development of mode choice model for gaza city · dr. essam almasri a thesis submitted in partial...
TRANSCRIPT
1
The Islamic University-Gaza
Higher Education Deanship
Faculty of Engineering
Civil Engineering Department
Development of Mode Choice Model for
Gaza City
(نموذج الختيار وسائل النقل في مدينة غزة تطوير)
Sadi I. S. Alraee
Supervised by
Dr. Essam Almasri
A thesis submitted in partial fulfillment of the requirements for the
Degree of Master of Science in Civil Engineering- Infrastructure
The Islamic University of Gaza- Palestine
June, 2012
I
بسم اهلل الرمحن الرحيم
درجات و يرفع اهلل الذين امنوا منكم و الذين أوتوا العلم
اهلل مبا تعملون خبري
صدق اهلل العظيم
(11)سورة اجملادلة
II
DEDICATION
To the soul of my mother God rest her soul; to my loving father who supported me all the way; to my wife for her unlimited support and encouragement; to my son whose innocent energy was and still a source of inspiration; to all my friends who stood beside me with great commitment; I dedicated this work hopping that I made all of them proud.
Sadi Alraee
III
ACKNOWLEDGMENT
First and foremost I would like to thank God for giving me inspiration, ability, and
discipline to make it through.
I would like to extent my sincere thanks to my supervisor Dr. Essam Almasri for all
his support and guidance during my thesis. His valuable suggestion and comments
always served me as a source of inspiration and encouragement
I would like to express my gratitude to the higher education division at faculty of
engineering for their administration and academic support.
My special thanks to all my friends and colleagues for their unlimited support and
encouragement.
Finally, I would like to thank my family for their support, love and for tolerating the
time I spend working with my research.
IV
ABSTRACT
Gaza city is considered one of the most densely populated areas in the world and it is
the most densely city in Gaza strip. The lack of efficient application of transportation
planning process leads to deficiency in adopting the suitable transport policies to
mitigate the transportation problems resulting from urbanization and rapid increase of
population. The mode choice model is probably the most important element in
transportation planning and policy making.
The aim of this study is to develop mode choice model for work trips in Gaza city and
therefore investigating the factors that affect the employed people’s choice for
transport modes. The revealed and stated preference mode choice models were
developed using about 2/3rd
of 552 questionnaires distributed for this purpose. The
rest 1/3rd
of questionnaires were used to validate the chosen models.
The results of this research show that the factors that significantly affect the choice of
transport modes for revealed model are: total travel time, total cost divided by
personal income, ownership of means of transport, distance, age, and average family
monthly income. The results also indicated that the travel time, fare divided by
personal income, frequency of service, age, average family monthly income and
distance are the factors that affect the mode choice for stated preference model. Both
revealed and stated preference models as illustrated in the results are able to predict
the choice behavior of employed people in Gaza city as the two models are valid at
95% confidence level.
This study can be used by transportation planners to predict the employed people’s
behavior and travel demand analysis in addition to study the possibility and feasibility
of introducing the bus services to the transport system in Gaza city. The developed
models can be used for predicting the future modal split by inputting predicted future
value of exploratory variables.
Key wards: Gaza City, Mode Choice, Transportation Planning.
V
ملخص البحث
قطاع مدنتعتبر مدينة غزة واحدًة من أعلى المناطق السكنية كثافًة في العالم كما أنها األعلى كثافًة من بين و لقد أدى الضعف في التطبيق الفعال آلليات و أسس تخطيط النقل و المواصالت الى القصور في تبني .غزة
و يعد . عداد السكانأسياسات نقل مناسبة للحد من المشاكل المرورية المرتبطة بالتمدن و الزيادة المطردة في
لية التخطيط و صنع السياسات في مجال النقل و نموذج اختيار وسائل النقل واحدًا من العناصر المهمة في عم .المواصالت
و يتمثل الهدف الرئيسي لهذه الدراسة في بناء نموذج رياضي الختيار وسائل النقل لرحالت العمل في مدينة غزة مقد تلهذا الغرض ف و. المستخدمة لوسائل النقلالعامين العوامل التي تؤثر في اختيار ما يترتب عليه من تحديد
باستخدام و ذلكفي مدينة غزة لرحالت العمل وسائل النقل رالختيا خر افتراضياواقعي و نموذج بناء
بينما استبيان 555و البالغ عددها التي تم توزيعها لهذا الغرض تاالستبياناالمعلومات التي تم جمعها من ثلثي . بناؤهاتم تيذج الااستخدم الثلث المتبقي في اختبار صحة النم
بشكل معنوي وضحت نتائج الدراسة المتعلقة بالنموذج الواقعي الختيار وسائل النقل أن العوامل التي تؤثرأو لقد الكلية للرحلة مقسومة على التكلفة الزمن الكلي للرحلة،: لوسائل النقل في مدينة غزة هي العاملين في اختيار
امتالك وسيلة نقل، المسافة،العمر و متوسط الدخل الشهري األسرة،متوسط الدخل الشهري للفرد الواحد في التي تؤثر وضحت النتائج أن العوامل أما فيما يتعلق بالنموذج االفتراضي الختيار وسائل النقل فقد أ. للعائلة
لكل فرد زمن الرحلة ، التعرفة مقسومة على الدخل الشهري : بشكل معنوي في اختيار العاملين لوسيلة النقل هي ن كال أو قد بينت النتائج . من العائلة ، التكرار ، العمر ، متوسط الدخل الشهري للعائلة و طول الرحلة
لوسائل النقل المستخدمة في القوى العاملةالنموذجين الواقعي و االفتراضي لديهما القدرة على التنبؤ باختيارات
%.55حيث انهما صحيحان عند مستوى ثقة مدينة غزة
من قبل مخططي النقل و المواصالت التي تم بناؤها وصت الدراسة بان يتم استخدام النماذج أو في النهاية باإلضافة الى استخدامها في دراسة امكانية و جدوى إدخال خدمة في اختيار وسائل النقل العاملينللتنبؤ بسلوك
كما يمكن استخدام نتائج هذه الدراسة للتنبؤ المستقبلي . الى نظام المواصالت في مدينة غزة النقل بالباصات . بأنماط وسائل النقل و ذلك من خالل ادخال القيم المستقبلية المتوقعة للمتغيرات االستكشافية
VI
TABLE OF CONTENTS
Detecation ............................................................................................................ 1
Aknoledgement .............................................................................................. I1
Abstract .............................................................................................................. III
Abstract (arabic) ........................................................................................... V
table of contents ........................................................................................ VI
list of abbreviations ................................................................................... X
List of figures .................................................................................................. XI
list of Tables ................................................................................................. XIII
Chapter 1: Introduction ........................................................................... 1
1.1Background ............................................................................................................... 1
1.2 Problem statement .................................................................................................... 3
1.3 Research objectives .................................................................................................. 4
1.4 Research Significance .............................................................................................. 5
1.5Research Scope: ........................................................................................................ 5
1.6 Research methodology ............................................................................................. 5
1.7 Thesis organization .................................................................................................. 6
Chapter 2: Literature Review ................................................................ 7
2.1 Introduction .............................................................................................................. 7
2.2 Background .............................................................................................................. 7
2.3Urban transportation planning process ................................................................... 10
2.3.1Classical Four-Step Model ............................................................................... 11
2.3.1.1 Trip Generation ........................................................................................ 11
2.3.1.2 Trip Distribution Model ........................................................................... 13
2.3.1.3 Modal Split............................................................................................... 14
2.3.1.4 Trip Assignment....................................................................................... 14
VII
2.4 Mode choice model ................................................................................................ 15
2.4.1 Overview and historical development of the mode choice model .................. 15
2.4.2 Factors influencing mode choice .................................................................... 16
2.4.3 Aggregate mode choice models ...................................................................... 17
2.4.3.1 Trip-End modal split models ................................................................... 17
2.4.3.2 Trip-Interchange modal split models ....................................................... 17
2.4.4 Disaggregate (Discrete) mode choice models ................................................. 18
2.4.5 Theoretical Framework for disaggregate mode choice models ...................... 19
2.4.6 Types of Mode Choice Models ....................................................................... 21
2.4.6.1 Logit Model ............................................................................................. 22
2.4.6.1.1 Binary Logit Models ............................................................................. 23
2.4.6..1.2 Multinomial Logit Models ................................................................... 26
2.4.6.2 Probit Model ............................................................................................ 27
2.4.6.3 General Extreme Value Model ................................................................ 29
2.4.7 Comparison of Modal Split Models ................................................................ 29
2.4.8 Model Estimation Techniques ........................................................................ 31
2.4.8.1 Maximum Likelihood Method ................................................................. 31
2.4.8.2 Least Squares Method .............................................................................. 33
2.5 Sampling and Data Collection ............................................................................... 33
2.5.1 Travel Survey Types ....................................................................................... 33
2.5.1.1 Household Travel Surveys ....................................................................... 34
2.5.1.2 Workplace Surveys .................................................................................. 34
2.5.1.3 Destination Survey ................................................................................... 34
2.5.1.4 Intercept Survey ....................................................................................... 35
2.5.2 Sampling Generation Methods ....................................................................... 35
2.5.2.1 Simple Random Sampling ....................................................................... 35
2.5.2.2 Stratified Random Sampling .................................................................... 36
VIII
2.5.2.3 Multi-stage Sampling ............................................................................... 37
2.5.2.4 Cluster Sampling ...................................................................................... 38
2.5.2.5 Systematic Sampling ................................................................................ 38
2.5.3 Revealed and stated preference survey ........................................................... 39
2.6 Previous case studies of mode choice modeling .................................................... 41
2.7 Summary ................................................................................................................ 44
Chapter 3: Research Methodology ................................................. 46
3.1Stages of the Study.................................................................................................. 46
3.2Study area…………………………………………………………………………50
3.3Target group ............................................................................................................ 50
3.4Design of Questionnaire ......................................................................................... 51
3.5 Sample Size Determination.................................................................................... 53
3.6 Pilot Study .............................................................................................................. 54
3.7 Preliminary analysis of questionnaire .................................................................... 54
3.8 Model Calibration and Comparison ....................................................................... 55
3.9Model Validation .................................................................................................... 56
Chapter 4: Results &Analysis ............................................................... 58
4.1 Introduction ............................................................................................................ 58
4.2 General Analysis of Data ....................................................................................... 58
4.2.1 Gender of respondents .................................................................................... 58
4.2.2 Status of respondents ...................................................................................... 59
4.2.3 Jobs of respondents ......................................................................................... 60
4.2.4 Age of respondents ......................................................................................... 61
4.2.5 Monthly income of respondents...................................................................... 62
4.2.6 Family size of respondents.............................................................................. 63
4.2.7 Ownership of transport modes ........................................................................ 64
4.2.8 Trip length ....................................................................................................... 65
IX
4.2.9 The means of transport usually used by the respondentsUsually used by the
respondents ..................................................................................................... 66
4.3 Relation between the mode of transport and socioeconomic characteristics ......... 68
4.3.1 Relation between the mode of transport and gender ....................................... 68
4.3.2 Relation between the mode of transport and marital status ............................ 70
4.3.3 Relation between the mode of transport and age ............................................ 72
4.3.4 Relation between the mode of transport and family size ................................ 74
4.3.5 Relation between the mode of transport and the monthly income.................. 76
4.3.6 Relation between the mode of transport and the Job ...................................... 78
4.3.7 Relation between the mode of transport and the ownership of means of
transport .......................................................................................................... 80
4.3.8 Relation between the mode of transport and the length of trip ....................... 82
4.4 Relation between the captive ridership and socioeconomic characteristics .......... 85
4.5 hypothetical questions ............................................................................................ 86
4.6 Importance of factors that affect mode choice ....................................................... 88
4.7 Calibration of revealed model ................................................................................ 89
4.8 Validation for revealed model ............................................................................. 106
4.9 Calibration of stated preference model ................................................................ 107
4.10 Validation of stated preference model ............................................................... 114
Chapter 5: Conclusions & Recommendations ........................ 115
5.1 Summary .............................................................................................................. 115
5.2 Conclusions .......................................................................................................... 116
5.2 Recommendations ................................................................................................ 118
References .................................................................................................................. 119
Annex1: Questionnaire in Arabic…………….………………...…123
ANNEX2: Questionaaire in english ……….…...………………….128
X
LIST OF ABBREVIATIONS
UTMS Urban Transport Model System
O-D Origin –Destination
SPME Single Path Matrix Estimation
MPME Multiple Path Matrix Estimation
AON All or Nothing
DM Discrete Choice Models
IIA Independent of Irrelevant Alternative
OSL Ordinary Least squares
SPSS Statistical Package for Social Science
RII Relative Important Index
ELM Easy Logit Model
LRTS Likelihood Ratio Test
PCBS Palestinian Central Bureau of Statistics
SP Stated Preference
CAPI Computer Assisted Personal Interviewing
XI
LIST OF FIGURES
Figure (2.1): Role of transport modeling in policy making ........................................... 8
Figure (2.2): Classical Four-Step Model ....................................................................... 9
Figure (2.3): Urban transportation planning process ................................................... 10
Figure (2.4): Example of a Simple Binary Logit Model .............................................. 23
Figure (2.5): Example of a nested Binary Logit Model ............................................... 25
Figure (2.6): Example of a simple multinomial Logit Model ...................................... 26
Figure (2.7): Example of a nested multinomial Logit Model ...................................... 27
Figure (2.8): Classification of mode choice models .................................................... 30
Figure (2.9): Example of Multistage Sampling Process .............................................. 37
Figure (3.1): Flow chart for research methodology ..................................................... 49
Figure (3.2): Study Area (Gaza city) ........................................................................... 50
Figure (4.1): Respondent’s gender ............................................................................... 59
Figure (4.2): Respondent’s status ................................................................................ 59
Figure (4.3): Respondent’s job .................................................................................... 60
Figure (4.4): Respondent’s age .................................................................................... 61
Figure (4.5): Respondent’s monthly income ............................................................... 62
Figure (4.6): Respondent’s family size ........................................................................ 63
Figure (4.7): Respondent’s ownership of transport means .......................................... 64
Figure (4.8): Trip length .............................................................................................. 65
Figure (4.9): The percent of different modes usually used by the respondents ........... 66
Figure (4.10): The number of captive and choice riders for different modes .............. 67
Figure (4.11): The percent of male and female riders for different modes .................. 69
Figure (4.12): Distribution of transport modes for marital status ................................ 71
Figure (4.13): Distribution of transport modes for age ................................................ 73
Figure (4.14): Distribution of transport modes over family size ................................. 75
Figure (4.15): Distribution of transport modes over monthly income ......................... 77
XII
Figure (4.16): Distribution of transport modes over job .............................................. 79
Figure (4.17): Distribution of transport modes over ownership of transport means ... 81
Figure (4.18): Distribution of transport modes over trip length .................................. 83
Figure (4.19): Distribution of riders’ choice for different levels (total sample) .......... 87
XIII
LIST OF TABLES
Table (2.1): Comparison of Common Mode Choice Models (Khan 2007) ................. 30
Table (3.1): Different levels of the hypothetical questions.......................................... 52
Table (4.1): Frequency table for respondent’s gender ................................................. 58
Table (4.2): Frequency table for respondent’s status ................................................... 59
Table (4.3): Frequency table for respondent’s job ....................................................... 60
Table (4.4): Frequency table for respondent’s age ...................................................... 61
Table (4.5): Frequency table for respondent’s monthly income .................................. 62
Table (4.6): Frequency table for respondent’s family size .......................................... 63
Table (4.7): Frequency table for respondent’s ownership of means of transport ........ 64
Table (4.8): Frequency table for trip length ................................................................. 65
Table (4.9): Frequency table for the modes of transport thatUsually used by the
respondents .................................................................................................................. 66
Table (4.10): Frequency table for the choice and captive riders ................................. 67
Table (4.11): Cross tabulation between the mode of transport and gender ................. 69
Table (4.12): Chi-square test for mode-gender relationship ........................................ 70
Table (4.13): Cramer’s V statistics for mode-gender relationship .............................. 70
Table (4.14): Cross tabulation between the mode of transport and marital status....... 71
Table (4.15): Chi-square test for mode-marital status relationship ............................. 72
Table (4.16): Cramer’s V statistics for mode-marital status relationship .................... 72
Table (4.17): Cross tabulation between the mode of transport and age ...................... 73
Table (4.18): Chi-square test for mode-age relationship ............................................. 74
Table (4.19): Cramer’s V test for mode-age relationship ............................................ 74
Table (4.20): Cross tabulation between the mode of transport and family size .......... 75
Table (4.21): Chi-square test for mode-family size relationship ................................. 76
Table (4.22): Cramer’s V test for mode-family size relationship ................................ 76
Table (4.23): Cross tabulation between the mode of transport and monthly income .. 77
XIV
Table (4.24): Chi-square test for mode-monthly income relationship......................... 78
Table (4.25): Cramer’s V test for mode-monthly income relationship ....................... 78
Table (4.26): Cross tabulation between the mode of transport and job ....................... 79
Table (4.27) :Chi-Square Tests for mode-job relationship .......................................... 80
Table (4.28): Cramer’s V test for mode-job relationship ............................................ 80
Table (4.29): Cross tabulation between the mode of transport and ownership of means
of transport ................................................................................................................... 81
Table (4.30) :Chi-Square Tests for mode-ownership of transport means relationship 82
Table (4.31): Cramer’s V test for mode-ownership of transport means relationship .. 82
Table (4.32): Cross tabulation between the mode of transport and length of trip ....... 83
Table (4.33):Chi-Square Tests for mode-trip length relationship ................................ 84
Table (4.34): Cramer’s V test for mode-trip length relationship ................................. 84
Table (4.35): Test of relationship between the mode choice and travel socioeconomic
variables ....................................................................................................................... 85
Table (4.36): Test of relationship between the captive ridership and travel
socioeconomic variables .............................................................................................. 86
Table (4.37): Distribution of riders’ choice for different levels .................................. 86
Table (4.38): Relative Importance Index and Rank of the factors that affect mode
choice ........................................................................................................................... 88
Table (4.39): Abbreviation and description of explanatory variables ......................... 89
Table (4.40): Estimation results of model_1 ................................................................ 91
Table (4.41): Estimation results of model_2 ................................................................ 92
Table (4.42): Estimation results of model_3 ................................................................ 94
Table (4.43): Estimation results of model_4 ................................................................ 95
Table (4.44): Estimation results of model_5 ................................................................ 97
Table (4.45): Estimation results of model_6 ................................................................ 99
Table (4.46): Estimation results of model_7 .............................................................. 100
Table (4.47): Estimation results of model_8 .............................................................. 102
XV
Table (4.48): Estimation results of model_9 .............................................................. 104
Table (4.49): Abbreviation and description of explanatory variable used in stated
preference model ........................................................................................................ 107
Table (4.50): Estimation results of model_S1 ........................................................... 108
Table (4.51): Estimation results of model_S2 ........................................................... 109
Table (4.52): Estimation results of model_S3 ........................................................... 110
Table (4.53): Estimation results of model_S4 ........................................................... 111
Table (4.54): Estimation results of model_S5 ........................................................... 112
1
Chapter 1: Introduction
1.1 Background
Gaza strip is located at the southern part of Palestine with area of 365 km2. It
composed of five governorates which are; Gaza, Middle, Northern, Khanyounis, and
Rafah governorate. According to the census conducted by the Palestinian central
bureau of statistics (PCBS) ; the total number of populations of Gaza strip at the mid
2011 is 1.59 millions. The percent of males is about 50.6% and the females represent
about 49.4% of the populations. According to these figures Gaza strip is considered
one of the most densely populated area in the world with 4356 inhabitants/km2. The
population pyramid for Gaza strip shows that the Palestinian community is a young
society where the percent of population in the range between 0-14 years is about
44.1% and the populations between 15-29 years is about 29.7% while the percent of
populations over 65 years represents about 2.4% of the populations. The percent
populations within the work age (over than 15 years) represents about 51.7%. The
participation of labor force is 38.1% of the populations within the work age. The
percent of unemployment is about 40.6% of the peoples within the labor force.
Gaza is the densely populated governorate in Gaza strip with a density of 7.5
inhabitants/dun. The area of Gaza is 72593 dounms and the number of population at
mid 2011 is 552,000 persons Gaza city is composed of eleven districts. Gaza has one
of the most highly rate of populations increase with a rate of 4% annually. The
participation of labor force in Gaza is about 36.4% of the peoples within the work age
and the percent of unemployment is about 38.3%.
The construction of modern paved roads in Gaza began in the last century. The road
network which was planned and constructed between 1936-1945 aimed at serving the
British security and logistics requirements in the second war. During the period
between 1947-1967 there is no significant improvement of the road networks. Only
minor improvement and construction of few roads leading to population centers
during the Egyptian period which ended in 1967. During the Israeli occupation for
Gaza strip, only minor improvements of road network have taken place. The roads
were constructed and improved for serving the Israeli settlements. During this period
a Minor improvements for the roads that constructed during the Egyptian period for
serving the Palestinian. In May 1994 after signing the Interim Agreement in Cairo, the
2
Palestinian National Authority takes its role in constructing Gaza strip. During this
period a highly improvement of roads construction has occurred but this integrated
with highly increasing of population due to coming back high number of refugees to
Gaza strip. During this period the development in roads construction does not
integrated with improvement in transportation planning process and policies.
Currently Gaza city is facing a congestion and transportation problems resulting from
a rapid increase in population. In order to solve the transportation problems relating to
congestion problem a good understanding for the travelers’ behavior is needed to help
in adopting the suitable transport policies for mitigating these problems.
The transport system in Gaza city depends on land transport which can be categorized
into private and public transport. Due to the limited income levels in Gaza city, a
public transport service is playing a major role in satisfying the mobility of the
population. The public transport is served by three different modes in Gaza city which
are Shared taxi, taxi and buses. The buses in Gaza city are classified into public and
private buses. The registered and licensed public buses served the regional connection
between Gaza city and the other governorates in Gaza strip. Private buses don’t have
fixed lines but most of its work is directed towards school and university students. As
far as private buses are concerned they don’t have stations nor terminals and they
regularly don’t enter the city center. Shared taxis are the mean mode of public
transport services in Gaza city which operate to provide short haul services within the
Gaza city. Taxis are available in Gaza city for point to point transport.
Because of the growing and complex problems of congestion and air pollution in
recent years, urban policymakers have begun to ask for more sophisticated decision
tools including models to forecast travel demand and its effect under various
circumstances (Abdel-Aty and Abdelwahab, 2001). In urban and regional areas
transport models are gaining more and more importance both for traffic and transport
planning, design and operation in modern-information guidance and management
system.
Travel demand forecasting is a process used in transportation planning to predict
future demand of a transportation facility. The transportation planning process has
four basic steps that are trip generation, trip distribution, mode choice, and travel
assignment. Trip generation estimates the number of potential trips starting or ending
in a given area. Trip distribution associates origin and destination to each trip
3
generated. Mode choice analyzes how the trips are split between available modes of
transportation based on the attractiveness of each mode. Finally, travel assignment
estimates volumes on different links of the transportation network (Richardson, 2003).
The choice of transport mode is probably one of the most important classic models in
transport planning. This is because of the key role played by public transport in policy
making. Almost without exception public transport modes make use of road space
more efficiently than the private car. Moreover, if some drivers could be persuaded to
use public transport instead of cars, the rest of the car users would benefit from
improved levels of service. It is unlikely that all car owners wishing to use their cars
could be accommodated in urban areas without sacrificing large parts of the fabric to
roads and parking space (Ortuzar and Willumsen, 2002).
Mode choice process is needed when there are two or more alternatives. For example,
an individual going to work might choose among driving (if a vehicle is available),
ride public transit, or walk. Decision for taking an alternative are usually based on
complex factors. For example, a parent might decide to let his/her child walk to
school after evaluating the distance or travel time between home and school, the status
of sidewalks along the route, the pedestrian safety along the route, and neighborhood
safety issues. Another parent might allow his/her child to walk to school because
other children in their neighborhood are also walking to school (Koppelman and Bhat,
2006).
Discrete choice models are widely used in transportation modeling during the last 25
years and they have played an important role in transportation modeling. The reason
towards the wide using of these models is its ability to provide a detailed
representation of the complex aspects of transportation demand, based on strong
theoretical justifications. The art of finding the appropriate model for a particular
application requires from the analyst both a close familiarity with the reality under
interest and a strong understanding of the methodological and theoretical background
of the model (Abdel-Aty and Abdelwahab, 2001).
1.2 Problem statement
Since the Israeli occupation of Gaza Strip in 1967, the transport sector has suffered
from deterioration in terms of quality and quantity. The occupation neglected the
construction of infrastructure projects that can improve the transport sector. After
4
signing the Oslo Agreement between Israelis and Palestinians and establishment of
Palestinian National Authority in 1994 there was a dramatic improvement in the
construction of road networks. But from another side there was a highly increase in
population and vehicles due to coming back of a lot of refugees to Gaza strip. The
transport policies that were adopted by transport planners were not sufficient for
solving the transport problems resulting from the increase of travel demands.
Gaza city is currently facing urbanization and economic growth, with this, demand for
private and public transport have been increasing. To meet the increasing of travel
demand without increasing the congestion problem there is a need for increasing the
use of high occupancy modes in addition to encourage the use of non-motorized
modes (walking and biking). This could not be done without understanding the
travelers’ needs and preference of using the modes.
In orders to adopt a suitable transport policies for solving the expected congestion
problem resulting from urbanization and economic growth, there is a need for
improving the transport planning process in Gaza Strip. One step should be improved
is mode choice modeling, which is considered very essential for predicting the future
growth for each mode in addition to specifying the factors that contribute the use of
each mode and shifting from one mode to another one.
Developing countries including Gaza Strip often use the mode choice models that are
developed by the developed countries. These models are not suitable to be used as the
original form because of the different conditions and circumstances in developing
countries. Therefore, there is a need to develop mode choice model for Gaza in order
to help in predicting the future demand for each mode of transport and adopting the
suitable transport policies to solve the congestion problem.
1.3 Research objectives
The main aim of this study is to develop a mode choice model for work trips in Gaza
city that can be used to simulate the behavior of individuals towards motorized and
non-motorized modes.
This main aim includes the following objectives:
1. To provide a quantitative explanation of the choices of travel modes for work
trips in Gaza city.
2. To study the factors affecting the mode choice.
5
3. To specify the most significant factors which affect mode choice.
4. To study the various types of mode choice models.
5. To choose the most suitable model.
6. To calibrate and estimate the chosen mode choice model.
7. To validate the developed models.
1.4 Research Significance
The main aim of this study is to develop mode choice model for work trips in Gaza
city. In addition to its application in transport modeling process as a travel demand
forecasting tool, this mode choice model can be used in:
1. The analysis of probable market share of motorized and non- motorized
modes.
2. The computation of modal choice elasticity.
3. Determination of time value for Gaza city residents.
1.5 Research Scope:
The scope of this study will be limited for work trips in only Gaza city. The reason for
this limitation is the time and financial constraints.
1.6 Research methodology
This study comprises six main phases of work as follows:
First phase:
The first phase is the literature review on mode choice modeling. The concentration
will be on discrete mode choice models as they are more efficient than conventional
models. The literature review should seek for case studies applied in cities of
developing countries especially in the cities that have similar conditions. Based on the
literature review, the transportation planning process that is appropriate to Gaza City
must be decided.
Second phase:
This phase relates to the process of selection of the travel attributes. It involves,
designing of initial (pilot) survey form and analysis of the survey data. This process is
important to determine the attributes which are most relevant to the travelers in the
study area. The resulting attributes will be included in the main survey.
6
Third phase:
This phase involves designing the final survey form and conducting the survey from
start to finish including selection of level of attributes, implementation of the survey,
the collection and analysis of data.
Fourth phase:
This phase includes calibrating and estimating of the utility functions for the Model.
Fifth phase:
This phase is preliminary concerned with the model validation.
Sixth phase:
This phase summarizes the main findings and conclusions from the study.
1.7 Thesis organization
This thesis will be organized into six chapters:
Chapter one presents the introduction chapter which includes background,
problem definition, objectives, scope of the study, significance of the study
and research methodology.
Chapter two reviews briefly the literature related to discreet choice behavior in
different field of research including the current literature on transportation
planning models, and aggregate and disaggregate mode choice models.
Chapter three describes the methodology and approach for the analysis and
evaluation of the results. It also describes the explanation of theoretical
foundation of the proposed mode choice methodology.
Chapter four describes the results of the descriptive analysis of the survey as
well as development of the mode choice model for work trips in Gaza city. It
begins with the estimation procedure, calibration and validation of the model.
Chapter five concludes the study with main findings from this behavioral
experiments and how the objectives of this study have been addressed. This
chapter includes conclusions and recommendations in addition to some
thoughts of future researches.
7
Chapter 2: Literature Review
2.1 Introduction
As the mode choice model is a part of transport planning process and the third step in
4-step travel demand forecasting, this chapter presents a state of the art literature
review on passenger mode choice model. The literature reviewed in section 2.2
includes background on transportation planning process. Section 2.3 discusses the
transportation planning process in some details with a brief description of the classical
4-step model. Section 2.4 illustrates the factors that affect the choice of transportation
modes and the approaches for modeling the mode choice. This section also presents
the types and estimation techniques for various types of mode choice models along
with selecting a particular discrete mode choice model in order to forecast the travel
demand behavior for this research. Section 2.5 presents the types of travel survey and
sampling generation methods. In addition, the difference between the revealed and
stated preference survey is discussed in this section. Section 2.6 presents for the
experience of some countries in the field of modeling the mode choice behavior for
various types of trips. Finally section 2.8 summarizes the main findings from the
literature review revealing the research framework design to forecast the travel
behavior of the study area.
2.2 Background
Modeling is one important part of the most decision making process. It is concerned
with the methods, be they quantitative or qualitative which allows us to study the
relationships that underlie the decision making. (Hensher and Button, 2000)
A model is defined as a simplified representation of the real world which concentrated
on certain elements considered important for its analysis from a particular point of
view (Qrtuzar and Willumsen, 2002).
A transport models can be defined as a simplified representation of the real world
usually implemented in computer software which describe the impact of transport
decisions. Transport models can cover whole countries, cities, areas or simply
individual junctions (European commission, 1996).
Transport models allow alternative solutions to problem to be tested before resources
are committed to implementing them. Models can also be used to:
8
Create traffic control systems which response to changing transport solutions,
allowing changing to control to be made automatically every few minutes.
Help understanding the full range of impacts which may result from transport
scheme.
Evaluate the costs and benefits of transport investments and so to prioritize
investments.
Transportation is very important for sustainable development of economics. Large
investments have been made in transportation planning and policy making in order to
forecast the future demand of travel. The forecasting needs to integrate between the
designing of existing transport system and the behavior of residents in the study area.
(Khan, 2007).Transportation modeling plays an important role in supporting
transportation planning and policy making as illustrated in Figure 2.1.
Figure (2.1): Role of transport modeling in policy making (Richardson, 2003)
PROBLEM
DEFINTION
System
Resources Objectives
TRANSPORT
MODELS
Criteria Consequences
Evaluation
Selection
Implementation Monitoring
Constraints
Alternatives
Data
Collection
9
The fundamentals of transport modeling were developed in the united stated during
the 1950s' and imported in the UK in the early of the 1960s'. Thereafter the following
20 years saw important theoretical development in the field of transport modeling
leading to further work in specific sub-areas. A contemporary dimension was the
development of transport mode choice models representing the behavior of travelers
of the study area. Since then the interest of this field as well as the growing
complexity has led to further development of various travel demand models. However
the most of these models trace their origin back to classical transport demand models,
the four-step models because of its overcharging framework and logical appeal. The
basic structure of the model is illustrated in Figure 2.2
Figure (2.2): Classical Four-Step Model (MCNally, 2000)
One of the most important aspects of the transportation modeling is to predict the
travel choice behavior which is the most frequently modeled travel decisions. It
involves specific aspects of human behavior dedicated to choice decisions.
Traditionally aggregate models are used in dealing with the travel choice behavior of
individual travelers; however the aggregate models have the limitation of forecasting
and estimating of travel choice with aggregated zonal data.
Disaggregate behavioral demand models which became popular during the 1980's
offer substantial advantages over the aggregate counterparts. Disaggregate behavioral
Trip Generation
Trip Distribution
Modal Split
Trip Assignment
11
models are based on the observed choices behavior of individual travelers. These
models considered that the demand is the result of several decisions of each individual
traveler. A discrete choice analysis is the methodology used to analyze and predict the
traveler decisions. The discrete choice model is mathematical functions which
estimate the probability of selecting individual travel choice based on the utility
maximization principle or relative attractiveness of competing alternatives (Qrtuzar
and Willumsen, 2002).
Revealed and stated preference survey data which contains data sets of individual
decisions, characteristics of the individuals and the alternative choices of the trip is
used to develop the discrete choice model (Qrtuzar and Willumsen, 2002).
2.3Urban transportation planning process
Transportation planning is undertaken at many levels from strategic planning to
project level planning at different geographic scales in any urban area. The urban
transport planning process can be classified into three phases as shown in Figure (2.3)
(Hanson, 1995).
Figure (2.3): Urban transportation planning process (Hanson, 1995)
11
The pre- analysis phase concerns with defining the current situation and problems
associated with the mode for trips. This is followed by the proposed solutions of
introducing new modes. The second part of this phase includes the data collection to
be used in technical phase and evaluation. Review of both secondary sources from
available reports and documents and the primary data collection is made as required.
The technical analysis phase concerns with using the mathematical description of
travel and related behavior to predict consequences of each scenario of transport
planning that is to be evaluated.
The post analysis phase comprises of predictions of the impacts of alternative plans
and policies. The purpose of these predictions is to inform decision making. The post
analysis phase of urban transportation planning includes evaluating the impacts of
alternatives, selecting the alternatives to be implemented and future programs
associated with in.
2.3.1Classical Four-Step Model
The Urban Transport Model System (UTMS) often referred as the 4-step model is
commonly used to predict the flows on the links of a particular transportation network
as a function of the land-use activity system that generates the travel (Hanson, 1995).
The model comprises of four sub-models as shown in Figure (2.2) that are employed
in sequential process: Trip generation, Trip distribution, Mode choice (or modal split)
and trip assignment.
2.3.1.1 Trip Generation
The trip generation stage of the classical transport model aims at predicting the total
number of trips generated by and attracted to each zone of the study area. Since, it
essentially defines the total travel in the study area, it is after trip generation analysis
that the transportation planner comes up with the vital figures about the total number
of trips generated and attracted by each zone, purposes of these trips, and the
travelling modes generally used for these trips.
Ortuzar and Willumsen (2002) have demonstrated common trip generation patterns on
the basis of following standard trip purposes,
• Work trips;
• Educational trips;
12
• Shopping trips; and
• Other trips (social, recreational, medical, bureaucratic trips etc.).
The most commonly used analytical technique to develop the trip generation patterns
of a study area is multiple linear regressions. In this technique, the dependent output
variable is assumed to have a linear dependence on the independent input variables,
which may or may not influence the trip generation, as shown in Equation 2.1.
EXXXY KK .......11110 (2.1)
Where,
,.....,2,1,0 K are coefficients of regression;
,.....,2,1,0 KX are independent input variables;
Y is dependent input variable;
E is error in estimating the output variable.
Definitions of the input and output variables vary with the type of linear regression
approach used in the research. Generally, two types of regression techniques are
applied in multi-modal transportation planning namely,
• Zonal-based Multiple Linear Regression; and
• Household-based Multiple Linear Regression.
The main difference between the two techniques is that the first is used to generate the
travel patterns on zonal basis, while the second does it at a household level.
Therefore, for zonal-based regression, Y is generally taken as the number of trips
generated for and attracted by each zone in the study area, while various independent
variables can be considered and tested for estimation purposes such as,
• Employment density of a zone1 (for work trips);
• School / university enrolment of a zone (for education trips); and
• shopping areas in a zone (for shopping, work, other trips).
Similarly, household-based regression tends to utilize various parameters associated
with a household, in order to estimate the regression coefficients, such as,
• Household size;
13
• Number of vehicles in a household;
• Number of adults in a household; and
• Number of workers and students in a household.
2.3.1.2 Trip Distribution Model
The trip distribution stage of the four-step model tends to provide a standard pattern
of trip making by linking the trip ends with the origins. The trip distribution model is
essentially a destination choice model and generates a trip table, for each trip purpose
utilized in the model as a function of activity-system attributes and network attributes.
This trip table, also commonly known as Origin-Destination Matrix (O-D Matrix),
provides a comprehensive illustration of the number of trips generated between
different zones of the study area. There are different traffic distribution algorithms for
forecasting the future O-D matrix which are: i) growth factor methods, ii) gravity
model iii) the entropy-maximizing approach, and iv) the proportional approach
(Qrtuzar and Willumsen, 2002).
A number of efforts have been made by transport researchers for developing efficient
and adaptive algorithms in order to optimize the O-D Matrix for achieving realistic
results. Nielsen (1994) presented two new methods for trip matrix estimation; namely
Single Path Matrix Estimation (SPME) and Multiple Path Matrix Estimation
(MPME), and demonstrated that the traffic models can be easily and cheaply
estimated using them. Three different approaches to O-D Matrix estimation were
reviewed and compared, in the context of transport planning, by Abrahamsson (1996)
who attempted to use the trip assignment parameters to calibrate the O-D matrix of
the study area. Later, Abrahamsson (1998) illustrated an O-D matrix for Stockholm,
Sweden that can reproduce the traffic counts, in terms of the number of trips
generated and attracted, using the previous distribution approaches improving the
accuracy of forecasting of O-D Matrices. Various computationally efficient
algorithms for estimating the trip distribution matrices were developed by Safwat and
Magnanti (2003) by using a simultaneous approach to develop a four-step model
rather than the conventional sequential method. Further, Ber-Gera and Boyce (2003)
developed a trip origin based algorithm for transportation forecasting models that
combine travel demand and network assignment variables in order to improve the
existing O-D flow models. Sherali et al. (2003) developed a non-linear approach to
14
estimate the O-D trip matrices by implicitly determining the path decomposition of a
network flow using a sequential linear programming approach.
2.3.1.3 Modal Split
Mode choice predicts the number of trips from each origin to each destination that
will use each mode of transportation. Clearly modal split has considerable
implications for transportation policy, particularly in large metropolitan areas (Qrtuzar
and Willumsen, 2002).
The issue of selecting the most appropriate travelling mode has always been a critical
issue in travel behavioral modeling, since it tells an individual about the most efficient
travelling mode available. Therefore, it is vital to develop and use models that are
receptive to those attributes of travel that influence a certain individual’s choice of
mode. The quantification of this interaction in terms of mathematical relationships is
known as modal split and the travel demand models are referred to as modal split or
mode choice models. Hence, the modal split assists a transport planner to assess the
impact of each urban element on mode choice and permits testing and evaluation of
various transportation schemes. This will be discussed in details in the next section
2.4
2.3.1.4 Trip Assignment
Trip assignment is the last stage of the four-step model, dealing with the allocation of
a given set of trip interchanges to a specific transport network. Its main objective is to
estimate the traffic volumes and the corresponding travel times or costs on each link
of the transportation system by the help of inter-zonal or intra-zonal trip movements
(determined by trip generation and distribution) and the travel behavior of the
individuals (determined by modal split).
The proportion of vehicles using each route between a particular origin-destination
pair depends upon a number of attributes and the alternative routes including travel
time, distance, number of stops / signals, aesthetic appeal etc. But travel time is the
attribute most commonly considered in network assignment models. There are
different traffic assignment algorithms which are: i) All-or-Nothing assignment
(AON), ii) Wardrop’s user Equilibrium assignment, iii) Method of successive
averages, iv) Stochastic user-equilibrium assignment (Qrtuzar and Willumsen, 2002).
15
Patriksson (1994) has presented a list of useful purposes of trip assignment in context
with transport planning namely,
• Assessing the deficiencies in the existing transportation system of the study area;
• Evaluating the effects of limited improvements and extensions to the existing
transportation systems;
• Developing construction priorities for the existing transportation system of the study
area; and
• Testing alternative transportation system proposals.
2.4 Mode choice model
2.4.1 Overview and historical development of the mode choice model
The model developed by Adam (1959) is one of the first modal split models to be
advised. Since mid 1960's many mode choice models for intercity travelers were
calibrated and used for prediction in various environments. Traditionally aggregate
models are used in dealing with the travel choice behavior of individual travelers;
however the aggregate models have the limitation of forecasting and estimating of
travel choice with aggregated zonal data.
The inability of aggregate data to explain the travelers’ behavior led to propose
another group of models called disaggregate models. These models require the data
that describes the behavior of an individual's characteristics and attitudes towards the
travel services provided by each mode.
Disaggregate travel demand models represent a recent innovation in travel forecasting
procedure. The earliest research into disaggregate mode choice models was done by
Warner (1962). The pioneering efforts were made during the period 1967-1969 used a
binary mode choice modeling with automobile as a base mode. Pioneers in this early
age of disaggregate modeling include, Quarmby (1967) [used discriminate analysis] ,
Elisco (1967) [used Probit analysis], Stopher (1969) [ used regression and
subsequently Logit analysis].
However it was found from the market research that besides the socioeconomic and
mode related characteristics and individual evaluates his choice depending upon the
level of service provided by the alternatives. This led to incorporation of factors such
as comfort, convenience, privacy and other mode related attitudinal indicators in the
16
models. One of the earliest efforts in this field was done by Ackoff (1965). His effort
was pioneering in considering psychological factors in mode choice. With the
development of transportation system the technology in various parts of the world,
attempts were made by prominent researcher in transportation planning to incorporate
the attitude of travelers in mode choice models.
2.4.2 Factors influencing mode choice
The factors influencing mode choice may be classified into three groups and a good
mode choice model should include the most important of these factors. These factors
are presented in Ortuzar, Willumsen, (2002) as follows:
a) Characteristics of the trip maker.
The following features are generally believed to be important:
• Car availability and/or ownership;
• Possession of a driving license;
• Household structure (young couple, couples with children, retired, singles, etc.),
• Income;
• Decisions made elsewhere, for example the need to use a car at work, take children
to school, etc;
• Residential density
b) Characteristics of the journey.
Mode choice is strongly influenced by:
• The trip purpose; for example, the journey to work is normally easier to undertake
by public transport than other journeys because of its regularity and the adjustment
possible in the long run;
• Time of the day when the journey is undertaken. Late trips are more difficult to
accommodate by public transport.
c) Characteristics of the transport facility.
These can be divided into two categories. Firstly, quantitative factors such as:
• Relative travel time: in-vehicle, waiting and walking times by each mode;
17
• Relative monetary costs (fares, fuel and direct costs);
• Availability and cost of parking.
Secondly, qualitative factors which are less easy to measure, such as:
• Comfort and convenience;
• Reliability and regularity;
• Protection, security.
2.4.3 Aggregate mode choice models
There are two basic ways of modeling aggregate behavior namely, aggregate and
disaggregate approaches. The aggregate approach directly models the aggregate share
of all or a segment of decision makers choosing each alternative as a function of the
characteristics of the alternatives and socio-demographic attributes of the group.
There are two types of aggregate mode choice mode namely, trip-end modal split
models and Trip-interchange modal split models
2.4.3.1 Trip-End modal split models
Traditionally, the objective of transportation planning was to forecast the growth in
demand for car trips so that investment could be planned to meet the demand. When
personal characteristics were thought to be the most important determinants of mode
choice, attempts were made to apply modal-split models immediately after trip
generation. Such a model is called trip-end modal split model. In this way different
characteristics of the person could be preserved and used to estimate modal split. The
modal split models of this time related the choice of mode only to features like
income, residential density and car ownership. The advantage is that these models
could be very accurate in the short run, if public transport is available and there is
little congestion. Limitation is that they are insensitive to policy decisions as example:
Improving public transport, restricting parking etc. would have no effect on modal
split according to these trip-end models (Ortuzar, Willumsen, 2002).
2.4.3.2 Trip-Interchange modal split models
This is the post-distribution model; that is modal split is applied after the distribution
stage. This has the advantage that it is possible to include the characteristics of the
journey and that of the alternative modes available to undertake them However, they
18
make it more difficult to include the characteristics of the trip maker as they may have
already been aggregated in the trip matrix (or matrices). It is also possible to include
policy decisions. This is beneficial for long term modeling. (Ortuzar, Willumsen,
2002).
The important limitation of these models is that they can only be used for trip
matrices of travelers who have a choice available to them. This often means the
matrices of car available persons, although modal split can also be applied to the
choice between different public transport modes (Ortuzar and Willumsen, 2002).
The models have little theoretical basis and therefore their forecasting ability must be
in doubt. They also ignore a number of policy sensitive variables like fares, parking
charges and so on. Further, as the models are aggregate they are unlikely to model
correctly the constraints and the characteristics of the modes available to individual
households (Ortuzar and Willumsen, 2002).
2.4.4 Disaggregate (Discrete) mode choice models
The disaggregate approach is to recognize that aggregate behavior is the result of
numerous individual decisions and to model individual choice responses as a function
of the characteristics of the alternatives available to and socio-demographic attributes
of each individual. Disaggregate mode choice models have substantial advantages
over the aggregate models for predicting the consequences of transportation policy
measures that affect mode choice. The advantages and useful proprieties of
disaggregate models have presented by (Ortuzar and Willumsen, 2002, Koppel and
Bhat, 2006, and Siddiqui, 1999) as follows:
1. The disaggregate approach explains why an individual makes a particular
choice given her/his circumstances and is, therefore, better able to reflect
changes in choice behavior due to changes in individual characteristics and
attributes of alternatives. The aggregate approach, on the other hand, rests
primarily on statistical associations among relevant variables at a level other
than that of the decision maker; as a result, it is unable to provide accurate and
reliable estimates of the change in choice behavior due changes in service or
in the population.
2. Disaggregate models avoid biases inherent in aggregate models.
19
3. Disaggregate models more efficient than aggregate one in terms of data and
computational requirements.
4. Disaggregate models can be developed using data less than one tenth of that
required by that aggregate models
5. The disaggregate approach, because of its causal nature, is likely to be more
transferable to a different point in time and to a different geographic context, a
critical requirement for prediction.
6. Discrete choice models are being increasingly used to understand behavior so
that the behavior may be changed in a proactive manner through carefully
designed strategies that modify the attributes of alternatives which are
important to individual decision makers. The disaggregate approach is more
suited for proactive policy analysis since it is causal, less tied to the estimation
data and more likely to include a range of relevant policy variables.
7. .DM models allow for a more flexible representation of the policy variables
considered relevant for the study.
8. The coefficients of the explanatory variables have a direct marginal utility
interpretation (i.e. they reflect the relative importance of each attribute).
2.4.5 Theoretical Framework for disaggregate mode choice models
A proposed framework for the choice process is that an individual first determines the
available alternatives; next, evaluates the attributes of each alternative relevant to the
choice under consideration; and then, uses a decision rule to select an alternative from
among the available alternatives (Koppel and Bhat, 2006)
An individual is visualized as selecting a mode which maximizes his or her utility
(Khan, 2007). The utility of a travelling mode is defined as an attraction associated to
by an individual for a specific trip. Therefore, the individual is visualized to select the
mode having the maximum attraction, due to various attributes such as in-vehicle
travel time, access time to the transit point, waiting time for the mode to arrive at the
access point, interchange time, travelling fares, parking fees etc. This hypothesis is
known as utility maximization.
As a matter of computational convenience, the utility is generally represented as a
linear function of the attributes of the journey weighted by the coefficients which
attempt to represent their relative importance as perceived by the traveler. A possible
21
mathematical representation of a utility function of a mode m is shown in Equation
(2.2) as,
(2.2)
Where,
is the net utility function for mode m for individual i;
, …, are k number of attributes of mode m for individual i; and
, ………, are k number of coefficients (or weights attached to each
attribute) which need to be inferred from the survey data.
As with deterministic choice theory, the individual is assumed to choose an
alternative if its utility is greater than that of any other alternative. The probability
prediction of the analyst results from differences between the estimated utility values
and the utility values used by the traveler
The choice behavior can be modeled using the random utility model which treats the
utility as a random variable, i.e. comprising of two distinctly separable components: a
measurable conditioning component and an error component. Therefore
(2.3)
Where,
is the systematic component (observed) of utility of mode m for individual i;
is the error component (unobserved) of utility of mode m for individual i.
For equation 2.3 to be correct, certain homogeneity is needed within the population
under study. In principle, it is required that all the individuals share a universal set of
alternatives and face the same constraints. Furthermore, in practical modeling work,
the difference between the socioeconomic characteristics of similar groups of
individuals is usually ignored (Ortuzar and Willumsen, 2002). Although this approach
makes the whole process simple overall, there is still a possibility of occurrence of
severe differences among various groups of people. This can be handled by
segmenting the entire set of individuals into separate utility functions for each group
of more similar individuals so that individual characteristics could be omitted from
the utility function.
21
By ignoring the attributes of the decision maker, the systematic component of the
utility can be treated as a function of attributes of available modes only. Therefore, a
single utility function can be visualized to exist for all individuals. Similarly, the error
component of the utility can also be considered independent of socioeconomic
characteristics for the same reason. Assuming that the error component has zero mean
and an extreme value distribution (Khan, 2007). The net utility function can be given
as:
(2.4)
Thus, if there are M number of total travelling modes available, the probability of an
individual selecting mode m, such that m Є M, is based on its associated utility
function Um, such that,
(2.5)
Where,
Um represents utility of travelling alternative m; and
Ui represents utility of any travelling alternative in the set of available
travelling modes.
Summarizing the theory of utility maximization presented in Equation 2.5, every
alternative associates a certain utility with itself determined by its various attributes
and an individual is supposed to select the alternative possessing the highest utility.
However, it is impractical to assume that the effects of all the variables in an
individual’s decision regarding the selection of a travel mode are perfectly
understood. The beauty of a random utility model is that it possesses the power to
estimate the effects of the observed variables without fully concerning that of the
unobserved ones incorporating all of them into the error component of the model, as
shown in Equation 2.4.
2.4.6 Types of Mode Choice Models
As mentioned above, Random Utility are the most used discrete choice models for
transportation applications. They have three different families of models depends
upon the functional form of the error term distribution:
1. Logit model
22
2. Probit Model
3. General Extreme Value Model.
2.4.6.1 Logit Model
Logit models are the most commonly used modal split models in the area of
transportation planning, since they possess the ability to model complex travel
behaviors of any population with simple mathematical techniques. The mathematical
framework of logit models is based on the theory of utility maximization (Ben-Akiva
and Lerman, 1985). There are three basic types of logit models depend on whether the
data or coefficients are chooser-specific or choice-specific .Multinomial logit model
has chooser-specific data where coefficients vary over the choices. Conditional logit
model has choice-specific data where the coefficients are equal for all choices. Mixed
logit model involves both types of data and coefficients (Siddiqui, 1999)
Briefly presenting the framework, the probability of an individual i selecting a mode
n, out of M number of total available modes, is given as,
Where,
Vin is the utility function of mode n for individual i;
Vim is the utility function of any mode m in the choice set for an individual i;
Pin is the probability of individual i selecting mode n; and
M is the total number of available travelling modes in the choice set for
individual i.
The theoretical framework of logit models is based on three main assumptions
regarding the error term , as shown in Equation 2.4. The assumptions are listed as
follows,
• is Gumbel distributed;
• is independently distributed; and
23
• is identically distributed.
All these three assumptions serve as the main postulates of the structure of logit
models. The first assumption of the random component being Gumbel distributed
indicates that all the utilities associated with the travelling modes should be
considered as a linear sum of attributes and has the same scale parameter (Ben- Akiva
and Lerman, 1985). The last two assumptions are normally grouped together to be
referred to as a property of Independence of Irrelevant Alternatives (IIA property),
simply meaning that all the travel modes used in modeling the travel behavior are
independent of each other.
Logit models are generally classified into two main categories namely binary and
multinomial logit models. Binary choice models are capable of modeling with two
discrete choices only, i.e. the individual having only two possible alternatives for
selection, where as the multinomial logit models imply a larger set of alternatives.
2.4.6.1.1 Binary Logit Models
The mathematical framework of a binary logit model can be represented by
simplifying of Equation 2.6 with the total number of available alternatives limited to
two, i.e. M = 2. An example of a binary logit model is shown in Figure 2.4 where the
choice set contains car and public transport as two competing alternatives.
Fig (2.4): Example of a Simple Binary Logit Model
24
Simplifying Equation 2.6, the probability of individual i selecting the mode m out of
two available travelling modes m and n is given as,
Or,
Where,
Vim is the utility function associated to alternative m for individual i;
Vin is the utility function associated to alternative n for individual i;
Pim is the probability that alternative m will be selected by individual i; and
Pin is the probability that alternative n will be selected by individual i.
The principle limitation of binary logit model mentioned above is that it relies on the
random or unexplained elements for each mode being independent (not correlated).
When there are groups of more similar or correlated modes, the assumption of having
an independent and identical error term across all the modes does not always remain
valid.
Nested (hierarchical) logit model can be used to relax the constraints of a simple logit
model by allowing correlation among utilities of the alternatives in a common group.
The nested logit model is constructed by grouping similar modes into hierarchical or
nests. Each nest, in turn, is represented by a composite alternative which competes
with the others available to the individual. The common example the choice between
the private car, bus and rail which can be represented as shown in Figure 2.5
25
Figure (2.5): Example of a nested Binary Logit Model
The theoretical framework of the nested logit model is based on the same assumptions
as the multinomial logit model, except that the correlation of error terms is assumed to
exist among various modes. Due to the tree structure of these models, Equation 2.6 is
reassessed, for trees having two levels, as,
(2.10)
(2.12)
(2.13)
26
Where,
C (i) is a set of lower-level alternatives that each form part of the higher-level
alternative i;
R is the set of higher-level alternatives;
Xj|I is the measured attractiveness of alternative j conditional on i;
Xi is the measured attractiveness of alternative i; and
Hi is the scale parameter.
The nested binary logit model has disadvantages over the simple binary logit model ,
mainly due to calibration difficulties but with increased the computing power theses
issues have been largely overcome. The choice of the most appropriate nesting
structure could, in theory, be a problem but practitioners have derived workable
nesting arrangement through time.
2.4.6..1.2 Multinomial Logit Models
The multinomial logit model can be derived from binary logit model to deal with to
deal with more than two modes. The multinomial logit model is categorized into
simple and nested multinomial logit models based on the characteristics of the
available travelling alternatives in the choice set. The examples of simple and nested
multinomial logit models are presented in Figures 2.6 and 2.7 respectively
Figure (2.6): Example of a simple multinomial Logit Model
27
Figure (2.7): Example of a nested multinomial Logit Model (Khan, 2007)
The multinomial logit models use the same mathematical framework as shown in
Equations 2.2 to 2.14 and are generally estimated using maximum likelihood method.
2.4.6.2 Probit Model
In certain cases where the utilities of some alternatives are correlated in a complex
way, the using of multinomial logit models can make incorrect forecasts regarding the
probabilities of mode shares when the attributes associated to one or more travelling
alternatives are varied. In these cases the probit can be used as one of the possible
methods to overcome this problem. The essential difference between the probit model
and the logit model is that the cost function coefficients in a probit model are random
(about normal distribution) compared to mean values in a logit model. The
multinomial probit model can address some of the problems associated with assuming
constant costs across individuals.
Similar to logit models, the probit model is based on random utility theory,
representing the utility function as the sum of the systematic component and an error
component. The model follows normal distribution for error terms and does not work
under the strict assumptions as that of logit models The standard equation for the
utility and the probability of an alternative i has the form (Horowitz, 1991 cited in
Khan, 2007) as shown in Equation 2.15 and 2.16,
28
(2.15)
And,
(2.16)
Where the covariance matrix of the Normal distribution associated to this latter model
has the form:
(2.17)
Where,
Ui is the utility of alternative i;
V is the systematic (observed) component of the utility function;
ε is the error (unobserved) component of the utility function;
xi is the vector of observed attributes of alternative i; and
s is the vector of observed characteristics of the individuals of the study area.
is the variance
The main disadvantage of the probit model is that it is very difficult to specify
mathematically especially for cases with more than three alternatives are available.
Due to this complexity the transport planners generally prefer using logit models as
they possess simple mathematical framework and can accurately model the travel
behavior of a study area. Ghareib (1996) estimated the travel behavior for different
cities of Saudi Arabia using logit and probit models and concluded that the logit
models are superior to their probit counterparts in terms of their goodness-of-fit
measures and tractable calibration. Dow and Endersby (2004) later supported his
findings by concluding that the logit models should always be preferred over probit
29
models and the latter should only be utilized if the travel behavior of the targeted
population to be determined is observed to be complexly correlated
2.4.6.3 Generalized Extreme Value Model
Generalized extreme value (GEV) models were developed as an important
simplification of multinomial logit models based on the stochastic utility
maximization. Although there exist a limitless number of possible models within this
class, only a few have been truly explored. This model is based on a function G(y1,
y2, …, yJn), for y1, y2, …, yJn ≥ 0, that has to satisfy certain conditions discussed in
detail in Ben-Akiva and Lerman (1985). The basic equation of the model is given as,
Where,
V is the systematic (observed) component of the utility function;
is the degree of homogeneity; and
Pn(i) is the probability of individual n selecting alternative i.
In addition to the three modal split models discussed above, there are a few discrete
choice models which can be referred as the generalizations of logit models, namely
Random Coefficient Logit, Tobit and Ordered Logistic models. Due to the occurrence
of high limitations in the specifications and estimation complexities of these models,
they are rarely put into practice by transport planners. A detailed mathematical
framework of these models is presented in Ben-Akiva and Lerman (1985).
2.4.7 Comparison of Modal Split Models
The generation of travel profile of the study area and determination of choice sets is
the first step in modal split modeling. The determination of choice sets size play an
important role in selecting the appropriate mode choice model in order to forecast the
travel behavior of the study area. If the choice set consists of two travelling modes, or
two sets of travelling modes, a binary modal split model can be applied. Contrarily,
31
multinomial modal split models can be selected for bigger choice sets. This
classification of the discrete mode choice models on the basis of the choice set is
illustrated in Figure 2.8
Figure (2.8): Classification of mode choice models (Khan, 2007)
The main difference among the three most common mode choice models namely,
Logit, probit, and general extreme value models are shown in Table (2.1) identifying
the main distinguishing factors among the specifications and applications of these
models.
Table (2.1): Comparison of Common Mode Choice Models (Khan, 2007)
Item Logit Model Probit Model General Extreme
value model
Basic Hypothesis Extreme value
distribution Normal distribution
Multivariate
extreme value
distribution
Major constraints
Error term should
be identically and
independently
distributed
Error term need not
be identically and
independently
distributed
Error terms need
not necessarily be
identically and
independently
distributed
31
Model
Formulation Simple Complex Complex
Model Estimation Simple Complex Complex
Introduction of
Access Modes
Model formulation
and calibration
becomes complex
to a small degree
Model formulation
and calibration
becomes highly
complex
Model formulation
and calibration
becomes highly
complex
Applications High Limited Limited
Accuracy High Low Low
The table mentioned above illustrates the reasons stand behind the using Logit models
among transportation planners for estimation and forecasting of travel behavior
although the specifications developed for logit models associate certain limitations
due to the IIA property. The main reasons for choosing them are their simple model
formulation and estimation techniques. Other mode choice models such as probit and
general extreme value models have relaxed the IIA restriction at the cost of
possessing highly complex mathematical structure and computational estimation.
Therefore, the logit models continue to remain dominant in the transport modeling
field (khan, 2007).
2.4.8 Model Estimation Techniques
Generally the most estimation techniques that are used for estimating the discrete
mode choice models, namely the maximum likelihood and least squares method.
These methods are used in order to infer the values of the unknown coefficients ,
…, shown in Equation 2.2,. Brief discussions of these methods are presented in the
next Sections. A detailed literature of the theoretical framework, applications and
limitations of these models is presented in (Ortuzar and Willumsen, 2002).
2.4.8.1 Maximum Likelihood Method
The method of maximum likelihood is the most common procedure used for
determining the estimators in simple and nested logit models. Stated simply as,
32
"The maximum likelihood estimators are the values of the parameters for which
the observed sample is most likely to have occurred" Ben-Akiva and Lerman (1985).
The procedure for maximum likelihood estimation involves two important steps:
1. developing a joint probability density function of the observed sample, called the
likelihood function, and
2. Estimating parameter values which maximize the likelihood function.
The likelihood function for a sample of ‘I’ individuals, each with ‘M’ alternatives are
defined as follows:
Where,
L is the likelihood the model assigns to the vector of available alternatives;
is the probability that individual i chooses alternative m.
is chosen indicator (=1 if j is chosen by individual i and 0, otherwise)
The values of the parameters which maximize the likelihood function are obtained by
finding the first derivative of the likelihood function and equating it to zero. The most
widely used approach is to maximize the logarithm of L rather than L itself. It does
not change the values of the parameter estimates since the logarithmic function is
strictly monotonically increasing. Thus, the likelihood function is transformed to a
log-likelihood function and is given as,
The first derivative of the logarithm of likelihood function can be represented as
shown in Equation 2.21
33
The maximum likelihood is obtained by setting Equation 2.21 equal to zero and
solving for the best values of the parameter vector, . to insure this is the solution
for a maximum value provided that the second derivative is negative definite.
Given the mode choice data, most existing estimation computer programs estimate the
coefficients that best explain the observed choices in the sense of making them most
likely to have occurred. Standard commercial packages such as ALOGIT are
generally implied for estimating logit models, mostly due to their capability of
handling complex nested logit structures, both linear and non-linear.
2.4.8.2 Least Squares Method
The method of least squares is generally stated as,
"The least square estimators are the values that minimize the sum of squared
differences between the observed and expected values of the observations". (Ben-
Akiva and Lerman, 1985)
The coefficients of regression are estimated by the basic objective function F which
is given by Equation (2.1),
The desired coefficients are estimated by taking (k+1) derivatives of equation 2.22
and solving for (k+1) unknowns. This method is usually called the Ordinary Least-
Squares (OLS). Generally, the least-squares estimators are unbiased under general
assumptions. However, it should be noted that the least-square method works
consistently and efficiently for linear models only, and can surmise erroneous
coefficients’ values in case of complex model specifications. Therefore, due to its
higher applications, the maximum likelihood method is generally preferred over the
least square method by the transport statisticians and planners.
2.5 Sampling and Data Collection
2.5.1 Travel Survey Types
The estimation of mode choice models requires collecting of travel and trip related
(including the actual mode choice of the traveler). These data are generally obtained
34
by surveying a sample of travelers from the population of interest. The most common
types of surveying methods are household, workplace and intercept surveys.
Koppelman. and Bhat, 2006).
2.5.1.1 Household Travel Surveys
The household travel surveys involve contacting respondents in their home and
collecting information regarding their household characteristics (e.g., number of
members in household, automobile ownership, etc.), their personal characteristics
(such as income, work status, etc.) and the travel decisions made in the recent past
(e.g., number of trips, mode of travel for each trip, etc.). Historically, most household
traveler surveys were conducted through personal interviews in the respondent’s
home. Currently, most household travel surveys are conducted using telephone or
mail-back surveys, or a combination of both. It is common practice to include travel
diaries as a part of the household travel survey. Travel diaries are a daily log of all
trips (including information about trip origin and destination, start and end time, mode
of travel, purpose at the origin and destination, etc.) made by each household member
during a specified time period. This information is used to develop trip generation,
trip distribution, and mode choice models for various trip purposes. Recently, travel
diaries have been extended to include detailed information about the activities
engaged in at each stop location and at home to provide a better understanding of the
motivation for each trip and to associate trips of different purposes with different
members of the household. Also, in some cases, diaries have been collected
repeatedly from the same ‘panel’ of respondents to understand changes in their
behavior over time.
2.5.1.2 Workplace Surveys
The workplace surveys involve contacting respondents at their workplace. The
information collected is similar to that for household surveys but focuses exclusively
on the traveler working at that location and on his/her work and work-related trips.
Such surveys are of particular interest in understanding work commute patterns of
individuals and in designing alternative commuter services.
2.5.1.3 Destination Survey
Destination Surveys involve contacting respondents at their destinations. Similar
information is collected as for workplace surveys but the objective is to learn more
35
about travel to other types of destinations and possibly to develop transportation
services which better serve such destinations.
2.5.1.4 Intercept Survey
Intercept Surveys “intercept” potential respondents during their travel. The emphasis
of the survey is on collecting information about the specific trip being undertaken by
the traveler. Intercept surveys are commonly used for intercity travel studies due to
the high cost of identifying intercity travelers through home-based or work-based
surveys. In intercept surveys, travelers are intercepted at a roadside rest area for
highway travel and on board carriers (or at carrier terminals) for other modes of
travel. The traveler is usually given a brief survey (paper or interview) for immediate
completion or future response and/or recruited for a future phone survey. A variant of
highway intercept surveys is to record the license plate of vehicles and subsequently
contact the owners of a sample of vehicles to obtain information on the trip that was
observed. Intercept surveys can be used to cover all available modes or they can be
used to enrich a household or workplace survey sample by providing additional
observations for users of infrequently used modes since few such users are likely to be
identified through household or workplace surveys.
2.5.2 Sampling Generation Methods
Sample generation is regarded as a vital step in travel demand modeling since the
modal split models are generally estimated using the data collected by surveying a
sample of respondents from the targeted population. Therefore, it is essential that the
sample generated for the research is representative of the characteristics of the
population of the study area. Inappropriate sample generation can lead to erroneous
modeling results involving biased estimated coefficients and non-representative travel
behavior forecasts.
2.5.2.1 Simple Random Sampling
Simple random sampling is the simplest approach out of all sample generation
techniques and is the basis of all other random sampling methods. In this method, a
totally random sample is chosen from the target population, using a sampling frame
with the units numbered. Since the sampling is totally random, every member of the
target population set has an equal probability of being selected. Therefore, if the set of
36
target population contains N number of members, and the sample is supposed to have
n members, provided that n ε N, the probability to generate the sample in n number of
draws, using simple random sampling, is presented in Equation 2.23 as,
Where,
NPn is the probability to select n number of members from a set of N members,
such that n ε N.
This method is also known as random sampling without replacement .Although this
method is simple; it becomes highly impractical for larger sample sizes. Ampt and
Ortuzar (2004) proved that the method often produces highly variable results from
repeated applications for high sample sizes. Therefore, the method is only applicable
for generating small sample sizes and is limited to simple sampling approaches.
2.5.2.2 Stratified Random Sampling
In stratified random sampling, the targeted population is split into distinct
subpopulations, known as strata. These strata are classified on the basis of various
factors of relevant interest to the survey and obtained by the simple random sampling
within each stratum. For example, for a mode choice survey, the strata can be
categorized on the basis of the users of various travelling modes, i.e. the individuals
using private cars and public transport. Similarly, the classification can also be done
on the basis of various socioeconomic conditions of the households such as structure,
age groups and income-levels. Chang and Wen (1994) explain that if the entire
population contains N units, then stratified random sampling can be done by dividing
it into L number of nonoverlapping strata such that,
Where,
N1,2, … , L are the number of units in each strata L.
37
Whilst stratified sampling is useful, in general, to ensure that the correct proportions
of each stratum are obtained in the sample, it becomes highly significant in
identifying relatively small sub-groups within the population. Therefore, it
enormously increases the precision of the estimates of attributes of the targeted
population of a study area. However, considerable prior information regarding the
attributes of the population should be known before generating the sample.
2.5.2.3 Multi-stage Sampling
Multi-stage sampling is a random sampling technique for study areas with large
populations. It is based on the process of selecting a sample in two or more successive
contingent stages. It proceeds by defining aggregates of the units that are subjects of
the survey, where a list of the aggregates is easily available or can be readily created.
Richardson et al. (1995) explained the process of multi-stage sampling within
Australian context by splitting it into five distinct stages as shown in Figure 2.9
Figure (2.9): Example of Multistage Sampling Process (Richardson et al., 1995)
38
The major disadvantage of multi-stage sampling is its low level of accuracy of the
parameter estimates for a given sample size as compared to that estimated using a
simple random sample for the same study area. However, the reduction in accuracy is
often traded off against the reduction in costs and efficiency in administration of the
sampling process that the multi-stage sampling associate. Hossain et al. (2003) proved
this argument by presenting various population models based on different sampling
techniques, out of which the most efficient method, in terms of application and
economy, was found to be multi-stage sampling.
2.5.2.4 Cluster Sampling
Cluster sampling is a slight variation of multi-stage sampling where the targeted
population is first divided into clusters of sampling units, and then sampled randomly.
The units within the cluster are either selected in total or else sampled at a very high
rate. Detailed literature on the theoretical framework of the method, along with some
useful examples, is presented in Stehman (1997).Similar to multi-stage sampling;
cluster sampling can also be highly economical and administratively efficient as
compared to simple random sampling, especially for study areas with large
populations. Additionally, if the study areas are well-defined, a transport modeler can
easily manage to have a high degree of quality control on the conduct of the
interviews. However, the main disadvantage, like multi-stage sampling, continues to
be the less accuracy in estimating the coefficients for any given sample size as
compared to that estimated using simple random sampling.
2.5.2.5 Systematic Sampling
Systematic sampling is perhaps the most widely known non-random sampling
technique among the transport modelers. The method involves selecting each kth
member of the targeted population. The first member is chosen randomly and then,
after every kth interval, another member is selected to be part of the sample. For
example, if the targeted population contains N members and the desired sample size is
n, then after selecting the first member randomly, the other members are selected
every N/nth interval. However, this constraint does not need to be strictly enforced
and can be modified by the modeler according to the level of model complexity. In
study areas where the size of the targeted population is very large or almost infinite,
Stopher (2000) suggested that every twentieth member of the set should be selected as
39
part of the sample. Although systematic sampling is the easiest and simplest sampling
method known, it possesses various limitations. First, and most importantly, the
sample set generated using systematic sampling generally contains various biases
because the targeted populations sometimes exhibit a periodicity with respect to the
parameter being measured. This causes the resulting sampling set to be significantly
biased towards that certain parameter. The second limitation is the scenario in which
the resulting sample set may not effectively represent the users of a certain travelling
mode. This situation generally occurs in enormously populous study areas where there
is assorted practice of travelling modes and the transport modelers unconsciously
ignore these users, causing bias in the sample set.
2.5.3 Revealed and stated preference survey
Many researchers have attempted to model travel demand and travelers’ behavior
using revealed preference and/or stated preference survey data. These two techniques
are used a complementary tools to elicit the preferences of the decision. The revealed
preference and stated preference techniques conveniently provide data for the
development of disaggregate travel forecasting models..
Revealed Choice data describes current observed travel patterns and costs and hence
is a very accurate picture of current modal choice.
The stated preference techniques allow the deficiencies of revealed preference data to
be overcome by testing hypothetical transport alternatives in interviews. This
technique becomes an attractive option in transportation modeling since it presents the
decision-makers choice and behavioral pattern under different hypothetical scenarios.
Many researchers use this technique to understand the unpredictable behavior of
decision-makers under conditions that are new or hypothetical.
Hensher (1994) states that there are three types of questionnaire that can be used
instated preference studies. These are: ranking, choice or rating .In a choice
questionnaire, the task is simpler for the respondent. The respondent simply chooses
the hypothetical combination of attributes that is most favor able to him or her and the
researcher has an actual prediction of the respondent’s choice in a hypothetical
situation. In a ranking questionnaire, respondents must order the hypothetical
situations in order of preference. In a rating questionnaire, the task becomes more
complicated as respondents must be able to order their responses in order of
41
preference but they must also be able to indicate how much they prefer one alternative
over others.
Revealed preference and stated preference studies each have their advantages and
disadvantages .According to Swait et al.(1994) the main advantage of using a revealed
preference study is that it can represent current market situations better than stated
preference studies. In revealed preference studies, the choices that are made by
respondents are known outcomes, although they are dependent on the respondent’s
perceptions of Attribute levels, which may or may not be accurate (Hensher,1994).
Stated preference studies are less constrained than revealed preference studies and
allow us to look at potential changes(Swait e tal.,1994).Stated preference studies
allow us to examine how decision-making varies as different types of attribute
profiles and levels are considered(Hensher,1994).Stated preference techniques were
originally popularized by the work of Louviere and Hensher (1983) and Davidson
(1973) in the1970s and 1980s who demonstrated how researchers could examine trip
makers answers to hypothetical combinations of attribute levels for travel modes. In
stated preference studies, outcomes are potential outcomes (Hensher, 1994).
According to Wang et al.(2000),stated choice and stated preference methods have
limits, however .They are limited by a respondent’s ability to understand the
hypothetical situations with which they are presented and to provide reliable answers.
Wang et al. (2000) argue that if hypothetical situations are far removed from the
respondent’s daily experience, the stated preference study will result in poor models
and inaccurate results. Therefore, stated preference studies should have some relation
to the real world. Wangetal (2000) also make recommendations regarding
strengthening stated preference models by some type of fusion with revealed
preference models.
The stated preference technique has become a convenient tool for researchers to
analyze data especially for the non-existing scenarios. Ortuzar and Willumsen (2002)
have summarized the main features of stated preference technique as follows:
i. It is based on the elicitation of respondents’ statements of how they would
respond to different hypothetical (travel) alternatives;
ii. Each option is represented as a ‘package’ of different attributes like travel
time, price, headways, reliability and so on;
41
iii. The researcher constructs these hypothetical alternatives so that individual
effect of each attribute can be estimated; this is achieved using experimental
design techniques that ensure the variations in the attributes in each package
are statistically independent from one another;
iv. The researcher has to make sure that interviewees are given hypothetical
alternatives they can understand, appear plausible and realistic, and relate to
their current level of experience;
v. The respondents state their preferences towards each option by either ranking
them in order of attractiveness, rating them on a scale indicating strength of
preference or simply choosing the most preferred option from a pair or group
of them;
vi. The responses given by individuals are analyzed to provide quantitative
measures of the relative importance of each attribute.
2.6 Previous case studies of mode choice modeling
Many discrete case studies have been applied around the world in the area of mode
choice modeling. The studies aimed at developing a discrete choice model that fits
with the study area. In this research we present some studies in cities of developed
and developing countries in order to benefit in developing a suitable mode choice for
our country.
Al Ahmadi (2006) developed intercity mode choice models for Saudi Arabia. These
models indicated that in-vehicle travel time, out of pocket cost, number of family
members travelling together, monthly income, travel distance, nationality of traveler,
and number of cars owned by family played the major role in decision related to
intercity mode choice.
Khan (2007) estimated various nested logit models for different trip length and trip
purpose using data from stated preference (SP) survey. A unique computer assisted
personal interviewing (CAPI) instruments was designed using motorized and non-
motorized travelling modes in the SP choice set. Additionally a unique set of access
modes for bus on bus way was also generating containing hypothetical modes such as
secure park and ride facilities and kiss and ride drop-off zones. He found from the
final model estimation that the travel behavior forecasted for regional trip makers is
considering different from that for local trip makers. The regional travelers for work
42
were found not to perceive the non- motorized modes as valid alternatives to car,
possibly due to longer trip length. The value of time (VoT) determined for local work
trip makers was 16.5 A$/hr was also found to be higher than that of regional work trip
makers (11.7 A$/hr).
Siddiqui (1999) modeled the non-motorized modes (walk and bike) along with the
traditional motorized (auto and transit). A nested multinomial modal split models
were developed for simulating the p.m. peak period home based work, home based
school, and other trips purposes for the national capital region. Model and individual
characteristics considered important in mode choice were identified and testing using
sensitivity analysis. The results indicated that travel cost, travel time, and travel
vehicle ownership are important factors in motorized modes where as travel time,
gender, member of bikes, and population density are significant variables for non-
motorized modes.
Adjaka (2009) analyzed factors that influence parents’ decision in choosing
transportation modes for schoolchildren in district of Colombia. Multinomial logistic
regression was used to predict the share between choice modes. The factors that found
to be most significant in predicting mode choice included distance home to school,
student's grade, school's encouragement of walking and biking and walking fun for
schoolchildren.
Abdelwahab and Abdel-Aty (2001) developed mode choice models for Florida,
USA. The mode choice model was estimated as three level nested Logit structure. The
overall model utilized full information maximum likelihood estimation. Among the
significant variables that entered into model are: transit access time, transit waiting
time, number of transfer, in-vehicle travel time, fare and household car ownership.
Ewing, Schroeer and Greene (2004) analyzed the relationship between school
modes and factors that influence the choice of a given mode. The data used for their
study was from travel diary surveys of students in grades K-12 from Gainesville,
Florida. The mode choice was developed using multinomial Logit model. The
following factors have significant impacts on school mode choice: distance home-
school, built-environment between home and school and household incomes. Students
who lived closer to their schools were more likely to walk and bike to and from
school. The presence of sidewalks and crosswalks encouraged walking, but did not
43
affect the biking decision. The authors also found that vehicle availability, which is
related to income, made walking and biking less attractive compared to the other
modes. Other factors such as school size, school enrolment, and land use density were
not significant in predicting the modes.
Mc Donald (2008) studied the impact of distance in school mode choice using
elementary and middle school students’ data from the US Department of
Transportation 2001 National Household Survey. The findings suggested that among
factors that influence mode choice, distance between 14 school and home had the
strongest impact. Children were likely to walk when the distance is less than one mile.
Gender and ethnicity had minor influence on the mode choice. The results also
indicated that density around school zone and neighborhood increases walking and
biking to and from school. The study concluded that a better integration of land use,
transportation, and school planning can encourage walking and biking to and from
school.
Yarlagadda and Srinivasan (2008) used the San Francisco Bay area Travel Survey
data of 2000 to model travel behavior of children, and the interdependency between
parents and children in mode choice. The targeted populations of students were under
eighteen year olds. The authors modeled school mode choice using multinomial logit
model. The results showed that characteristics of students such as age, gender, and
ethnicity and characteristics of their parents such as employment and flexibility in
working hours have strong impacts on the mode choice. The research found distance
between home and school as the primary barrier to the choice of walking to or from
school. The authors also found that the significance of the explanatory variables were
not the same for trip to school and trip from school.
Kweon, Shin, Folzenlogen and Kim (2006) investigated environmental factors that
encourage walking and biking to school. The researcher used household surveys from
College Station, Texas. The authors limited their research to 2-mile radius around
each school, areas defined as walk zones. The results indicated that students walk
more in neighborhoods where there are mature trees and bike more in neighborhoods
with sidewalks. The average walking distance was 0.71 miles. The average biking
distance was 0.93 miles. The majority of children living beyond one mile from their
schools used motorized modes.
44
Rhoulac (2005) analyzed factors that affect K-8 children school mode in Wake
County, North Carolina. The data was from household surveys, and the alternative
modes considered in the study were school bus vs. private automobile. The results
suggested that mode choice for children’s school trips were influenced mainly by
factors such as number of students in a household, student’s grade, parents’ perceived
safe mode, and the convenience of driving
Almasri (2011) investigated the factors that affect travel choice of shared taxi versus
bus for Palestinian university student trips. The results of this study indicated that
factors that are significantly affecting the mode choice of students are: family income
divided by family size, weighted travel time, out of vehicle travel time divided by
distance and cost divided by natural logarithm of income. The results also show that
the age and gender variables are statistically insignificant and it could be dropped
from the model.
2.7 Summary
The mode choice model is the third step of the classical four-step model and it plays
an important role in travel demand forecasting. There are different factors that affect
the choice of transportation modes which can be categorized into three groups
namely; factors related to the characteristics of trip maker such as car availability and
possessing of driving license, factors related to the characteristics of journey such as,
time of day and types of trips, and factors related to characteristics of transport
facilities such as cost, travel time and waiting time. This chapter illustrated that there
are two approaches for modeling the choice of transportation modes which are
aggregate and disaggregate approaches. Disaggregate (discrete) approach are widely
used because of its advantages which overcome the problems facing the aggregate
one. There are three types of disaggregate mode choice models namely; logit model.
Probit model and general extreme value model. Among these types the logit model is
the most widely used for calibration the mode choice because it is simple in terms of
formulation and estimation of the model in addition to its accuracy compared with the
other types. The estimation techniques that are used for estimation the mode choice
model are maximum likelihood and least squares methods. The maximum likelihood
is the most common technique used in determining the estimators for simple and
nested logit model so this technique was adopted in this research for estimating the
mode choice model. Among the four methods of travel survey which are household
45
survey, workplace survey, destination survey, and intercept survey; the workplace
survey method was used for collecting the data in this study. The reviewing of
literature show there are five methods of sample generation. These methods are:
simple random method stratified random method, cluster sampling method, multi
stage sample method and systematic sample method. Among these methods the
simple random method was used for sample generation in this research. The
experiences of some countries in modeling the mode choice for different types of trips
which were reviewed in this chapter illustrated that the factors related to transport
policies and the factors related to the characteristics of trip maker are the most
important factors in guiding the choice behavior of travelers.
46
Chapter 3: Research Methodology
3.1 Stages of the Study
In order to achieve the main aim and objectives of this research the work is divided
into six main phases as can be seen in Figure 3.1
First phase:
The first phase is the literature review on mode choice modeling. The concentration
will be on discrete mode choice models as they are more efficient than conventional
models. The literature review should seek for case studies applied in cities of
developing countries especially in the cities that have similar conditions. Based on the
literature review, the transportation planning process that is appropriate to Gaza City
must be decided.
Second phase:
This phase relates to the process of selection of the travel attributes. It involves,
designing of initial (pilot) survey form and analysis of the survey data. This process is
important to determine the attributes which are most relevant to the travelers in the
study area. The resulting attributes will be included in the main survey.
Third phase:
This phase involves designing the final survey form and conducting the survey from
start to finish including selection of level of attributes, determine the sample size and
sample space, implementation of the survey, the collection and analysis of data.
Fourth phase:
This phase includes calibrating and estimating of the utility functions for the Model
and chooses the best model by comparing the models with regard to the coefficient
estimates of the variables and their overall fit.
Fifth phase:
This phase is preliminary concerned with the model validation.
Sixth phase:
This phase summarizes the main findings and conclusions from the study.
47
FIRST STAGE:
REVIEW THE LITERATURE
SECOND STAGE:
DESIGN OF INITIAL SURVEY QUESTIONNAIRE
Transportation
Planning Process
(Four Step Model)
Types of Mode
Choice Models
Model Estimation
Techniques
Sampling and
Data Collection
Design of initial
survey questionnaire
Pilot Study
Analysis of pilot
study
Contd..
.
48
THIRD STAGE:
DESIGN OF FINAL SURVEY QUESTIONNAIRE AND DATA COLLECTION
FOURTH STAGE:
CALIBRATION OF MODEL
Design of Final
survey questionnaire
Determination of
Sample Size
Distributing and
collecting of
questionnaire
Analysis of Data
Calibration
Model 1
Calibration
Model 2
Calibration
Model N
Comparison of
Models in terms of
a. Coeff- Estimators
b. t – Statistics
c. Stnd error
d. Overall fit
Contd..
.
49
FIFTH STAGE:
VALIDATION OF MODEL
SIXTH STAGE:
CONCLUSION AND RECOMMENDATION
Figure (3.1): Flow chart for research methodology
Validation of the
chosen Model
a. Likelihood Ratio test LRTS Calculation
b. Estimation of Prediction Ratio
Comparison of
LRTS with
Critical Chi-
Square Value @
95% confidence
level
Conclude the main Findings
Recommendations
51
3.2 Study area
The study are in this research is Gaza city which represents the largest part of Gaza
governorate. Gaza is the largest governorate after Khanyounis with area of 72593
dounms. The number of population in Gaza at mid 2011 is about 552,000 inhabitants.
This put Gaza as the most densely governorate with a density of 7.5 cap/dun (PCBS,
2007). Gaza city is composed of eleven districts as shown in Figure 3.2. These
districts are old city, Rimal, Zeiton, Shujaiyya, Alsabra, Aldarag, Alnasser, Tuffah,
Sheikh Radwan, Sheikh Ajlin, Tel Alhawa, Alshatia camp, and ALawda.
Figure (3.2): Study Area (Gaza city) (MoG, Planning department, 2005)
Gaza city is suffering from congestion problem resulting from urbanization and the
rapid increase of population where the population is growing about 4% a year. This
put the transportation planners toward a big challenge to adopt efficient transport
policies contribute in solving this problem.
3.3 Target group
This research targeted the employed people they live and work in Gaza city.
According to the statistics of PCBS the number of population within the work age
(>15 year) in Gaza strip at mid 2008 is 744,000 inhabitants represents about 51.7% of
the total number of population in Gaza strip. The labor force participation in Gaza
strip is about 38.1% of the population within the work age while the rest percent
which represents about 61.9% is outside the labor force. The outside labor force
51
population includes housewives (46.7%), students and trainee (37.9%), old/ill
(10.5%), and neither working or looking for work (4.9%). The percent of employed
people from the population within the labor force in Gaza strip is about 59.4%, of
which about 53% is full employed and 6.4% is underemployed. The percent of
unemployment in Gaza strip represents about 40.6% of the population within the
labor force
The census of PCBS referred that the participation of labor force in Gaza represents
about 36.4 % from the population within the work age; 63.4% for males and the rest
for females. The percent of employed people represents about 61.7% of the
population within the labor force, of which, about 57% is full employed and 4.7% is
underemployed. The percent of unemployment represents about 38.3% of the
population within the labor force. The labor forces in Gaza are divided into
governmental employee, UN employee, private sector employee, wage workers, self
works and business men. .
3.4 Design of Questionnaire
Questionnaire was developed in order to collect the data required for calibrating the
mode model for work trips in Gaza city. The questionnaire was divided into four parts
as can be seen below:
1. Part one: which includes the social and economical information about the
respondents such as (gender, age, job, income, family size, ownership of private
car, ownership of motorcycle, ownership of bicycle……etc).
2. Part two: This focuses on the factors that affect the mode choice. The respondents
were asked to indicate their perception on the importance of twelve well organized
factors that affect the choice of transportation mode for work trips in Gaza city.
These variables are: age, gender, average monthly income, travel cost, travel time,
waiting time, weather conditions, privacy, comfort, health status, and trip length.
A five point Likert scale ranging from (1: very low important to 5: very high
important) was adopted to analyze the importance of factors that affect the choice
of transportation mode.
3. Part three: This focuses on the trip characteristics. For the purpose of this study
the daily trips which constitute home-work trips have been included. In this part
the information covered is relating to travel behavior of individual for his/her daily
52
trips such as (the mode usually used, travel time, travel cost, monthly fuel
consumption, license and maintenance cost ….. etc.).
4. Part Four: this concern on the hypothetical choice of the respondents. The
respondents were asked to rate his/her preferred modes among three different
modes (shared taxi, mini bus, and bus) according to three different levels of
attributes as shown in the Table 3.1
Table (3.1): Different levels of the hypothetical questions
Level Attribute/mode Bus Minibus Shared taxi
Level 1
Travel time 40% more than
shared taxi
30% more than
shred taxi -
Travel cost 50% less than
shared taxi
25% less than
shared taxi --
Frequency Every 40
minutes
Every 20
minutes
Every 5
minutes
Level 2
Travel time 30% more than
shared taxi
15% more than
shred taxi -
Travel cost 40% less than
shared taxi
15% less than
shared taxi --
Frequency Every 30
minutes
Every 15
minutes
Every 5
minutes
Level 3
Travel time 20% more than
shared taxi
10% more than
shred taxi -
Travel cost 30% less than
shared taxi
10% less than
shared taxi --
Frequency Every 15
minutes
Every 8
minutes
Every 5
minutes
53
3.5 Sample Size Determination
Survey is intended to estimate the true value of one or more population
characteristics. In order to draw inference from a sample that will accurately reflect
the population careful attention must be given to determining the needed sample size.
Many efforts were done to determine the minimum sample size that accurately
reflects the population characteristics. The central limit theorem is on the heart of
these efforts. Kish (1995) showed that the minimum sample size can be calculated
using the following equation
Where,
is a total number of population
is a sample size from finite population
is a sample size from infinite population
The Sample size from infinite population can be calculated using the following
equation
Where,
is the variance of population elements
is the standard error of sampling population
For 95% confidence level and 10% error the sample size can be calculated as a
function of coefficient of variation cv
As the value of is very small comparing with N thus the ratio of
is very small
thus the sample size of finite population n can be taken as the same value of . For
cv=1 the minimum sample size is 384.
54
Green (1991) recommended minimum sample size N for Multinomial logistic
regression with N>50+8m where m is the number of predictors (factors). For twelve
factors that affect the mode choice model in Gaza city, the minimum sample size for
this study is 146.
For the purpose of this study 700 questionnaires were distributed for work trips. The
random sample method was adopted in this study. Of the 700 questionnaires were
distributed 552 questionnaires are valid.2/3rd
of theses questionnaires were used for
calibration of model and the rest were used for the validation process.
3.6 Pilot Study
A pilot survey is a complete run through of the actual survey done over a small set of
population in order to the level of credibility of instrument, data coding, and data
recording. In this study 20 questionnaires were distributed to experts in transportation
field and for a chosen sample of population. The objective of pilot study was to verify
the completeness of questionnaire. The following items are a summary of major
observations based on pilot study:
1. Some questions were added to the different parts of questionnaire such as:
The fuel consumption per month for private car (Part III)
The number of kilometers that the private car cut per 1 liter of fuel (Part
III).
The distance factor was added to the factors list (part II).
The engine type of private car (part I).
The ownership of motorcycle and bicycle (part I).
2. One question was omitted from questionnaire as suggested by the respondents.
These questions were considered impractical or unrealistic which is:
The distance between home and work (part III).
3. Some questions were rearranged in order to give more suitable and considered
meaning such as questions 1,2,3 in part III and the question in part IV
3.7 Preliminary analysis of questionnaire
This is the fourth phase in the present study. This focuses on the determination of
choice and captive riders for various travel and socioeconomic characteristics. The
statistical analysis software (SPSS) was used to perform the analysis. The important
55
travel and socioeconomic characteristics of the captive travelers affecting mode
choice were also determined with the help of chi-square and Cramer’s test.
The relative important index (RII) was used to determine the relative importance of
various factors that affect the mode choice. The (RII) can be calculated using the
following equation
Where,
W is the weight given to each factor by the respondents and range from 1-5
A is the highest weight =5
N is the total number of respondents
The RII was used to rank the different factors that affect the mode choice in order to
cross-compare the relative importance of the factors as perceived by the respondents.
3.8 Model Calibration and Comparison
This is the fifth phase in the present study. The concept of utility theory is used for
calibration process. The basic approach in this theory is that the individual select an
alternative that maximizing his/her utility. The mathematical details of this theory as
well as the available procedures for model calibration were explained in chapter two.
A multinomial logit model which relates the utility of the alternatives to the
probability of choice is used to calibrate the model. The EASY LOGIT software
package was used in this study to calibrate the desired models. The package used the
maximum likelihood technique to calibrate logit model. The outputs of this package
include various statistical performance indicators which are:
1. t-test and associated significance of parameters estimators.
2. Log of likelihood function value LL (β) at its maximum.
3. Log of likelihood function value LL (0) when all parameters are zero. In other
words all alternatives have equal probability of being chosen.
4. Goodness of fit index roh-square (ρ2) that measures the fraction of an initial
likelihood value explained by the model, which can be calculated as
56
5. Corrected goodness of fit index roh-bar square similar to (ρ2) but corrected
for the number of parameters estimated which is calculated as follows
Where, K is the number of parameters estimated in the model.
The signs of parameters are also checked. The signs of parameters have to be logical
for instance if travel time and travel cost coefficients have positive signs then a
decrease in these variables will decrease the demand and vise versa which is illogical.
Mode choice models were calibrated on 2/3rd
of the data and the remaining 1/3rd
is
reserved for validation. The calibrated models were compared with respect to
statistical performance indicators. Among the most important one that are studied are
the signs of coefficient, the goodness of fit, the adjusted goodness of fit, and t- test.
3.9 Model Validation
After the calibration process is completed and the models have been compared,
validation of mode choice model is checked. Approximately 1/3rd
of the reserved data
sets were used for this purpose. The validity of the model was tested by LRTS
(likelihood ratio test) and estimation of prediction ratio.
The null hypothesis formulated for the purpose is as follows:
H0: there is no difference between the observed and predicted behavior i.e. there is
no difference between the parameter vectors obtained from calibration data
and the validation data
H0: βi = βj , where
βi , βj are the estimated parameter vectors of the model obtained from calibration and
validation data ( same specification is needed for this test)
To obtain the LRTS value the coefficient of variables of particular model will be
restricted and the ELM program is executed with validation data. The program
outputs two log-likelihood values. The first value is the one computed by restricting
the coefficient of the calibrated model while the second is the one when the
parameters are unrestricted for validation data. The LRTS value can be obtained by
the following
57
Where,
represents the likelihood ratio test statistics which restricts the
parameters estimated from data j to be used to predict mode share in
data i for same specifications
is log likelihood ratio value when the parameters are restricting in data
j
is log likelihood ratio value when the parameters are unrestricted in
data j
The LRTS tests discussed is distributed as chi-square with k degrees of freedom
where k is the number of model parameters. If LRTS value is less than critical chi-
square value @ 95% confidence level and degree of freedom equal to k then for that
particular case the null hypothesis can’t be rejected otherwise it is rejected.
58
Chapter 4: Results &Analysis
4.1 Introduction
This chapter describes the results of the descriptive analysis of the survey as well as
the calibration and validation for revealed and stated preference models. Section 4.1
presents the results of general analysis of data. Section 4.2 discusses the relation
between the mode choice and socioeconomic variables. Section 4.3 discusses the
relation between the captive ridership and socioeconomic variables. Section 4.4
presents the results of the travelers’ choice for the hypothetical questions. Section 4.5
involves the calculation of the relative importance index and ranking the factors that
affect the mode choice. Sections 4.6 and 4.7 discuss the calibration and validation for
revealed mode choice model. Finally sections 4.8 and 4.9 describe the calibration and
validation for stated preference model.
4.2 General Analysis of Data
Initially frequency tables were obtained on a whole datasets to determine the
distribution of travelers for various travel and socioeconomic characteristics. The
results of this analysis are summarized in the form of frequency tables and pie charts.
4.2.1 Gender of respondents
The distribution of travelers for their gender can be seen in Table and Figure (4.1). As
can be seen in the table about 68.5% of the respondents are male and 31.5% are
female.
Table (4.1): Frequency table for respondent’s gender
Frequency Percent Valid
Percent
Cumulative
Percent
Valid Male 378 68.5 68.5 68.5
Female 174 31.5 31.5 100.0
Total 552 100.0 100.0
59
Figure (4.1): Respondents’ gender
4.2.2 Marital status of respondents
Table and Figure (4.2) presented the distribution of travelers for their status .As can
be seen in the table about 20.8% of the respondents is single and 79.2% are married.
Table (4.2): Frequency table for respondent’s status
Frequency Percent Valid
Percent
Cumulative
Percent
Valid Single 115 20.8 20.8 20.8
Married 437 79.2 79.2 100.0
Total 552 100.0 100.0
Figure (4.2): Respondents’ marital status
61
4.2.3 Jobs of respondents
The distribution of respondents’ job can be seen in Table and Figure (4.3). the results
reported in the table show that about 40.2% of the respondents are governmental
employee, 25% are private sector employee,13.6% are UN employee , 3.8% are
business man, 15.8 % are waged workers and 1.6 works on others job.
Table (4.3): Frequency table for respondent’s job
Frequency Percent Valid
Percent
Cumulative
Percent
Valid Governmental employee 222 40.2 40.2 40.2
Private sector employee 138 25.0 25.0 65.2
UN employee 75 13.6 13.6 78.8
Business man or special
works 21 3.8 3.8 82.6
waged worker 87 15.8 15.8 98.4
Others 9 1.6 1.6 100.0
Total 552 100.0 100.0
Figure (4.3): Respondents’ job
61
4.2.4 Age of respondents
The distribution of respondents’ age was presented in Table and Figure (4.4). As can
be seen in the table the large percent of workers lies in the age category from 25-45
years which represents about 60% of the whole sample.
Table (4.4): Frequency table for respondent’s age
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid 18-25 38 6.9 6.9 6.9
26-30 107 19.4 19.4 26.3
31-35 114 20.7 20.7 47
36-40 130 23.6 23.6 70.6
41-45 85 15.3 15.3 86
46-50 55 9.9 9.9 95.9
51-55 17 3.0 3.0 98.9
>55 6 1.1 1.1 100.0
Total 552 100.0 100.0
Figure (4.4): Respondents’ age
62
4.2.5 Average family monthly income of respondents
The distribution of respondents’ average family monthly income can be seen in Table
and Figure (4.5). The results reported in the table show the majority of workers have a
monthly income between 1500-4000 NIS which represents about 73% of the whole
sample.
Table (4.5): Frequency table for respondent’s monthly income
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid less than 1000 NIS 16 2.9 2.9 2.9
1001-1500 NIS 56 10.1 10.1 13.0
1501-2000 NIS 111 20.1 20.1 33.2
2001-3000 NIS 181 32.8 32.8 65.9
3001- 4000 NIS 115 20.8 20.8 86.8
4001-5000 NIS 46 8.3 8.3 95.1
more than 5000 27 4.9 4.9 100.0
Total 552 100.0 100.0
Figure (4.5): Respondents’ monthly income
63
4.2.6 Family size of respondents
The distribution of respondents’ family size was presented in Table and Figure
(4.6).the collected data was categorized into three categories. The results show that
the majority of employed people have a family size between 1-6 persons which
represents about 69.6% of the whole sample. While the employed people with a
family size bigger than 10 persons represents about 1.5 % of the sample.
Table (4.6): Frequency table for respondent’s family size
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid 1-6 386 69.9 69.9 69.6
7-10 158 28.6 28.6 98.5
>10 8 1.5 1.5 100.0
Total 552 100.0 100.0
Figure (4.6): Respondents’ family size
64
4.2.7 Ownership of transport modes
Table and Figure (4.7) presented the distribution of transport means owned by the
respondents. As can be seen in the table about 16.5% of the respondents have a
private car, 20.3% of the respondents have a motorcycle, 4% have a bicycle and the
rest have no means of transport.
Table (4.7): Frequency table for respondent’s ownership of means of transport
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid Private car 91 16.5 16.5 16.5
motorcycle 112 20.3 20.3 36.8
Bicycle 22 4.0 4.0 40.8
No means 327 59.2 59.2 100.0
Total 552 100.0 100.0
Figure (4.7): Respondents’ ownership of transport means
65
4.2.8 Trip length
The distribution of trip length can be seen in Table and Figure (4.8). The result
shown in the table below illustrate that the majority of the sample has a trip length lies
between 0.3-4.0 km which represents about 75.2% of the whole sample. While the
trips which have a length more than 6.0 km represents about 2% of the sample.
Table (4.8): Frequency table for trip length
Frequency Percent Valid
Percent
Cumulative
Percent
Valid 0.30 -1.0KM 37 6.7 6.7 6.7
1.10 - 2.0 KM 121 21.9 21.9 28.65
2.10 -3.0 KM 164 29.7 29.7 58.3
3.10 – 4.0 KM 93 16.9 16.9 75.2
4.10 – 5.0 KM 77 13.9 13.9 89.1
5.10 – 6.0 KM 49 8.9 8.9 98.0
> 6.0 Km 11 2.0 2.0 100.0
Total 552 100.0 100.0
Figure (4.8): Trip length
66
4.2.9 The means of transport usually used by the respondents
The distribution of the various modes that are usually used by the respondents was
reported in Table and Figure (4.9).
Table (4.9): Frequency table for the modes of transport that
Usually used by the respondents
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid private car 76 13.8 13.8 13.8
shared taxi 245 44.4 44.4 58.2
Taxi 39 7.1 7.1 65.2
motorcycle 90 16.3 16.3 81.5
Bicycle 13 2.4 2.4 83.9
Walking 89 16.1 16.1 100.0
Total 552 100.0 100.0
Figure (4.9): The percent of different modes usually used by the respondents
As can be seen from the table and chart above the large percent of the respondents
usually use the shared taxi to go their work by 44.4% of the sample while the using of
67
bicycle comes on the last order by 2.4%. The percents for different modes shown in
the table and chart above include the people of one choice “captive”. In order to study
the effect of these people on the model, the questionnaire form includes a question
about the captive riders as follows: “ Would you considered using modes of
transport other than the one you usually used to go to your work?” if the answer
with no , then the respondent considered as “ captive rider”. The Table and Chart 4.10
show the frequency and percent of captive and choice riders.
Table (4.10): Frequency table for the choice and captive riders
CAPTIVE
/CHOICE Total
Choice Captive
MODES Private Car 63 13 76
Shared Taxi 141 104 245
Taxi 28 11 39
Motorcycle 81 9 90
Bicycle 13 0 13
Walking 70 19 89
Total 396 156 552
Figure (4.10): The number of captive and choice riders for different modes
68
The results reported in the table and chart above show that the percent of the captive
riders is about 28.2% of the respondents while the rest 71.8% considered as a choice
riders. The above table and chart also illustrate that the percent of captive riders
increases in the Shared Taxi users in comparing with the other modes of transport.
The percent of captive riders in the users of this mode represents about 42.4% . This
result seems to be realistic because there is a high percent of shared taxi users don’t
have their own means of transport ( private car, motorcycle, bicycle) and in certain
circumstances the other modes of transport that can be available for them such as
(taxi, and walking) are not be suitable for them to use. As example if the traveler does
not have his own mode of transport and the distance between the home and work is
long and the traveler’s income is low so he has no choice other than Shared Taxi
because the fare of taxi is very high comparing with his income and the distance is
very long to walk.
4.3 Relation between the mode of transport and socioeconomic characteristics
In order to better understand of the relation between the choice of the mode of
transport and the socioeconomic characteristics of the respondent, a cross tabulation
were performed between the modes of transport that are usually used by the
respondents and the different social and economical characteristics of the respondents
and the results were analyzed. A chi-square test was applied to cross tabulation
between the mode variable and socioeconomic variables. The null hypothesis stated
that there is no relationship between the mode of transport and the socioeconomic
variables. The null hypothesis will be rejected if the significance level is less than 5%
for 95% confidence interval. Cramer’s V statistics was used to give indication of the
strength of the relationship which range in value between 0-1 i.e. the higher is better.
4.3.1 Relation between the mode of transport and gender
The distribution of the transport modes that are usually used by the respondents
versus the gender of the respondents can be seen in Table and Figure 4.11.The table
and chart below illustrate that the using of motorcycle and bicycle modes are limited
to males because of the social habits of the Palestinian society. The table also shows
that the females are likely to use the taxi and shared taxi modes of transport more than
the males do as the percent of the taxi and shared taxi users from female is about
69
13.8% and 59.8 % respectively while this percent is about 4% and 37.3% of the male
users.
Table (4.11): Cross tabulation between the mode of transport and gender
Count Gender Total
Male Female
MODES private car 51 25 76
shared taxi 141 104 245
Taxi 15 24 39
motorcycle 90 0 90
Bicycle 13 0 13
walking 68 21 89
Total 378 174 552
Figure (4.11): The percent of male and female riders for different modes
The results of chi-square test and Cramer’s V statistics which show the relation
between the gender and the mode of transport that is usually used by the respondent
can be seen in Tables 4.12 and 4.13 respectively.
71
Table (4.12): Chi-square test for mode-gender relationship
Value df Asymp. Sig.
(2-sided)
Pearson Chi-Square 79.901a 5 .000
Likelihood Ratio 108.487 5 .000
Linear-by-Linear
Association 23.243 1 .000
N of Valid Cases 552
Table (4.13): Cramer’s V statistics for mode-gender relationship
Value Approx.
Sig.
Nominal by
Nominal Phi .380 .000
Cramer's V .380 .000
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
The results presented in the tables above show that the null hypothesis which stated
that there is no relationship between the gender and the mode of transport that is
usually used by the respondent can be rejected for 95% confidence level because the
significance level is less than 0.05. The Cramer’s V statistics is 0.38 as can be seen in
Table 4.13 which means that there is a weak relationship between the gender and the
mode of transport that are usually used by the respondent.
4.3.2 Relation between the mode of transport and marital status
The distribution of the transport modes that are usually used by the respondents
versus the marital status of the respondents was presented in Table 4.14 and Figure
4.12. As can be seen from the table and chart the percent of single users of taxi,
motorcycle, and walking modes is bigger than the percent of married users with a
percents of 10.43%, 22.61%, and 20.8% respectively. The table also illustrates that
the percent of married users of private car mode is bigger than the percent of single
users with a percent of 16.25 %. The result that can be concluded from the table and
71
chart shown below is that the effect of marital status will be significant on private car
and motorcycle modes more than the other modes.
Table (4.14): Cross tabulation between the mode of transport and marital status
Count Status Total
Single Married
MODES private car 5 71 76
shared taxi 48 197 245
taxi 12 27 39
motorcycle 26 64 90
bicycle 0 13 13
walking 24 65 89
Total 115 437 552
Figure (4.12): Distribution of transport modes for marital status
For testing the relationship between the marital status and the choice of transport
mode, a chi-square test was applies to cross tabulation between the transport mode
variable and marital status variable. The results of chi-square test and Cramer’s V
72
statistics which show the relation between the marital status and the mode of transport
that is usually used by the respondent were presented in Tables 4.15 and 4.16
respectively.
Table (4.15): Chi-square test for mode-marital status relationship
Value Df Asymp. Sig.
(2-sided)
Pearson Chi-Square 20.918a 5 .001
Likelihood Ratio 25.573 5 .000
Linear-by-Linear
Association 7.473 1 .006
N of Valid Cases 552
Table (4.16): Cramer’s V statistics for mode-marital status relationship
Value Approx.
Sig.
Nominal by
Nominal Phi .195 .001
Cramer's V .195 .001
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
As can be seen from the tables above the null hypothesis which says that there is no
relationship between the marital status and the choice of transport mode can be
rejected for 95% confidence level because the significance level is less than 0.05. The
Cramer’s V statistics is 0.195 as can be seen in Table 4.16 which means that there is a
very weak relationship between the marital status and the choice of transport mode.
4.3.3 Relation between the mode of transport and age
The distribution of the transport modes that are usually used by the respondents over
the age of the respondents can be seen in Table 4.17 and Figure 4.13. As can be seen
from the figure, the percent of private care users increases as the age increases
because the old people prefer to have privacy and comfort in transport modes which
are available in private car. The relationship between the age and the choice of
73
transport mode can be tested by applying chi-square test to cross tabulation between
the transport mode variable and age variable. The results of chi-square test and
Cramer’s V statistics which show the relation between the age and the mode of
transport that is usually used by the respondent were presented in Tables 4.18 and
4.19 respectively.
Table (4.17): Cross tabulation between the mode of transport and age
Count Age Total
18-30 31-40 41-50 >50
MODES private car 3 29 33 11 76
shared taxi 70 108 60 7 245
Taxi 11 24 4 0 39
motorcycle 33 37 19 1 90
Bicycle 0 5 10 0 13
Walking 28 30 14 4 89
Total 145 233 140 23 552
Figure (4.13): Distribution of transport modes for age
74
Table (4.18): Chi-square test for mode-age relationship
Value Df Asymp. Sig.
(2-sided)
Pearson Chi-Square 331.141a 205 .000
Likelihood Ratio 294.428 205 .000
Linear-by-Linear
Association 17.025 1 .000
N of Valid Cases 552
Table (4.19): Cramer’s V test for mode-age relationship
Value
Approx.
Sig.
Nominal by
Nominal
Phi .775 .000
Cramer's V .346 .000
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
The results of statistical tests show that the null hypothesis which stated that there is
no relationship between the age and the choice of transport mode can be rejected for
95% confidence level because the significance level is less than 0.05. The Cramer’s V
statistics is 0.346 as can be seen in Table 4.19 which means that there is weak
relationship between the age and the choice of transport mode.
4.3.4 Relation between the mode of transport and family size
The distribution of transport modes that are usually be used by the respondents over
the family size of the respondents can be seen in table 4.20 and chart 4.14. In order to
study the relationship between the family size and the choice of transport mode, a chi-
square test was applied to cross tabulation between the transport mode variable and
age variable. Tables 4.21 and 4.22 show the results of chi-square test and Cramer’s V
statistics.
75
Table (4.20): Cross tabulation between the mode of transport and family size
Count Family size Total
1-6 7-10 >10
MODES private car 37 39 0 76
shared taxi 182 60 3 245
Taxi 37 2 0 39
Motorcycle 61 27 2 90
Bicycle 1 10 2 13
Walking 68 20 1 89
Total 386 158 8 552
Figure (4.14): Distribution of transport modes over family size
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
40.00%
45.00%
50.00%
private car shared taxi taxi motorcycle bicycle walking
1-6 persons
7-10 persons
>10 persons
76
Table (4.21): Chi-square test for mode-family size relationship
Value df
Asymp. Sig.
(2-sided)
Pearson Chi-Square 139.101a 55 .000
Likelihood Ratio 118.196 55 .000
Linear-by-Linear
Association .717 1 .397
N of Valid Cases 552
Table (4.22): Cramer’s V test for mode-family size relationship
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
As can be seen from the tables above the null hypothesis stated that there is no
relationship between the family size and the choice of transport mode can be rejected
for 95% confidence level because the significance level is less than 0.05. The
Cramer’s V statistics is 0.224 as can be seen in table 4.19 which means that there is
weak relationship between the age and the choice of transport mode.
4.3.5 Relation between the mode of transport and the monthly income
The distribution of transport modes that are usually be used by the respondents over
the average family monthly income of the respondents was presented in Table 4.23
and Figure 4.15. As can be seen from the chart the percent of travelers using private
car and taxi increases as the monthly income increases and the percent of riders using
motorcycle and walking modes decreases as the monthly income increases. In order to
study the relationship between the monthly income and the choice of transport mode,
a chi-square test was applied to cross tabulation between the transport mode variable
Value
Approx.
Sig.
Nominal by
Nominal Phi .502 .000
Cramer's V .224 .000
N of Valid Cases 552
77
and the average family monthly income variable. The results of chi-square test and
Cramer’s V statistics can be seen in Tables 4.24 and 4.25 respectively.
Table (4.23): Cross tabulation between the mode of transport and monthly income
Figure (4.15): Distribution of transport modes over monthly income
Count Average monthly income Total
less than
1000
NIS
1001-
1500
NIS
1501-
2000
NIS
2001-
3000
NIS
3001-
4000
NIS
4001-
5000
NIS
more
than
5001
Modes private car 0 1 2 11 18 20 24 76
shared taxi 5 14 47 98 67 12 2 245
taxi 0 0 1 3 23 12 0 39
motorcycle 5 22 30 33 0 0 0 90
bicycle 0 0 7 6 0 0 0 13
walking 6 19 24 30 7 2 1 89
Total 16 56 111 181 115 46 27 552
78
Table (4.24): Chi-square test for mode-monthly income relationship
Value df Asymp. Sig.
(2-sided)
Pearson Chi-Square 369.977a 30 .000
Likelihood Ratio 340.922 30 .000
Linear-by-Linear
Association 113.342 1 .000
N of Valid Cases 552
Table (4.25): Cramer’s V test for mode-monthly income relationship
Value
Approx.
Sig.
Nominal by
Nominal Phi .819 .000
Cramer's V .366 .000
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
According to the results shown in the tables above, the null hypothesis stated that
there is no relationship between the average family monthly income and the choice of
transport mode can be rejected for 95% confidence level because the significance
level is less than 0.05. The Cramer’s V statistics is 0.366 as can be seen in Table 4.25
which means that there is weak relationship between the average family monthly
income and the choice of transport mode.
4.3.6 Relation between the mode of transport and the Job
The distribution of transport modes that are usually used by the respondents over the
job of the respondents can be seen in Table 4.26 and Chart 4.16. As can be seen from
the chart, the large percent of private car users locate in businessman category
(85.7%) which has a high income to enable them to own their private vehicles while
the large percent of motorcycle users is from the waged workers (44.8%) that have a
relatively low income so they tend to use a cheap means of transport. In order to study
the relationship between the job and the choice of transport mode, a chi-square test
79
was applied to cross tabulation between the transport mode variable and the job
variable. The results of chi-square test and Cramer’s V statistics can be seen in Tables
4.26 and 4.27 respectively.
Table (4.26): Cross tabulation between the mode of transport and job
Count Job Total
Governmental
employee
Private sector
employee
Un
employee
Business
man or
special
works
waged
worker Others
Modes private car 25 14 19 18 0 0 76
shared taxi 121 60 37 2 20 5 245
taxi 8 27 3 0 0 1 39
motorcycle 31 17 2 1 39 0 90
bicycle 6 1 0 0 6 0 13
walking 31 19 14 0 22 3 89
Total 222 138 75 21 87 9 552
Figure (4.16): Distribution of transport modes over job
81
Table (4.27) :Chi-Square Tests for mode-job relationship
Value df Asymp. Sig.
(2-sided)
Pearson Chi-Square 245.437(a) 25 .000
Likelihood Ratio 212.538 25 .000
Linear-by-Linear
Association 17.410 1 .000
N of Valid Cases 552
Table (4.28): Cramer’s V test for mode-job relationship
Value Appro
x. Sig.
Nominal by
Nominal Phi .667 .000
Cramer's V .298 .000
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
As can be seen from the tables above the null hypothesis which stated that there is no
relationship between the job and the choice of transport mode can be rejected for 95%
confidence level because the significance level is less than 0.05. The Cramer’s V
statistics is 0.298 as can be seen in Table 4.27 which means that there is weak
relationship between the job and the choice of transport mode.
4.3.7 Relation between the mode of transport and the ownership of means of
transport
The distribution of transport modes that are usually used by the respondents over the
ownership of transport means of the respondents can be seen in Table 4.29 and Chart
4.17. As can be seen from the chart, the large percent of shared taxi and walking users
don’t have their own mean of transport (PC, MC, BC) while the large percent of
private car, motorcycle and bicycle users have their own means of transport which are
usually used to go to work .In order to study the relationship between the ownership
of transport means and the choice of transport mode, a chi-square test was applied to
81
cross tabulation between the transport mode variable and the ownership of transport
means variable. The results of chi-square test and Cramer’s V statistics can be seen in
Tables 4.30 and 4.31 respectively.
Table (4.29): Cross tabulation between the mode of transport and ownership of means
of transport
Count availability of
private car
availability of
motorcycle
availability of
bicycle Total
Yes No Yes No Yes No
Modes private car 72 4 0 76 0 76 76
shared taxi 10 235 12 233 1 244 245
Taxi 4 35 0 39 0 39 39
motorcycle 2 88 90 0 4 86 90
Bicycle 0 13 1 12 12 1 13
Walking 3 86 9 80 5 84 89
Total 91 461 112 440 22 530 552
Figure (4.17): Distribution of transport modes over ownership of transport means
82
Table (4.30) :Chi-Square Tests for mode-ownership of transport means relationship
Transport Mean Private Car Motorcycle Bicycle
Value df
Asymp.
Sig. (2-
sided)
Value df
Asymp.
Sig. (2-
sided)
Value df
Asymp.
Sig. (2-
sided)
Pearson Chi-Square 393.474 5 .111 425.719 5 .111 278.647 5 .111
Likelihood Ratio 318.174 5 .111 395.713 5 .111 93.619 5 .111
Linear-by-Linear
Association 96.211 1 .111 49.921 1 .111 31.673 1 .111
N of Valid Cases 552 552 552
Table (4.31): Cramer’s V test for mode-ownership of transport means relationship
Private Car Motorcycle Bicycle
Value Appro
x. Sig. Value
Appro
x. Sig. Value
Appro
x. Sig.
Nominal by
Nominal Phi 8440. .111 8780. .111 7110. .111
Cramer's V 8440. .111 8780. .111 7110. .111
N of Valid Cases 552 552 552 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
As can be seen from the tables above the null hypothesis stated that there is no
relationship between the ownership of transport means and the choice of transport
mode can be rejected for 95% confidence level because the significance level is less
than 0.05. The Cramer’s V statistics is 0.844 , 0.878 , and 0.710 for private care,
motorcycle and bicycle respectively which means that there is strong relationship
between the ownership of transport means and the choice of transport mode.
4.3.8 Relation between the mode of transport and the length of trip
The distribution of transport modes that are usually used by the respondents over the
distance between the home and the work of the respondents can be seen in Table 4.32
and Chart 4.18. As can be seen from the table and chart below the using of non-auto
modes (walking and bicycle) increases as the length of trip decreases while the using
of auto modes ( private car, taxi, shared taxi, and motorcycle) increases as the length
83
of trip increases. In order to study the relationship between trip length and the choice
of transport mode, a chi-square test was applied to cross tabulation between the
transport mode variable and the distance between home and work variable. The
results of chi-square test and Cramer’s V statistics can be seen in Tables 4.33 and 4.34
respectively.
Table (4.32): Cross tabulation between the mode of transport and length of trip
Count Monthly income Total
0.3-1.00
KM
1.1-2.00
KM
2.1-3.00
KM
3.1-4.00
KM
4.1-5.00
KM
5.1-6.00
KM
more
than 6
KM
Modes private car 1 6 23 29 10 7 0 76
shared taxi 2 41 93 31 40 29 9 245
taxi 0 6 14 10 8 1 0 39
motorcycle 0 19 19 20 18 12 2 90
bicycle 1 3 6 2 1 0 0 13
walking 33 46 9 1 0 0 0 89
Total 37 121 164 93 77 49 11 552
Figure (4.18): Distribution of transport modes over trip length
84
Table (4.33):Chi-Square Tests for mode-trip length relationship
Value df Asymp. Sig.
(2-sided)
Pearson Chi-Square 527.932a 280 .000
Likelihood Ratio 455.502 280 .000
Linear-by-Linear
Association 99.438 1 .000
N of Valid Cases 552
Table (4.34): Cramer’s V test for mode-trip length relationship
Value
Approx.
Sig.
Nominal by
Nominal
Phi .978 .000
Cramer's V .437 .000
N of Valid Cases 552
a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.
As can be seen from the tables above the null hypothesis stated that there is no
relationship between the trip length and the choice of transport mode can be rejected
for 95% confidence level because the significance level is less than 0.05. The
Cramer’s V statistics as can be seen in table 4.34 is 0.437 which means that there is
medium relationship between the trip length and the choice of transport mode.
The Table 4.35 shown below summarizes the chi square, Cramer’s V, and
significance of the factors that affect the mode choice. As can be seen in the table,
among the ten factors which affect the mode choice, the ownership of transport mean,
the distance between home and work, gender, age, and monthly income have the
strongest relationship as they have the highest Cramer’s V value while the marital
status, family size and job have weak relationship with the choice of transport mode
as they have low Cramer’s V value.
85
Table (4.35): Test of relationship between the mode choice and travel socioeconomic
variables
Factor Chi-square Cramer’s V Sig. ( 2-sided)
Gender 79.9 0.380 0.000
Marital Status 20.918 0.195 0.001
Age 331.101 0.346 0.001
Family Size 139.101 0.224 0.000
Monthly Income 369.977 0.366 0.000
Job 245.43 0.298 0.000
P.C. ownership 393.474 0.884 0.000
Motorcycle ownership 429.709 0.878 0.000
Bicycle ownership 278.647 0.710 0.000
Distance 527.932 0.437 0.000
4.4 Relation between the captive ridership and socioeconomic characteristics
To study the relation between socioeconomic variables and the captive ridership, the
chi-square test was used. This test was applied to cross tabulations between the
variable captive (which categorized into captive and choice) and the socioeconomic
variables. The null hypothesis to be tested stated that is that there is no relationship
between the captive and socioeconomic variable used. The null hypothesis will be
rejected if the significance level is less than 0.05 for 95% confidence interval.
Cramer’s V statistics was used to give indication of the strength of the relationship
which range from 0-1 i.e the higher the better. The Table 4.36 shown below
summarizes the chi-square, Cramer’s V statistics, and the significance of travel
socioeconomic variables which affect the captive ridership. As can be seen from the
table there are six variables which affect the captive ridership where the null
hypothesis can be rejected at a confidence level greater than 95%. These variables are
gender, job, P.C. ownership, motorcycle ownership, bicycle ownership and distance
(trip length). Among these variables the distance and gender variables seem to have
the strongest relationship as they have the highest Cramer’s V values.
86
Table (4.36): Test of relationship between the captive ridership and travel
socioeconomic variables
Factor Chi-square Cramer’s V Sig. ( 2-sided)
Gender 29.79 0.232 0.000
Marital Status 1.097 0.045 0.295
Age 30.668 0.236 0.881
Family Size 7.004 0.113 0.799
Average monthly Income 6.883 0.112 0.332
Job 18.047 0.181 0.003
P.C. ownership 10.497 0.138 0.001
Motorcycle ownership 28.35 0.227 0.000
Bicycle ownership 9.026 0.128 0.001
Distance 90.060 0.404 0.003
On the other hand, marital status, age, family size, and average family monthly
income were not found to have an effect on the captive ridership where the null
hypothesis can’t be rejected at a confidence level greater than 95%.
4.5 Hypothetical questions
The distribution of travelers’ choice for the hypothetical questions mentioned in the
questionnaire shown in [Appendix 1] can be seen in Table 4.37 and Figure 4.19
Table (4.37): Distribution of riders’ choice for different levels
87
Figure (4.19): Distribution of riders’ choice for different levels (total sample)
As can be seen from the table and chart above the large percent of travelers prefer to
use shared taxi so they put it as the first choice for different levels while the large
percent of travelers don’t prefer to use bus so they put it as the last choice in different
levels. The table and chart also show that the percent of travelers who put minibus as
their first choice increases as the frequency decreases although the reduction in cost
decreases. This result illustrates that the travelers are more sensitive to frequency
which give indication about the waiting time rather than the cost and in vehicle travel
time because the travel cost and in vehicle travel time using shared taxi is relatively
small so any reduction in cost or increasing of in vehicle travel time will not be
significant if the travel waiting time or frequency is large. This will illustrate the large
percent of travelers who put minibus in the third level as the first choice (which
represent about 33.2% of total sample) as the frequency for minibus is 8 minutes
which is relatively near to those of shared taxi although the reduction of cost is only
10 % less than the cost of travel by shared taxi.
88
4.6 Importance of factors that affect mode choice
The relative importance index and the rank of the factors that affect the mode choice
can be seen in Table 4.38
Table (4.38): Relative Importance Index and Rank of the factors that affect mode
choice
Factor RII Rank
Health status 0.884 1
Distance (trip length) 0.806 2
Waiting time 0.788 3
Weather conditions 0.786 4
Travel time (IVTT) 0.755 5
Travel cost 0.752 6
Monthly income 0.748 7
Age 0.703 8
Gender 0.703 8
Comfort 0.671 9
Privacy 0.666 10
The results shown in the table above show that health status, distance, waiting time,
weather condition, in vehicle travel time, and cost is the most important factors that
affect the mode choice so these factors have to be taken into account when calibrating
the model. Some factors like health status and weather condition will be taken into
account through adding constant for the utility functions of the modes to reflect the
unobserved conditions which are difficult to account. These results will be compared
with the results which will be obtained from the model to stand on the consistency
between the results obtained from direct question for respondents about the factors
that affect their mode choice and the results obtained from indirect question which
can be concluded through calibration process.
89
4.7 Calibration of revealed model
On the basis of descriptive analysis of the data, there are six modes to be considered
for modeling the mode choice for work trips in Gaza city which are private car, shared
taxi, taxi, motorcycle, bicycle, and walking. Different specifications have been
evaluated to determine which specifications which replicate the data for work trips in
Gaza city. The list of variables that have been used in this research include the
variables that have been found from the literature review such as in vehicle travel time
(IVTT), out of vehicle travel time (OVTT), total travel time (TT), total travel cost
include the depreciation for private car and motorcycle modes (TC), age of
respondent (AGE), Gender of respondent (GENDER), ownership of transport means
(OWTM), family income (FINC), and distance (DIST). Composition variable such as
family income over family size and total cost over family income also have been used
to modify the impact of pure level of service variables. The list of variables that have
been used in model calibration with their abbreviations are presented in Table 4.39
Table (4.39): Abbreviation and description of explanatory variables
Variable Description
IVTT In vehicle travel time in minutes (generic variable)
OVTT Out of vehicle travel time in minutes (generic variable)
TT Total travel time in minutes (generic variable)
TC Total travel cost in ILS (generic)
GENDER Gender of respondent ( I if female and 0 otherwise)
AGE Age of respondent in years
DIST Length of the trip in kilometers
OWTM Ownership of transport means ( 1 if have transport mean and 0
otherwise)
FINC Average family monthly income in ILS
PINC Average monthly income per person in ILS
TC/PINC Total cost over person income in ILS
Constant Mode specific constant
91
Different model specifications were tested based on prior experience in intercity mode
choice modeling and the impact of introducing additional explanatory variables. The
specifications of the model are formulated based on the following criteria:
Wrong sign coefficient variables were dropped from the model.
Variables with insignificant coefficients were dropped from the model except the
level of service variables (travel time and travel cost).
Some variables with insignificant coefficient were considered based on its
improving the statistics of the model.
The level of service variables were considered in different forms (strait forward as
cost and travel time) or in ration form such as cost over income.
Sets of variables with high correlation were considered.
Some of intuitively important variables which have been dropped from the model
were reconsidered.
The mode specific constants were considered in spite of the significance of
coefficients of the variables.
The first model which has been estimated includes total travel time (TT) and total
travel cost (TC) as a generic variables which means that an increase of one unit of
travel time or travel cost has the same impact on the modal utility for all modes. The
distance variable (DIST) considered as a specific variable for bicycle and walking
modes. The private car mode was considered as a base mode when adding constants
for the mode utilities. The utilities for different modes can be written as the following:
(4.1)
The results of estimation shown in Table 4.40 illustrate that the estimated coefficients
of cost and travel time variables have a negative sign as expected which means that
the utility of modes decreases as the travel time and travel cost increase.
91
Table (4.40): Estimation results of model_1
Parameters MODEL_1 (MNL)
Estimated Value t -statistics
Generic Parameters
TT -0.0915 -2.7645
TC -0.5148 -1.6013
Alternative Specific Parameters
CONSTANT S_Taxi -0.2971 -0.4490
CONSTANT Taxi -0.1575 -0.1235
CONSTANT Motorcycle 0.1316 0.1641
CONSTANT Bicycle 0.8295 0.3938
CONSTANT Walking 2.5356 2.3450
DIST Bicycle -0.2001 -0.2716
DIST Walking -1.6014 -3.8219
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -120.9370
Rho Squared w.r.t. Zero 0.3666
Rho Squared w.r.t Constants 0.1638
Adjusted Rho Squared w.r.t. Zero 0.3195
Adjusted Rho Squared w.r.t Constants 0.1316
Number of Cases 368
Number of iterations 41
Estimation status Converged, with contants, with zeros, valid lic.
The results of estimation also show that both travel time (TT) and distance for
walking (DIST) have a large absolute values of t-statistics of 2.764 and 3.82
respectively which are greater than critical t- value at 90% confidence level (1.65),
this result leads to reject the null hypothesis that these variables have no effect on
modes utilities. Although the t-statistics of travel cost variable is lower bit than critical
t-value at 90% confidence level it should be included in the model because it
considered as a policy variable (Qrtuzar and Willumsen 2002). The lack in
significance of the alternative specific constants for shared taxi, taxi, motorcycle and
bicycle is immaterial since the constants represent the average effect of all variables
not included in the model and always should retain in the model despite the fact they
don’t have well understood behavior interpretation (Koppelman and Bhat 2006). The
t-statistic on bicycle specific distance variable is less than critical t-value even at 90%
confidence level, suggestion that the effect of distance on bicycle utility may not
92
differentiate it from the reference mode (private car mode) so this variable is
considered to drop from the model.
The removing of this variable (DIST) from the utility of bicycle leads to model_2
where the utility of bicycle can be written as the following while the utilities for the
other modes still they are in model_1:
(4.2)
The results of estimation for model_2 can be shown in Table 4.41
Table (4.41): Estimation results of model_2
Parameters MODEL_2 (MNL)
Estimated Value t-statistics
Generic Parameters
TT -0.092 -2.7823
TC -0.5074 -1.5892
Alternative Specific Parameters
CONSTANT S_Taxi -0.2818 -0.4280
CONSTANT Taxi -0.1836 -0.1444
CONSTANT Motorcycle 0.1408 0.1761
CONSTANT Bicycle 0.3512 0.3161
CONSTANT Walking 2.5407 2.3520
DIST Walking -1.5937 -3.8141
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -120.9767
Rho Squared w.r.t. Zero 0.3665
Rho Squared w.r.t Constants 0.1636
Adjusted Rho Squared w.r.t. Zero 0.3246
Adjusted Rho Squared w.r.t Constants 0.1381
Number of Cases 368
Number of iterations 13
Estimation status converged, with constants, with zeros, valid lic.
The results for two models show that both have a good goodness of fit measures ρ2.
To compare the two models a likelihood test was applied in order to study the impact
of exclusion a distance specific variable from the utility of bicycle mode. The null
hypothesis to be tested stated that there is no impact of a distance specific variable on
the mode choice decision
93
(4.3)
The statistical test that a distance specific variable (DIST) for bicycle mode has no
effect on the choice decision has a chi square value of
(4.4)
Where,
LLR is the log likelihood of the restricted model
LLU is the log likelihood of the unrestricted model
With one degree of freedom ( one variable was constrained to zero), the null
hypotheses can’t be rejected even at low confidence level 90% where the chi square
value << critical chi square value at 90% confidence level (2.71) thus the distance
specific variable can be excluded from the bicycle utility.
In order to improve the statistics of model_2, new intuitive variables were added to
the model. Average monthly income for family variable (FINC) was added to the
utility function of taxi, bicycle, motorcycle and walking and the model was estimated
and labeled as model_3. The utility functions for different modes can be written as the
following
(4.5)
The results of estimation presented in Table 4.42 illustrate that both travel time and
travel cost variables have a correct negative sign of coefficients and they are
statistically significant at 90% confidence level where the t-statistics value are greater
than the critical one (1.645). The results also show that both average family monthly
income variables for taxi and bicycle are statistically insignificant at significance level
greater than 0.1 where the t-statistics value is less than the critical one at 90%
94
confidence level so the null hypothesis that these variables have no effect on choice
decision can’t be rejected even at 90% confidence level so these two variables are
suggested to drop from the mode.
Table (4.42): Estimation results of model_3
Parameters Model_3 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1301 -3.4088
TC -0.6816 -1.8606
Alternative Specific Parameters
CONSTANT S_Taxi 0.0371 0.0513
CONSTANT Taxi -1.3403 -0.6259
CONSTANT Motorcycle 3.3519 1.8833
CONSTANT Bicycle -71.8416 -0.0073
CONSTANT Walking 5.9025 3.3668
DIST Walking -1.9979 -4.2545
FINC Taxi 0.0005 1.1787
FINC Motorcycle -0.0016 -2.2157
FINC Bicycle 0.0512 0.0065
FINC Walking -0.001 -2.7333
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -106.2369
Rho Squared w.r.t. Zero 0.4436
Rho Squared w.r.t Constants 0.2655
Adjusted Rho Squared w.r.t. Zero 0.3808
Adjusted Rho Squared w.r.t Constants 0.2098
Number of Cases 368
Number of iterations 16
Estimation status converged, with contants, with zeros, valid lic.
The dropping of average family monthly income variable (FINC) from the utility
functions of bicycle and taxi modes leads to new model labeled as model_4 where the
utilities of taxi and bicycle modes can be written as the following while the utilities
for the other modes still as they are in the previous one.
(4.6)
The results of estimation for the model mentioned above are reported in Table 4.43
95
Table (4.43): Estimation results of model_4
Parameters Model_4 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1068 -2.9849
TC -0.4878 -1.5109
Alternative Specific Parameters
CONSTANT S_Taxi 0.032 0.047
CONSTANT Taxi -0.1777 -0.1378
CONSTANT Motorcycle 3.973 2.3196
CONSTANT Bicycle 1.2022 1.0167
CONSTANT Walking 6.3621 3.7658
DIST Walking -1.9967 -4.1707
FINC Motorcycle -0.0017 -2.3869
FINC Walking -0.0011 -3.1419
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -113.6093
Rho Squared w.r.t. Zero 0.405
Rho Squared w.r.t Constants 0.2145
Adjusted Rho Squared w.r.t. Zero 0.3526
Adjusted Rho Squared w.r.t Constants 0.1739
Number of Cases 368
Number of iterations 13
Estimation status converged, with contants, with zeros, valid lic.
The results reported in the table above show that both travel time and travel cost have
negative sign of coefficients as expected. The results also show that travel time (TT),
distance (DIST), average family monthly income for motorcycle (FINC motorcycle) , and
average family monthly income for walking (FINC walking) have absolute t-statistics
value of 2.98, 4.17,2.38, and 3.14 respectively which are greater than the critical t-
value at 95% confidence level so the null hypothesis stated that these variables has no
effect on choice decision can be rejected at significance level greater than 0.05. Even
though the travel cost variable (TC) has a low absolute t-statistics value of 1.51 which
is less than the critical t-value even at 90% confidence level it will retain in the model
because it is considered as level of service variable. In order to study the effect of
adding average family monthly income variable to both motorcycle and walking
utilities on the statistics measures of the model, this model was compared with
model_2. The comparison of the two models show that the goodness of fit measures
96
for this model ,
, and
were improved to 0.405, 0.2145, 0.3526, and 0.1739
compared with 0.3665, 0.1636, 0.3246, and 0.1381 for model_2.
The likelihood ratio test was applied to between modl_4 and mode_2 to test the
hypothesis that the average family monthly income variable (FINC) for motorcycle
and walking modes can be excluded from the model
(4.7)
The statistical test has a chi square value of - - - which
is greater than critical chi square value even at 99.9 % confidence level with two
degrees of freedom ( two variables were constrained to zero) (13.82). This result leads
to reject the null hypothesis stated that these variables have no effect on choice
decision and accordingly they could not be excluded from the model.
For further improvement the statistics of model_4 the ownership of transport means
variable (OWTM) was added to the utility of shared taxi and the new model was
labeled as model_5. The utilities for different modes can be written as the following:
(4.8)
Table 4.44 presents the estimation results of model. As can be seen from the table
both travel cost and travel time coefficients have a correct negative sign. The results
also show that the total travel time (TT), ownership of transport means (OWTM S_taxi),
distance (DIST walking) and average family monthly income for motorcycle and
walking modes (FINC Motorcycle, FINC Walking) variables have a large absolute t-
statistics values of 2.774, 1.828, , 3.98, 2.53, and 3.04 respectively which are greater
than the critical t-value at 90% confidence level (1.645) . This result leads to reject the
null hypothesis stated that these variables have no effect on choice decision. Although
total cost (TC) variable has absolute value of t-statistic (1.4138) which is lower bit
97
than critical t-value at 90% confidence level, it will retain in the model as it is
considered as a policy variable.
Table (4.44): Estimation results of model_5
Parameters Model_5 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1039 -2.7774
TC -0.4579 -1.4138
Alternative Specific Parameters
CONSTANT S_Taxi 0.6291 0.8152
CONSTANT Taxi 0.1739 0.1332
CONSTANT Motorcycle 4.4475 2.4906
CONSTANT Bicycle 1.3541 1.1244
CONSTANT Walking 6.5117 3.8294
OWTM S_Taxi -0.9876 -1.8281
DIST Walking -1.8958 -3.9829
FINC Motorcycle -0.0019 -2.5321
FINC Walking -0.0011 -3.0413
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -111.9227
Rho Squared w.r.t. Zero 0.4138
Rho Squared w.r.t Constants 0.2262
Adjusted Rho Squared w.r.t. Zero 0.3562
Adjusted Rho Squared w.r.t Constants 0.1785
Number of Cases 368
Number of iterations 12
Estimation status converged, with contants, with zeros, valid lic.
The goodness of fit measures for this model ,
, and
were improved to
0.4138, 0.2262, 0.3562, and 0.1785 compared with 0.405, 0.2145, 0.3526, and 0.1739
for model_4.
To study the hypothesis involving with the exclusion of OWTM variable from the
model, the likelihood ration test was applied between model_4 and model_5. The null
hypothesis to be tested stated that the OWTM variable can be excluded from the
model
(4.9)
98
The statistical test has a chi square value of - - - - thus
the null hypothesis can be rejected at significance level greater than 0.1 and one
degree of freedom (one variable OWTM was constrained to zero). According to this
result the OWTM variable has to be retaining in the model.
To study the effect of social characteristics on mode choice, gender variable
(GENDER) was added to both taxi and walking utilities and the age variable (AGE)
was added to motorcycle, bicycle and walking utility functions. The new model was
labeled as model_6 with the following utility functions
(4.10)
The results of model estimation were reported in Table 4.45. The results show that the
travel time and travel cost variables have a correct sign of estimators and they are
statistically significant at 90% confidence level where the t-statistics value is greater
than the critical one. The results also show that the OWTM S_taxi, DIST walking,,
FINCmotorcycle, FINC walking, and AGE bicycle are statistically significant at significance
level greater than 0.1 with a t-statistics values of 1.65, 4.4189, 2.7413, 2.8678, and
2.4589 respectively so the null hypothesis that these variables have no effect on mode
choice can be rejected at significance level greater than 0.1. Gender variable for both
taxi and walking modes and age variable for motorcycle and walking modes are
statistically insignificant at 90% confidence level as the t-statistics for these variables
have values of 0.2333, 1.1818, 0.9684, and 0.0109 which are less than the critical t-
value at 90% confidence level (1.645). So that the null hypothesis that these variables
have no effect on choice decision can’t be rejected at the specified significance level
thus these variables are suggested to be excluded from the model.
99
Table (4.45): Estimation results of model_6
Parameters Model_6 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1129 -2.8731
TC -0.5919 -1.7198
-- Alternative Specific Parameters
CONSTANT S_Taxi 0.6138 0.7658
CONSTANT Taxi 0.6095 0.4515
CONSTANT Motorcycle 3.1076 1.2218
CONSTANT Bicycle -16.8546 -2.314
CONSTANT Walking 7.1476 3.3614
GENDER Taxi 0.1401 0.2333
GENDER Walking -0.9417 -1.1818
AGE Motorcycle 0.0714 0.9684
AGE Bicycle 0.4376 2.4589
AGE Walking -0.0004 -0.0109
OWTM S_Taxi -0.9442 -1.6562
DIST Walking -2.1831 -4.4189
FINC Motorcycle -0.0025 -2.7413
FINC Walking -0.0011 -2.8678
-- Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -106.7392
Rho Squared w.r.t. Zero 0.441
Rho Squared w.r.t Constants 0.262
Adjusted Rho Squared w.r.t. Zero 0.3572
Adjusted Rho Squared w.r.t Constants 0.1797
Number of Cases 368
Number of iterations 18
Estimation status converged, with contants, with zeros, valid lic.
The exclusion of gender variable from the utility functions of taxi and walking modes
in addition to the exclusion of age variable from the utility functions of motorcycle
and walking modes leads to new model labeled as model_7. The results of estimation
for the model which were reported in Table 4.46 show that all the variables have a
correct sign of estimators and they are statistically significance at a confidence level
of 90% as the t-statistics for them are greater than critical t-value at significance level
greater than 0.1 so the null hypothesis stated that these variables have no effect on
choice decision can be rejected at that confidence level.
111
Table (4.46): Estimation results of model_7
Parameters Model_7 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1205 -3.0795
TC -0.574 -1.6783
Alternative Specific Parameters
CONSTANT S_Taxi 0.7542 0.9483
CONSTANT Taxi 0.6375 0.4745
CONSTANT Motorcycle 4.7429 2.5529
CONSTANT Bicycle -15.9109 -2.3185
CONSTANT Walking 6.9207 3.8584
AGE Bicycle 0.4105 2.4881
OWTM S_Taxi -0.9837 -1.7851
DIST Walking -2.0619 -4.3152
FINC Motorcycle -0.0021 -2.6665
FINC Walking -0.0011 -3.0245
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -107.9959
Rho Squared w.r.t. Zero 0.4344
Rho Squared w.r.t Constants 0.2533
Adjusted Rho Squared w.r.t. Zero 0.3716
Adjusted Rho Squared w.r.t Constants 0.1981
Number of Cases 368
Number of iterations 18
Estimation status converged, with contants, with zeros, valid lic.
In order to test the hypothesis involving with exclusion of gender variable (GENDER)
from the utility functions of taxi and walking modes in addition to exclusion of age
variable from the utility functions of motorcycle and walking modes, the likelihood
ratio test was applied between model_6 and model_7. The null hypothesis can be
written as the following
(4.11)
The statistical test has a chi square value of - - - -
which is less than critical chi square value at significance level greater than 0.1 with
four degrees of freedom ( four variables were constrained to zero) (7.78) . According
111
to this result the null hypothesis can’t be rejected even at 90% confidence level so
these variables seems to have no effect on choice decision.
For studying the possibility of excluding age variable from the utility function of
bicycle mode, the likelihood ratio test was applied between model_7 and model_5.
The statistical test has a chi square value of - - - - .
With this condition the null hypothesis that this variable can be excluded from the
model can be rejected even at significance level greater than 0.01 with one degree of
freedom as the chi square value (7.854) is greater than the critical value (6.63).
The decision maker related variables such as average income, ownership of transport
means, family size and others can be included in the models by two approaches. The
first is to include them as specific variables to each or some of alternatives (except for
the reference alternative. All the models reported to this point used this approach for
inclusion of decision maker related variables in the models. The other approach is to
include such variables through interaction with mode attributes such as dividing cost
by income to reflect decreasing the importance of cost by increasing the annual
income.
To take this issue into account, the travel cost variables in the previous model
(model_7) was replaced by cost over person income variable (TC/PINC).The new
modes was labeled as model_8 with the following utility functions
(4.12)
The results of estimation which are presented in Table 4.47 show that travel time and
cost over personal income variables have a correct sign of estimators. The results also
show that except ownership of transport means (OWTM) which is statistically
significant at 90% confidence level, all the variables are statistically significant at
significance level greater than 0.05 as they have an absolute value of t-statistics larger
112
than critical t-value at 95% confidence level (1.96). This result leads to reject the null
hypothesis that these variables has no effect on choice decision at significance level
greater than 0.1 (90% confidence level)..
Table (4.47): Estimation results of model_8
Parameters Model_8 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.1299 -3.264
TC/PINC -227.5075 -2.7647
Alternative Specific Parameters
CONSTANT S_Taxi 1.0773 1.4379
CONSTANT Taxi -0.5826 -0.987
CONSTANT Motorcycle 4.9794 2.7175
CONSTANT Bicycle -15.3013 -2.2932
CONSTANT Walking 7.366 4.2017
AGE Bicycle 0.3979 2.4455
OWTM S_Taxi -1.0626 -1.9061
DIST Walking -2.16 -4.3961
FINC Motorcycle -0.0021 -2.6153
FINC Walking -0.001 -2.7697
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -105.8891
Rho Squared w.r.t. Zero 0.4454
Rho Squared w.r.t Constants 0.2679
Adjusted Rho Squared w.r.t. Zero 0.3826
Adjusted Rho Squared w.r.t Constants 0.2122
Number of Cases 368
Number of iterations 15
Estimation status converged, with contants, with zeros, valid lic.
The goodness of fit measures for this model ,
, and
were improved to
0.4454, 0.2679,0.3826, and 0.2122 compared with 0.4344, 0.2533,0.3716, and 0.1981
for model_7 respectively.
To compare model_7 with model_8 a non-nested hypothesis test was applied for this
purpose as the two models have the same number of variables. The null hypothesis to
be tested stated that the lower roh-square model is the true model. In non-nested
hypothesis test the adjusted roh-square is used to test the hypothesis. The null
113
hypothesis can be rejected at significance level SL determined by the following
equation
(4.13)
Where,
is the adjusted roh-square relative to the zero model with higher value.
is the adjusted roh-square relative to the zero model with lower value
is the standard normal cumulative distribution function
As model_8 has better goodness-of-fit than model_7 then the null hypothesis to be
tested is that the model_7 is the true model. The significance level to reject the null
hypothesis can be calculated as follows
As the significance level calculated from the equation above is less than 0.05 then the
null hypothesis that model_7 is the true model can be rejected at significance level
greater than 0.05. The result is consistence with the theory that the importance of
travel cost decreases as the income increases.
The above formulated models assume that the utility of in vehicle travel time (IVTT)
and out of vehicle travel time is (OVTT) is equal; however the workers may be more
sensitive to one of them than the other. In order to take this issue into account the total
travel time was disaggregated into two parts namely in vehicle travel time and out of
vehicle travel time and a new model was formulated with the following utilities
(4.14)
114
The new model was labeled as model_9 and the results of estimation for the model
can be seen in Table 4.48.
Table (4.48): Estimation results of model_9
Parameters Model_1 (MNL)
estimated value t-statistics
Generic Parameters
IVTT -0.2939 -3.873
OVTT -0.0647 -1.4002
TC/PINC -261.2765 -3.0919
Alternative Specific Parameters
CONSTANT S_Taxi 0.6629 0.8788
CONSTANT Taxi -0.7432 -1.2693
CONSTANT Motorcycle 5.7455 2.9782
CONSTANT Bicycle -13.8246 -1.867
CONSTANT Walking 7.5244 4.1174
AGE Bicycle 0.3913 2.2187
OWTM S_Taxi -0.9586 -1.6886
DIST Walking -0.9354 -1.3755
FINC Motorcycle -0.0024 -2.8245
FINC Walking -0.0011 -2.7988
Model Statistics
Log Likelihood at Zero -190.9438
Log Likelihood at Constants -144.6343
Log Likelihood at Convergence -102.2029
Rho Squared w.r.t. Zero 0.4647
Rho Squared w.r.t Constants 0.2934
Adjusted Rho Squared w.r.t. Zero 0.3967
Adjusted Rho Squared w.r.t Constants 0.2301
Number of Cases 368
Number of iterations 15
Estimation status converged, with contants, with zeros, valid lic.
The results shown in the table above show that in vehicle, out of vehicle travel time,
and total travel cost variables have a correct negative sign of estimators. The results
also show that all the variables except out of vehicle (OVTT) and distance variables
(DIST) are statistically significant at 90% confidence level where the absolute value
of t-statistics are greater than critical t-value (1.645) so that the null hypothesis that
these variables have no effect on choice decision can be rejected at level of
significance greater than 0.1. Although OVTT, and DIST are statistically insignificant
at 90% confidence level, caution should be taken before removing it from the model
as the dropping it may adversely affect the significance of other variables.
115
As can be seen from the results of estimation the in vehicle travel time has a larger
coefficient than out of vehicle travel time , this results contradict with the results of
some researches such as (Almasri 2011) which concluded that the travelers are more
sensitive to out of vehicle travel time than in vehicle travel time. The good
accessibility for the trips in Gaza city and the short access, egress and waiting time of
the trips may explain this result.
The test of hypothesis of equal value of in and out of vehicle travel time, the
likelihood ratio test was applied between model_8 and model_9. The statistical test
has a chi square value of - - - - . This result leads
to reject the null hypothesis and accordingly reject the constraints imposed by
model_8 at a significance level greater than 0.05 with one degree of freedom as the
chi square value calculated above is larger than the critical chi square value at 95%
and one degree of freedom (3.84).
By comparing the above formulated models it is clear that model_8 and model_9 have
the best goodness of fit measures among the estimated models. Although model_9 has
goodness of fit measures ,
, and
of value 0.4647, 0.2934, 0.3967, and
0.2301 respectively which is better than the goodness of fit measures for model_8
which has a goodness of fit measures ,
, and
of 0.4454, 0.2679, 0.3826,
and 0.2122 respectively , but this model suffer from shortage represents on that some
of variables in this model ( out of vehicle (OVTT), and distance (DIST) are
statistically insignificant even at 90% confidence level. So based on the criteria which
were mentioned in the methodology in chapter 3 for comparing the models and
choosing the most satisfactory one, model_8 seems to be the most satisfactory one for
representing the behavior of employed people in choosing the mode of transport in
Gaza city as this model has a correct sign of estimators and all the variables are
statistically significant at 90% confidence level in addition it has a good goodness of
fit measures while some of variables in model_9 which is better than it in goodness of
fit measures are statistically insignificant at 90% confidence level. The utility
functions for the model can be written as following
116
4.8 Validation for revealed model
Model validation is considered very important process to evaluate the performance of
the calibrated model and its ability to predict modal split for data other than that used
for calibration process. The validation process is tested on three different phases. The
first phase is the test of reasonableness validation process which was tested during the
calibration process depending on the expected sign of estimators. All models with a
wrong sign of estimators would not consider as a valid model. Based on this criterion,
it is clear that model_8 which was chosen as the most satisfactory model for work
trips in Gaza city is considered as a valid model because the travel time and travel
cost variables have correct sign of estimators (negative signs).
The second phase of validation process is the statistical validation test which is
conducted by the likelihood ratio test (LRTS). This test was conducted for model_8
using about 1/3rd
of the data sets (184 observations). The details for this test were
discussed in chapter three. The results of this test show that the calculated chi square
was - - - - . With twelve degrees of freedom (number
of restricted coefficients) as indicted in equation 4.12, the calculated chi square value
can’t lead to reject the null hypothesis stated that there is no difference between the
predicted and observed behavior because the calculated chi square value is less than
critical chi square value at 95% confidence level and twelve degrees of freedom
(21.026).
The last phase for validation process is calculated the prediction capability of the
calibrated model (model_8). To calculate the prediction ratio, the utility for each trip
maker was calculated then the probability of each alternative (mode) was estimated.
The alternative with the highest probability is predicted to be the chosen mode for that
particular. The number of travelers correctly predicted was summed up to each
alternative and compared to yield the prediction value. The calculated prediction
value was 0.69 which means that the model is capable to predict about 69% of the
choices of the trip makers’ correctly.
117
4.9 Calibration of stated preference model
The stated preference model was calibrated using the data collected from the answers
of respondents about the hypothetical questions in part V of questionnaire. On the
basis of data analysis there are three modes to be considered for modeling which are
shared taxi, minibus and bus modes. The main objective for this model is to
investigate the market share when introducing bus service to transport system in Gaza
city. The list of variables that have been used in model and their abbreviations can be
seen in Table 4.49
Table (4.49): Abbreviation and description of explanatory variable used in stated
preference model
Variable Description
TT In vehicle travel time in minutes (generic variable)
FARE Total travel cost in ILS (generic)
FREQ Service frequency in minutes (generic variable)
GENDER Gender of respondent ( I if female and 0 otherwise)
AGE Age of respondent in years
DIST Length of the trip in kilometers
FINC Average family monthly income in ILS
PINC Average monthly income per person in ILS
FARE/PINC Total cost over person income in ILS
Constant Mode specific constant
The criteria that have been used for calibrating the stated preference model are the
same that have been used in revealed model and mentioned above.
The first model has been estimated (model_S1) includes travel time (TT), fare of trip
(FARE), and service frequency (FREQ). Constants have been added to bus and
minibus modes. The utility functions for different modes can be written as the
following
118
(4.15)
The results of estimation for this model have been reported in Table 4.50
Table (4.50): Estimation results of model_S1
Parameters model_S1 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.4093 -3.4477
FARE -3.0676 -4.316
FREQ -0.06 -3.3729
Alternative Specific Parameters
CONSTANT Minibus -0.6597 -3.3386
CONSTANT Bus -1.549 -3.9467
Model Statistics
Log Likelihood at Zero -542.7145
Log Likelihood at Constants -380.0125
Log Likelihood at Convergence -359.5908
Rho Squared w.r.t. Zero 0.3374
Rho Squared w.r.t Constants 0.0537
Adjusted Rho Squared w.r.t. Zero 0.3282
Adjusted Rho Squared w.r.t Constants 0.0456
Number of Cases 494
Number of iterations 13
Estimation status converged, with contants, with zeros, valid lic.
The results of estimation shown in the table above indicated that all the variables have
a correct sign of coefficients and they are statistically significance at 95% confidence
level as the t-statistics for these variables are greater than critical t-value at 95%
confidence interval (1.96), hence the null hypothesis that these variables have no
effect on choice decision cab be rejected at significance level greater than 0.05. The
goodness of fit measures for this model is considered low where the , and
for
this model are 0.0537 and 0.0456 respectively.
For improvement the statistics of the previous model a new explanatory variables
were added to the utility functions for different modes. The age variable (AGE),
average family monthly income variable (FINC), and gender variable (ENDER) were
added to minibus and bus modes. The new model was labeled as model_S2. The
utilities for different modes can be written as the following
119
(4.16)
The results of estimation for model_S2 presented in Table 4.51 show that the
goodness of fit measures was improved dramatically compared with model_S1 where
the values of ,
, and
for this model are 0.5323, 0.3321, 0.512, and 0.3068
respectively while these values for model_S1 are 0.3374, 0.0537, 0.3282, and 0.0456
respectively.
Table (4.51): Estimation results of model_S2
Parameters model_S2(MNL)
Estimated value t-statistics
Generic Parameters
TT -0.4118 -2.708
FARE -2.7787 -3.0092
FREQ -0.0762 -3.1972
Alternative Specific Parameters
CONSTANT Minibus 1.2998 1.9289
CONSTANT Bus 3.1458 2.595
GENDER Minibus 0.4946 1.8102
GENDER Bus -0.4395 -0.8498
AGE Minibus 0.0603 3.5379
AGE Bus 0.1036 3.78
FINC Minibus -0.0019 -8.021
FINC Bus -0.0044 -8.1287
Model Statistics
Log Likelihood at Zero -542.7145
Log Likelihood at Constants -380.0125
Log Likelihood at Convergence -253.8182
Rho Squared w.r.t. Zero 0.5323
Rho Squared w.r.t Constants 0.3321
Adjusted Rho Squared w.r.t. Zero 0.512
Adjusted Rho Squared w.r.t Constants 0.3068
Number of Cases 494
Number of iterations 15
Estimation status converged, with contants, with zeros, valid lic.
111
As can be seen from the table above TT, FARE, FREQ, AGE, and FINC variables
have expected sign of coefficients and they are statistically significant at 95%
confidence level so the null hypothesis that these variables have no effect on choice
decision can be rejected at this confidence level ,hence these variables should be
retain in the model. In contrast gender variable (GENDER) has unexpected sign of
coefficient for minibus and it is statistically insignificant for bus mode even at 90%
confidence level so the GENDER variable is suggested to remove from the model.
The dropping of gender variable from the model leads to new model labeled as
model_S3. The results of estimation for this model can be seen in Table 4.52.
Table (4.52): Estimation results of model_S3
Parameters model_S3 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.3862 -2.6172
FARE -2.6531 -2.89
FREQ -0.0761 -3.25
Alternative Specific Parameters
CONSTANT Minibus 1.4636 2.2021
CONSTANT Bus 2.935 2.4903
AGE Minibus 0.0588 3.4747
AGE Bus 0.1082 3.9707
FINC Minibus -0.0018 -7.8992
FINC Bus -0.0045 -8.1862
Model Statistics
Log Likelihood at Zero -542.7145
Log Likelihood at Constants -380.0125
Log Likelihood at Convergence -256.7447
Rho Squared w.r.t. Zero 0.5269
Rho Squared w.r.t Constants 0.3244
Adjusted Rho Squared w.r.t. Zero 0.5103
Adjusted Rho Squared w.r.t Constants 0.3044
Number of Cases 494
Number of iterations 13
Estimation status converged, with contants, with zeros, valid lic.
For testing the null hypothesis stated that gender variable can be excluded from the
utility functions of minibus and bus modes the likelihood ratio test was applied
between model_S2 and model_S3. The statistical test has a chi square value of
- - - - . With this condition the null hypothesis can’t
111
be rejected at significance level greater than 0.05 with two degrees of freedom as the
chi square value (5.85) is less than the critical value (5.99) hence the gender variable
can be dropped from the model.
The theory and the practices in mode choice modeling indicated that the importance
of cost decreases as the monthly income increases. To take this issue into account the
fare variables in the previous model was replaced by fare over person income variable
(FARE/PINC). The new model was labeled as model_S4. The results of estimation
for the model were reported in Table 4.53.
Table (4.53): Estimation results of model_S4
Parameters model_S4(MNL)
Estimated value t-statistics
Generic Parameters
TT -0.2106 -1.7393
FREQ -0.0689 -2.9793
FARE/PINC -404.2799 -2.4162
Alternative Specific Parameters
CONSTANT Minibus 1.7941 2.6785
CONSTANT Bus 4.1263 3.631
AGE Minibus 0.0477 2.8001
AGE Bus 0.0743 2.5139
FINC Minibus -0.0018 -7.6681
FINC Bus -0.0042 -7.5742
Model Statistics
Log Likelihood at Zero -542.7145
Log Likelihood at Constants -380.0125
Log Likelihood at Convergence -258.0411
Rho Squared w.r.t. Zero 0.5245
Rho Squared w.r.t Constants 0.321
Adjusted Rho Squared w.r.t. Zero 0.508
Adjusted Rho Squared w.r.t Constants 0.301
Number of Cases 494
Number of iterations 20
Estimation status converged, with contants, with zeros, valid lic.
The results shown in the table above show that all the variables have expected sign of
coefficients and they are statistically significant at significance level greater than 0.05
except the travel time variable which is statistically significant at significance level
greater than 0.1.in order to compare this model with model_S3 the non-nested
hypothesis test was applied as the two models have the same number of variables
112
where model_S4 is the lower model and model_S3 is the higher model. The null
hypothesis to be tested stated that model_S4 is the true model. The significance level
for the test of model_S4 being true is - - - -
= [-1.58]=0.057 which is >0.05. This result can’t lead to reject the null hypothesis at
significance level greater than 0.05 and implies that Model_S4 with fare over income
variables is the true model. This result is consistent with the theory and practices
indicated that the value of money decreases as the income increases.
For further improvement of the statistics of the previous model, a distance (DIST)
variable was added to the utility function of bus mode. The reason for adding this
variable to bus mode is the hunch generated from the descriptive analysis of data that
this variable will be significant for bus mode as it is noted that the large percent of
travelers who choose the bus mode as their preferred mode have a long trips. The
results of estimation (model_S5) were reported in Table 4.54.
Table (4.54): Estimation results of model_S5
Parameters model_S5 (MNL)
Estimated value t-statistics
Generic Parameters
TT -0.3213 -2.5138
FREQ -0.0518 -2.127
FARE/PINC -372.5231 -2.003
Alternative Specific Parameters
CONSTANT Minibus 1.8179 2.7047
CONSTANT Bus 1.1165 0.7961
AGE Minibus 0.0493 2.8485
AGE Bus 0.069 2.2217
DIST Bus 0.6234 3.6936
FINC Minibus -0.0018 -7.6099
FINC Bus -0.0039 -6.6471
Model Statistics
Log Likelihood at Zero -542.7145
Log Likelihood at Constants -380.0125
Log Likelihood at Convergence -250.54
Rho Squared w.r.t. Zero 0.5384
Rho Squared w.r.t Constants 0.3407
Adjusted Rho Squared w.r.t. Zero 0.5199
Adjusted Rho Squared w.r.t Constants 0.318
Number of Cases 494
Number of iterations 18
Estimation status converged, with contants, with zeros, valid lic.
113
The results in the table above show that all the variables are statistically significant at
significance level greater than 0.05 which means that these variables have significant
effect on choice decision. In addition the TT, FARE/PINC, and FREQ have a negative
sign of estimators which is consistent with the expected sign. The distance variable
has a positive sign which matches with the expectation that the desire of using bus
mode increases as the trip length increases and also consistent with the results of
descriptive analysis of data. The goodness of fit measures for this model was
improved compared with model_S4 as the value of ,
, and
for this model
are 0.5384, 0.3407, 0.5199, and 0.318 respectively while these values are 0.5245,
0.321, 0.508, and 0.301 respectively for model_S4. In order to test the hypothesis that
the distance variable can be excluded from the model, this model was compared with
model_S4. The likelihood ratio test was applied between the two models where the
distance variable in model_S4 was constrained to zero. The statistical test has a chi
square value of - - - - which is greater than the critical
chi square value at 95% confidence level so the null hypothesis can be rejected at
significance level greater than 0.05 hence distance variable can’t be excluded from the
model.
By comparing the above estimated models, it is clear that model_S5 has the best
goodness of fit measures with a value of 0.5384, 0.3407, 0.5199, and 0.318 for ,
,
and
respectively. in addition all variables in this model have a correct sign of
coefficients and they are statistically significant at confidence level of 95%, hence this
model can be chosen as the most satisfactory model to describe the behavior of
travelers in choosing the mode of transport for work trips in Gaza city based on the
criteria which were mentioned in the methodology for comparing the models and
choosing the most satisfactory one. The utility functions for different modes can be
written as the following
114
4.10 Validation of stated preference model
To test the validation of the stated preference model, the same approach was used as
in revealed model. The test of reasonableness indicated that the model which was
chosen as the most satisfactory one (model_S5) is considered as a valid model
because all the variables included in this model have correct signs of coefficients.
The likelihood ratio test (LRTS) which is the second phase of validation process was
conducted on the chosen model (model_S5) using about 1/3rd
of the data sets. The
result of this test show that the calculated chi square value was
- - - - which is less than the critical chi square
value at 95% confidence level and ten degrees of freedom ( the number of variables in
the chosen model), therefore the null hypothesis that there is no difference between
the predicted and observed behavior can’t be rejected at this significance level.
The prediction capability for the model which is the last phase of the validation
process indicated that the calculated prediction ratio of the model is 0.80 which means
that the model is capable to predict about 80% of the choices of the trip makers’
correctly.
115
Chapter 5: Conclusions & Recommendations
5.1 Summary
The main objective of this research was to investigate the factors that affect the choice
behavior of the employed people and developing a mode choice model for work trips
in Gaza city. For achieving the above mentioned aim the work in this research was
divided into six phases. The first phase involving reviewing the literature in the field
of urban transportation planning and mode choice modeling. The subjects which were
discussed in this phase, included background about transportation planning process
and travel demand forecasting, historical development of mode choice models, types
of mode choice models and comparison among them, the estimation techniques used
in mode choice models, and sampling and data collection methods. In addition
previous case studies for mode choice models were presented in this phase. According
to the literature review the logit model was chosen to be used for calibrating the mode
choice for work trips in Gaza city because it is simple in terms of formulation and
estimation in addition to its good accuracy compared with the other types of the
models. The factors that affect the mode choice models and which will be included in
questionnaire were determined according to the literature review.
The second phase involving the design of initial survey questionnaire and conducting
of pilot study and analysis of results for pilot study. In the third phase of this study the
final questionnaire was designed by adding, removing or editing questions in the
initial questionnaire based on the results of the pilot study. This phase also includes
determination of sample size, distributing the questionnaire to the target group and
analysis of the results.
The fourth phase included calibration of mode choice model for work trips in Gaza
city using both revealed and stated preference data. The ELM software was used.
Nine models were calibrated for revealed data and five models for stated preference
data and the most satisfactory one was chosen based on the goodness of fit measures
and t-statistics of the attributes. The model which yielded significance coefficients
estimators for all variables and high goodness of fit values was selected as the best
one which was Model_8 for revealed data and Model_S5 for stated preference data.
116
The fifth phase involved validation of the selected model based on three different
levels. The first level in validation process is the test of reasonableness of the selected
model which is testing during the calibration process where all the models with wrong
signs of coefficients will not be considered as a valid model. The next level of
validation process is the statistical test of the null hypothesis that there is no
difference between observed and predicted behavior using likelihood ratio test. The
last level of validation process is the calculation of prediction ratio for the selected
model to test the prediction capability of the model. The results of validation test for
the selected revealed and stated preference models in this research show that they are
reasonable because the two models have correct signs of coefficients. The results also
indicated that there is no difference between the predicted and observed behavior
where the calculated chi square values for both revealed and stated preference models
are 16.25 and 9.226 respectively which are less than the critical chi square values at
95% confidence level and 13 and 10 degrees of freedom respectively. The prediction
ratio for revealed and stated preference models are 0.69, and 0.80 respectively.
The last phase of this research introduce summarize for the main finding and
conclusions and recommendations for the study.
5.2 Conclusions
Based on the findings of the research the following conclusions can be drawn
1. Based on the descriptive analysis of the data, the ownership of transport means,
trip length, age, monthly income and gender can be considered as the most
important factors that affect the mode choice as they have the strongest Cramar’s
V value while marital status, family size, and job seem to have low effects on
choice decision of travelers as they have low values of Cramer’s V statistics.
2. There are six factors that affect the captive ridership which are gender, job, private
car ownership, motorcycle ownership, bicycle ownership, and distance.
3. Among the variables that affect the captive ridership, distance and gender
variables seem to have the highest effect on captivity as they have the highest
Cramer’s V statistics.
4. For revealed model, the total travel time, total travel cost divided by personal
income, ownership of transport means, age, distance and average family monthly
income are the factors that affect the mode choice of employed people in Gaza
117
city. While the gender and out of vehicle time are statistically insignificant at 90%
confidence level so they are excluded from the model.
5. The utility functions for the selected revealed model for different modes have the
following formulas
6. For stated preference model, the travel time, ratio of fare over personal income,
frequency of service, age, average monthly family income, and distance have an
effect on mode choice decision of employed people as they are statistically
significant at 95% confidence level while the gender variable has no effect on
mode choice decision as it is statistically insignificant even at 90% confidence
level.
7. The utility functions for the selected stated preference model for different modes
have the following formulas
8. The developed revealed at stated preference models are able to predict the choice
behavior of employed people in Gaza city as the two models are valid at 95%
confidence level.
9. The prediction ratio for revealed model is 0.69 while the prediction ratio for stated
preference model is 0.80.
118
5.3 Recommendations
Several recommendations have emerged from this research
1. Using the developed revealed model in travel demand analysis and in developing
transport policies for Gaza city.
2. Using the developed stated preference model in studying the possibility and the
feasibility of introducing the bus services for transport system in Gaza city.
3. Using the developed stated preference model in establishing the time table and in
determining the appropriate fare for bus services in Gaza city.
4. Awareness campaigns should be implemented to encourage young people for
using a bicycle mode.
5. In case on introducing a bus service to transport system in Gaza city awareness
campaigns may be needed to encourage the young people for using bus modes.
6. Further study for developing mode choice models for trips other than work trips
such as social, recreational and study trips.
7. Studying the effect of captive travelers on mode choice models.
8. Calibrating the mode choice using probit and generalized extreme model and
comparing them with logit model.
119
References
1. Abdel-Aty, M. and Abdelwahab, H. (2001), calibration of nested mode choice
model for Florida, Final research report, University of central Florida.
2. Abrahamsson, T. (1996). Network Equilibrium Approaches to Urban
Transportation Markets - Combined Models and Efficient Matrix Estimation.
PhD Thesis, The Royal Institute of Technology, Stockholm, Sweden.
3. Abrahamsson, T. (1998). Estimation of Origin-Destination Matrices Using
Traffic Counts: An Application to Stockholm, Sweden, pp of Travel Behavior
Research: Updating the State of Play, edited by Ortuzar, Hensher and Jara-
Diaz. Elsevier Science Ltd., Oxford, UK.
4. Ackoff, R.L. (1965), Individual Preferences for Various Means of
Transportation, Management Science Center, Transport Study Center,
University of Pennsylvania, Pennsylvania, USA.
5. Adams, W.T. (1959), Factors Influencing Mass Transit and Automobile
Travel in Urban Areas, Public Transport, (30), pp 256-260.
6. Adjaka, K.A. (2009). Safe routes to school mode choice modeling and
application, master thesis, Washington D.C., Howard University.
7. Al-Ahmadi, H.M. (2006), Development of Intercity Mode Choice Models for
Saudi Arabia, JKAU: Eng.Sci, Vol. 17 No. 1 , pp. 3-12.
8. Almasri, E.H. (2011). Factors affecting travel choice of shared taxi versus bus
for Palestinian university student trips. International Review of Civil
Engineering (I.RE.C.E) no.1 , vol.(2)
9. Ampt, E. and Ortuzar, J. de D. (2004). On Best Practice in Continuous Large
Scale Mobility Surveys. Transport Reviews, 24 (3), pp 337-363.
10. Ben-Akiva, M.E. , and Lerman S.R. (1985), Discrete choice analysis: theory
and applications to travel demand, MIT press, Cambridge, Massachusetts,
USA.
11. Chang, H.J. and Wen, B.S. (1994). The Distribution of an Actual Sample Size
Increment in Stratified Random Sampling. Communications in Statistics
Theory and Methods, 23 (6), pp 1735-1742.
121
12. Davidson, D. (1973).Forecasting traffic on STOL .Operation Research
Quarterly Vol. 24,pp. 561–569.
13. Dow, J.K. and Endersby, J.W. (2004). Multinomial Probit and Multinomial
Logit: A Comparison of Choice Models for Voting Research. Electoral
Studies, 23
14. European commission 1996, Modelling of Urban Transport, Final report,
Brussels.
15. Ewing, R., Schroeer, W. & Greene, W. (2004). School location and student
travel: analysis of factors affecting mode choice. Journal of the Transportation
Research Board No. 1895, Washington D.C. pp. 55-63.
16. Ghareib, A.H. (1996). Estimation of Logit and Probit Models in a Mode
Choice Situation. Journal of Transportation Engineering, 122 (4), pp 282-290
17. Green, S. B. (1991). How many subjects does it take to do a regression
analysis? Multivariate Behavioral Research, 26, pp. 499-510.
18. Hanson, S. (1995), The Geography of Urban Transportation, 2nd
edition.
19. Hensher, D.A. and Button K.J. (2000), Handbook of Transport Modeling,
edited by Hensher and Button , Elsevier Science Ltd, Oxford, U.K. PP 1-10.
20. Hensher,D.A.,(1994).Stated preference analysis of travel choices :the state of
practice. Transportation Vol. 21,pp. 107–133.
21. Hossain, M.I., Hossain, M.Z., Ahmed, M.S. and Ali, M.A. (2003). A Class of
Predictive Estimators in Multi-Stage Sampling Using Auxiliary Information.
International Journal of Information and Management Sciences, 14 (1), pp 79
22. Khan,O. (2007), Modelling Passenger Mode Choice Behavior Using
Computer Aided Stated Preference Data, P.H.D. thesis, Queensland University
of technology.
23. Kish, L. (1995). Survey Sampling (65th edition). John Wiley and Sons Inc.,
New York.
24. Koppelman, F.S. and Bhat, C. (2006), A Self Instruction Course in Mode
Choice Modeling: Multinomial and Nested Logit Models ,U.S. Department of
Transport, Federal Transit Administration.
121
25. Kweon, B., Shin, W., Folzenlogen, R. & Kim, J. (2006). Children and
transportation: identifying environments that foster walking and biking to
school. Research report, Texas Transportation Institute.
26. Lisco, T.E. (1967), The Value of Commuters Travel Time: A study in Urban
Transportation, University of Chicago, Chicago, USA.
27. McDonald, N.C. (2008). Children’s mode choice for the school trip: the role
of distance and school location in walking to school. Transportation Vol. 35
No.2, 2008, pp. 23-35.
28. MCNally, M.G. (2000), Handbook of Transport Modelling, 3rd
edition, edited
by Hensher and Button, Elsevier Science Ltd., Oxford, U.K.
29. Nielsen, O.A. (1994). Two New Methods for Estimating Trip Matrices from
Traffic Counts. Paper presented at 7th International Conference on Travel
Behavior, Santiago, Chile.
30. Palestinian Central Bureau of Statistics. "Population, Housing and
Establishment Census 2007". Main Indicators by Locality Type. Ramallah –
Palestine, January, (2009)
31. Patriksson, M. (1994). The Traffic Assignment Problem: Models and
Methods. VSP, Utrecht, The Netherlands.
32. Qrtuzar J.de D. and Willumsen L.G. (2002), Modelling Transport, 3rd
edition,
John wiely and Sons Ltd, New York.
33. Quarmby, D.A. (1967), Choice of Travel Mode for Journy to Work: Some
Findings, Journal of Transport Economics and Policy, (1),pp 273-314.
34. Rhoulac, T.D. (2005). Bus or car: the classic choice in the context of school
transportation. Transportation Research Board 84th Annual Meeting CD-
ROM, Washington, D.C.
35. Richardson A.J. (2003), Creative Thinking about Transportation Planning,
paper presented at 82nd
annual meeting of transportation research board,
Washington DC, USA.
36. Richardson, A.J., Ampt, E.S. and Meyburg, A.H. (1995). Survey Methods for
Transport Planning. Eucalyptus Press, Melbourne, Australia.
122
37. Safwat, K.N.A. and Magnanti, T.L. (2003). A Combined Trip Generation,
Trip Distribution, Modal Split, and Trip Assignment Model in the
Automobile. Classics in Transport Analysis, (7), pp 336-352.
38. Sherali, H.D., Narayanan, A. and Sivanandan, R. (2003). Estimation of
Origin–Destination Trip-Tables Based on a Partial Set of Traffic Link
Volumes. Transportation Research Part B: Methodological, 37 (9), pp 769-
855.
39. Siddiqui, N.H. (1999), Nested Logit Models for Motorized and Non-
Motorized Modes, Master thesis, Karachi, N.E.D. University of Engineering
and Technology.
40. Stehman, S.V. (1997). Estimating Standard Errors of Accuracy Assessment
Statistics under Cluster Sampling. Remote Sensing of Environment, 60 (3), pp
258-269.
41. Stopher , P.R. (1969), Probability Model of Mode Choice for The Working
Journey, Highway Research Record, 283,pp 57-65.
42. Stopher, P.R. (2000). Survey and Sampling Strategies, pp 229-252 of
Handbook of Transport Modelling, edited by Hensher and Button. Elsevier
Science Ltd., Oxford, UK.
43. Swait,J.,Louviere,J.J.,Williams,M.,(1994).A sequential approach to exploiting
the combined approach of stated preference and revealed preference data:
applications to freight shipper choice. Transportation Vol. 21,pp. 135–152.
44. Wang,D.,Borgers,A.,Oppewal,H.,Timmermans,H.,(2000).A stated choice
approach to developing multi-faceted models of activity behavior.
Transportation Research Part A: Policy and Practice Vol. 34, pp. 625–643.
45. Warner, S.L. (1962), Stochastic Choice of Mode in Urban Travel: a study in a
binary choice, North Western University Press, Evanston, Illinoise.
46. Yarlagadda, A.K. & Srinivasan, S. (2008). Modeling children’s school travel
mode and parental escort decisions. Transportation, Vol. 35 No.2, pp. 201-
218.
123
ANNEX1: QUESTIONNAIRE IN ARABIC
غزة- اإلسالميةالجامعة كلية الهندسة
قسم الهندسة المدنية برنامج ماجستير البنية التحتية
Development of mode choice model for Gaza city
بناء نموذج الختيار وسائل النقل في مدينة غزة
العاملينيهدف هذا االستبيان إلى دراسة واقع المواصالت في مدينة غزة و العوامل التي تؤثر في اختيار نموذج رياضي وصواًل إلى بناء ....(مشي , دراجة , مواصالت عامة, سيارة خاصة) لوسائل النقل
.في مدينة غزة الختيار وسائل النقل لرحالت العمل
يرجى التكرم بمأل االستبيان بالحقائق المناسبة و الدقيقة قدر اإلمكان حيث أن هذه المعلومات سوف تستخدم لغرض البحث العلمي فقط و سوف يتم المحافظة على سريتها و إحاطة المشاركين بنتائج
. الدراسة فور االنتهاء من إعدادها
إعداد الباحث
الراعي إبراهيمسعدي
إشراف
عصام المصري. د
1122 فبراير
124
اديةالمعلومات االجتماعية و االقتص: األولالجزء
هذه األسئلة هي لغرض البحث العلمي فقط و سوف يتم المحافظة على سريتها و لن يتم ذكر اسم صاحبها في .البحث
أنثى [ ]ذكر [ ]: الجنس -2
.سنة................................... : العمر -1
متزوج [ ]أعزب [ ]: الحالة االجتماعية -3
: المهنة -4
موظف وكالة الغوث [ ] موظف قطاع خاص [ ] موظف حكومي [ ] ..............( ......حدد) أخرى [ ] عامل بأجر يومي [ ] رجل أعمال تاجر أو [ ]
................................. :(المسئول عنهم) التي تعيلها عدد أفراد األسرة -5
ال [ ]نعم [ ] : مركبة خاصة ؟ لديكهل يوجد -6
ديزل [ ]بنزين [ ]ما هو نوع محركها؟ , إذا كانت اإلجابة بنعم
ال [ ]نعم [ ] : ؟ دراجة نارية لديكهل يوجد -7 ال [ ]نعم [ ] : ؟ دراجة هوائية لديكهل يوجد -8
......................بجوار ...............رع الشا...................الحي :عنوان السكن كاماًل -9
......................بجوار ................الشارع ...................الحي :عنوان العمل كاماًل -21
:لعائلتك متوسط الدخل الشهري -22
شيكل 0511-0110من [ ] شيكل 0111اقل من [ ] شيكل 0111-5110من [ ] شيكل 5111-0510من [ ] شيكل 5110أكثر من [ ] شيكل 5111-0110من [ ] شيكل 0111-0110من [ ]
125
العوامل المؤثرة في اختيار وسيلة النقل: الثانيالجزء , مركبة خاصة ) التي يستخدمونها النقل لوسائل العاملينفيما يلي عدد من العوامل التي قد تؤثر في اختيار
الرجاء تحديد مدي أهمية كٍل من هذه العوامل من وجهة نظرك , ( الخ.... ,دراجة نارية , سيارة أجرة
العوامل المؤثرةمهم بدرجة
قليلة جدا
مهم بدرجة
قليليه
مهم بدرجة
متوسطة
مهم بدرجة
كبيرة
مهم بدرجة
كبيرة جدا
العمر
الجنس
الدخل الشهري
(األجرة) تكلفة الرحلة
زمن الرحلة
زمن االنتظار لوسيلة النقل
األحوال الجوية
الخصوصية في وسيلة النقل
الراحة داخل وسيلة النقل
امتالك سيارة خاصة
الحالة الصحية
العملالمسافة بين البيت و
خصائص الرحلة: الثالثالجزء
(واحدة فقط إجابةالرجاء اختيار ) للذهاب إلى العمل؟وسيلة المواصالت التي تستخدمها معظم األحيان .2 (طلب) تاكسي [ ] سيارة أجرة [ ] سيارة خاصة [ ]
األقدام مشيًا على [ ] دراجة هوائية [ ]دراجة نارية [ ]
؟( غير التي حددتها في السؤال السابق) للذهاب إلى العملهل تستخدم وسائل نقل أخرى أحيانًا .1 ال [ ]نعم [ ]
يمكنك اختيار أكثر من ) الرجاء تحديد هذه الوسائل ؟ , "نعم "عن السؤال السابق إذا كانت اإلجابة .3
(إجابة (طلب) تاكسي [ ] سيارة أجرة [ ] سيارة خاصة [ ] مشيًا على األقدام [ ] دراجة هوائية [ ]دراجة نارية [ ]
126
الرجاء القيام بتعبئة البيانات المطلوبة و المتعلقة بزمن و تكلفة الرحلة لجميع وسائل النقل التي .4 (التي قمت بتحديدها سابقًا في السؤال األول و الثالث ) تستخدمها للعودة من الجامعة
ممشيًا على األقدا - أ .ةدقيق.................. مكان العملإلى البيت الزمن المستغرق من
دراجة هوائية - ب .دقيقة.................. مكان العملإلى البيتالزمن المستغرق من
.شهر /شيكل .................. تكلفة الصيانة للدراجة
سيارة خاصة -ج .دقيقة.................. مكان العملإلى البيتالزمن المستغرق من
.كيلو متر................ المسافة بين البيت و مكان العمل شهر/لتر................ متوسط االستهالك الشهري من الوقود
كيلو متر............. المسافة التي تقطعها المركبة باستخدام ا لتر من الوقود .شهر /شيكل................. تكلفة الصيانة للمركبة .شهر /شيكل ..................تكلفة ترخيص المركبة .شهر /شيكل.................. تكلفة التامين للمركبة
سيارة أجرة -د .دقيقة .................. المكان الذي تستقل منه المركبةإلى البيتالزمن المستغرق من
.دقيقة.................. الزمن المستغرق في انتظار المركبة .دقيقة.................. المركبة الزمن المستغرق أثناء الرحلة داخل
دقيقة............... مكان العملالزمن المستغرق من لحظة نزولك من المركبة حتى وصولك إلى .شيكل ................. األجرة المدفوعة
(طلب) تاكسي -ه .دقيقة.................. الزمن المستغرق في انتظار التاكسي
.دقيقة.................. خل التاكسي الزمن المستغرق دا .شيكل ................. األجرة المدفوعة
دراجة نارية -و .دقيقة.................. مكان العمل إلى البيتالزمن المستغرق من
كيلو متر................ المسافة بين البيت و مكان العمل شهر/لتر.......... ......متوسط االستهالك الشهري من الوقود
كيلو متر............. باستخدام ا لتر من الوقود الدراجةالمسافة التي تقطعها شهر /شيكل................. تكلفة الصيانة للمركبة .شهر /شيكل ..................تكلفة ترخيص الدراجة شهر /شيكل.................. تكلفة التامين للدراجة
127
االختيارات االفتراضية : الرابعالجزء الرجاء ترتيب , افترض انه سوف يتم إدخال خدمة النقل بالباصات إلى وسائل النقل المستخدمة داخل مدينة غزة
حسب ( سيارة أجرة, باص صغير, باص كبير) وسائل النقل المفضلة لديك في حالة توفر الخيارات التالية فقط :المعطيات التالية
المستوى األول :اوالً
سيارة أجرة باص كبير باص صغير وسيلة النقل \العامل
عاااان % 30زيااااادة بنساااابة زمن الرحلة
زمااااان الرحلاااااة بواسااااا ة
سيارة األجرة
عن زمان % 41زيادة بنسبة
الرحلااااااة بواساااااا ة ساااااايارة
األجرة
-
عن أجرة % 25اقل بنسبة تكلفة الرحلة
السااااافر بواسااااا ة سااااايارة
أجرة
عاان أجاارة %51اقاال بنساابة
- السفر بواس ة سيارة أجرة
الزمن بين قدوم وسيلتي نقل
(التكرار) متتاليتين دقائق 5كل دقيقة 41كل دقيقة 21كل
............................ الخيار الثالث........................ الخيار الثاني....................... الخيار األول
الثاني المستوى: ثانياً
يارة أجرة س باص كبير باص صغير وسيلة النقل \العامل
عاااان % 21زيااااادة بنساااابة زمن الرحلة
زمااااان الرحلاااااة بواسااااا ة
سيارة األجرة
عااااان % 30زيااااادة بنسااااابة
زمن الرحلة بواسا ة سايارة
األجرة
-
عن أجرة % 15اقل بنسبة تكلفة الرحلة
السااااافر بواسااااا ة سااااايارة
أجرة
عاان أجاارة %41اقاال بنساابة
- السفر بواس ة سيارة أجرة
الزمن بين قدوم وسيلتي نقل
(التكرار) متتاليتين دقائق 5كل دقيقة 31كل دقيقة 15كل
......................... الخيار الثالث......................... الخيار الثاني........................ الخيار األول
الثالث المستوى: ثالثاً
سيارة أجرة باص كبير باص صغير وسيلة النقل \العامل
عاااان % 10زيااااادة بنساااابة زمن الرحلة
زمااااان الرحلاااااة بواسااااا ة
سيارة األجرة
عااااان % 20زيااااادة بنسااااابة
زمن الرحلة بواسا ة سايارة
األجرة
-
عن أجرة % 11اقل بنسبة تكلفة الرحلة
السااااافر بواسااااا ة سااااايارة
أجرة
عاان أجاارة %31اقاال بنساابة
- السفر بواس ة سيارة أجرة
الزمن بين قدوم وسيلتي نقل
(التكرار) متتاليتين دقائق 5كل دقيقة 15كل دقائق 8كل
........................... الخيار الثالث........................ الخيار الثاني........................ الخيار األول
128
ANNEX2: QUESTIONNAIRE IN ENGLISH
The Islamic University-Gaza Higher Education Deanship
Faculty of Engineering Civil Engineering Department
Development of mode choice model for Gaza city
This questionnaire aims to study the situation of transport system and the factors that
affect the employed people’s choice for transportation modes (Private car, shared taxi,
walking,….etc) in Gaza city reached to build a mathematical mode choice model for
work trips in Gaza city.
Please fill the attached questionnaire with accurate facts, knowing that this
information will be used for the purpose of scientific study only and will be treated
confidently.
Prepared by researcher
Eng. Sadi AL-raee
Supervised by
Dr. Essam Almasri
February 2011
129
Part I: socioeconomic information
The questions bellow are for statistical purpose only , they will be confidential and no
individual will be identified in the research
1. What is your gender?
[ ] Male [ ] Female
2. What is your age? ------------------- years.
3. What is your status?
[ ] Single [ ] Married
4. What is your occupation?
[ ] Governmental employee. [ ] Private sector employee
[ ] UN employee. [ ] business man.
[ ] others (specify) ----------------------------------.
5. What is your family size? ----------------------- Persons.
6. Do you have a private car?
[ ] yes [ ] No
If yes, what is your car type?
[ ] Diesel [ ] Gasoline
7. Do you have motorcycle?
[ ] yes [ ] No
8. Do you have bicycle?
[ ] yes [ ] No
9. What is your home a dress?
--------------------------------------------------------------------------------------
10. What is your work address?
-----------------------------------------------------------------------------------------------
11. What is average family monthly income?
[ ] Less than 1000 ILS. [ ] 1001-1500 ILS. [ ] 1501-2000 ILS.
[ ] 2001-3000 ILS. [ ] 3001-4000 ILS [ ] 4001-5000 ILS.
[ ] More than 5000 ILS.
131
Part II: factors affecting mode choice
To what extent do you think that the following Factors are important in choosing the
transport mode for work trips
Attributes Very low
important
Low
important
Medium
important
High
important
Very high
important
Age
Gender
Income
Cost of travel (fare)
Travel time
Waiting time
Weather conditions
Having some
privacy
Comfort
Availability of
private car
Health status
Part III: Trip Characteristics
1. What is the mode of transport you usually use to go to your work?
[ ] Private car. [ ] Shared taxi [ ] Private taxi
[ ] Motorcycle [ ] bicycle. [ ] walking
2. For your work trips, did you ever use modes other than the mode you usually use to
go to your work?
[ ] yes [ ] No
If yes, please specify the other modes (you can choose more than one choice)
[ ] Private car. [ ] Shared taxi [ ] Private taxi
[ ] Motorcycle [ ] bicycle. [ ] walking
131
3. Please give your estimation for travel time and cost for the modes you can use in
your trip? ( fill only for the modes you can use)
a) walking mode
Travel time ……………………..minutes.
b) bicycle mode
Travel time ……………………..minutes.
Maintenance cost ……………… ILS/month
c) private car mode
Travel time …………………...…....minutes.
Average monthly fuel consumption …………… liters
No of kilometers cut using 1 liter ……………... km
Maintenance cost ………………………..… ILS/month
License fees ……………………………….. ILS/month
Insurance fees ……………………………....ILS/month
d) Shared taxi mode
Time from home to station …………………...…....minutes.
Time waiting a taxi ……………………………..… minutes.
In-vehicle travel time ……………………………....minutes.
Time from station to work ………………………….minutes.
fare ……………………….… ILS
e) taxi mode
Time waiting a taxi ……………………………..… minutes.
In-vehicle travel time ……………………………....minutes.
Fare ……………………….… ILS
f) motorcycle mode
Travel time …………………...…....minutes.
Average monthly fuel consumption …………… liters
No of kilometers cut using 1 liter ……………... km
Maintenance cost ………………………..… ILS/month
License fees ……………………………….. ILS/month
Insurance fees ……………………………....ILS/month
132
Part V: stated preference survey
Assume that a new bus service will be introduce to transport system in Gaza city,
please rate your preferred modes according to the following conditions:
1. Level 1
Attribute/mode Mini bus Bus Shared Taxi
Journey time 30% more than
shared Taxi
40% more than
shared taxi -
Journey cost 25% less than
shared taxi
50% less than
shared taxi -
Service frequency Every 20 minutes Every 40 minutes Every 5 minutes
Please Rank your choice:
First choice …………… Second choice ……………… Third choice…………..
2. level 2
Attribute/mode Mini bus Bus Shared Taxi
Journey time 20% more than
shared Taxi
30% more than
shared taxi -
Journey cost 15% less than
shared taxi
40% less than
shared taxi -
Service frequency Every 15 minutes Every 30 minutes Every 5 minutes
Please Rank your choice:
First choice ……………Second choice ……………… Third choice ………………..
3. level 3
Attribute/mode Mini bus Bus Shared Taxi
Journey time 10% more than
shared Taxi
20% more than
shared taxi -
Journey cost 10% less than
shared taxi
30% less than
shared taxi -
Service frequency Every 8 minutes Every 15 minutes Every 5 minutes
Please Rank your choice:
First choice …………… Second choice ……………… Third choice …………...