (19)). The latter denotes how easy it is to
participate in the job market and therefore stands for the job opportunity force
mentioned above. The formulation is based on equation (1) in section 3.
<LFS Labor <WH Worker
AIMLMPT Attractiveness for family size> housing i>
internal migration labor
<L Labor i>
multiplier perception time k
<LT Labor total>
AIMLMP
HLR Housing/labor <WHPD Worker
<L Labor k> Attractiveness for
<L Labor i> ratio i housing population
internal migration CAIMLMP Change density> <U Underemployed
labor multiplier in AIMLMP
perceived k i>

LAHM Labor arrival

housing multiplier i (17)
LIBZM Labor internal LUR
AIMLM Attractiveness for
between zones (20) Labor/underemployed
internal migration labor
movement i k ratio i
multiplier k LAUM Labor arrival
underemployed
AIMLM Attractiveness for multiplier i (18)
internal migration labor <L Labor i>
LIOMN Labor internal multiplier i
out-of-zone movement (16)
normal
LLAM Labor locational LCZA Labor
attractiveness multiplier comparison zone i to
(19) i average

AAALJAZ Average
accessibility attractiveness
AALJZ Accessibility
to labor jobs all zones
attractiveness to labor
jobs zone i

LJ Labor jobs k

<LJ Labor jobs i>

<TDF Trip
distribution function
i>

Figure 24: The attractiveness concept and the inter-zonal migration

LIBZM Labor internal between zones movement i k (20 in figure 24) is based on the
formulation in equation (4) in section 3, in this equation the perceived attractiveness
of k has a shorter time lag to represent the diffusion of information. This reflects the
fact that it is quicker to access information about zones within the city than it is to
access information about other cities.

The net zone movements are expressed as the net difference of the migration to a zone
and the migration out of a zone. The migration between the different zones is
formulated as suggested above in equation (4).

32

Back to the Top