skip to Main Content




Urban Transport Research Group, University of Cape Town, Private Bag,

Rondebosch, 7701

It is almost 50 years since the first comprehensive urban transport studies took place in
the United States (in Detroit (1953) and in Chicago (1956)). During that time there have
been changes to the processes adopted, but this paper argues that the general approach
to transport analysis in South Africa remains fundamentally the same as in those first
studies – an aggregate four-stage computerized transport model.1

Over time there has
been a steady increase in the complexity and apparent sophistication of these models.
They are in common usage both locally and internationally, and many variants of computer
modelling suites have been developed. However, despite the considerable cost of model
development, and their short shelf life, there has been surprisingly little critical review of
their performance in South Africa. It would appear that computer modelling is accepted as
a mainstay of transport planning efforts, and the need for a model goes largely
In this paper the historical development of transport planning is traced with a particular
focus on the development of transport planning models. Criticisms of conventional
transport planning models are highlighted (section 2). The focus of the paper then moves
to South Africa. The evolution of transport planning models is traced (section 3), and
comment on them is outlined (section 4). In section 5 the current South African policy
environment is detailed and the role of conventional aggregate four-stage transport
modelling in that context is discussed (section 6). Finally some alternatives to conventional
models are briefly described.
Evolution of Transport Planning Models
In this section the evolution of transport planning models is briefly traced, through a
consideration of the policy developments and socio-economic environments which have
influenced transport model changes. In this way it is possible to identify four periods of
model development:
• 1950s – 1960s: developments in response to accelerated highway construction and
advances in computing;
• 1970s – 1980s: developments in response to criticisms of aggregate methods;

The four stages of the model include: trip generation (how many trips will be made); trip distribution (where will they
go); modal split (what mode will be used); and trip assignment (what route will be taken).

• 1980s – 1990s: developments in response to criticisms of static, trip-based
• 1990s: developments in response to environmental pollution, and policy shifts
towards travel demand management (Behrens, 2002).
Each of these is considered briefly in turn.
Prior to the 1950s travel analysis used data from traffic counts to assess the use of
transport systems. Whilst such approaches were adequate when considering present-day
issues, any predictions were coarse and were based on a consideration of historical
trends. During the 1950s highway construction, especially in the United States accelerated
rapidly and with it came a requirement for more sophisticated prediction tools, in order that
pavements could be designed, and the economic impacts of such schemes be assessed
and prioritised more accurately. The development of computers during this period provided
obvious tools to process the large quantities of data required to model entire urban
systems. The pioneers of these first models were mainly engineers with a ‘positivist’
outlook, that is the belief that relationships such as those found in natural science could be
extended to urban systems (e.g. that of gravitational attraction which influenced the
development of trip distribution models). They assumed that it would be possible to
forecast human behaviour with some degree of accuracy. The key method to emerge
during this period of rapid development was the ‘aggregate four-stage model’.
The pioneering American studies in the 1950s established procedures which were
institutionalised in the 1960s via the ‘3C process’. From the time of the 1962 Federal Aid
Highway Act, urban areas had to undertake a ‘continuing, comprehensive and co-operative
transportation planning process’ in order to receive funding. The technical travel
forecasting processes required were described in various manuals published by the
Bureau of Public Roads.
The assumptions of the developers of aggregate four-stage models were that:
• it was possible to predict a future land-use pattern independently of changes to the
transport system;
• it was possible to predict travel behaviour based on household data averaged over
a zone;
• relationships between household characteristics and travel behaviour would remain
steady over (long periods of) time;
• travel decisions were made principally on the minimization of travel time and cost;
• interzonal, average weekday, peak hour vehicular trips provided an adequate
picture for the purposes of transport system improvements.
Criticisms of Transport Planning Models
By the mid 1960s opposition to the transport developments of the previous decade had
started to emerge (Atkins, 1977; Healey, 1977; Batty 1994; Harris, 1994; Klosterman,
1994; Pas, 1990). This concern heightened in the early 1970s when a series of political
crises across the developed world, including the end of the post-war boom, fuel shortages
and damagingly high inflation in the UK led to citizen unrest over many issues, including
transport. This was especially the case in inner city urban areas where highway
construction had displaced some communities. A new type of study emerged during this
period (Allen, 1985) which was less computer dependent, more open to public
participation, and included a broader range of evaluation criteria, such as environmental
and equity issues. (The previous focus had been largely on economic appraisal). However,

although the process of planning was amended slightly during this time, the computer
models used were fundamentally the same as the aggregate four-stage models developed
in previous decades. In 1973 Lee produced a damning critique of the large-scale urban
models common at that time, within which he highlighted seven fundamental flaws:
• hypercomprehensiveness (trying to do too many things at once);
• grossness (too aggregate to be meaningful or useful);
• hungriness (requiring vast amounts of data);
• wrong-headedness (with a poor match between theory used in the models and
actual human behaviour);
• complicatedness (the outputs were difficult to interpret, and required adjustments to
get realistic results);
• mechanicalness (the computers introduced errors due to rounding); and
• expensiveness.
This criticism from the United States was later joined by criticism from the United Kingdom.
Atkins (1977) memorably stated that:
“we have a series of excessively complicated and expensive models using
unsubstantiated and biased techniques to provide information of dubious accuracy for
answering the wrong questions”. He arrived at his conclusion following a review of
transport planning studies undertaken in the UK to that date.
In response to the criticisms three new analytical methods emerged: land-use transport
interaction models (which modelled the impact of transport changes on the land-use
system over time, and vice-versa); disaggregate methods (which constructed travel-choice
models for individuals rather than for households or zones) and micro-simulation methods,
(an improvement on the aggregate assignment procedure, to take account of driver
behaviour at vehicle or ‘platoon’ level).
Despite these improvements to transportation modelling technique during the 1970s and
early 1980s, practitioners tended to remain with the familiar aggregate four-stage model
and in 1986 Atkins was able to compile another damning compilation of criticism. These
critical voices commented on the redundancy, inefficacy and wastefulness of mainstream
methods in transport planning.
In the 1980s and 1990s further modelling developments emerged in response to the
critics: dynamic methods and activity-based methods. Dynamic methods were developed
in response to the ‘static’ nature of most preceding approaches. Travel analysis had been
based on data from one cross-section of time, but critics argued that since transport
behaviour across generations and sectors of the population vary over time, so transport
responses will change over time also. However these changes, it was argued, are more
complex than we may assume, and not under a constant state of re-appraisal as the
economic utility model assumes. Rather there are critical periods (such as a house,
employment or school change) when transport is reappraised, and these adjustments
profoundly impact all subsequent travel in that household. Dynamic methods collate
longitudinal data in order to properly address this topic.
Activity-based methods emerged in the late 1970s and early 1980s in response to the
limited behavioural theory underpinning the conventional aggregate four-stage models. In
activity-based approaches the observation of the trip is replaced by a detailed
consideration of the activity which leads to the trip. It was argued that it is not possible to
understand (and so to change) travel behaviour without a much deeper understanding of

the human lives impacted by the transport system under study. Hence activity-based
methods are crucially different from aggregate methods in that they focus on household
and personal activity scheduling rather than trips as isolated events.
The final period of development during the 1990s saw fundamental changes in transport
policy in the developed world, including the passing of the Clean Air Act Amendments
(CAAA, 1990) and the Intermodal Surface Transportation Efficiency Act (ISTEA, 1991) in
the US. These two pieces of legislation provided for (in the case of CAAA) actions (such
as enhanced public transport, the provision of HOV lanes and congestion pricing) to
counteract pollution problems, and, (in the case of ISTEA) unprecedented flexibility in the
transfer of funds between highway and public transport investment. CAAA and ISTEA
challenged the transport professional to be able to accurately forecast transport emissions,
and in response to this the US government launched a substantial programme of work in
transport modelling: The Travel Model Improvement Programme (TMIP). Although the
work is still in progress, a microsimulation model of some detail has been developed –
TRANSIM – which should allow air emissions to be forecast at a fine level of spatial detail.
(Lakshanan, 1998).
Despite these developments, and in the face of growing concern over the realism of the
forecasts, mainstream practice in local government and in most consultancy practice has
remained firmly entrenched in the use of aggregate four-stage models. Concern over this
caused the new Professor of Transport Policy at University College London to state in his
inaugural lecture “our ability to treat the new policies analytically; to understand their
effects; to assess their costs and benefits; is seriously hindered by our inheritance of a
tool-kit that is bright, impressive, of unchallengeable intellectual achievement, and wrong”
(Goodwin, 1997).
Such statements were part of growing dissenting voices against mainstream practice in
the UK during the 1990s. The start of this dissent can be traced back to the ‘New Realism’
conference of 1991. The conference acknowledged the impossibility of increasing the
capacity of the road network to address both the forecast traffic levels, and the
environmental problems associated with such traffic. There was a realisation (supported
by the SACTRA report of 1996) that in congested areas increases in road capacity were
likely to induce more traffic than would otherwise be the case. Thus the conclusion was
that if supply could not be made to meet demand, then demand would have to be changed
to meet supply. Hence, the need for demand management and the realisation that
‘different policies will result in different forecasts’ (Goodwin, 1997). The aggregate methods
still widely used in practice had not been designed to evaluate demand-management
policies and fell far short of what was required, hence the development of ‘strategic
appraisal methods’, discussed in more detail later.
In the next section of this paper the adoption of transport models over time in South Africa
is considered.
Transport modelling in South Africa began in earnest in the 1960s. From then until the
early 1980s data was widely collected in order to calibrate MINI-TRAMP and DELTRAN
four-stage models. These models were applied in most major urban centres and were
usually developed for typical morning peak periods. During this time there was also a
limited application of land-use-transport interaction models. From the mid-1980s to the
early 1990s some attention was turned to the analysis of the impacts of apartheid on the

travel patterns of black workers. Some of these studies used activity-based methods or
collected longitudinal data whilst others used more conventional household surveys or
screen line counts.
During the early 1990s there was renewed interest in general data collection in order to
calibrate newly acquired four-stage modelling software – EMME/2. Whilst internationally
there are numerous software packages available, it has been shown that South Africa has
almost blanket use of EMME/2. (Davies, Rontiris and De Roodt, 1995) Most of the models
still in use have been calibrated for the morning peak period.
Apart from the general criticisms of aggregate four-stage models highlighted above, there
has also been some limited concern voiced in South Africa regarding the appropriateness
of these models to local conditions. Specifically, Davies et al (1995) pointed to the
following problems:
• The importance of walking is sometimes overlooked;
• There is poor interaction between public transport and private transport models;
• There is a lack of qualified professionals;
• Technology transfer is “black-box” rather than a source of knowledge;
• The combi-taxi mode is not easily matched with EMME/2; and
• Social stresses inhibit good data collection required for accurate modelling.
Other authors, writing about the developing world specifically, have raised major concerns
regarding the appropriateness of the aggregate four-stage model for developing world
conditions. Dimitriou (1990:169) suggests that many of the problems with the four-stage
model can be traced back to the assumptions underpinning the earliest models, which are
largely inappropriate for developing world conditions. For example, the US developers of
the 1950s saw the urban transport problem as mainly one of how to overcome motorised
traffic congestion. This is sensible in a country where the majority of residents are vehicle
owning. The converse is true in developing countries. In addition, early developed world
models were not used to study informal transport and so this essential travel mode is
treated in an ad-hoc manner in most models. Aggregate four-stage models assume some
long term stability in the variables affecting travel demand. This presumption is especially
questionable in a rapidly growing developing country.
More recently, Vasconcellos (2001) has been equally critical of the application of
aggregate four-stage models in developing countries. He points out that not only is the use
of transport models flawed from technical and ideological standpoints, but there are also
problems with how the outputs are used in appraisal. Full environmental appraisal (which
would include a full safety analysis; disruption and costs to non-motorised transport users)
are generally not present. The attribution of monetary costs to accidents and time, which is
necessary for economic appraisal, becomes particularly problematic in the developing
world where there are large variations in values due to, for example, extreme differences
in incomes.

This section outlines the current South African transport policy environment, which points
towards likely areas of planning effort over the next decade. Moving South Africa, the
detailed policy plan which distils the more general policy statements of the National White
Paper, and helps to identify areas which (if Moving South Africa is to be implemented) will
need to be investigated more widely:
• Lack of affordable access to the public transport system;
• An ineffective public transport system, ageing vehicle fleets and poor cost recovery
from subsidised fares;
• Increasing car dependence for some segments due to lack of alternatives and
relatively cheap car travel;
• A combination of apartheid spatial distortions and current dispersion which result in
high commuting distances.
Moving South Africa goes on to suggest that there is a need to resist further dispersion,

promote investment in public transport rather than road building and to consolidate non-
motorised modes. In the light of this, three action sets are proposed:

• Densification of transport corridors, which should improve levels of service through
corridor-based public transport offering increased speeds and frequencies;
• Optimal deployment of modes, which requires customer-based planning “matched
to local travel patterns…and the preferences of specific customer segments”;
corridor supportive infrastructure investment such as priority or dedicated bus and
taxi ways and intermodal transfer stations and tough road space management
“through a combination of controls and pricing, backed by improvements in the
public transport system”.
• Improving firm level performance in the provision of urban transport services.
(Wilkinson and Behrens, 2002)
What does all of this mean for urban transport analysis and modelling? It points towards
an increased emphasis on:
• walking, as an important mode for those without access to motorised transport; as a
mode used to access public transport; and as an alternative for choice users to
• public transport and taxi systems;
• detailed knowledge of customer needs
• integration of modes; and
• demand management
within a framework of integrated land-use and transport planning, aiming towards
densification. Clearly these are not areas of strength for the aggregate four-stage model,
but what are the alternatives? The next section discusses possible alternatives to the
conventional approaches.
The transport policy of South Africa has been lauded (De Saint Laurent, 1998), and it
certainly appears to have many elements which, if implemented, could lead to fundamental
change in transport conditions. The key to change, then, is operationalising the policy, that
is, moving the policy from words in a report to action on the ground. In order to
operationalise the policy there must be a re-orientation of all elements involved in
implementation towards policy goals. Transport planners working for government (either

directly, or through private consultancy), are key actors in the operationalisation of policy.
We suggest that most transport planners in South Africa today would support the broad
aims of the transport policy, even though there may be some disagreement over the detail.
However, even those practitioners who are fully supportive of policy will be confounded in
their aim of implementing policy if they continue to use only the aggregate four-stage
modelling approach described above. The reasons for this are that the conventional
modelling process acts as a filter to precisely the type of information and knowledge
required in order to sensibly implement the local policies.
With respect to public transport, conventional modelling treats public transport as an
adjunct to vehicular models, where the focus is on operational characteristics of private
vehicle flow such as flow size, flow speed and delays to the neglect of information
regarding waiting times, queue sizes, passenger comfort, ease of transfer and safety –
matters known to be important to public transport users.
A customer-focused plan requires that the detailed needs of customers are known and
addressed. In order to look at customers in the first instance, the developers of the Moving
South Africa project had to commission new surveys in order to supplement the poor level
of information available at that time. Indeed, it has recently been found that, as a result of
a focus on motorised traffic, commutes and morning peaks in past travel surveys, there is

very little published information in South Africa regarding non-home-based, non-work, off-
peak and non-motorised trip-making. Regarding matters of equity conventional models fall

far short of ideal. At best they can provide average journey times for peak period trips
across different types of users. A detailed consideration of equity requires at the very least
an explicit consideration of walking trips, and of trips taken by those not working, or not of
working age. Such trips may well be infrequent, irregular and non-symmetric and so will
tend to be either missed, or misinterpreted, by conventional modelling tools.
A major thrust of recent policy, demand management, was not conceptualised at the time
conventional models were developed. These conventional models were developed to
address questions of ‘how much?’ and ‘where?’ road capacity should be built. They were
never intended to consider whether capacity should be added at all, or if any other
(demand management) techniques could be adopted. Hence the four-stage model using
fixed trip matrices and exogenous growth factors will inevitably answer ‘yes’ when asked if
roads will become over capacity in future. In a self-fulfilling prophesy, the four stage
models built to consider road infrastructure improvements will implicitly tend to promote
roads, since alternatives are not easily part of the four stage modelling process.
Given this discussion, it seems clear that there is a need for some alternative approaches,
and these are discussed in the final section of this paper.
The damning critique of conventional practice and the re-orientation of transport policy in
South Africa requires practitioners to undertake a fundamental re-evaluation of the tools
which they are employing. Vasconcellos (2001) argues that the developing world requires
a ‘sociological approach’ to transport, which focuses on individuals instead of vehicles;
lives instead of trips; within a framework of equitable provision. The tables below contrast
the prevailing technical approach with the proposed sociological approach. It can be seen
from this comparison that a sociological approach brings to the fore information, which
highlights inequities, but which is often masked in conventional approaches due to
insufficient or over aggregated data.

It can be seen that such a sociological approach would require detailed, disaggregated
data from individuals as opposed to vehicular travellers. An activity-based approach could
supply such data. Activity diary surveys produce rich personal and household data, and
activity scheduling models enable a much wider range of possible behavioural responses
to transport system changes to be predicted (e.g. travel substitution, trip inducement, trip
suppression). Importantly these techniques have the capability of identifying those
individuals and households with greatest transport disadvantage, and of analysing the
activity scheduling and trip-making consequences of demand management strategies2
However, activity scheduling models are unlikely to be of practical use for some time, and
even when refined they would still suffer from data-hungriness and costliness. A rather
more cost-effective approach to forecasting individual responses would be to use Stated
Preference techniques, which have already received some attention in South Africa. These
can assist decision-makers in understanding how various categories of users may respond
to proposed change, at reasonable cost when compared with other (network-based
modelling) approaches. (Although Stated Preference is also not without its problems).
Strategic policy appraisal models also have promise as they predict the potential impact of
different travel demand management policy scenarios on travel behaviour quickly and
(relatively) cheaply. They are essentially multi-modal ‘equilibrium’ models in which the
demands for different modes of transport are balanced with the available supply, using an
iterative process. (Roberts, 1997)
In conclusion, the conventional four-stage aggregate transport planning model will form
part of the planners toolkit for a while to come. It is useful for providing the data required
for cost-benefit evaluations, and for assessment of operational efficiency, but the
underpinning assumptions behind this type of model are at odds with New Realist
transport planning and no amount of refinement to these models will enable them to fully
meet the requirements of contemporary policy.
This paper draws heavily on Chapter 2 of the forthcoming PhD thesis of Roger Behrens:
Matching networks to needs: Travel needs and the configuration and management of local
movement networks in South African cities, and on Wilkinson, P. and Behrens, R. (2001)
South African urban passenger transport policy and practice. Working Paper 4 of the
Urban Transport Research Group, University of Cape Town.

Back To Top