This article was downloaded by: [The University of Manchester Library]

On: 18 November 2014, At: 11:08

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,
UK

Transportation Planning and

Technology
Publication details, including instructions for authors
and subscription information:
[Link]

A Unified Framework for

Selecting a Travel Demand
Forecasting Model for
Developing Countries
a
Hannibal Bwire
a
Department of Transportation and Geotechnical
Engineering, Faculty of Civil Engineering and the Built
Environment, College of Engineering and Technology ,
University of Dar es Salaam , Dar es Salaam, Tanzania
Published online: 20 Jun 2008.

To cite this article: Hannibal Bwire (2008) A Unified Framework for Selecting a Travel
Demand Forecasting Model for Developing Countries, Transportation Planning and
Technology, 31:3, 347-368, DOI: 10.1080/03081060802087809

To link to this article: [Link]

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the
information (the “Content”) contained in the publications on our platform.
However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or
suitability for any purpose of the Content. Any opinions and views expressed
in this publication are the opinions and views of the authors, and are not the
views of or endorsed by Taylor & Francis. The accuracy of the Content should
not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions,
claims, proceedings, demands, costs, expenses, damages, and other liabilities
whatsoever or howsoever caused arising directly or indirectly in connection
with, in relation to or arising out of the use of the Content.
 This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan, sub-
licensing, systematic supply, or distribution in any form to anyone is expressly
forbidden. Terms & Conditions of access and use can be found at http://
[Link]/page/terms-and-conditions
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014
 Transportation Planning and Technology, June 2008
Vol. 31, No. 3, pp. 347
368

ARTICLE

A Unified Framework for Selecting

a Travel Demand Forecasting
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Model for Developing Countries

HANNIBAL BWIRE
Department of Transportation and Geotechnical Engineering, Faculty of Civil
Engineering and the Built Environment, College of Engineering and Technology,
University of Dar es Salaam, Dar es Salaam, Tanzania

(Received 27 October 2006; Revised 10 July 2007; In final form 28 March 2008)

ABSTRACT The ability to judge and select a model that is appropriate for a
particular application is considered to be one of the most important aspects in
contemporary transport planning. However, there is no suitable procedure for
the systematic selection of a model that is most appropriate for meeting the needs
and requirements of a particular planning task. Although there is little literature
on the criteria for model assessment and selection methodologies, none can
support systematic evaluation of different models versus quality of obtainable
data versus efforts for data provision. Such deficiencies support the need for
further guidance on a model selection procedure for developing countries where
efforts for data provision are highly susceptible to higher sampling and
measurement errors. This study presents a unified framework for the systematic
model selection process. Evaluation of the framework for a case study of Dar es
Salaam city in Tanzania evidences its benefits and applicability.
KEY WORDS: Travel demand forecasting model; model selection; developing
country; transport planning; data quality

Correspondence Address: Department of Transportation and Geotechnical Engineering, Faculty of

Civil Engineering and the Built Environment, College of Engineering and Technology, University
of Dar es Salaam, P. O. Box 35131, Dar es Salaam, Tanzania. Email: bw_hann@[Link]

ISSN 0308-1060 print: ISSN 1029-0354 online # 2008 Taylor & Francis
DOI: 10.1080/03081060802087809
 348 Hannibal Bwire

Introduction
With the increase in travel demand and traffic management problems,
travel demand forecasting models are being employed increasingly to
make informed decisions about the operational improvements to the
existing transportation system and the design and performance of
future transportation systems. The main advantage of using travel
demand forecasting models for such purposes is that they are capable of
capturing the interactive effects of different components of the system
under study. For that, they provide a mechanism to predict the impact
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

of various policies and programmes on travel. As a result, over the past

decades a number of competing models have emerged. Their sophis-
tication ranges from simple mathematical formulae to complex
modelling software.
In view of the developments in modelling approaches, several authors
have been of the opinion that some of the approaches appear to be
promising for application to cities in developing countries. However, it
is also well known that none of the state-of-the-art modelling as well as
state-of-the-art of practice is free from some limitations. In fact a model
that is appropriate to a particular application may be inappropriate in a
different application context. Therefore before proceeding to use
models one needs to select a model that meets the specific and unique
analysis requirements and constraints of the project at hand or of a
particular planning department. However, there is no procedure for the
systematic selection of a model that is most appropriate for meeting the
requirements and constraints of a particular planning context.
Although there is very little literature about model assessment and
selection procedure, none has identified and incorporated criteria for
assessment of methods for data provision. Consequently, most past
studies do not provide a suitable procedure that can support systematic
evaluation of a pool of potential models versus quality of obtainable
data versus efforts for data provision. Such deficiencies support the
need for further guidance on model selection procedure to developing
countries where efforts for data provision are susceptible to higher
sampling and measurement errors.
This paper bridges the above gap by presenting a unified framework
for a model selection procedure that can be used to support a
systematic model selection process. The framework is based on a set
of qualitative and quantitative criteria for assessment of models and
methods of data collection. In order for the framework to be useful to
different applications, it is focused on a more instructive selection
procedure rather than a prescriptive one. The selection procedure is
applied to a case study of Dar es Salaam city in Tanzania to illustrate its
benefits and its applicability in evaluating the potential for trip
 Travel Demand Forecasting Model for Developing Countries 349

generation modelling. Section ‘Background’ provides the background

to this study. Section ‘Model Selection Procedure’ presents the frame-
work of the model selection procedure while Section ‘Evaluation of the
Framework’ evaluates the framework based on a case study of Dar es
Salaam in Tanzania. Section ‘Summary and Conclusions’ provides a
summary and conclusions.

Background
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Since the formal introduction of the conventional four-stage travel

demand forecasting, the model has witnessed a number of highly
critical reviews. Subsequently, improvements have been made to the
four-stage modelling approach and new modelling approaches have
emerged. Few of the approaches have already been tried in developing
countries’ cities (Stopher & Wilmot, 1979; Takyi, 1990; Vougioukas
et al., 1991; Bussiere & Rice, 1995; Arasan et al., 1996) while other
approaches appear to be promising for application (Willumsen, 1981;
Khan & Willumsen, 1986; Yanaguaya, 1993; Kane & Behrens, 2002;
McDonald et al., 2002). However, there is very little literature
(Elangovan & Crouch, 1992) which has contributed to the criteria
and procedure for model assessment and selection for developing
countries. As a result, criteria for assessment of data quality and efforts
for data provision is lacking in existing literature. Instead available
literature such as Elangovan and Crouch (1992) proposed criteria of
‘use of secondary data’, ‘cost for whole modelling process’ and
‘suitability for developing countries’ only. The selection procedure
proposed by such a study does not provide for a systematic evaluation
of a pool of models versus data quality versus efforts for instituting
different methods of data collection.
Likewise, studies from developed countries (Islam et al., 1995;
Bonsall, 1997; Boyce, 1997; Lohse & Schneider, 1997; Horowitz &
Farmer, 1999; Pilz et al., 1999; Yue & Yu, 2000; Ortuzar &
Willumsen, 2001), which have contributed to criteria for model
assessment and selection procedure, also lack criteria for assessment
of data quality and efforts for data provision. However, the studies
provide a broad range of other issues to consider when evaluating
models.
The problem of the lack of criteria for assessment of methods of data
collection (data quality and efforts for data provision) is further
compounded by the lack of a well-established methodology for
assessing methods of data collection. However, there are many studies
which have dealt with single or few indicators of data quality while
other studies (Richardson et al., 1995; Cambridge Systematics, 1996)
focused on issues that should be taken into account so as to institute
 350 Hannibal Bwire

quality methods of data collection. These studies provide a broad range

of indicators of data quality and factors which have a profound
influence on data quality and efforts for data provision.
Developing countries have limited financial resources and number of
transport planning professionals, and lack experience in transport
modelling. In addition, data collection activity in these countries is
highly vulnerable to higher sampling and measurement errors. Thus,
the above deficiencies argue in favour of further guidance on model
selection procedure for developing countries. This viewpoint is
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

consistent with the notion that the ability to judge and select a model
which is most suitable for a particular application is one of the most
important elements of contemporary transport planning process
(Ortuzar & Willumsen, 2001). Despite the continued discussion and
struggle over how transport planning should be conducted in develop-
ing countries, such guidance support efforts to evolve appropriate
planning techniques to these countries.

Model Selection Procedure

The main purpose of the unified model selection procedure is to enable
a systematic model assessment process with a view to selecting the most
appropriate model that meets the needs and unique requirements of a
particular planning context. This can only be achieved through a
framework of a model selection process that ensures consistency and
logical process in model assessment. Thus, the framework is based on a
set of selection criteria. As indicated in the previous section, the criteria
are classified into two main levels of assessment: assessment of methods
of data collection and assessment of models.
The framework is best suited to evaluating a pool of known models.
It is intended to be more instructive rather than prescriptive in order for
it to be applicable to different planning contexts. Consequently, its
efficiency and effectiveness is mainly dependent upon the actual
information needs of different assessment steps versus the informa-
tion/data provided.

Assessment of Methods of Data Collection

The assessment of methods of data collection is subdivided into two
levels of assessment. The first level involves data quality assessment
while the second level focuses on assessment of efforts for data
provision.
In the data quality assessment level, there are three sub-levels of
assessments:
 Travel Demand Forecasting Model for Developing Countries 351

. General Assessment. This seeks to provide basic information needed

to inform the qualitative and quantitative assessments of data
quality. It provides general information about methods of data
collection. It comprises criteria which are not good measures of data
quality but which seek to provide information that is directly related
to the general usefulness of data.
. Qualitative Assessment. This examines factors that influence the
quality of methods of data collection.
. Quantitative Assessment. At this level, data quality is assessed in
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

terms of quantitative measures.

The second level of assessment involves the assessment of efforts for

data provision. While the main concern of the design and institution of
methods of data collection has been the issue of ensuring high-quality
data, different methods of data collection need varying efforts to ensure
that each step of data production process is at high quality. This implies
that the quality of data depends also on the quality of different steps of
data production process. The quality of data is thus built in or
destroyed at various interrelated steps of data production. It is therefore
important to assess efforts required to maintain each step of data
collection activity to its highest quality. This viewpoint is in line with
the ‘Total Survey Design’ concept (Dillman, 1978 cited in Cambridge
Systematics, 1996). Several methods of data collection are available for
a particular data collection activity. The quality of data that can be
gained from each method may vary as well as the efforts required to
provide the data. Therefore, criteria for assessment of efforts for data
provision aim at examining factors which influence the total costs for
providing the required data. This amounts to the determination of the
resources, in terms of the total cost, required to provide the required
data.
Criteria for the above assessment levels are provided in Table 1.

Assessment of Models
A distinction is made at the outset between a model and modelling
software/planning software package. A model is a simplified represen-
tation of a real-world system, valid from a particular perspective. It is
considered to represent a transport system reasonably well provided
most exogenous conditions remain more or less stable (Ortuzar &
Willumsen, 1992). On the other hand, the terms modelling software or
planning software package represent a computer-based model(s)
implemented for practical application (Bonsall, 1997). This distinction
is utilised only where the term modelling or planning software or model
system is used, otherwise the term model is meant to carry both
 352 Hannibal Bwire

Table 1. Criteria for assessment of methods of data collection

Assessment level Criteria

Data Quality Assessment

General Assessment Survey purpose and objectives
Data types
Survey duration
Consistency of definitions and classifications
Data adjustment methods
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Reasons for the selection of a particular method for data

collection
Problems and strategy to overcome them
Qualitative Assessment Level of aggregation of data
Target population
Sampling unit
Quality of sampling frame
Sampling method
Sample selection procedure
Choice or design of measurement instrument
Quality control procedure
Quantitative Assessment Sample size
Coverage error
Sampling error
Observational errors
Statistical comparison of survey results with secondary
data
Cross validation of survey results
Unit non-response
Assessment of Efforts for Characteristics of the study’s area of interest
Data Provision
Experience and knowledge from past surveys
Quality of potential methods for data collection versus
desirable data quality
Resource requirements versus resource available

meanings. Thus, the assessment of models can be subdivided into the

following five levels of assessment:
. Model Data Needs Assessment. This is concerned with data needs
assessment. It focuses on assessing data input requirements of
models.
. Model Functional Assessment. This level seeks to ascertain major
functions of a model.
. Model Reliability Assessment. It aims at examining the validity of
model’s underlying theory and basis for the structure and development
 Travel Demand Forecasting Model for Developing Countries 353

on one hand, and the ease with which the analyst can apply the model
and properly validate application procedures on the other hand.
. Modelling Efforts Assessment. This level focuses on determining
efforts required to implement, operate, and to maintain a model.
. Modelling Software Assessment. Aims at ascertaining whether
modelling software possesses certain kind of features and properties
that provide ease with which the analyst can apply it to achieve the
intended functions.
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Criteria concerning the above assessments are listed in Table 2.

Interaction among Assessment Levels

Based on a synthesis of the criteria listed in Tables 1 and 2 for
assessment of methods of data collection and models, respectively,
Figure 1 postulates a framework for unifying some levels of assessment
of models and methods of data collection. While the data-needs criteria
for assessment of models seek to identify data input requirements of
models, the criteria for assessment of data quality of methods for data
collection seek to establish the quality of data to be provided by a
particular method for data collection. It can be noted from Tables 1
and 2 that the data needs assessment level for model assessment has
some criteria for assessment in common with the data quality
assessment level for methods of data collection. On the basis of this
interaction, the data needs assessment level for model assessment
process can therefore be subdivided into two sub-levels of assessment.
The first sub-level involves those criteria which do not interact in any
way with the criteria for assessment of data quality of methods for data
collection. The second sub-level involves those criteria which are in
common with the data quality assessment criteria for assessment of
methods for data collection.
Likewise, it can also be noted from Tables 1 and 2 that the level for
assessment of modelling efforts has some criteria for assessment in
common with the level of assessment of efforts for data provision.
While the criteria for assessment of modelling efforts seek to determine
efforts required to implement, to operate, and to maintain a particular
model; the criteria for assessment of efforts for data provision provide
the former assessment level with the assessment results of efforts for
data provision.
The aforementioned interactions among the different assessment
levels serve to illustrate the peril of emphasising quality modelling
approaches on one hand, while on the other the same emphasis is not
placed on quality methods of data collection versus efforts for data
provision. Thus, Figure 1 postulates a framework for unifying the
 354 Hannibal Bwire

Table 2. Criteria for assessment of models

Assessment level Criteria

Model Data Needs Assessment

Sub-level 1 Total number of input data variables included/required by a
model
Type and number of missing or redundant input data
variables
Type and number of variables that are difficult to forecast
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Type and number of unavailable or difficult to provide

input data variables
Sub-level 2 Level of aggregation of data as compared to that of
available or obtainable data
Quality of available data/methods for data collection
Sample size requirements of a model
Model Functional Relevance of included/required variables as compared to
Assessment the desired options
Type of responses/options allowed in the model as
compared to the desired ones
Possibility of a model to allow inclusion of new variables
Level of detail of analysis permitted in the model
Scenarios allowed in the model as compared to the desired
ones
Usefulness of outputs to other planning tasks
Model Reliability Understandable and relevant model theory, assumptions
Assessment and formulation
Type, number, and relevance and flexibility of default
parameters and/or supporting functions
Strength and limitation
Level of accuracy of model outputs
Scope/range, type and degree of detail of the results to be
achieved
Sensitivity tests allowed in the model
Modelling Efforts Efforts for obtaining existing model codes
Assessment
Efforts for programming or coding the model
Efforts for acquiring modelling software and licence
Hardware requirements
Model system operation costs
Efforts for model system integration
Efforts for model system modification
Efforts for data provision
Efforts for data integration
Estimation and calibration efforts
The availability of professionals and training needs
 Travel Demand Forecasting Model for Developing Countries 355

Table 2 (Continued)

Assessment level Criteria

Modelling Software Modelling software design philosophy

Assessment
Comprehensiveness and clarity of modelling software
documentation and transparency relative to its application
to a range of simulation conditions and in depicting input
and output data
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Type of models implemented

Flexibility to incorporate additional models
Types of estimation/calibration procedures implemented or
which can be implemented
Model software run time
Maturity of modelling software
Friendliness and flexibility of user interfaces and support
provided
The inclusion of input and output functions and their
capabilities to store and develop input data and outputs,
respectively, at any spatial scale of analysis
Scale of application of modelling software
The relative ease with which existing data and database can
be accepted by the models
Type of data models incorporated/to be supported

process of assessment of models versus data quality versus efforts for

data provision.

Unified Framework
A unified framework for the model selection process is illustrated in
Figure 2. It postulates a set of selection decisions that should be made
on various steps of the assessment process. Thick lines with arrows
indicate a link between main assessment steps or decision dialogs while
thin lines with arrows show iteration process between the steps.
Step 1: Study Area Objectives, Analysis Requirements and Constraints. The
first step is to specify study objectives, analysis requirements, and to
identify constraints of the study. For analysis requirements, clear
specifications have to be made with regard to the following:
. Modelling purpose
. Level of detail of analysis at which the model must operate
. Types of responses desirable in the model
. Desirable level of detail and accuracy of model outputs
 356 Hannibal Bwire

Assessment Levels of Data

Models Assessment Levels
Collection Methods

Data Quality Assessment

[Link] Assessment
Data Needs Assessment
[Link] Assessment
[Link] Assessment

Functional Assessment
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Model Reliability Assessment

Assessment of Efforts for Data

Modelling Efforts Assessment
Provision

Modelling Software Assessment

Figure 1. Interactions among assessment levels

. The need to support continuous planning efforts

. Capacity building requirements of a transport-planning unit

Constraints of the study are usually multi-faceted ranging from time

available for the study to staff and money allocated for the study. As
indicated in Figure 2, dashed lines emanating from Step 1 with arrows
pointing to various steps of the process signify that the resource and
information base provided at this step influences the extent to which
other steps of model selection process will be.
Step 2: Available Data. This step involves examination of the available or
existing data in relation to the specified level of detail of analysis at
which the model must operate and type of responses required in the
model. For whatever data is available, one will certainly wish to
ascertain its quality before deciding to acquire the data. This implies
that the available data must be subjected to quality assessment (Step 3).
Step 3: Data Quality Assessment. This step involves a qualitative
assessment of the available data in relation to the criteria listed in
Table 1.
Step 4: Pool of Relevant Models. The quality of available data and
information base from Step 1 coupled with the user’s familiarity with
different models will enable the analyst to identify a pool of models that
are relevant to the study (Step 4). In addition, this step determines the
 Travel Demand Forecasting Model for Developing Countries 357

Step 1
Study Area Objectives, Analysis
Requirements and Constraints

Step 2
Available Data

Step 4
Pool of Relevant
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Models Step 3
Data Quality
Assessment

Some/All
No
Suitable or
Available?

Yes Step 6
Assessment of Efforts for Data Provision
Steps 5 & 7
Acquisition &
Models Assessment Processing of New Data
Available Data

Step 8
Choice of Modelling Tool

Modelling Programming General Purpose Manual

Software Software Software Computation

Step 9 Step 11
Pool of Relevant Software Testing of Modelling Tool

Step 10 End
Most Appropriate
Modelling Software Assessment
Model(s)

Figure 2. Unified framework of model selection process

 358 Hannibal Bwire

number of models that will be subjected to further assessment in the

next steps.
Step 5: Model Assessment (Assessment of Data Needs, Functionality and
Reliability). This step deals with a detailed assessment of the models
identified in Step 4. It involves the first three levels of assessment shown
in Figure 1, that is, data needs assessment, functional assessment, and
model reliability assessment.
The first level of data needs assessment has to be performed in two
sub-levels, as elaborated in the previous section. The first sub-level
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

involves the criteria (of type and number of redundant input data
variables, total number of input data variables, input data variables
which are difficult to forecast, and unavailable or difficult to provide
input data) for assessment of models, which do not interact in any way
with those in Step 3. The second sub-level involves the remaining
criteria, which do interact with some criteria used in Step 3, as listed in
Tables 1 and 2. It is during this level of assessment conclusions on the
results from the assessment of quality of data in Step 3 will be referred
to in order to enable the evaluation of quality of available data versus
data needs of models. An arrow from Step 5 pointing to Step 3 and vice
versa illustrates this process. However, if it happens that the quality of
available data does not match with data needs of models then a new
pool of models will be needed. An arrow pointing from Step 5 to Step 4
represents such a decision. Conversely, if the quality of available data
matches with data needs of models then the assessment process will
proceed to the remaining two levels of assessment, that is, functional
assessment and model reliability assessment.
Overall assessment results of Step 5 in combination with the resource
and information base provided from Step 1 will result into either
finding that all the identified models under assessment are not
appropriate or all/some of the identified models are appropriate to
enable the continuation of the selection decision process. The former
decision is represented by an arrow from Step 5 pointing to Step 4 while
the latter decision is represented by an arrow pointing to the dialog box
with the question of ‘some/all suitable or available?’ The former
iteration, in effect, re-initiates the selection process from Step 4 by
identifying a new pool of models for consideration. On the other hand,
the latter decision is expected to provide clear statements with regard to
gaps or sufficiency in data requirements. In this regard, whichever
answer is provided to the question in the dialog box, it will lead to the
assessment of efforts for data provision.
Step 6: Assessment of Efforts for Data Provision. This step involves the
assessment of efforts to provide the available data or new data. Even if
some data may be available, one will certainly wish to ascertain efforts
 Travel Demand Forecasting Model for Developing Countries 359

required to acquire it and, if necessary, to process it to the required level

of detail. In the event that new data are required, information has to be
provided on the quality of data that can be gained from a particular
method of data collection under consideration and on the efforts
required to provide the data. The information need is with respect to
the criteria for assessment of data quality as shown by an arrow from
this step pointing to Step 3 and from Step 3 pointing back to this step.
Details regarding the determination of efforts required to acquire and
to provide existing data or new data are with regard to the criteria
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

provided in Table 1. An arrow from this step pointing to Step 7

represents an optimal decision with regard to the necessary efforts
required to provide the required data.
Step 7: Models Assessment (Assessment of Modelling Efforts). Step 7
involves the assessment of modelling efforts. Feedback of information
from Step 6 and Step 1 allows the completion of the assessment at this
step as per the criteria of assessment listed in Table 1. Likewise, overall
assessment results of this step in combination with the resource and
information base provided from Step 1 will influence decision to be
made at this level. In the event that the required effort is beyond reach
then a pool of other relevant models will be required as shown by an
arrow from Step 7 pointing to Step 4 while an arrow from this step
pointing to Step 8 represents the decision to proceed to Step 8. The
former iteration, in effect, re-initiates the selection process from Step 4
by identifying a new pool of models for consideration. Moreover, an
arrow from this step pointing to Step 4 also represents a situation when
new types of models are to be considered in order to allow the
consideration of specific types of modelling tools (in the following
steps) as well. Such a consideration will be heavily influenced by the
analyst’s familiarity with different modelling tools. It can be noted from
Figure 2 that assessment results from this step will enable the analyst to
select a particular modelling tool (Step 8) and to identify relevant
software (Step 9).
Step 8: Choice of Modelling Tool. The arrow pointing from Step 8 to Step
11 represents a decision to apply general-purpose software or manual
computation. Otherwise, a decision to employ specialised transport
planning modelling software or programming software amounts to
identifying a pool of relevant software (Step 9). Decision to select a
particular modelling tool will depend on the flexibility of information
base from Step 1, assessment results from Step 7, and the analyst’s
familiarity with different modelling tools.
Step 9: Pool of Relevant Software. This step deals with the selection of a
pool of relevant software which either comprises of or can support
 360 Hannibal Bwire

programming of all appropriate models, as identified in Step 7. In the

event that a pool of software is established, they should be assessed in
terms of the criteria for assessment of modelling software (Step 10).
Step 10: Modelling Software Assessment. This step deals with the
assessment of modelling software with respect to the criteria of
assessment listed in Table 2. Thus, an arrow from this step pointing
to Step 7 represents a situation when other modelling tools have to be
considered in view of efforts required to apply them (Step 7) before
considering a new pool of relevant software in Step 9. Otherwise, the
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

selection process has to proceed to Step 11 where modelling software

has to be tested.
Step 11: Testing of Modelling Tool. Step 11 of the model selection process
requires testing of the selected modelling tool. In this case a modelling
tool can be a tool for manual computation, general-purpose software,
models implemented into programming software, or specialised trans-
port planning modelling software. The purposes of testing a tool
include to:
. Test a model solution procedure or algorithms
. Test model system run times
. Gain an understanding of a particular model with respect to a
particular function/task
. Review and check the characteristics of modelling software with
respect to the criteria listed in Table 2

Successful practical testing of the tool renders it to be the most

appropriate modelling tool and hence the models it contains. Other-
wise, as indicated by an arrow from this step pointing to Step 8 and
Step 9, a new choice of modelling tool or pool of software, respectively,
has to be made. If a new pool of relevant software is proposed, then
effort to acquire, operate, and to maintain the software has to be
assessed. This process is represented by an arrow from Step 9 pointing
to Step 7. The iteration may, in effect, also re-initiate the selection
process back to Step 4 from Step 7 as shown by an arrow from Step 7
pointing to Step 4.

Evaluation of the Framework

The assessment procedure described in the previous section was used to
demonstrate how one can use the selection procedure step by step to
evaluate models versus quality of data versus efforts for data provision.
The demonstration is focused to evaluate the potential for application
of trip generation models to a case study of Dar es Salaam in Tanzania.
 Travel Demand Forecasting Model for Developing Countries 361

Two commonly used trip generation modelling approaches
category

analysis models and regression models
were examined and comments
and evaluation results are described below.

Step 1: Study Area Objectives, Analysis Requirements and Constraints

In this evaluation, however, no particular problem relevant to the case
study area was identified. Therefore no specific study area objectives,
analysis requirements and constraints were identified.
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Step 2: Available Data

Common variables for trip generation models are attributes of socio-
economic and demographic. Data available on the case study area is
provided in Table 3.

Step 3: Data Quality Assessment

In view of the criteria for assessment of data quality listed in Table 1,
the criterion of level of aggregation of data was the most applicable
here with regard to the data types identified above because it was not
possible to obtain information concerning other criteria for assessment
of data quality. Table 3 presents a range of available data in relation to
their level of aggregation versus availability. As shown in Table 3,
almost all the data variables that are available are limited at zonal level
only.

Step 4: Pool of Relevant Models

In this step, the following most common approaches for trip generation
modelling were considered:

Table 3. Assessment of secondary data

Available data Zonal level Household level Person level

Demographic
Age â â
Gender â â
Population â
Socio-economic
Household size â
Employment â
 362 Hannibal Bwire

. Regression models
. Category analysis models
Step 5: Model Assessment (Assessment of Data Needs, Functionality
and Reliability)
Data Needs Assessment. Data needs assessment was carried out at two
sub-levels of assessment. The first sub-level involved the assessment of:
. Total number of input data variables included/required by the
models
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

. Type and number of missing or redundant variables in the models

identified
. Type and number of input data variables that are difficult to forecast
. Type and number of unavailable or difficult to provide input data
variables

The second sub-level of assessment involved assessing models in

relation to the following criteria which are listed in Table 2:
. Level of aggregation of data as required by the models compared to
that of available/obtainable data
. Quality of available data/methods for data collection (results of Step
3)
. Sample size requirements of models (amount of data)

As far as there was no presumed model formulation, then all variables

that are generally considered to explain trip-making behaviour of
people were taken into account under this assessment disregarding
whether they are available or not. Table 4 provides a summary of the
assessment of variables which are commonly used in trip generation
modelling in correspondence to the information from the data quality
assessment provided in Table 3.
Double ticks in the table indicate increased difficulty. It can be noted
from the table that both models face almost similar data problems in
terms of absence of data. The table also shows that both models can
easily be implemented at zonal and person levels. However, there is
some flexibility of implementing a regression model at the zonal and
person levels as compared to a category model in view of the difficulty
involved in providing and forecasting data. It is also well known that
category models requires extremely high amount of data (sample size
requirements) for their calibration as compared to regression models.
Functional Assessment. The next level of assessment involves the
assessment of model functionality. Table 5 presents a summary of
assessment results. Of the criteria listed in Table 2, the following
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

Travel Demand Forecasting Model for Developing Countries 363

Table 4. Data needs assessment of trip generation models
Zone Household Person

Difficult Difficult Difficult Difficult Difficult Difficult

Unavailable to provide to forecast Unavailable to provide to forecast Unavailable to provide to forecast

Regression Analysis Model

Person income â â â â â â â â â
Household income â â â â â â â â â
Car ownership â â â â â
Employment â â â â â â
Population â â â
Household size â â â â
Age â
Gender â
Categories of Category Model
Person income â â â â â â â â â â
Household income â â ââ â â â â â â ââ
Car ownership â â â â â â â
Employment â â â â â â â
Population â â â â
Household size â â â â â â
Age ââ ââ â ââ â â ââ ââ
Gender ââ ââ â ââ â â ââ ââ
 364 Hannibal Bwire

Table 5. Functional assessment of trip generation models

Regression Model Category Model

Zone Household Person Zone Household Person

Inclusion of new variables â â â

Level of analysis permitted â â â â â â
Usefulness of outputs â â â â â â
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

criteria listed in Table 5 were used to perform model functional

assessment. Other criteria were found not applicable due to the fact
that this evaluation process was not constrained to a particular
presumed model formulation and analysis requirements. As shown in
Table 5, it is well known that regression models are more attractive
when one considers including new variables in the models in future
time.
Model Reliability Assessment. The most important criteria applicable here
was that of strength and limitation of models. Other criteria were
found not applicable due to the fact that this evaluation process was
not constrained to a particular presumed model formulation. The
following were identified to in order to signify the relevant strengths
and limitations of the two models:
. Demand of data
. Multicollinearity of variables
. Inclusion of new variables in the models in the future

The former and the latter issues have already been discussed in the
previous assessment. On the other hand, multicollinearity poses a
problem in regression models while in category models it is less
problematic.

Step 6: Assessment of Efforts for Data Provision

At this stage it will have already be known or established if some or all
of the required data are available and if all are suitable. Generally,
efforts for acquiring existing data should be taken into account as well
as efforts for providing new data. For this particular evaluation
example, these two issues were not dealt with in detail as far as the
existing data was accessed free of charge for research purposes and
similar resources (time and manpower) were used to collect travel data
from home interviews using three methods of data collection (trip-
based method, out-of-home activity-based method, and fully activity-
based method). All the surveys were performed using same manpower
 Travel Demand Forecasting Model for Developing Countries 365

on equal number of days. Analysis of travel data showed that data

obtained from full activity-based methods was of higher quality (less
observational errors) than the other two methods.

Step 7: Model Assessment (Assessment of Modelling Efforts)

This step involves assessing models in terms of efforts required to
implement the most appropriate models. Both models can be used for
trip generation modelling. However, the implementation of category
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

analysis model requires more efforts, as noted in Step 5. This limitation

makes regression models to be more attractive and potential for
application. This means if, in the model assessment Step 5, weights
for each of the assessment criteria group would have been defined then
the regression model would have been the top-rated model.

Step 8: Choice of Modelling Tool

Available general software such as SPSS and Microsoft Excel has built-
in functions, which are capable to implement regression models.

Step 11: Testing of Modelling Tool

Few data were input in Excel and a multiple regression model was
obtained. The purpose of running the Excel program was mainly to test
if a model can be obtained and no further efforts was made to compare
and check the properties and features of the program in accordance
with other criteria listed in Step 11 and Table 2.

Evaluation
The application of the framework revealed four key issues that need
attention. The first is that identifying a pool of models or software can
be very difficult, as it is constrained by time available and experience of
the analysts.
Second, the application of the framework helps to identify deficien-
cies in the present data versus the models and the consequences of
selection decisions. Not only that this allows models and data to be
evaluated for their deficiencies, but also serve as a basis of needs for
further development and research.
Third, the framework for model selection process is mainly based on
a qualitative assessment process. It is only in Step 11 where the top-
rated model or models can be subjected in real-world conditions. The
benefit of this methodology is that it is less expensive to use the
 366 Hannibal Bwire

framework as different models which are not at hand can be assessed

and evaluated.
Fourth, as it is noted in Step 7, it is possible to define relative weights
for each of the criteria groups for model assessment in accordance with
their relative importance to the whole model system. Weighting can
also be applicable to Steps 3 and 10 of the model selection process. As
an example, an agency or analyst might wish to acquire an improved
model system with regard to a particular model feature or property. In
this case, a criteria group relevant for the assessment of this property or
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

feature would weight relatively more as compared to other criteria

groups.

Summary and Conclusions

This study has identified an improved methodological approach for
selecting a model that is most appropriate for a particular planning
study. The study highlighted a close interaction between model
assessment and assessment of methods for data collection. Through
this interaction, the study has integrated model assessment and
assessment of methods for data collection into a unified framework
for model selection process.
The efficiency and effectiveness of the framework to support and
guide systematic selection decisions of a model that is most appropriate
for a particular application is mainly dependent upon the actual
information needs of different assessment steps versus the informa-
tion/data provided. It provides a systematic approach to the entire
model selection process and that it helps in setting forth the myriad
consequences or the questions of ignoring the interactions inherent in
the model selection process. It also provides a basis for a systematic
integration of trade-offs between the whole cost of efforts to apply a
model and the available resources for a particular study. In other
words, it sets forth the basis for qualifying a model as the most
appropriate for a particular application or for raising questions in
search of the appropriateness of the model.
A major strength of the model selection process is that it integrates
systematically the tradeoffs of the assessment of methods of data
collection and model assessment. The process therefore offers a
systematic approach to assess the credibility of any decision to adopt,
to keep, or to revise a particular model. Thus, it is hoped that the
11-step framework developed in this study is useful to any analyst or
agency seeking to select, to acquire, or to revise a model for a particular
application. The application of the framework revealed four key issues
that need attention, as identified in the previous sub-section. The
framework is flexible in such a way that it is possible to modify or add
 Travel Demand Forecasting Model for Developing Countries 367

further criteria for assessment within any assessment criteria group.

This is because the criteria groups are based on either model
performance elements, for the case of model assessment, and indicators
of data quality and factors that affect quality of methods for data
collection.
As part of demonstration of the applicability of the model selection
process, the study has provided useful insights into the potential
applicability of trip generation models to the case study area of Dar es
Salaam. Deficiency in the existing secondary data has been highlighted
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

in relation to the application of regression and category analysis

models. In view of the deficiencies in the present data, regression model
has proved to be the most appropriate model for practical application.
It should also be noted that the results indicate that these models can
appropriately be implemented at zonal and person levels of analysis.

References
Arasan, V. T., Wermuth, M. & Srinivas, B. S. (1996) Modelling of stratified urban trip
distribution, ASCE Journal of Transportation Engineering, September/October, pp. 342
349.
Bonsall, P. W. (1997) Principles of transport analysis and forecasting, in: C. A. O’Flaherty (Ed.),
Transport Planning and Traffic Engineering, pp. 103
131 (London: Arnold).
Boyce, D. (1997) A practitioner’s guide to urban travel forecasting models, in: 12th International
EMME/2 Users’ Conference, October, San Francisco, CA.
Bussiere, Y. & Rice, R. G. (1995) Simplified travel demand forecasting: A comparison of
applications in developed and developing countries. Paper prepared for presentation at the
Simplified Travel Demand Forecasting Techniques Session of the 1995 Annual Conference of
the Transportation Association of Canada, October, Victoria, BC.
Cambridge Systematics Inc. (1996) Travel Survey Manual, prepared for US DOT and US EPA,
TMIP Program, Washington DC, July.
Elangovan, T. & Crouch, F. O. (1992) Towards simplified transport planning techniques for cities
in developing countries, in: J. O. Ortuzar (Ed.) Simplified Transport Demand Modelling, A
Compilation of Major Papers Selected from PTRC’s Summer Annual Meetings and Courses
1984
1989, Perspectives 2, pp. 19
32 (London: PTRC).
Horowitz, A. J. & Farmer, D. D. (1999) Statewide travel forecasting: A critical review,
Transportation Research Record, 1685, pp. 13
20.
Islam, M. N., Seneviratne, P. N. & Kalakota, K. R. (1995) Traffic management during road
closure, Transportation Research Record, 1509, pp. 46
54.
Kane, L. A. & Behrens, R. B. (2002) Transport planning models: An historical and critical review,
in: 21st South African Transport Conference South Africa, Pretoria, 15
18 July.
Khan, M. A. & Willumsen, L. G. (1986) Modelling car ownership and use in developing countries,
Journal of Traffic Engineering and Control, November, pp. 554
560.
Lohse, D. & Schneider, R. (1997) Vergleichende Untersuchungen der Aggregierten und
Disaggregierten Verkehrsplanungsmodelle, Final Report, Heft 3, TU-Dresden, Dresden.
McDonald, N. C., Porter, C. D. & Gorham, R. (2002) Availability of travel forecasting methods
and data for air quality planning in developing countries, Transportation Research Record,
1859, pp. 110
116.
 368 Hannibal Bwire
Ortuzar, J. de D. & Willumsen, L. G. (1992) Monitoring of transport plans using low cost data
and advanced modelling techniques, in: J. O. Ortuzar (Ed.) Simplified Transport Demand
Modelling, A Compilation of Major Papers Selected from PTRC’s Summer Annual Meetings
and Courses 1984
1989, Perspectives 2, pp. 121
131 (London: PTRC).
Ortuzar, J., de, D. & Willumsen, L. G. (2001) Modelling Transport (3rd edn) (Chichester, West
Sussex: Willey).
Pilz, A., Plank-Wiedenbeck, U., Willems, F., Axhausen, K. W., Hayes, S., i Renau, L. I., Segarra,
E., Boltze, M., Jaehrling, H., Zhang, X., Allekotte, D., Stahnke, G., Retzko, H.-G., Bell, M.
G. H., Saleh, W., Thorpe, N. & Wright, S. (1999) Analysis and development of tools for
assessing traffic demand management strategies: Guidelines on the use of tools for assessing
Downloaded by [The University of Manchester Library] at 11:08 18 November 2014

TDM strategies, in: European Commission 4th Framework Programme, Brussels.

Richardson, A. J., Ampt, E. S. & Meyburg, A. H. (1995) Survey Methods for Transport Planning
(Oakland, CA: Eucalyptus Press).
Stopher, P. R. & Wilmot, C. G. (1979) Work-trip mode choice models for South Africa,
Transportation Engineering Journal of ASCE, 105(TE6), pp. 595
608.
Takyi, I. K. (1990) Trip generation analysis in a developing country context, Transportation
Research Record, 1285, pp. 9
21.
Vougioukas, E. M., Harajli, A. H. & Timberlake, R. S. (1991) Appropriate traffic modelling
techniques for developing cities: The case of the Oman Capital Area, in: M. Heraty (Ed.),
Urban Transport in Developing Countries
Lessons in Innovation, Perspectives 1, pp. 83
88
(London: PTRC).
Willumsen, L. G. (1981) Appropriate transport planning techniques for developing countries, in:
S. Yerrell (Ed.) Transport Research for Social and Economic Progress, pp. 341
352
(Aldershot: Gower Press).
Yanaguaya, W. (1993) The use of U.K. transport models in developing countries, Traffic
Engineering and Control, 10, pp. 476
479.
Yue, P. & Yu, L. (2000) Travel Demand Forecasting Models: A Comparison of EMME/2 and
QRS II Using a Real-World Network, Report No. SWUTC/00/167901-1, prepared for US
DOT, University Transportation Centers Program, Washington, DC.