SIMULTANEOUS DEMAND MODEL FOR PASSENGER TRAVEL - A case study

SIMULTANEOUS DEMAND MODEL FOR PASSENGER TRAVEL - A case study of Indonesia

Dail Umamil Asri Chief of Road Infrastructure Section Directorate of Transportation National Development Planning Agency Republic of Indonesia Jl. Taman Surapati No. 2 Jakarta Pusat 10310 INDONESIA Fax: +62-21-314-8550 E-mail: [email protected]

Yoriyasu SUGIE Professor of Transportation Studies Faculty of Engineering Hiroshima University 1-4-1 Kagamiyama Higashi-Hiroshima 739-8527 JAPAN Fax: +81-824-24-7826 E-mail: [email protected]

Abstract: Considering the condition of Indonesia’s transportation systems and the socio-economic fluctuations, it is necessary to incorporate regional socio-economic parameters into transport policy formulation which can take into account various uncertainties in the future. Therefore, this research proposes applicable models which are relatively flexible and simple to use to estimate inter provincial passenger travel demand in Indonesia and to overcome the variation of available modes between province pairs, and inter-mode competition. The resulted models is utilizing simultaneous demand models approach, in which socio economic variables, regional conditions, and existing transport systems are simultaneously included and the calibration process is only conducted in a single step. Key Words: Simultaneous demand model, Single step, Transport policy, Economic fluctuation, and Inter-mode competition. 1. INTRODUCTION AND RESEARCH OBJECTIVES

Indonesia is a vast archipelago country covering thousands of islands and over huge distances. Geographic, topographic, and socio-economic condition strongly influenced regional transport characteristics. As a developing country, Indonesia is characterized by rapid growth of population, wide socio-economic variations, and transport policy changes. Due to the high degree relationship, it is necessary to incorporate regional socio-economic parameters into transport policy formulation which can take into account various uncertainties in the future. So far, there is still shortcoming in policy analysis exercises which take into account the effects of socio economic fluctuations and transport policy changes on inter regional travel demand pattern and inter mode competition such as the regional economic condition, variation of available modes between province pairs, improvement in transportation attributes and introducing of a new mode.

The standard approach of transport demand modeling is well known as the Sequential Model consisting of trip generation, trip distribution, modal split and trip assignment. The models are analyzed sequentially, where the output of the first sub-model (trip generation) is used as input to the second sub-model (trip distribution) then the output of the second sub-model (trip distribution) is used as input to the third sub-model (modal split). But, the simultaneous demand models handle the stages and integrate them into a single model. By this approach, the calibration process is merely conducted in a single step. Furthermore, socio economic variables, regional conditions, and existing transport systems are simultaneously included in the model.

869

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.4, October, 2003

mailto:[email protected]

The latest Indonesia National Transport Origin and Destination (OD) Survey was conducted in 1996. Regardless of inherent weaknesses and limitations, the survey results should be properly utilized and explored as still being the best available data compiling inter regional transport demand both for freight and passengers to analyze inter regional travel demand and patterns in Indonesia. The objectives of this research are firstly, to identify the determining socio economic factors of passenger travel demand for inter-provincial travel in Indonesia. Secondly, to encourage an appropriate model usage that is relatively flexible and simple in analysis to estimate inter regional passenger travel demand in Indonesia. 2 . SIMULTANEOUS DEMAND MODEL Major innovations in intercity transport modeling were made in the 1960s in association with the Northeast Corridor Transportation Project of the United States. Most models developed for that project as well as many other intercity models are simultaneous rather than sequential. With respect to the variables included, the models are regarded as econometric models. It should be noted that the simultaneous models might be regarded as an extension of the gravity model. Within simultaneous models both total travel and modal split are responsive to policy changes and this approach intuitively better represents the behavior of travel decision-making.

Normally, only three out of the four stages explicitly appear in these models. They compute the number of trips between each pair of regions by each mode (i.e. trip generation, distribution, and modal split), but do not handle the route choice (assignment). This implicitly assumes that for every pair of regions or cities only one route is available for each mode. This is generally a realistic assumption since unlike dense street networks in urban areas the intercity traffic especially for medium or long distances rarely has alternative routes available. So, for this reason the models are called “simultaneous demand models” or some time “direct demand models”. The simultaneous approach is widely applied for inter urban or inter city travel modeling and rarely applied for urban travel modeling. Rare examples for the urban travel modeling usage are those suggested by Kraft and Wohl (1967), Wong (1969), and Talvitie (1973). So far, three approaches have evolved for the analysis of intercity transport demand: the multi-modal approach, the abstract mode model, and the mode-specific model. 2.1 Multi modal Approach

The underlying concept of the multi modal approach as proposed by Kraft-SARC model (Ortuzar & Willumsen, 1995) recognizes that the demands for travel by different modes are related and should be analyzed simultaneously. The multi-modal models of intercity travel postulate that the demand for travel between a pair of cities by a particular mode is a function of the socioeconomic characteristics of the cities and of the supply attributes of that mode, plus those of all other modes available. As such, the multi-modal model is a combined “destination” and “mode choice” model with an implicit assumption that these choices are made simultaneously. These types of models often suffer from co-linearity problems. Other shortcomings are in the use for forecasting. Impacts of

870


introducing a new mode, for example, cannot possibly be predicted since this new mode will most likely have quite different characteristics.

2.2 The Abstract Mode Model

According to Quandt & Baumol (1966), the “abstract mode theory presupposes that individuals are characterized by a modal neutrality …” implying that “a modally neutral person chooses among modes purely on the basis of their characteristic and not on the basis of what they are called”. This approach is quite appealing especially dealing with the introducing of a new mode. Since any mode is expressed by its services rather than its name, estimates of parameter from existing modes can be used to forecast the impacts of a new mode.

The second hypothesis is that the model to be formulated as a compound model, covering the total demand for travel between a pair of points by each available mode or in other words, travel demand by mode from i-j pair, can be determined by considering the direct- and cross-elasticity of demand (Stopher and Meyburg, 1975). Various formulations of the abstract mode model have been calibrated and tested by Monsod (1966,1967), Quandt & Young (1969), Young (1969), Crow & Longeot (1972), and Sjafruddin (1997). 2.3 Mode-specific Approach

In contrast to the abstract mode model, the specific mode model considers “names” of the mode important besides their attributes. Therefore, by this approach estimated parameters are specific for each mode. Mclynn & Waronka (1969) developed such kinds of models. For the Indonesian case, Sjafruddin (1997) has applied the model formulated by Mclynn for intercity passenger travel demand in Java-Sumatra corridor.

Furthermore, Koppelman et al (1984) can be reviewed for a more complete historical perspective of significant intercity modeling efforts. Safwat (1987) and Daly (1999) developed disaggregate direct demand model for intercity passenger in Egypt and Denmark respectively. Meanwhile, the latest interesting development is a “direct demand model” utilizing co-integrated analysis in Spain (Milan et al, 1999). 3. DEVELOPED SIMULTANEOUS DEMAND MODEL FRAMEWORK 3.1 General Specification

The general specification of the simultaneous demand model approach is expressed as:

Tijm = K f(.) g(.) h(.) (1)

where : Tijm represents number of trips between i and j by mode m K represents an overall constant f(.) represents socio economic term g(.) represents the general impedance term h(.) represents the inter modal competition term.

871


The specification of the model is based upon the following arguments:

1. The total trip between a zone pair is a function of socio-economic characteristic and the transport attributes services available between them.

2. General impedance variables are indexes of the general difficulty or disutility of traveling on the considered route. One of these indexes is set for each travel characteristic. The presence of the general impedance variables in determining the demand for one mode reflects the fact that the model is formulated in terms of attributes of all the modes. The total demand for all modes increases when the characteristics of one mode are improved.

3. Inter modal competition variables reflect the variation of demand for one mode as a function of the performance of the mode with regard to some characteristics, compared to the performance of all other modes in service on the route. Then, in trip decision-making, a traveler chooses the mode of transport based upon the comparison of certain characteristics of the available modes.

Reviewing some previous models, the Crow and Longeot model approach (1972) is interesting in its use to predict the impacts of introducing a new mode or services improvement. From a practical point of view this model approach has some advantages because the model parameters estimated become less in number since these parameters could be commonly adopted for available modes. So, based upon the mathematical function of equation (1) and using approach applied by Crow and Longeot, the developed models are expressed as follows. 3.2 Abstract Mode Model Model-1

321

321210 )()()()()(γγγ

βββααα

=

ijg

ijm

ijg

ijm

ijg

ijmijgijgijgijjiijm F

FHH

CC

FHCIPPeT (2)

where:

Pi and Pj = population in zone i and zone j in thousand population, respectively Ii and Ij = income in zone i and zone j in thousand Rupiahs, respectively

ji

jjiiij PP

IPIPI

+= = weighted mean of per capita income in zone i and j

Cijm = travel cost between zone i and zone j by mode m in thousand Rupiah Hijm = travel time between zone i and zone j by mode m in hours Fijm = frequency of the services between i and j by mode m per two-weeks αo, α1, α2, β1, β2, β3, γ1, γ2, and γ3 = model parameters.

)(),(),( ijgijgijg FHC are three “general impedance” variables; (Cijg) is an index of travel cost, (Hijg) is an index of travel time, and (Fijg) is an index of frequency of the services between zone i and j. Exactly, (Cijg) and (Hijg) are weighted mean of transport attributes by frequency of the services. Meanwhile, (Fijg) its self is mean of the frequency of services specified as following:

872


∑

∑

=

== M

kijk

M

kijkijk

ijgC

F

CF

1

1 ,

∑

∑

=

== M

kijk

M

kijkijk

ijgH

F

HF

1

1, and

M

FM

kijk

ijgF∑== 1 (3)

Where, M = the number of serving modes between i and j Model-2

321

321210(*)(*)(*))()()()()(

γγγ

βββααα

=

ijg

ijm

ijg

ijm

ijg

ijmijgijgijgijjiijm F

FHH

CC

FHCIPPeT (4)

(*) means that all specifications are similar with those in Model-1, except denominators of the competition terms are modified including only competing modes as following.

∑

∑

≠=

≠== M

mkkijk

M

mkkijkijk

C

F

CF

ijg

,1

,1(*)

, ∑

∑

≠=

≠== M

mkkijk

M

mkkijkijk

H

F

HF

ijg

,1

,1(*)

, and )1(,1(*)

−=∑

≠=

M

FF

M

mkkijk

ijg (5)

where, m = the serving mode between i and j 3.3 Specific Mode Model Although the travel time, travel cost, and service frequency will often influence a traveler’s choice of mode, many other factors such as travel privacy, freedom, convenience, and safety must be considered as well. Due to the existence of those non-measurable variables and different nature for each mode, it is reasonable to assume that a mode specific-constant term of such multimode is necessity and should vary with the mode in question when the value of all other variables remain the same. In addition, the dummy income variable to expresses an hypothesis that the income elasticity for that travel mode will significantly different from income elasticity for travel by other modes. So, we get Model-3 and Model-4 in which all specifications are similar with those in Model-1 and Model-2, except incorporating dummy variables on the specific constant and income variables. Model-3

321

32122100 )()()()()(

γγγ

βββααααα

= ++

ijg

ijm

ijg

ijm

ijg

ijmijgijgijg

Aijji

Aijm

F

F

H

H

C

CFHCIPPeT mm (6)

Model-4

321

32122100***)()()()()(

γγγβββααααα

= ++

ijg

ijm

ijg

ijm

ijg

ijmijgijgijg

Aijji

Aijm

F

F

H

H

C

CFHCIPPeT mm (7)

873


where: A = Dummy variable , where A=1 in the case of mode m, or A = 0 otherwise. αo, αom, α1, α2, α2m , β1, β2, β3, , γ1, γ2, and γ3 = model parameters. Actually, all proposed models resemble Crow-Longeot model. However, some modifications have been made as described below:

• In the term of general impedance, we employed the weighted mean of travel time, travel cost, and service frequency, instead of the weighted geometric mean in the Crow and Longeot Model.

• Meanwhile, in Model-2 and Model-4, in the terms of competitiveness, the denominator is the weighted mean of travel times and travel cost by services frequency excluding the mode in question (only including competing modes).

4. DATA STRUCTURE The 1996 National O-D survey data as the best available data was adopted as the basis and starting point for the passenger travel demand analysis. The data from the 1996 Nastonal O-D survey comprised of inter provincial and inter district trip matrices for person trip, vehicle trip, and freight trip. The patterns of demand were based on the most significant long-distance mode of travel used for a journey. For example, if a traveler used walk-bus-train-taxi for a journey, then the mode selected in the analysis of demand would have been train, since it would have been considered to be the most significant strategic mode used. However, the O-D survey data also has several shortcomings. One of those was that O-D journey purpose by mode was not obtained. The available information is journey purpose by mode generated from an origin zone (regardless specific destination zone) and journey purpose by mode attracted into a zone (regardless specific destination zone). Nevertheless, it is possible to undertake any analysis of modal shift by total purpose due to improvements in transport modes, changes in tariff levels, travel time, frequency, etc. Referring to these limitations, the paper will focus on trip by all purposes and assumes that number of trips of the O-D pairs is summary of generation and attraction. Alternative sources of information would need to be reviewed to complete travel time and travel cost data of each mode through other previous studies, several transport companies, the Central Bureau for Statistic, the Ministry of Communications, and the Ministry of Settlements and Regional Infrastructure. 4.1 Data Grouping

A nationwide coverage, which consists of 27 provinces, has been chosen for the study with zoning based on availability and reliability of data. Zoning for the study areas follows administrative boundaries. A province is essentially defined as one zone. The main reasons are that available statistical data are given by administrative boundaries and the inter-provincial passenger movement is also compiled by province. The data collected includes passenger trips, travel cost, and service frequency of each mode. It is difficult to find similar numbers of modes available in all province pairs. In

874


several cases, a province pair is only connected by 2 or 3 modes. Thus, competition among the serving mode is identified by grouping. Then, the travel could be grouped as follows: Sea vs. Air; Car vs. Bus vs. Railway mode; Car vs. Bus vs. Air vs. Sea mode; and Car vs. Bus vs. Sea vs. Air vs. Rail mode. With regarding to the above classification, then 870 samples of data are obtained and identified as input data for the proposed Models. In practice, few locations where strategic modal choice in the Indonesian transport system is available, with the possible exception of the North Java corridor, where road, rail, sea and air transport all provide inter-urban/provincial services. Even here, there may be modal alternatives, the choice is seldom real. Each transport mode provides services to meet demands from different segments of the market as shown in Figure 1.a and Figure 1.b.

Annual Trip vs. Distance

1.E+00

1.E+01

1.E+02

1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

1.E+08

1.E+09

0 1000 2000 3000 4000 5000Distance (in kms)

Ann

ual T

rip

Air Mode Sea Mode

Annual Trip vs. Distance

1.E+00

1.E+01

1.E+02

1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

1.E+08

1.E+09

0 1000 2000 3000 4000 5000Distance (in kms)

Ann

ual T

rip

Railway Bus Non-Bus

Car

Car

Figure 1.a. Trip Length Distribution of Car, Figure 1.b. Trip Length Distribution of Air Bus, and Railway Mode and Sea Mode 4.2 Socio-economic Variables

Based on preliminary correlation analysis on factors generating passenger travel demand, the socio-economic characteristics selected in the models are population and income per capita (in this case is represented by Gross Regional Domestic Product per capita excluding oil and gas) of each province. Population is the most obvious factor influencing the number of trips between province i and province j. This formulation implies that the total trip is a function of the number of potential pairs of individuals between the two populations. The income variable is designated to represent the purchasing ability of the population. This variable appears in the formulation as a function of the weighted mean Gross Regional Domestic Product (GRDP) per capita excluding oil & gas (population being the weights), Iij. These socio-economic data were collected from the standard regional data provided on a provincial basis collected from related institutions.

875


4.3 Network Models

For the purpose of the travel demand development, network models are based upon the 1996 existing transport system in Indonesia obtained from several document produced by related institutions such the Ministry of Communication and the Ministry of Settlements and Regional Infrastructure. In a simultaneous formulation of inter provincial travel demand models it is implicitly assumed that there is only one route available between any province pairs for each mode. Therefore, if between particular provinces pair in the network more than one route is available, the route with the shortest travel time is chosen in accordance with the “all-or-nothing” concept. The three network models are:

• Road Network (bus and car), the road network model is established based on the existing main artery or artery roads.

• Railway network is only that in Java and one location in Southern Sumatera. The reason is recently no inter-provincial rail traffic in North Sumatera and West Sumatera.

• Air transport network is based upon scheduled airlines that were operational in Indonesia in 1996.

• Sea transport network is based upon scheduled passengers shipping that were operational in 1996.

5. ESTIMATION OF THE MODEL USING NATIONAL ORIGIN AND

DESTINATION SURVEY IN INDONESIA.

All the models are calibrated against all the available trips data. Basically, all the equations are non-linear in the parameters. But, those can be transformed into log-linear forms. The estimated parameters are obtained by taking logarithms on both sides and by using the ordinary least square regression (OLS). The OLS is employed to estimate the parameters of models based on the Gauss-Newton iterative algorithm using Times Series Processor (TSP) Software version 4.3. The result of estimation is shown in Table 1 & 2. Table 1 and 2 summaries the statistical output of Proposed Models. All models show that:

Table 1. Estimation Result of Model-1 and Model-2

Model-1 Model-2 No. Variables Expected Sign

Coefficient tvalue Coefficient tvalue Ln (Constant) -2.084 -1.67 -4.068 -3.25 Ln (Population) + 0.715 18.00 0.717 17.92 Ln (Income) + 1.056 8.44 1.177 9.35 Ln (Mean cost) - -1.456 -25.12 -1.326 -9.35 Ln (Mean time) - Ln (Mean frequency) + Ln (Relative cost) - -0.862 -16.32 Ln (Relative time) - -1.054 -15.25 Ln (Relative frequency) + 0.338 15.61 0.246 16.97

% RMSE of Log Form 15.5 % 15.6 % Multiple R 0.814 0.810 F 338.72 330.56 R2 adjusted 0.660 0.655

Number of samples 870 870 Note : RMSE is Root Mean Square Error

876


• Population and income per capita contributes significant positive explanatory power to the travel demand.

• By introducing dummy on Constant variable, Model-3 and Model-4 shows that air mode has lower demand than the other mode.

• By introducing dummy on Income variable, as Model-3 and Model-4 shows that car and air mode has higher positive elasticity for income.

• Incorporating dummy variables, of course, caused reduction of significance level (F) and R2

adjusted -values. Meanwhile, models not incorporating dummy variables have better performance in level significance (F) but are poorer in their interpretive accuracy.

Table 2. Estimation Result of Model-3 and Model-4

Model-3 Model-4 No. Variables Expected

Sign Coefficient tvalue Coefficient tvalue

Ln (Constant) Ln (Population) + 0.676 18.91 0.706 21.29 Ln (Income) + 0.441 3.10 0.459 5.04 Ln (Mean cost) - -0.347 -3.60 -0.795 -8.44 Ln (Mean time) - -0.932 -7.14 -0.932 -3.68 Ln (Mean frequency) + Ln (Relative cost) - -0.372 -4.01 Ln (Relative time) - -1.225 -9.06 -0.255 -3.04 Ln (Relative frequency) + 0.306 7.86 0.119 3.97 Constant Dummy for : • Car 2.268 1.73 • Bus 2.008 1.50 0.530 2.38 • Air -12.917 -8.28 -13.622 -8.51 • Sea 1.614 1.18 • Rail 1.823 1.00 Dummy for Income • Car 0.214 4.94 • Air 1.418 6.25 1.602 7.72 % RMSE of Log Form 13.7 % 14.6 % Multiple R 0.858 0.837 F 217.60 201.29 R2 adjusted 0.686 0.697

Number of samples 870 870 Note : RMSE is Root Mean Square Error

• The positive values of dummy variables on air transport and car in respect to

income implies that the elasticity of air and car travel demand is more sensitive than the others modes.

• Model-3 shows better performance than Model-1, 2, and 4 in the term of % RMSE of Log Form, Multiple R, and level of significance (F)

6. THE ACCURACY OF THE MODEL

The accuracy of the models is examined by utilizing the value of R2 statistic to compare the predicted trips against the actual trips to ascertain how close they are. Here, Model-3 as candidate selected model using mode specific model approach will be compared with Model-1 using abstract model approach. Figure 2 compares the accuracy Model-1 and Model-3 according to Total trips data (including car, bus, sea, air, and rail mode).

877


R2 = 0.155 (Model- 1)

R2 = 0.748 (Model- 3)

1.E+00

1.E+02

1.E+04

1.E+06

1.E+08

1.E+10

1.E+00 1.E+02 1.E+04 1.E+06 1.E+08 1.E+10

Actual Trip

Pred

icte

d T

rip

Model-1Model-3Model-1Model-3

Figure 2. Accuracy of Model-1 and Model-3 according to Total Trip

Meanwhile, Figures 3 and 4 show the accuracy of Model-1 and Model-3 according to each travel mode. Those figures clearly illustrate improvement accuracy as the effect of incorporating dummy variables in Model-3 rather than Model-1. As a result, the accuracy of the models illustrated by comparing the predicted trips demand with the actual trips increased through incorporating dummy variables and this result performs better.

1.E+00

1.E+01

1.E+02

1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

1.E+08

1.E+09

1.E+10

1.E+00 1.E+02 1.E+04 1.E+06 1.E+08 1.E+10Actual Trip

Pred

icte

d T

rip

∆Car Mode; R2 =0.876− Bus Mode; R2 =0.696 + Air Mode ; R2 =0.137 ο Sea Mode ; R2 =0.019

Rail Mode; R2 =0.524

Figure 3. Accuracy of Model-1 according to each travel mode Based upon that result, the Model-3 has the better performance than Model-1, 2, and 4. Therefore, Model-3 is considered the best and will be utilized for further analysis, where from the calibration result is expressed as following equation:

878


1.0E+00

1.0E+01

1.0E+02

1.0E+03

1.0E+04

1.0E+05

1.0E+06

1.0E+07

1.0E+08

1.0E+09

1.0E+10

1.E+00 1.E+02 1.E+04 1.E+06 1.E+08 1.E+10

Actual Trip

Pred

icte

d T

rip

∆ Car Mode; R2 =0.905 − Bus Mode; R2 =0.846 + Air Mode ; R2 =0.347 ο Sea Mode ; R2 =0.017

Rail Mode; R2 =0.592

Figure 4. Accuracy of Model-3 according to each travel mode

Tijm = K (f.) g(.) h(.) where:

K= where α=2.268 for car, 2.008 for bus, -12.917 for air, 1.614 for sea, αeand 1.823 for rail.

441.0676.0 )()((.) ijji IPPf = (for all modes except air mode) 519.1676.0 )()((.) ijji IPPf = (in case of air mode)

932.0347.0(.) −−= ijgijg HCg 306.0225.1

(.)

=

−

ijg

ijm

ijg

ijm

F

F

H

Hh

7. SENSITIVITY ANALYSIS AND FORECASTING ABILITY

Sensitivity analysis involves assigning alternative assumptions of the variables and evaluating their effect on the predictions. How far variation in the assumption affects the predictions can be recognized. Besides, the sensitivity analysis may also be considered as an attempt to test the behavior of predictions in response to variation in assumed factors. By this way, we can investigate how sensitively the predictions will react to changes or improvements of such factors as public transport fares, higher speed, frequency of services, etc. Source of uncertainty can be indicated including estimated parameters, model structures, and their underlying assumptions. But here, we focus on modal attributes as these are the variables which resemble the transport system performance as impacts of transport policy changes. The sensitivity of both total demand and modal split is investigated and

879


illustrated by employing the selected model (Model-3). For example, the illustrations are given by Figure 5.a until Figure 7.b to variations in travel cost, travel time, and frequency of services in the case of bus mode. This illustration based on 1996 base year data.

Sensitivity to Variation in Bus Travel Cost

0

50

100

150

200

250

300

350

400

450

0.8 1 1.25 1.5 2 3 4

(In

Mill

ions

)

Factor Variation

Ann

ual T

rip (C

ar,B

us,T

otal

Trip

)

0

5

10

15

20

25

30

35

40

(in M

illio

ns)

Ann

ual T

rip (R

ail,

Sea,

Air)

Rail

Bus

Car

Air

Sea

Total Trip

Sensitivity to Variation in Bus Travel Cost

0%

20%

40%

60%

80%

100%

120%

140%

160%

180%

0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4Factor Variation

Chn

ages

of T

rip

SeaAirCarTotal TripBus

Figure 5.a. Sensitivity of Model-3 Figure 5.b. Sensitivity of Model-3

Sensitivity to Variation in Bus Travel Time

0

50

100

150

200

250

300

350

400

450

0.6 0.8 1 1.25 1.5 2

( in

Mill

ions

)

Factor Variation

Ann

ual T

rip (C

ar,B

us,T

otal

Trip

)

0

5

10

15

20

25

30

35

40

(in M

illio

ns)

Ann

ual T

rip (R

ail,

Sea,

Air)

Rail

Bus

Car

AirSea

Total Trip

Sensitivity to Variation in Bus Travel Time

0%

20%

40%

60%

80%

100%

120%

140%

160%

180%

0.8 1.2 1.6 2Factor Variation

Chn

ages

of T

rip

Bus

RailCarSea Air

Total Trip

Figure 6.a. Sensitivity of Model-3 Figure 6.b. Sensitivity of Model-3 The impact of transport attributes changes (travel time, travel cost, and frequency of service) on demand of other modes clearly illustrated. The increasing in the frequency of bus will attract some travelers from other modes. In addition, the improvement in the bus travel time will also attract some travelers from other modes. The presence of cost as

880


the general impedance variable in determining demand for one mode reflects that the total demand for all modes decreases when the cost characteristic of one mode increase. Intuitively, we would expect that increasing in the cost of a mode could attract some travelers to other modes. It is not well known whether the right effect (time and frequency) will always dominates the wrong effect (cost). More over, for the number of frequencies (influencing the weights) might lead to different conclusion.

Sensitivity to Variation in Bus Frequency

0

50

100

150

200

250

300

350

400

450

0.8 1 1.25 1.5 2

(in M

illio

ns)

Factor Variation

Ann

ual T

rip (C

ar,B

us, T

otal

Trip

)

0

5

10

15

20

25

30

35

40

(in M

illio

ns)

Ann

ual T

rip (R

ail,

Sea,

Air)

Bus

Rail

Car

Air

Sea

Total Trip

Sensitivity to Variation in Bus Frequency

0%

20%

40%

60%

80%

100%

120%

140%

160%

180%

0.8 1.2 1.6 2Factor Variation

Chn

ages

of T

rip

Bus

Air

SeaCarRail

Total Trip

Figure 7.a. Sensitivity of Model-3. Figure 7.b. Sensitivity of Model-3

50%

75%

100%

125%

150%

175%

200%

225%

1996 1997 1998 1999 2000 2001 2002 2003Year

Chn

ages

of T

rip

AirBusRailSeaTotal trip

0

50

100

150

200

250

300

350

400

1996 1998 2003

(in M

illio

ns)

Ann

ual T

rip (C

ar,B

us,T

otal

Trip

)

0

5

10

15

20

25

30

35

40

(in M

illion

s)A

nnua

l Trip

(Rai

l, Se

a, A

ir)

Rail

Bus

Car

AirSea

Total Trip

(Base year) (Prediction) (Prediction)

Car

Figure 8.a. Annual Trip Prediction Figure 8.b. Changes of Annual Trip Prediction The explanatory variables are exogenous to the model and therefore are determined outside the model. Thus, the exogenous variables are assumed first by referring to recent trends as far as possible. Their exogenous variables are two kinds, i.e., socio-economic variables and modal attributes. Projection of socio-economic variables which include population and GRDP per capita of provinces are possible to be made based on current

881


historical data, whilst some hypothetical sets are established about the future development of transport systems. Based on the economic growth forecasts and on existing movement patterns and characteristics, estimates of base year and future travel demand for passengers were made for each mode. Forecasts using intermediate rates data of economic growth (TSSS, 2001) were made for the further years to 2003. The resulting prediction of 1998 and 2003 travel demand for passenger travel demand of each mode at the national level is illustrated by Figure 8.a and Figure 8.b. All the figures show that travel by road (car and bus) nationwide is the largest mode of all trips followed by Rail, Sea and Air mode. Considering the reasonability of prediction, Model-3 appears to perform well. It matches the observation that when economic crisis hit Indonesia in 1998 demand for air travel and of all modes experienced downturn. 8. CONCLUSION The travel demand model developed in this study appears promising for application in evaluating impacts of socio-economic changes and/or transport system improvement on the travel demand considering all available travel modes. Either impacts of the changes over a given time period or impacts of different policy alternatives are evaluated as well. In addition, such formulation deals with Indonesian condition which face the differences of available travel mode in competition among the province pairs.

Referring to the data limitation problems, this result should be supposed as a preliminary research. Whenever a more complete data base becomes available, further works need to be done with considering following matters:

The travel survey is recommended to collect classification of trip purposes and income groups for every mode. With these classifications, the effect of inhomogeneous of travelers may be reduced.

Related to the level aggregation, the next works is recommended to conduct based on smaller units, the regions and cities. With these classifications, the impact of regional characteristic could be identified: for example, the possibility of aggregating several regions while considering population density, and then, route assignment can be added as extension of this set.

This model could be developed and applied for freight transport demand as well to examine freight transport demand patterns in Indonesia.

ACKNOWLEDGMENT Special acknowledgment to Prof. Akimasa Fujiwara and Dr. Toshiyuki Okamura, who generously donated their time and expertise to the research and also their forbearance and encouragement. I would also like to thank Dr. Christ Summer and Dr. Timothy Curtis Winn, to whom I am grateful for the depth of his commentary and the detail of his editing.

882


REFERENCES a) Books and Books Chapter OECD (1974). Urban Traffic Models: Possibility for Simplifications. Organization for Economic Cooperation and Development, Road Research Group, Paris. Stopher, P.T. and A.H. Meyburg (1975). Urban Transportation Modeling and Planning. Lexington Books, Massachusetts, Toronto b) Journal papers Coto-Millan, P., J. Banos-Pino and V. Inglada (1999). Railway Inter-city Passenger Transport in Spain: A Co-Integration Analysis. Selected Proceedings of the 8th World Transport Research, Vol. 3, pp. 161-169 Crow, R.T. and J. Longeot (1972). A Forecasting Model of the Market for New Aircraft: The Case of the Concorde. Transportation Research, Vol. 6, pp. 211-214 Crow, R.T., K.H. Young and T. Cooley (1973). Alternative Demand Functions for “Abstract” Transportation Modes. Transportation Research, Vol. 7, pp. 335-354 Daly, A., C. Rohr and G. Jovicic (1999). Application of Models Based on Stated and Revealed Preference Data for Forecasting Passenger Traffic between East and West Denmark. Selected Proceedings of the 8th World Transport Research, Vol. 3, pp. 121-134 Gaurdy, J.I. and M.J. Wills (1978). Estimating The Functional Form of Travel Demand Models. Transportation Research, Vol. 12, pp. 257-289 Kraft, G. and M. Wohl (1967). New directions for Passenger Demand Analysis and Forecasting. Transportation Research, Vol. 1, pp. 205-230 Quandt, R.E. (1968). Estimation of Modal Split. Transportation Research, Vol. 2, pp. 41-50 Quandt, R.E. (1976). The Theory of Travel Demand Model. Transportation Research, Vol. 10, pp. 411-413 Quandt, R.E. and K.H. Young (1969). Cross-Sectional Travel Demand Models: Estimates and Tests. Journal of Regional Science, Vol. 6, pp. 201-214 Quandt, R.E. and W.J. Baumol (1966). The Demand for Abstract Transport Modes: Theory and Measurements. Journal of Regional Science, Vol. 6/2, pp. 13-26 Sjafruddin, A. (1997). The Development of Inter city Passenger in Java – Sumatera Corridor (in Indonesian). Jurnal Teknik Sipil, Vol. 4, No. 2, pp. 91-100. Talvitie, A.P. (1973). An Econometric Model for Downtown Work Trips. Chicago Area Transportation Study, Document No. 341,031, 1973; and A. P. Talvitie, “ A Direct

883


Demand Model for Downtown Work Trips,” Transportation, 1973 (a), Vol. 2, pp. 121-152. Wilson, A.G. (1967). A Statistical Theory of Spatial Distribution Models. Transportation Research, Vol. 1, pp. 253-269 Wong, S.T. (1969). Multivariate Analysis of Urban Travel Behavior in Chicago. Transportation Research, Vol. 3, pp. 345-363 Young, K.H. (1969). An Abstract Mode Approach to the Demand for Travel. Transportation Research, Vol. 3, pp. 443-461 Yu, J.C. (1970). Demand Model for Intercity Multimode Travel. Transportation Engineering Journal, Proceeding of the American Society of Civil Engineers, T.E. 2, pp. 203-218 c) Papers presented to conferences Elangovan, T. and F.O. Crouch (1989). Towards Simplified Transport Planning Techniques for Cities in Developing Countries. Proceeding of Seminar, PTRC 17th Summer Annual Meeting, pp. 171-188 d) Other documents Asri, D. U. (2001). Passenger Travel Demand Estimation Using Simultaneous Demand Models: A Case Study of Indonesia. Master Thesis, Graduate School for International Development and Cooperation, Hiroshima University, Japan BPS (1999). Statistical Yearbook of Indonesia. Jakarta Hall, H. B. (1995). Times Series Processor (TSP). Reference Manual and User Guides. TSP International, Palo Alto, USA Koppelman, F.S., G-K. Kuah, and M. Hirsh. Review of Intercity Passenger Demand Modeling: Mid-60’s to the Mod-80’s, The Transportation Center, Northwestern University, Evanston, Illinois, 1984 Ministry of Communication (2001). Transport Sector Strategic Study. Halcrow Fox and Associates, Final Report, Jakarta Ministry of Communication (1997). Survey on Origin and Destination of National Transportation Fiscal Year 1996/1997. Final Report, LAPI-ITB, Bandung. National Development Planning Agency (1998). Development in Numbers (PDA). Office of the Deputy Chairman for Regional Development, Jakarta National Development Planning Agency (2000). Regional Development in Numbers (PDDA). Office of the Deputy Chairman for Regional Development, Jakarta __________, Public Transport Tariff in the last 25 Years. Tempo, Vol. 11, April 2000, Jakarta

884


SIMULTANEOUS DEMAND MODEL FOR PASSENGER TRAVEL - A case study

Documents