able of Contents
The Dynamic Urban Model: Transport and Urban
Development
John Swanson
Steer Davies Gleave
28-32 Upper Ground, London
United Kingdom
SE1 9PD
Tel: +44 20-7919-8500
Fax: +44 20-7827 9850
Email: j.swanson@ sdgworld.net
Abstract
This paper describes a model built to simulate the interaction between transport, land-
use, population and economic activity in an urban area. It provides a brief discussion
of how the need for such a model arose, and of other modelling methods that have been
used to address the problem, touching on the contribution of Forrester’s original Urban
Dynamics.
It describes how Forrester’s model has been adapted and extended to allow a zonal
representation of an urban area, coupled with a full representation of road and public
transport networks allowing choice of mode and route. A discussion is provided about
how hierarchical logit models are used to model choices by travellers on the networks.
Finally brief descriptions are given of four applications of the model, three of them
consultancy assignments carried out by Steer Davies Gleave, and the other a
hypothetical town that has been used to develop and test the model.
Transport and economic activity in urban areas.
The question of whether transport investment contributes towards economic well-being,
or more frequently, towards economic regeneration, in urban areas has long taxed
economists and transport analysts. The claim is often made in support of proposed
transport schemes in the UK: the investment is needed to boost the local economy, to
create jobs, to regenerate the area. In fact while the idea that transport should be related
somehow to economic activity seems grounded in common sense, the evidence for the
nature of that relationship is far from clear. A UK government advisory committee
produced a very substantial report on the topic in 1999! drawing on research from
around the word, and concluding that, at least in developed economies with existing
transport systems, the case was not proven. ‘Necessary but not sufficient’ seems to be
the most that can be said about transport investment; some authors even question
whether it is necessary.
1 “Transport and the Economy”, The Standing Advisory Committee on Trunk Road Assessment
(SACTRA), Department of Environment, Transport and the Regions, 1999.
The topic has of course been a rich source of research and modelling work over the
years. Most writers agree that urban economies are complex ‘systems’, but then do not
go on the use system models to try to study them. The analytic framework tends to be
grounded in equilibrium economics and techniques, even when it is apparent that the
problem is essentially dynamic’. The earliest large scale transport models, known as
four- stage models and still in use, were equilibrium models, operating in four sequential
steps: trip generation (estimating how many trips each household will make);
distribution (working out where they will go); mode choice (car, public transport etc);
and assignment, in which the trips are assigned to routes through the networks. An
iterative procedure is used to cycle round until equilibrium, or something close to it, is
reached.
Land-use and transport interaction models (LUTI models) take this even further. They
usually comprise a transport model, with some or all of the four stages, and a land use
model that attempts to allocate households and economic activity to each zone in the
model in response to changes in the network, and then to describe how conditions on
the transport network will change in response, and so on. Typically the procedure will
be something like: describe the land use patterns in the starting year; run the transport
model until it reached equilibrium; revise the land use patterns in the light of the new
transport conditions in some subsequent year; run the transport model until it reaches
equilibrium for that year; revise the land-use patterns, etc.
These models are usually cumbersome and expersive to set up, requiring large amounts
of data. They run slowly, and may not in fact reach equilibrium. Their developers often
stress the ever- finer segmentation and micro-detail these models? provide, but for all
this their reputation is not very high.
For system dynamicists the objection to such models is not that just they are technically
inefficient, but that they seem to fail to address the central issue: that our major towns
and cities are in states of constant flux, the product of many interacting forces acting on
short and long time spans with feedbacks also operating on many time scales. Their
dynamism arises because of tensions caused by non-equilibrium, and it requires a
dynamic model, a system dynamic model, to describe the processes. This is what the
Dynamic Urban Model attempts to do.
System Dynamics and Urban Economics
Forrester’s Urban Dynamics is well known in the SD community. His model described
the growth and eventual stagnation of an urban area over very long time periods of up to
200 years, but it actually said little about transport. There was an implicit assumption
that an adequate degree of accessibility existed in the simulated city, but nothing
? As for example in ‘Transport Investment and Economic Development’ by David Banister and Joseph
Berechman, UCL Press, London, 2000, page 38. After explaining numerous ways in which travel
behaviour can change dynamically, they then say: “ This discussion implies that, in modelling the effects
of transport improvements on the local economy, it is necessary to carry it out within an equilibrium
framework, so that the changes in total demand for infrastructure facilities, resulting from changes in
travel patterns and rates and in spatial location, are equilibrated with network supply.”
3 For instance, we were given access to a LUTI model built recently that had over 30 household
categories and 24 types of employed resident. These models are often strong on cross sectional detail like
this, but weaker on structure.
explicitly stated. It was essentially a two-zone model: the city itself, and the outside
world.
Nathan Forrester has taken the idea further, by constructing a zonal model with
transport costs between the zones represented by a function of the straight-line distances
between them. Congestion arose as the numbers of movements increased; while each
zone was treated rather like its own example of the original Urban Dynamics city.
The Dynamic Urban Model (the DUM) described here borrows many ideas from
Forrester’s original. The idea of ‘attractiveness’ as a location in which to live or do
business remains, and many of the mechanisms used can be traced back to that model.
However where the new model gains is:
¢ It allows the modelled area to be divided into zones, each of them stocked with
people, houses, businesses, etc;
¢ It provides a full representation of the transport networks in a manner that is
consistent with more traditional transport modelling. This includes software
written to convert transport network models built using commonly used
transport packages into a format suitable for the DUM;
¢ It makes use of a hierarchical logit structure to handle choices of travel mode,
route, and willingness to travel at all;
¢ It allows multiple classes of households, people, houses, business sectors etc,
each with their own preferences;
* It is set up to read initialising data describing the urban area from an extemal
Excel database.
The DUM has been built to be as general as possible, so that it can readily be adapted to
new applications.
A Model of Transport and the Economy
If transport is to impact on the economy of a region or urban area, then it must be via
the patterns of accessibility it permits, in that it allows movements of people or goods
within economically acceptable times and costs.
Several types of accessibility might be expected to apply:
¢ For the workforce, there is access to employment, or more strictly to
employment that matches skills;
« For employers there is access to a workforce, which affects ease of recruitment;
and
« Access to markets and suppliers, both within the modelled area and beyond it.
In general we might say that the longer or more expensive a trip, the fewer people will
be willing to undertake it. This gives rise to the idea of a deterrence function, relating
the cost of travel to willingness to undertake the trip. Figure 1 shows a deterrence
function calibrated using the DUM in an application in Merseyside, in England. It
relates to travel to work, and shows how the proportion of people willing to commute
falls as the commute time increases. (In fact the implementation of this curve is via a
hierarchical logit formulation, discussed further below.)
Figure 1: A deterrence curve
1
0.8
0 10 20 30 40 50 60
Drive time (minutes)
Given a deterrence curve and information about the costs and times of travel between
each pair of zones, then the workforce willing to access a given zone, i, is given by:
Accessible workforce , => workforce, .f (cost of travel between zones i and j)
J
Here f() is the deterrence function that retums the proportion of people willing to travel
given the cost involved’. This is simply the workforce in each of the other zones
multiplied by the proportion willing to travel to it®. Similarly the number of jobs
accessible from a given location is given as:
Accessible jobs, => Jobs, - f (cost of travel between zones i and j)
Access to markets and supply chains is rather more awkward. For the retail sector,
customers are members of the public, so one obvious measure is the number of people
living within accessible range of a site, calculated in a similar way as above. More
generally however businesses do business with each other, and business customers and
suppliers may be located within the modelled area or outside it.
For each location the numbers of accessible businesses can be calculated in a similar
way as above, given a deterrence function. Connections with the outside world are
often important, but the model, by definition, does not know anything about the world
beyond its own geographic boundary, and for the time being all that can be done is to
include large zones whose role is to represent the world beyond the region of interest.
* Some measures of that cost are discussed below
5 Travel within a zone is a slightly special case, for here we will not usually have a representation of the
transport networks. However it is still possible to calculate accessibility functions within a zone, and the
same principal applies.
More information about how transport costs are calculated is given further below, but
using this simple notion of accessibility, we can now look at some of the main
components of the model.
Figure 2 illustrates some of the primary linkages in the model.
Figure 2: Some linkages in an urban area
workforce
elsewhere
. el tin a
Jobs in other
locations “access to
\ markets and
transport investment are |
NO (tet
transport activity NA ") chain
workforce
ae )
* accessible
vs a ——
+ attractiveness net inward ‘Skiled attractiveness as
asplacetolive + migration + Lworkfored business location
+
housing
availablity
net business
startups
Businessoh
in area
availability
of units
Housing
ook dy housin
consrucion & seal a
See tJ Land available fdr
~~ vet [> consinction
Tee f
land-use
policy
Transport is represented here as a stock, right at the top of the picture. If this stock is
added to, travel times (more generally, the generalised cost of travel) will fall, so that
the number of accessible jobs, from the point of view of residents, will increase.
Similarly, access to markets and supply chains will tend to improve. Both of these then
affect the attractiveness of the town as somewhere to live and to do business, and
consequently the net migration rate and the business start- up rate.
Attractiveness as a place to live is also affected by the availability of housing, and of
course if the migration rate increases the availability of housing will fall, unless more is
built. Similarly from the point of view of businesses, attractiveness is considered to be
affected by access to a workforce, to markets and supply chains, and by the availability
of suitable premises.
Finally, at the bottom of the diagram, we have land-use policy and the construction of
houses and business premises. The rates at which developers will add to the stock of
houses or business units is affected by the availability of land on which to build,
vacancies, and their view of expected demand.
The Dynamic Urban Model
The Dynamic Urban Model is built in Vensim, with a supporting database in Excel.
The area being modelled is divided into zones, and each zone is stocked with:
¢ The total land area, the land initially occupied by business premises and by
housing, and optionally, the areas of bnd allocated for business and housing
development;
« The numbers of housing units;
¢ The number of households;
¢ The number of business premises;
¢ The number of businesses.
The following sections say more about how each is handled.
The Housing Sector
This is a simplified version of Forrester’s model. Housing units are built, age,
refurbished and, eventually, demolished. The rates at which these events occur depends
upon conditions as they might be seen by developers, and there is a lag between the
instantaneous conditions arising and construction activity taking place.
The factors taken into account are:
¢ The available land;
* The current number of vacancies. High vacancies discourage new construction,
and vice versa;
¢ Market trends, and indicated by the smoothed net household migration rate for
the zone.
The model can distinguish between different types of housing as necessary. The
differences lie in:
¢ The density at which they are built; and
¢ The preferences of different types of household for types of house.
A co-flow (not shown) is used to keep track of the average age of the housing stocks in
each zone. The interpretation of age is a little loose, for it does not mean the time since
originally built, but since last refurbished. If renewal of the stock ontinues at an
adequate rate the average age will remain steady; if the renewal rate falls, the stock
starts to deteriorate and become less attractive; if the renewal rate rises the stock
improves and becomes more attractive.
Figure 3: Housing construction
ey
Building houses houses demolished
+
‘actual housiha
constructioi actual housit
Changes to housingL_tale {demolition re Changes to housing
construction rate demolition rate
initial actual housing initial actual housing
construction rate Base fractional housing demolition rate
construction rate <time to alter housin
time to alter housing gaa a Raenineecart
construction rate target housing
construction rate target housing
¥7 demolition rate
target housing construction
rate multiplier target housing
demolition rate
/ oe
<Housins
Migration
Again, this is a simplified version of Forrester’s model. Households migrate in and out
of each zone, at rates that are affected by how attractive the zone is to live in. Factors
taken into account are:
« Access to suitable employment. Although this is derived from the accessibility
measures described above, the measure actually used is the average time for
people out of work to find employment, which is an output of the recruitment
sector described below. The number of accessible jobs or vacancies is not
something people are likely to know; the time to find work is something they
will experience, or see others experiencing, and they can form at least an
approximate estimate of it;
* The availability of housing. The measure is the vacancy rate, which is used as a
proxy for price. Each household type has its own preferences for types of
housing, and the availability measure is calculated taking this into account;
* The age of the housing stock, generated by the co- flow described above.
A lag is used to introduce delay between instantaneous conditions and people
perceiving them and taking action.
Figure 4: Household migration
Househol¢-—=—$_—
—$ $< iowseto Outward migration oO
Inward migration
actual inward base household actual outward
migration ratag___———— Migration rate migration rate
average household
dwell time in months
Perceived
attractiveness for
households change in perception
as place to live
Time for awareness
initial perceived to spread
attractiveness for Instantaneous
households attractivent
Ts multiplier
Housing availability \
ype>
Different types of household can be defined. They differ in terms of their preferences
for types of house, and the mix of people in them. People can be defined in terms of
their employment skills, and there is a correspondence between types of people and
types of job offered by employers. Not all household members need be in the
workforce, that is they are not all necessarily available for work.
Each type of household also has a specified level of car ownership (cars per household)
and fraction of its members that have a driving licence. They will also have a specified
average monthly expenditure on retail goods.
Construction of business units
Structurally this is similar to the construction of housing. Developers respond to the
attractiveness of the zone as a development site, taking into account the availability of
land, current vacancies and trends in business activity.
As with housing, there can be several different types of business unit. In practical terms
their difference lies in the density at which they are built, the preferences of each type of
business for premises, and the average floor space of each.
A co-flow is used to keep track of the average ‘age’ of the stocks of units exactly as
with housing.
Businesses
Structurally these are handled in a similar way as household migration. Businesses start
up and close down in response to conditions in the modelled area. Different types of
business can be defined, each with their own preferences for suitable premises, and
profile of employees. The mix of employees is described in the same terms as types of
people, above.
The attractiveness of the area a somewhere to do business varies in response to the
availability of suitable premises (ie the vacancy rates); the availability of a suitable
workforce; the condition of the stock of business units; access to markets and supply
chains.
The measure of availability of a suitable workforce is the job vacancy rate. This derives
from the accessibility calculations described above, but again it seems better to use a
measure employers are likely to have direct experience of, the difficulty of recruiting
staff, than a more abstract measure like accessible workforce. Vacancy rates are
calculated in the recruitment sector, described below.
Measures of access to markets and suppliers were discussed above. To recap, there are
two types:
¢ For retail, access to residents;
* Access to other businesses located within the model.
In fact for retail the model estimates how the monthly household expenditure will be
distributed across the accessible retail floor space, and seeks to expand or contract the
retail sectors depending on how the revenues compare to a target required earnings level
per square metre per time period.
Recruitment
Figure 5 illustrates the core of the recruitment cycle. This is the mechanism that
generates the travel to- work patterns, and handles the process by which people move
between jobs.
The stock ‘live and work’ is a three dimensional array, indexed on ‘home zone’, ‘work
zone’ and person type. It stores information about where people live and work, and is,
by implication, the travel to work trip matrix®, The ‘person type’ index allows us to
keep track of different types of people, categorised by employment skill.
In time people leave their jobs, either voluntarily or because their employer has closed
down. They then join a stock of ‘job seekers”’.
® Travel to work trips are the only trips explicitly present in the model. Other trips are absent, but
accessibility measures are still constructed for them. Retail, and business proximity to other businesses
are examples of this.
7 The model assumes that everyone leaves their current employment before looking for another job. This
is not true, of course, but for the time being is a reasonable approximation. The main implication is that
the number of job seekers is an overestimate of unemployment. The diagram also excludes the effect of
migration, which will affect the numbers of people in work and seeking work.
Figure 5: The recruitment cycle
Average time in Lost jobs in home
Job zones
ive ant Job seekers
x = work Z Pr concer weighted by
its
average time to sores weld
find work
retum to work accessible seekers at
each work zone
a time to fill a post
Meanwhile the employers will have vacancies for which they wish to recruit staff. For
each employer, there is a pool of accessible job seekers living within range, and the
employer will recruit from among them. These accessibility weights are given by the
deterrence function, described more fully below. The model recruits job-seekers back
into employment from each zone in proportion the their accessibility; it will tend to
recruit more heavily from adjacent zones than from those further away.
The effect of this process is that travel to work patterns change as the transport costs
change. Reductions in transport costs will tend to increase the range people will travel,
and the trip patterns become longer and more dispersed - a pattern that has been seen in
urban areas in many countries as car ownership rises and the highway networks are
expanded.
This module generates two important indicators:
* The average vacancy rate for employers, which is used as a measure of access to
a workforce; and
¢ The average time to find work, used as a measure of access to employment.
Transport Networks
It is now possible to explain how the model represents the transport system.
In fact the model is capable of representing transport networks in two ways: by zone to
zone generalised costs, or by a link based network model.
Most trips, at least by mechanised transport, will involve time and money. Generalised
cost is a commonly used measure that combines both of them linearly as in:
Generalised cost = monetary cost + travel time * value of time.
The ‘value of time’ converts time to money ‘equivalent’. It represents the rate at which
people trade off money and time, and is widely used in transport analysis. Research
methods are available to give estimates of the value of time, which can differ by type of
person and type of trip.
Sometimes generalised time is used instead. This is given by:
Generalised time = travel time + monetary cost / value of time.
More generally, other terms might be included. Public transport trips for instance might
involve several stages: walking, waiting, travelling, interchanges etc, and these can all
be incorporated into the generalised cost or time.
All transport modelling packages can calculate the generalised times or costs between
all pairs of zones, differentiating between car and public transport. These can be read
directly into the Dynamic Urban Model.
A moment's thought will reveal that this way of working loses sight of individual links
in the network, with two immediate consequences. First although we will be able to
estimate movements between pairs of zones, we cannot say what volumes of traffic flow
on any given link, and, consequently, it is not possible to build in any congestion or
crowding response.
Second, we cannot change the characteristics of individual sections of road, or test the
imposition of such things as link based tolls within the model. This can only be done by
going back to the network model and recalculating and outputting the generalised cost
matrices again.
However the Dynamic Urban Model is capable of representing public and private
transport networks, link by link. It does not have a path building routine, but rather the
available routes between any pair of zones must be defined externally as a sequence of
links. This is coded in the model in an array, as in:
Route[i,j,link_no,route_no] = 1 if the route number ‘route_no’ between
zones i and j contains the link ‘link no’, and
zero if not.
This is fairly tedious to set up by hand, but we have developed software that can take
the outputs from conventional transport models, which are very good at path finding
through networks of many thousands of links, and write it out in a format the Dynamic
Urban Model can read directly. Each link also has a set of properties: its length, free
flow speed and capacity for roads, plus such things as frequency, fare and capacity for
public transport.
This works well, but of course slows the model down significantly for a network of any
size, and arguably introduces a level of detail that is not necessary for longer term
studies. The choice between the two types of model depends on the purpose for which
the model is to be used. For longer term studies designed to identify transport strategies
that will best aid the local economy, the generalised cost matrix approach is probably
adequate; for shorter term studies in which the impacts of specific network changes are
to be examined, the network models are needed.
Logit Models
Logit models are commonly used to describe how choices about transport are made,
such as choice of route and of mode. Logit models propose that given a set of available
alternatives, the probability of any one of them being chosen is given by:
expU;)
> xp(U)
J
where Uj; is the utility of alternative i. Utility is related to the attributes of each
altemative, which in the context of transport will be the characteristics of the journey:
time, cost etc. In practice we cannot do more than postulate a functional form for
utility, and most commonly a linear form is chosen, such as:
P, =
Uj; = a.Time + b.fuel cost; + c.parking charges;
where a, b and c are parameters to be estimated®. Of course other terms might be
included if relevant: quality of road surface, reliability, congestion etc have all been
used in transport studies.
The appeal of the logit model is that given a set of alternatives, it retums a probability
of each being chosen and that rises or falls as the utility rises or falls. Figure 6 gives an
example of a logit choice model, where the choice is between two alternatives, A and B.
The horizontal axis is the difference in the utilities of A and B, while the vertical scale
gives the probability of choosing A. As the utility of A rises, relative to B, the
probability of choosing A rises. Parameters can be chosen to vary the slope of the curve
and the point at which it crosses the vertical axis.
® This is, of course, a multiple of the generalised cost.
Figure 6: A logit curve.
go ——]
ae
L.
06 VA
oS VA
an A
37
——
5
-10
Utility A minus Utility B
An extension of this idea is the hierarchical logit model. This applies where alternatives
can be grouped together in clusters, or ‘nests’ of similar alternatives. Indeed these
models are often called nested logit.
To see how this can be used, consider Figure 7. This illustrates a hierarchy of choices
that might be considered to apply for someone considering travel between two
locations. Right at the top, there is a decision about whether the trip is simply too far or
too expensive, taking into account all the available transport options. The model will
retum a proportion of people who would or would not be willing to make the trip, given
the transport conditions.
At the next level, for those who are willing to make the trip, is a choice between mode,
in this case between car, public transport or walk. In each case the choice is made
taking into account the attributes of each available route.
Finally, at the lowest level, having chosen a mode, there may be a choice of which route
to use, out of those available (except, in this case for walk).
Figure 7: The Hierarchical Logit Structure for Travel Choices
Too far
/\
Routel
Route 2
Route 3
In range
Public transport
PT PT.
Route 1
Walk
Right at the bottom, the choices between routes can be modelled using a logit model as
described above. Above this the choices are handled by a further logit model that
responds to a measure of the total utility of each mode assessed over all available
routes’, This measure, known as the logsum, is given by:
Logsum, =0- Li exp(U , 4
J
This gives the logsum for branch i, where j counts across all the sub- options available in
branch i, and @ is a parameter whose value lies between 0 and 1. It can be regarded as a
utility for each available mode. Finally a similar formulation is used right at the top to
describe the proportion of people willing to travel or not.
The model uses this structure to model decisions about travel to work. The branch at
the top of the logit tree is especially interesting because it is used to generate the
deterrence function for travel to work, as shown in Figure 1. The effect is that for any
pair of zones, as the travel times and costs increase fewer people are willing to make the
trip, and they fall out of recruitment range for employers; conversely, as those times and
costs fall, more people come into range.
These models require a number of parameters, but these can be obtained in several
ways. First, many of them have been borrowed from earlier research. For example, the
‘value of time’, which is the relative weighting given to travel time and money, is
extensively researched, and suitable values are known.
Second, some parameters have been chosen on the grounds of the logit models they give
rise to. For instance, it is relatively easy to plot the logit curves in Excel, and select
parameter values that give reasonable curves.
Third, several parameters were chosen so that they generated results in the SD model
that were consistent with known information, such as travel to work matrices. In fact
the Dynamic Urban Model has been used to calibrate deterrence curves for travel to
work, essentially by adjusting the parameter values incrementally until travel to work
patterns and mode choices are generated that correspond closely to those actually
observed in practice in the UK.
A dynamic logit model
An obvious criticism of the logit model as described above is that it supposes that
everyone has full information about all the altematives available to them, that this
information can be specified in advance, and that choices are made immediately. In
practice none of these things is likely to apply. People do not necessarily seek out
information about all the options open to them unless they have to, and in transport the
° See for example the Ben Akiva and Lerman reference for a very extensive discussion of these models
and the theory underlying them.
choices they make are likely to impact directly on the conditions experienced due to
congestion or crowding.
The model handles this by monitoring two sets of choice proportions: the ‘ideal’
proportions predicted by the logit model on the basis of instantaneous conditions, and
the choices actually being made by travellers at the same point of simulated time. These
proportions need not be the same, but we assume that as people learn about conditions
and adapt their behaviour, the actual choices will tend towards the ideal. This is a
common device in system dynamics models: the logit models deliver the target shares,
and the actual shares are gradually adjusted towards them.
Figure 8 illustrates how this is handled in the context of route choice for car trips. The
current actual route choice proportions are held as stocks. From these it is possible to
calculate the current number of people travelling on each route, and hence on each link
in the network. These numbers are then converted into flows (‘trips to work per hour’)
simply by dividing by a time period: for travel to work this would typically be two
hours, or so.
Figure 8: A dynamic logit model to model choice of driving route.
target route s!
Car route utility
Car route utilities scaler beta
actual rout
aa change in share
Sear vaves Car route
= generalised times base link times
routes
Time to adapt to
a changes
congestion functic
route times
current link, Capacity time
“times i
ti capacity
capacity ratio
Averaging period for Trips to work per aPacity ratig,
. work trips hour
Each link has a capacity, an ability to handle traffic flow. If the flow rises above this
the effect is congestion, and the traffic speeds drop. By comparing the actual flows with
the capacities on each link it is possible to calculate an adjustment to the travel times,
and hence a total, current, route drive time.
Other terms can be added in at this point to allow for such things as congestion charges.
However eventually we reach a point where the utility of each available route between
every pair of zones can be calculated, on the basis of current levels of congestion, and a
logit model can then calculate the target choice proportions. The actual route choice
proportions are adjusted towards these targets in simulated time as people leam and
adapt. As they do so of course, they may change the congestion conditions on the road,
or crowding on public transport, and hence the target choice proportions.
It is worth noting here that congested route assignment, which this is, is notoriously
difficult to handle in traditional equilibrium models because of the feedback loops.
Within a system dynamics framework it is relatively simple to formulate and
implement.
Implementation
This section outlines some applications of the model that have been made over recent
years. The model is, or course, still evolving, and is being adapted as circumstances
allow.
An early version of the model, built in Ithink, was used in a study of Hastings, on the
south coast of England. Following that the model was completely rebuilt in Vensim,
and has now been used in a study of Merseyside (Liverpool) in North West England,
and, currently in a major study of the north east of England. A test application based
around a hypothetical town has also been constructed.
Hastings
Hastings is a town on the south coast of England. Although the South East is the most
prosperous part of the country, Hastings has long been in decline. Its older industries of
tourism and fishing have largely gone, and it is too far from London for any significant
commuting. It suffers however from severe through traffic passing along the coast, and
a by-pass had been proposed for many years, along with improved links to the north,
towards London. The djective of this study was to assess the by-pass (plus other
transport schemes); a significant sub- objective was to assess the extent to which new
road links would help regenerate the town’s economy, a claim frequently made by
advocates of the schemes. The by-pass would have provided access to a pocket of land
on the edge of the town, largely inaccessible now, that could be released for
development, and the argument was that this would provide the impetus for the
regeneration needed.
Conventional transport models were built, and programs of surveys and interviews with
businesses in the town carried out. The first DUM was built to explore the dynamics of
what might happen if the roads were built. In summary, the findings were:
¢ That the roads would increase the range of commuting, both into and out of the
town, at least for people with access to cars. This happened because of the
reduced congestion delays and the increased speeds on the by-pass, giving car-
owning residents of Hastings access to employment opportunities elsewhere.
However it also provided new competition for what jobs there were in Hastings,
and this put people in the town without a car at increased disadvantage. The
total volume of car travel increased, mainly due to increased driving distances,
but there was little impact on the levels of economic activity in the town;
¢ With the land released for development, the model did predict that it would
eventually be developed, with houses and business units. New jobs would be
created, and commuters would be drawn in to the site from a fairly wide area -
not just Hastings. However there was little improvement in economic activity in
the town itself, and in some scenarios jobs there were actually lost as activity
transferred to the new site.
It might be added that this scheme was hugely controversial, for it passed through an
area of outstanding natural beauty, and the government’s position at the time was a
presumption against road building in such locations. The ‘pro’ lobby was very strong
and active, and it was certainly the case that the problems due to through-traffic in the
town were bad. In the event however the government decided not the build the by- pass.
Part of the argument was that the model (and other work) had shown there would be
increased business activity on the town edge and increased migration to take up the new
housing and jobs on offer, but because of this employment benefits would go to
primarily to new migrants to the town, not to existing residents. The objective of
raising their welfare was not shown to have been achieved.
There is an interesting footnote to the story. A few months after the decision, it was
announced that £125m was to be made available to support a regeneration programme
for the town. This was the money that would have been spent on the by-pass: now it
will be spent on refurbishment, internet technology and a new University in the town
centre.
Hypothetical town
A hypothetical town has been set up largely to test the model as it is developed, but
which is capable of demonstrating interesting dynamics. This town consists of seven
zones, with simple road and public transport networks kinking them. There is no
transport link with the outer world.
Figure 9 illustrates the town. It lies on the coast (the design was inspired by Hastings,
but it is not that town). The red lines represent the roads linking the zones, and the
green lines a small public transport network. The links are directional: the model
recognises the distinction between link 1a, for example, which caries traffic in the
direction of the arrow show, and its reverse, which caries traffic in the opposite
direction.
Figure 9: Hypothetical town
The model is initialised so that it is in reasonably stable equilibrium, so without some
interventions not much changes over ten simulated years. Various tests can then be
implemented.
As with many countries, road pricing is being considered in the UK, usually to reduce
congestion, and in fact has recently been introduced in London. Increased public
transport capacity is usually proposed to accompany the charges (as is the case in
London). The Dynamic Urban Model is well suited to test the likely impacts of road
and congestion pricing, so the first test described here is to impose a fixed cost of £5.00
on every car with a destination in zone 6, the town’s central area.
Figure 10 shows the effect on the total number of jobs in the town compared to the base
tun: there is a moderate decline, in response to what is effectively an increase in the cost
of doing business. Figure 11 shows how this looks in Zone 6, for finance and service
businesses: a slight decline, compared to the base, followed by recovery. However it
can be seen in Figure 12 that the effect on retail is very severe, with a drastic decline in
the number of retail businesses.
This can be compared to Figure 13, which shows the number of retail businesses in
Zone 1, on the edge of the town. Here there is an increase, albeit not large numerically.
Why does this happen for retail, but not other sectors (the finance and services sector
changes very little in Zone 1)?
It is due to a combination of two factors. First, the charge deters retail customers from
visiting zone 6, who then switch their spending to other locations. Second, it makes
recruitment of a staff harder for companies based on zone 6, and as people switch jobs
over time they tend to switch to locations with cheaper access. The result is that
conditions improve for businesses outside Zone 6, because now it is easier to recruit,
and they pick up more retail business: eventually the result is the shift shown.
The effect is less severe for finance and service businesses because they are less
affected by loss of customers than the retail sector, and in the longer term they also gain
from the premises vacated by retail businesses that leave, which they can make use of.
This is why after a period the number of finance and service businesses in zone 6
recovers.
Figure 10: Impact on total jobs of £5 fee in Zone 6
Total jobs
26,000
25,000
24,000
2000 2002 2004 2006 2008 2010
Year (year)
Total jobs : Parking charges ———————————_ Jobs.
Total jobs : Base Jobs
Figure 11:Finance & service businesses in Zone 6
Businesses
70
60
50
2000 2002 2004 2006 2008 2010
Y ear (year)
Businesses[z6, Finance and Services] : Parking charges- businesses
Businesses[z6,Finance and Services] : Base businesses
Figure 12: Retail businesses in Zone 6
Businesses
100
50
0
2000 2002 2004 2006 2008 2010
Y ear (year)
Businesses[z6,Retail] : Parking charges —--————_ businesses
Businesses[z6,Retail] : Base businesses
Figure 13: Retail businesses in Zone 1
Businesses
100
95
90
2000 2002 2004 2006 2008 2010
Year (year)
Businesses[z1,Retail] : Parking charges ———————_ businesses
Businesses[z1,Retail] : Base businesses
The question of mode shift is also of interest, since a shift towards public transport is
sometimes an explicit objective of proposed charging schemes. Figure 14 shows the
total daily revenue taken by the public transport operator from commuting trips.
Compared to the base, the charge generates more revenue for the operator, as might be
expected, due to the mode switching of commuters. The revenue rises, but then falls
slightly.
Figure 15 plots the change in daily ridership on one of the public transport links into the
central area (link 7a). It can be seen that there is an initial transfer to public transport,
but this dissipates surprisingly quickly and has virtually gone after four years. The
reason is that the people who transfer all have cars and, by and large, prefer to use them.
When they next change jobs, the cost of access to jobs outside zone 6 is less, and they
tend to switch their place of work, and in so doing move back to car use. In the rather
longer term this process is further encouraged by the movement of some businesses out
of zone 6 to the rest of the town.
This does not happen on all the links. The zones on the periphery have poorer links
with the rest of the town, and so the transfer to other jobs is less easy. The result is that
the public transport operator is still left with some revenue gain in the long run.
Figure 14: Total public transport revenue per day
Total PT revenue per day
4,000
3,000 ann
2,000
2000 2002 2004 2006 2008 2010
Y ear (year)
Total PT revenue per day : Base —————————— Pounds
Total PT revenue per day : Parking charges———————__ Pounds
Figure 15: Passengers per day on link 7a
Total PT trips on each link
600
200
2000 2002 2004 2006 2008 2010
Y ear (year)
Total PT trips on each link[17a] : Parking charges people
Total PT trips on each link[17a] : Base ———————_ people
Of course these results are not presented as generalisable to any town. This is after all a
rather small test town and £5 is a very high charge given the scale of the place.
Moreover the results in any case will depend on the distribution of residences and
businesses and the transport networks connecting them. What is claimed however is
that that the DUM offers an excellent platform for exploring the likely dynamics of road
pricing in an urban area, and for exposing the possibility of some unexpected effects,
such as the temporary nature of some of the changes.
Merseyside
This study, still under way at the time of writing, was designed to look at the
regeneration impacts of a new light rail scheme in the city. This was a larger
application of the model in some ways - it contains 140 zones, for example, and the
workforce and jobs were segmented by five types of skill. However the application was
also more restricted in that many of the dynamics, including population migration and
job creation, where switched off, in order to allow externally imposed numbers of jobs
and residents to be used. Zone to zone generalised costs for car, public transport and
walk, with and without the proposed light rail scheme, were all provided by extemal
conventional transport models.
The only significant dynamic left operating was the recruitment process, and the model
was used essentially to see whether the light rail would provide improved access to
employment opportunities for people living in severely deprived quarters of the city. As
a demonstration of the Dynamic Urban Model’s capabilities it is therefore of rather
limited value, except in demonstrating that it is capable of handling many zones.
However the following conclusions were drawn:
¢ For households without a car, located in the regeneration areas served by the
light rail, the number of accessible jobs increased by about 25%. The figure for
those with a car available was very slightly negative, due to the fact that there
was some increase in car travel times due to reduced road capacity for cars, but
these people continue to give less weight to public transport in their travel
choices;
* This generated a modest shift in the balance of advantage for people with and
without a car, causing a modest increase in the number of car less people finding
work;
¢ The overall increase in employment across the whole city was, however, almost
zero. As already noted, the model was not able to generate new jobs as a result
of the new transport. However there is always a pool of job vacancies, and it
was thought that the light rail might improve recruitment and reduce the size of
this pool. This did not occur, primarily because there was an excess of
unemployed people over available jobs, so that access to staff was not really a
constraint. In other words, such vacancies as there were were due to internal
delays in the recruitment processes of employers, and not numbers of accessible
job seekers. Improving the mmber of accessible job seekers did not, therefore,
reduce the vacancies, although it did shift the geographic distribution and social
mix of those who were in work.
The North East of England
The most recent application, under way at the time of writing, is in the North East of
England. The economy here was traditionally based on coal and heavy industry, but
most of this has gone. There are prosperous areas, but also very deprived areas in which
employment is high, housing and health is poor and all the problems that brings. We
were asked to look at how transport strategy might be designed to help stimulate
economic regeneration in the region, possibly coupled with other initiatives such as
training.
The model has been set up with 81 zones covering the region. Each zone is initialised
with information about:
«Its area, and the area allocated for housing or business use;
* The housing stock, split by five types of house;
¢ The number of households, split by five 6 types; and the structure of each type
of household;
¢ The numbers of businesses operating in the zone, in each of five categories;
¢ The stock of business premises, in which the businesses are located.
The area covered by the model has a workforce of roughly 1.1m and, currently, rather
fewer than 1m jobs. The model has been set up with a link-based representation of the
road network that includes the major roads, but not the smaller ones. This amounts to
rather more than 1,000 links. Public transport networks were not built, but zone-to-zone
gereralised costs were inferred on the basis of travel to work patterns and mode choices
as reported in the 1991 census.
The model operates in time periods of quarter years. It was run for 30 simulated years
to allow transient instabilities to settle out, and then subsequent test runs started from
that end point. The full network version of the model takes 1h 15m to simulate ten
years, using a 2 Ghz PC, so a second version of the model was also set up to provide a
more rapid tool, using fixed zone to zone generalised costs. This takes about ten
minutes to simulate ten years, and can be used to provide quick checks of broad based
policies, but is not capable of testing specific transport schemes.
This work is still under way at the time of writing, but the following broad conclusions
are emerging:
¢ While there are some deficiencies in the public transport networks and some
known congestion spots on the roads, the transport network in the region is quite
good. Consequently improvements to the network, while relieving some of the
congestion, does not lead to any very great increase in jobs!°;
¢ On the other hand the balance of advantage between residents can be shifted
when transport costs are reduced. Some areas of high unemployment and low
car ownership could be shown to suffer some increase in unemployment if
improvements to road transport led to increased external competition for those
jobs they did have.
Scenarios were tested in which economic growth was imposed externally, such as might
happen if the national economy grew. This was modelled by increasing the
attractiveness of the region for new businesses, thereby increasing the start-up rates.
10 This finding is at least consistent with empirical work carried out elsewhere, which has suggested that
in economies with well developed transport networks new schemes will not usually lead to increases in
jobs unless transport is clearly acting as a constraint
Growth targets were set of 2% of the existing stock per year, in addition to the start-ups
already generated by the model. This suggested the following:
* Congestion on the network rose, leading to congestion and some constraint on
job growth. However the growth in businesses also increased the attractiveness
of each zone for business start-ups, and this tended to counter the effect of the
congestion. In other words businesses were tolerant of the congestion because
they still had access to growing numbers of businesses. It could be argued that
something of this can be seen in successful cities everywhere;
« In many zones the target 2%pa growth could not be achieved because of other
constraints, primarily the availability of suitable premises to house the
businesses, which in tum was usually due to land shortage. To a lesser degree
the availability of a suitable workforce was also a constraint in some cases, but
rather as with Merseyside, the high unemployment levels meant that recruitment
was not, on the whole, a constraint. In these cases, relieving the transport
system did not help to generate new jobs.
Summary
The relationship between transport and the economy is of great interest and importance,
but is still little understood, and there is much disagreement about it. Part of the reason
for this is the sheer difficulty of obtaining good empirical data, following an
intervention to a transport system, and then of demonstrating convincingly that changes
to the economy can indeed be attributed to those interventions.
The topic is important because it is frequently claimed that new transport investment is
essential to help regenerate towns and cities, even though the mechanisms by which this
regeneration will arise are not always explained.
This paper has provided a brief overview of the problem, and of some of the methods
that have been used to model these interactions, frequently equilibrium models. It has
been argued that equilibrium models are not well suited to the problem, and that system
dynamics provides a more appropriate methodology. A model has been described that
builds on ideas borrowed from Forrester’s original Urban Dynamics work, but extends
it to a fully zonal model with road and public transport networks. Four applications of
the model, three ‘real’ and one hypothetical, have been described.
The model is still ‘work in progress’ and has more shortcomings than the author would
like. However it is argued that it represents a significant step forward in the field, and
is genuinely practical tool that can help policy makers understand the dynamics of their
town or city.
References
Banister D, Berechman J. 2000. Transport Investment and Economic Development.
UCL Press, London
Ben-Akiva M, Lerman SR. 1985. Discrete Choice Analysis. MIT Press, Cambridge,
MA
Forrester J, 1969. Urban Dynamics. MIT Press, Cambridge, MA
Forrester N, Private correspondence
Standing A dvisory Committee on Trunk Road Assessment (SACTRA). 1999. Transport
and the Economy. UK Department of Environment, Transport and the Regions,
London.
Back to the Top
