34
CHAPTER 5
DATA PREPARATION & PRELIMINARY ANALYSIS
5.1 DATA USED FOR TRIP GENERATION MODELLING
The data used for modelling can be broadly classified as
I. Travel data
II. Socio-economic data
5.1.1 Travel Data
The household interviews done for the Colombo Urban Transport Study Stage 1
(CUTS-1) surveys in 1995 are used to estimate the trip rates per household for each DSD
by the trip purpose and trip mode. Thereafter, the total trips per DSD were estimated by
multiplying the trip rates by the total number of households of that DSD.
Surveys have been carried out in 22 DSD's and a total number of 3,517
households have been interviewed. A household has been defined as an entity where
residents cook and eat together. Thus boarders and lodgers have been considered as
separate individual households. Therefore it is possible to have records of more than one
household for a single dwelling unit. According to the survey report, the enumerators
have attempted to select the sample such that it is representative of the houses in the
zone. A list of survey locations and sample size is given in Table 5.1.
The type of data collected include number of households in dwelling unit, number
of persons in a household, vehicle ownership details, occupation and educational details,
type of trips made by different modes and the trip purpose. A specimen survey form (SF
83) is given in Appendix 1.
35
AGA Division Households Interviewed AGA Division Households
Interviewed
Ratmalana 92 Gampaha 72
Wellawatta 90 Homagama 142
Dehiwela 83 Kollupitiya 105
Kotte 89 Minuwangoda 89
Maradana 115 Bandaragama 164
Kaduwela 73 Fort 90
Nugegoda 128 Negombo 113
Kolonnawa 91 Kesbewa 183
Wattala 98 Kochchikade 119
Moratuwa 160 Mahara 136
Borella 88 Maharagama 138
Ja-Ela 109 Kelaniya 54
Panadura 157 Mattakkuliya 93
Bambalapitiya 75 Grand Pass 90
Katana 92 Biyagama 74
Horana 131 Kalutara 117
Town Hall 67 Total 3,517
Table 5.1: Locations of Household Interviews and Sample Size
Since this particular survey was carried out for a different purpose, not all data is
useful for trip generation modelling. Further, since the surveys haven't been done on all
the DSD's within CMR, it is not possible to calculate the trip generations for the DSD's
where the data is not available. But when the models are calibrated, it is possible to
estimate the trip generations for the DSD's for which the data is not available, using the
models.
36
5.1.1.1 Sampling and Survey Errors
There are a number of erroneous and confusing entries in the survey data files.
These may be attributed to incorrect data entry at the field and at data coding. Some of
these errors include unusually high trip rates per person, incorrect household size, and
partly completed or missing records. The survey itself has been conducted in an
inconsistent manner. In some locations, the trip data of people have been collected for 5
previous weekdays whereas in other locations, it is for 2 days and at some places in was
for a single day. The survey report does not elaborate on these. Further, there were no
details on sampling method. It was only through the observation of the total reported trip
rates that it was able to establish the number of days the surveys have been done. Table
5.2 gives the number of days for which travel data from the household occupants have
been gathered at different survey locations. Upon identification of erroneous data, every
attempt was made to rectify the errors and include it in the analysis as omission of data
reduces the sample size.
AGA Division
Number of Days AGA Division Number
of Days AGA
Division Number of Days
Ratmalana 5 Ja-Ela 2 Fort 2
Wellawatta 2 Panadura 2 Negombo 2
Dehiwela 5 Bambalapitiya 2 Kesbewa 2
Kotte 5 Katana 2 Kochchikade 2
Maradana 5 Horana 2 Mahara 2
Kaduwela 5 Town Hall 2 Maharagama 1
Nugegoda 2 Gampaha 2 Kelaniya 1
Kolonnawa 2 Homagama 2 Mattakkuliya 1
Wattala 2 Kollupitiya 2 Grand Pass 2
Moratuwa 2 Minuwangoda 2 Biyagama 2
Borella 2 Bandaragama 2 Kalutara 2
Table 5.2: Travel Information Collection Duration at Different AGA's
37
5.1.2 Socio-Economic Data
Socio-economic variables that explains the trip generation such as population,
number of households, licensed vehicles and employment data were gathered from
publications by Department of Census and Statistics and Transport Database of the
University of Moratuwa.
In addition, the vehicle ownership was estimated using the household interview
data. The operational vehicle fleet data as compiled in the Transport Database of
University of Moratuwa is different from the interview based vehicle fleet as the
operational fleet also includes the vehicles owned by government institutions and
companies. This enables to select the most appropriate vehicle fleet estimation using two
different values. The percentage distribution of population by employment type was also
estimated using the information gathered from the household interviews.
5.2 DATA PREPARATION
Before commencing the model calibration, the household data and socio
economic data need to be arranged such that data is in the required format for calibration.
The following sections will elaborate on the procedure adopted for the preparation of data
for model calibration.
5.2.1 Household Trip Rates
The steps in the calculation of household trip rates are given below.
I. Checking of the household survey data files for incorrect entries.
II. Using Microsoft Excel formula option, selecting entries that record bus,
rail and private vehicle trip information separately.
III. Getting totals of households in the sample and the one-way trips for
different modes.
38
IV. If the trip data is available for more than one day, estimating the total trips
for one day by dividing total trips by the number of days during which
these trips have been made.
V. Calculating the one-way trip rate per household by dividing the total trips
by the number of households in the sample.
VI. The above steps should be repeated for the all required modes of travel for
each survey location.
VII. Formulation trip generation rate table.
Table 5.3 gives a sample of the outputs in the above mentioned calculation steps.
Calculated trip rates for different trip purposes are tabulated in Appendix 2 while total
trip generations by the purpose of the trip are given in Appendix 3.
DSD
(1)
Location
(2)
# of sample Households (corrected)
(3)
# o f days of survey
(4)
# of HBWT by bus
(corrected) (5)
# of trips per day
(6) (5)/(4)
HBWT per HH
(3)/(6)
Colombo Borella 104 2 64 32 0.308
Ja Ela Ja Ela 222 2 108 54 0.243
Kaduwela Kaduwela 85 5 133 27 0.318
Kelaniya Kelaniya 74 1 70 70 0.946
Table 5.3: Sample of Trip Generation Rate Calculation from Household Interviews
5.2.2 Socio-Economic Variables
The following soio-economic variables were considered for model calibration. All
these are at DSD level.
I. Population
II. Households
III. Vehicle ownership - motor cycles and car/van
IV. Population by employment categories
39
As the above data is not collected annually, it is necessary to extrapolate the available
data to the required year. For this purpose, suitable growth rates need to be applied.
Statistical abstracts published by the Central Bank and recent reports such as Colombo
Metropolitan Regional Structure Plan (CMRSP) carry the growth rates for the socio
economic variables under different economic scenarios. But these growth rates are
available at district level only. As such it is not feasible to estimate the growth of soio-
economic indices at DSD. Therefore the growth rates of the parent district are applied to
the constituent DSD's. The annual growth rates used (CMRSP, 1998) for different socio
economic variables are given in Table 5.4.
Socio-Economic Variable
District Socio-Economic Variable Colombo Gampaha Kalutara
Population 1.28 2.24 2.40
Household 2.25 2.00 1.50
Table 5.4: Annual Growth Rates Applied for Socio-economic Variables
The following table shows the calculated values of socio-economic variables for
some of the DSD's for the base year 1998. The year on which primary data available is
also given in the Table 5.5.
40
DSD District Population 1994
Households 1994
Population 1998
Households 1998
Maharagama Colombo 120,466 26,618 126,753 29,096
Homagama Colombo 140,825 32,687 148,175 35,730
Gampaha Gampaha 143,688 31,502 157,002 34,099
Negombo Gampaha 145,813 31,695 159,323 34,308
Horana Kalutara 108,112 25,158 118,870 26,702
Panadura Kalutara 140,793 28,655 154,804 30,413
Table 5.5: Calculation of Socio-economic Variables for the Base Year = 1998
5.2.3 Categorization of Zonal Population by Employment Type
The employment categories identified for the population over 5 years of age are listed
below. Population below five years of age are not considered in this classification as their
contribution to trip generation is insignificant. In addition, they are not in a position to
generate trips on their own.
I. School/university education
II. Professionals
III. Clerical
IV. Business
V. Agricultural
VI. Self Employment
VII. Unemployed
VIII. Retired
The data collected in the household interviews was used for the above classification
process. When the demand models are calibrated, which forecast trip generation as DSD
level, it was assumed that the population distribution by employment in the sample
represents that of the DSD. If the sample chosen is biased, (i.e. it does not represent the
41
general employment distribution of the population of the particular DSD then the above
assumption may lead to errors in the models.
5.2.4 Rail Station Density
This variable can be used as a measure on the accessibility to the rail for people. It
is calculated by dividing the total number of rail stations in a DSD by the area of that
DSD. Another way of representing the effect of rail mode is to give the rail line length
per unit area. But this does not give a measure of accessibility to rail, as accessibility is
represented by the availability of stations rather than the availability of rail line.
Station density variable has certain shortcomings such as depending on the
location of the stations within the DSD, sometimes stations may serve neighbouring
DSD's as well. In addition, the level of service of the rail service is not represented by
this variable. Despite the above-mentioned shortcomings, this variable is retained for the
trip generation models, as they are only concerned about the factors that influence the trip
generation. If a trip distribution analysis is carried out then it is necessary to include
supply statistics as well. It should be mentioned that this variable is analogous with
vehicle ownership variable. The private vehicle ownership variable also does not identify
how often the owners use the vehicles for travelling. Therefore, both of these variables
are semi-static in their form as modal share variables. Table 5.6 gives the DSD's that have
rail stations and their rail station density.
42
DSD Area (km2) Rail Stations Station Density
(1) (2) (3) (3/2)
Colombo 40.5 12 0.30
Homagama 77.5 5 0.06
Maharagama 17.0 4 0.24
Nugegoda 20.0 1 0.05
Dehiwela 20.5 3 0.15
Moratuwa 20.0 5 0.25
Gampaha 96.0 5 0.05
Katana 108.0 6 0.06
Negombo 22.0 4 0.18
Wattala 63.0 1 0.02
Ja Ela 61.5 9 0.15
Kelaniya 20.0 3 0.15
Mahara 99.0 1 0.01
Kalutara 76.0 3 0.04
Panadura 46.0 3 0.07
Table 5.6: Rail Station Distribution Data for CMR
5.3 PRELIMINARY ANALYSIS
Before the commencement of calibration of models, a set of preliminary analysis
was carried out to identify the relationship between dependent and independent variables.
The types of analysis performed are scatter plots, bar charts and pie charts. Analysis of
this nature are extremely useful as they give an insight to the type of relationship that
exist between the variable such as linear or exponential and make it easier to identify the
influential variables in the models. This screening of variables would result in selection
of the most appropriate variables for the model calibration. The following sections will
explain the different types of analyses carried out.
43
•
5.3.1 Percentage Distribution of Trip Generations by Trip Purpose
Figure 5.1 gives the observed home based trip generation percentages for bus
mode by trip purpose. It can be seen that the most number of bus trips are made for work
and other trips while educational trips also constitute a significant proportion of bus
travel. But one of the most noticeable observations of the household sample interview
survey is the relatively insignificant non-home based travel by bus mode. The possible
reasons for the low non-home based trips made by bus are discussed under section 5.3.2.4
of this report.
Non-home Based Trips
3 %
V Home Based Home Based Other Trips
36%
Home Based Educational
Trips 2 5 %
Figure 5.1: Percentage Distribution of Bus Trips by Purpose of Travel
5.3.2 Analysis of the Variation of a Single Variable
The observed values of the dependent variable, i.e. trip rates, are analyzed at DSD
level. Any of the significant variations observed are discussed and the possible reasons
are also mentioned. Here, the attention is drawn towards the socio-economic
characteristics of the DSD's , availability of alternative modes of travel and possible
errors in the data collection procedure in explaining the variations of the trip rates.
44
Similarly, the vehicle ownership rates and rail station density variations are also
analyzed on DSD level. The vehicle ownership is further classified as motor cycle
ownership and car ownership, which consist of cars, vans and jeeps. This is in the view
that these two vehicle ownership categories clearly represent two strata of income groups.
5.3.2.1 Home Based Work Trips per Household
The observed trip generation rates for home based work trips are graphically
displayed by means of a bar chart given in Figure 5.2. The variations of the trips
generation rates for different DSD's can be noted and the possible reasons can be
identified in terms of socio-economic parameters. This also helps to identify the errors in
the interview data, as some of the extraordinary variations that cannot be explained in
terms of socio-economic variables may be attributed to errors in the interviews.
It can be seen that Kaduwela and Mahara have the highest home based work trip
generation rates. The most convincing explanation for this is the lack of a rail station in
either of the DSD. As a result, all people who are captive to public transport service have
to use buses. Maharagama, Negombo and Panadura have the lowest work trip generation
rates. All three DSD's have a lower rate than Colombo. This observation is questionable
as Colombo has the highest vehicle ownership and the best rail facilities. Therefore the
errors in the survey sample are suspected to have contributed to the low work trips by
bus. Colombo DSD has reported a home based work trip generation of 0.42 trips by bus
per household. This figure is acceptable since higher income and resulting private vehicle
ownership as well as availability of a distributed rail network catering for employees who
prefer to avoid congested road usage may have comparatively reduced the home based
work trips by bus.
In general the following observations are conspicuous. The income level and
availability of rail transport has a negative impact on the amount of home based work
trips by bus. When there is no rail option, majority of the employees is confined to the
bus transport for home based work trips. Another observation is the DSD's located far
45
from the administrative capitals of the districts have reported a higher bus trip generation
rate. The average home based work trip generation rate by bus per household for CMR is
about 0.5 trips.
5.3.2.2 Home Based Educational Trips per Household
Home based educational trip generation rates are shown graphically in Figure 5.3.
The following observations are made regarding the household trip generation rates for
different DSD's.
There is a significant variation in home based educational trip generation rate
between DSD (e.g. Moratuwa 0.17 trips per hh, Kolonnawa 0.64 trips per hh). One of the
common observations for DSD's with a trip generation rate of 0.5 per hh or above is that
these DSD's do not have popular schools. As a result, many students have to travel to
main town centres where better schools are located. Therefore a comparatively higher
educational trip rate can be seen for these DSD's. Ja-Ela, Katana and Kelaniya DSD's
show a low home based educational trip generation rate despite not having popular
schools in them. One of the possible explanations is that there is a significant travel by
rail for school trips. The observed rail trip rates are given in Appendix 2. Similar to work
trips, there is comparatively lesser number of trips from DSD's Moratuwa, Panadura,
Maharagama and Nugegoda. The presence of popular schools as well as availability of
rail service and reasonably higher private vehicle ownership may have contributed
towards lowering the home based educational trips by bus.
A* V> V v° >* *̂ <** i>*
/////// s < Figure 5.2: Home Based Work Trip Generation Rates per Household (Bus
0.60
0.50
X 0.40 X i. v a
£ 0.30
0.20
0.10
0.00
* P J> J>* J>* . 1 * ! ^ V* J * J * V> ^ V > V « » <S? -J 1 ' 0> ^ < ^ V* <P
y / s y y / y y *• X * ^ p. ^ j , ^ j> ^> » ^ ^ j
.°" &> i<?
4r
Figure 5.3: Home Based Educational Trip Generation Rates per Household (Bus)
48
5.3.2.3 Home Based Other Trips per Household
The household trip generation rates for home based other trips by bus are given in
Figure 5.4. One of the noticeable observations is the very high trip generation rate of
Kolonnawa DSD. It is clear that some error in the household data has caused this result.
Therefore, this particular data is not included in the model calibration.
Previous research on home based other trips such as leisure trips reveal that when
the household income increases, these trips also increase (Kanafani, 1983). This
phenomenon is evident to a certain degree in the CMR data as well. For example,
Colombo DSD has a higher "other" trip rate than "work" or "educational" trips.
Homagama, Maharagama, Katana and Bandaragama DSD's have the lowest
"other" trip generation rates. Of these, Bandaragama has predominantly a rural setting
and the average income levels are low. As such the actual "other" trip generations are
less. On the other hand, Homagama, Maharagama and Katana, which may be, treated as
sub-urban areas. A significant share of the home based "other" trips in these DSD's may
be made using rail or private transport modes.
Figure 5.4: Home Based Other Trip Generation Rates per Household (Bus)
50
5.3.2.4 Non-Home Based Trips per Household
According to the household interview, the proportion of the non-home based trips
is very small, accounting only for 3% of the total trips as shown in Figure 5.1 of section
5.3.1. Similar to the home based other trips, non-home based trip generation is also
largely dependent on the income level of the households (Kanafani,1983). Generally
when the private vehicle ownership increases the non-home based trips also increase.
Conversely, most of the "average" people who are captive to public transport modes, are
engaged in professions that do not require additional travelling other than between place
of work and residence. As the variations of the non-home based trip generation rates are
very negligible, it is not possible to calibrate a model using regression method. Therefore,
a non-home based trip generation model for bus travel is not calibrated under this study.
5.3.2.5 Vehicle Ownership per Household
The vehicle ownership data is obtained from two different sources. One is the
Transport Database of the University of Moratuwa, which has been developed using the
annual vehicle licensing data. Other is the vehicle ownership data collected in the
household interviews of the CUTS - 1. The observed vehicle ownership rates from two
sources are given in Appendix 4. There is one noticeable difference between the two
databases. It is the inclusion of cars and motor cycles belonging to government and
private organizations in the Transport Database. But the CUTS - 1 based database solely
gives vehicles owned by households. Even though it is the actual vehicles owned by the
households that should be considered when modelling home based trip generation,
because of the apparent deficiencies in the interview sampling method, it was decided to
take the average value of the two databases for calibrations. However, when it comes to
the application of the models for other geographic regions, obtaining of average vehicle
ownership rates discussed above may become difficult. In such situations, it is suggested
to use the operational vehicle ownership rates suitably adjusted by a sample survey to
account for the institutions - owned vehicles.
51
The vehicle ownership is used as a proxy variable for income in transport demand
models. One of the major reasons for this is the difficulty in obtaining reliable data on
income at household level. Generally, as the income increases, the demand for travel also
increases (Kanafani, 1983). But as the income levels further increases, it is seen that
people shift from public and non-motorized forms of transport to motorized private
modes of transport. Therefore, vehicle ownership variable in a demand model explains
two scenarios. One is the increase in mobility with income and second is the change in
mode of transport. As a result, the coefficient of this variable in a demand model
ultimately shows the net effect of the two scenarios.
Figure 5.5 gives the private vehicle ownership distribution in CMR identified
separately for motor cycles and cars. As expected, Colombo DSD has the highest car
ownership rate. Nugegoda and Dehiwela DSD's also have more cars than motor cycles.
The car ownership level of Moratuwa DSD at 0.06 per household is very low. It can be
seen that car ownership level is a reasonable indicator of the average income level of the
DSD's. When the average car ownership is considered on district basis, Colombo has the
highest ownership, followed by Gampaha and Kalutara respectively.
As the car ownership increases, the proportion of the bus trips decreases. But car
ownership is not a very good variable when the trips are forecast at DSD level where the
income distribution is very large among the households. The actual proportion of the
population who owns cars (i.e. the higher income households') is a smaller percentage.
Therefore, taking average value of the car ownership for entire population does not
represent the accurate trip generation and modal share pattern for bus models.
The average motor cycle ownership is highest in the Gampaha District. This
suggests that motor cycle ownership represent the proportion of the medium income
population in the DSD. Due to relatively lesser capital and operating costs, most of the
' This study does not look into income classes in detail. If the average monthly income of the head of the household is more than Rs.20 000, it is considered as high income. This can be judged using the employment data given in interview survey.
52
middle income families can afford to have a motor cycle. As a result, motor cycle is a
definite competitor with public transport modes for home based "work" and "other" trips.
Therefore, the motor cycle ownership level here also explains any reduction in bus trip
rates.
At the same time it can be argued that majority of the people who use public
transport belongs to low and medium income groups. Since the need for mobility
increases as the income level changes from low to medium, as shown by the category
analysis on trip generation rates (Kanafani, 1983), high motor cycle ownership would
mean high trip generation rate for bus trips. This is further discussed in Section 6.4.1.2.
Even though motor cycle and car ownership explicitly describe the trip
distribution patterns of different income groups adequately, in the regression method of
model calibration where the entire population is considered, the attempts to include the
two as separate variables were failed. There is a high degree of correlation between these
variables as well. As such, an aggregate private vehicle ownership variable, formed by
adding the two variables, is used in the model calibration.
2 Medium income range is taken as Rs.5 000 - 20 000 for the head of the household.
v \ ° j& j& i ' \& jfr ~& i& ^ Vfr .tf > > ^ .£> .«> JST jr jr ^ J> j? J> J ^ J ^
Figure 5.5: Private Vehicle Ownership per Household (Average of CUTS-1 & License Data)
l i m b Hear
54
5.3.2.6 Rail Station Density Distribution
The station density is calculated as shown in Table 5.6. Figure 5.6 graphically
shows these values. It can be seen that except for Colombo DSD, other DSD's are not
adequately served by the rail stations. Even though the service frequency of rail is not
considered here, the accessibility to rail is represented by the station density. Since every
DSD does not have a rail station, in the model calibration, this variable enters the model
in exponential form. This is due to the fact that when there are no rail stations, the model
prediction should not be equal to zero. Typical format of such a regression model is given
in section 6.2.
5.3.3 Analysis of Variation of Two variables
In this section, the effect of variation of vehicle ownership and rail density on the
home based work trips taken as a percentage of total work trips is analyzed. Figure 5.7
gives the variation of percentage of home based work trips by bus with private vehicle
ownership. Definite pattern of variation cannot be identified here. This is mainly due to
the fact that work trip percentage is dependent not only on vehicle ownership but also on
the availability rail transport. In general, there is a decreasing trend in the percentage of
bus trips with the increase in vehicle ownership. This means as the vehicle ownership
increases, there is a shift of mode preference from public (bus) transport to private
transport. However, the observed bus modal share of Kalutara, Maharagama, Negombo,
Minuwangoda and Ja-Ela appears to be low.
Figure 5.8 shows the relationship between rail station density and percentage of
hbwt. It can be seen that when there are no rail stations, the bus modal share is very high.
The low home based work trip generation percentage may be attributed to sampling error.
As the rail station density increases, the bus modal share gradually decreases. Again,
Maharagama and Negombo data points seem to under predict the bus trips.
0.35
0.30
Figure 5.6: Observed Rail Station Densities in CMR
95.0
90.0 4> Biyagama
56
85.0
80.0
£. 75.0 H 2
es 5? 70.0
Bandaragama 4> Kaduwela
. • Kelaniya Kolonnawa Mahara
e Horana 4> • Kesbewa f
• Panadura # Homagama Katana
Nugegoda
65.0
60.0
55.0
4> Moratuwa
• Kalutara • Ja-Ela
Wattala
Dehiwela
Gampaha •
«> Minuwangoda
* Negombo Maharagama
Colombo
50.0 0.15 0.20 0.25 0.30 0.35
PV per HH
0.40 0.45 0.50 0.55
Figure 5.7: Modal Share of HBWT by Bus Vs Average Private Vehicle Ownership per HH
- 9 5 * -
90.0
Kesbewa
4> Biyagama
85.0
4> Kaduwela
80 0 + Nandaragama
Mahara Katana
75-0 • Horana
Kolonnawa u „ ™ o „ o m , ... + Panadura Homagama •
70.0 Nugegoda
; Gampaha 65.0 -) * # +
; Wattala Minuwangoda • Kalutara
60.0
55.0
Kelaniya
• Ja - Ela
• Dehiwela Negombo
Moratuwa
4> Colombo
Maharagama
o.i -0.05 0.05 0.1 0.15
Stations/km2
0.2 0.25 0.3 0.35
Figure 5.8: Percentage of H B W T by Bus Vs Rail Station Density