Open Access proceedings Journal of Physics: Conference series


 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

34 

 

The Application of Modal Split Using Revealed and Stated 

Preference Techniques: A study in Malang 

Muhammad Nurjati Hidayat1  

1 Water Resources Engineering Department, Universitas Brawijaya, Malang, 65145, 

Indonesia 

muhammadhidayat@ub.ac.id  

Received 05-02-2018; revised 29-03-2018; accepted 08-04-2018 

Abstract. In this study we investigate modal split and travel behaviour in Malang by conducting 

person trip survey in study area. The purpose is to understand respondents’ travel behaviour and 

their preferences in selecting mode of transport. This is carried out to understand what are 

respondents feel regarding their perception on mode of transportation that available to them. The 

data being used are Revealed Preference (RP) and Stated Preference (SP) data. The first data 

based on present situation of respondents (including respondents’ characteristics and daily travel 

information), while the second one is hypothetical scenario that has not available in present 

condition. These data then compared and analysed using Multinomial Logit Model (MNL).  

Keywords: Travel behaviour, Revealed Preference, Stated Preference, Multinomial Logit Model 

1. Introduction 
In modelling travel demand, actual behaviour of user usually estimated using Revealed Preference 

(RP) data by employing discrete choice analysis (e.g. Ben-Akiva and Lerman [1]). However, RP data 

may have shortage in estimating individual choice due to the following reasons [2]: 

- Preferences for non-existing services are not provided in RP data 
- The set of alternatives considered by an individual may be ambiguous 
- There are some errors in estimating service attributes 
- There are similarities in attributes or lack of variability, or both 
In order to lighten those drawbacks, a survey with hypothetical choice scenario and fully controlled 

alternatives need to be done. It called Stated Preference data that widely used by researchers for travel 

demand [3] and [4], and also in marketing research [5], as well. 

This paper investigates the idea of using RP and SP data simultaneously because they have 

complementary characteristics. Unknown reliability of SP data is explicitly considered and its objective 

is to yield more reliable travel demand model on combining RP and SP data rather than analysing RP 

and SP data separately. The key features of this method are bias identification (the effect of new services 

that are not recognizable by individual in RP data), efficiency (all preferences on available data is jointly 

estimated), and bias correction [6]. 

In RP data, trade-offs among certain attribute cannot be estimated accurately. For example, the 

correlation between travel cost and travel time in RP data may resulting insignificant parameter 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

35 

 

estimates for their coefficients. However, an SP survey designed based on little or low correlation 

between these attributes may presenting additional information on their trade-offs.  

The aim of this paper is to perform the effectiveness of the combined RP/SP estimation method by 

an application to predict the mode choice in Malang (East Java) if new mode of transport is introduced. 

2. Material and Methods 

2.1. Study Area 
This research located in Malang City (urban area) and Malang Regency (rural area) which is unified 

in Greater Malang territory (Batu City excluded in this research). Malang is the second biggest city in 

East Java having population of 820,243 residents, while Malang Regency having 2,459,982 residents. 

Total area is 110.06 and 3,534.86 km2, respectively. Government planning to introduce commuter train 

as a new public transportation in order to reduce the congestion level. Another reason is there is no 

public transport that connects directly from northern to southern part of Malang. By using the current 

transportation, it would take longer time rather than using private vehicle. On the ground of those 

reasons, the study of travel demand is needed. 

2.2. Modelling Approach 
Discrete choice model is a technique in which decision makers choose an alternative from choice set 

of available alternatives. It identifying pattern made by an individual facing available choice set. 

Discrete choice model postulates that the probability of individual choosing an alternative is based on 

their socioeconomics and level of attractiveness to the alternative. To represent the attractiveness of the 

alternative the utility function is constructed. Generally, utility derived from individual characteristics.  

Utility is value indicator for an individual, which, generally, derived from the attributes of 

alternatives or sets of alternatives. The utility theory states that an individual will select an alternative 

in choice set that maximize his/her utility. The theory states that it contains function of attributes of 

alternative and individual characteristics[7]. In the newest study, the decision utility (mode choice) could 

be affected by characteristics the built environment (diversity, design and density) and travel modes 

(travel cost and time)[8]. 

The utility function, U, has a property that an alternative is chosen if its utility is greater than utility 

of other alternatives in individual’s choice set. This function can be expressed as: 

𝑈𝑖𝑛 ≥ 𝑈𝑗𝑛, ∀𝑗 ∈ 𝐶𝑛, 𝑖 ≠ j          (1) 

𝑈𝑖𝑛 = 𝑉𝑖𝑛 + 𝜀𝑖𝑛      (2) 

which could be interpreted as the utility of alternative i for individual n is greater or equal to alternative 

j in individual’s choice set.  

Utility function can be represented by two components: an observable or representative part Vin which 

is a function of the measured attribute, and a random term ε which reflect unobservable part of individual 

and also error made by modeler. Two individuals with same attribute, x, in observable part and having 

the same choice set probably select different choice, and some individual may not always choose the 

best alternative. 

2.3. Model Specification 
RP and SP model types are considered. The RP model reflecting behaviour of individual explained 

in utility function, whereas SP model is the result of SP response. Suppose mode choice model utility 

function for RP model with following equation: 

 

𝑈𝑖𝑛
𝑅𝑃 = 𝑣𝑖𝑛

𝑅𝑃 + 𝜀𝑖
𝑅𝑃 = 𝛼′𝑤𝑖

𝑅𝑃 + 𝛽′𝑥𝑖
𝑅𝑃 + 𝜀𝑖

𝑅𝑃       (3) 

 

and the choice of decision maker given by: 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

36 

 

𝑑𝑖𝑛
𝑅𝑃 = {

1, if 𝑈𝑖 = max𝑗=1, . . , I{𝑈𝑖 }

0, otherwise
           (4) 

where i = 1, . . , I is number of alternatives; wi and xi are vector of attributes of alternative; α and β are 

parameter to be estimated; and ɛi is error term. Similar to utility function of RP model, the utility function 

of SP model is: 

 𝑈𝑖𝑛
𝑆𝑃 = 𝑣𝑖𝑛

𝑆𝑃 + 𝜀𝑖
𝑆𝑃 = 𝛽′𝑥𝑖

𝑆𝑃 + 𝛾 ′𝑧𝑖
𝑆𝑃 + 𝜀𝑖

𝑆𝑃        (5) 

and the choice of decision maker given by: 

 

𝑑𝑖𝑛
𝑆𝑃 = {

1, if 𝑈𝑠 = max𝑗=1, . . , I{𝑈𝑖 }

0, otherwise
          (6) 

 

From the framework above we assume that SP response is the most preferable alternative chosen by 

respondent. Thus we have equation as follow: 

 

𝑈𝑐𝑛 ≥ 𝑈𝑎𝑛          (7) 
𝑈𝑡𝑛 ≥ 𝑈𝑐𝑛          (8) 

 

The subscripts c, a, and t represent the current mode, alternative mode, and commuter train, 

respectively. The term 𝑥𝑖
𝑅𝑃 and 𝑥𝑖

𝑆𝑃 in both model implies the common variable in RP and SP model and 

term γ’z is specific to SP model that may contain SP biases. The level of random noise in RP and SP 

data is presented by the variance of the disturbance term ε. If the data source has different noise level, it 

can be shown as follow: 

Var(𝜀𝑖𝑛
𝑅𝑃 ) = 𝜇2Var(𝜀𝑖𝑛

𝑆𝑃 ), ∀𝑖, 𝑛         (9) 

 

With the framework described above, the choice probability of alternative using multinomial logit 

model can be expressed as follow: 

𝑃𝑛(𝑖) =
𝑒xp(𝑣𝑖𝑛)

∑ 𝑒xp(𝑣𝑗𝑛)
𝐼𝑛
𝑗=1

       (10) 

2.4. Combination of RP/SP Models 
Since our concern is actual behaviour of respondents, prediction only using RP model. Hence, utility 

component being used is: 

𝑣𝑖𝑛 = �̂�
′𝑥𝑖𝑛 + �̂�

′𝑤𝑖𝑛    (11) 
 

Note that �̂� is calculated using both RP and SP data. In SP questions, hypothetical services are to be 
included for predicting demand, thus the term in SP model will be: 

 

𝑣𝑖𝑛 = �̂�
′𝑥𝑖𝑛 + �̂�

′𝑤𝑖𝑛 + +�̂�
′𝑧�̅�𝑛   (12) 

 

where 𝑧�̅�𝑛 represents hypothetical attributes related to the policy changes and �̂�
′is an estimation on 

parameter of 𝑧�̅�𝑛. In equation above, RP and SP utility can be combined because new scale parameter 
(µ) is introduced with the purpose to adjust RP and SP scale parameters. 

3. Result and Discussion 

3.1. Field survey 
In this research we conducting two times survey, namely: preliminary survey and primary survey. 

The first survey conducted to ensure the population is being sampled, evaluate the questionnaire, and 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

37 

 

the result is being evaluated. This survey also helps to avoid the problem of collecting large of useless 

or incorrect data in primary survey because of ineffective constructing questionnaire and improper 

sampling preparation. This survey targeting car, microbus, motorbike, and train users with 80 

respondents. 20 passengers interviewed each of those modes, thus we understand their trip behaviour. 

Private vehicle (car and motorbike) users were interviewed in gas station, traffic light or when the user 

stops near the streets. Microbus users interviewed on-board and the rests was sampling people on train 

stations. 

The second survey, primary survey, was conducted with small revision from preliminary survey. The 

sample being taken in this survey is 360 respondents for car, microbus, and motorbike. We eliminate 

train user because in preliminary survey there were only few respondents who use train as their mode 

for working purpose. 

3.2. Result 
The survey result is displayed in table 1. Number of respondent who travelled from Lawang or 

Kepanjen to city center is dominated by male (61%) compared to female (39%). Sequentially, the age 

of respondents are 25-45 years old (58%), 46-65 years old (22%), less than 25 years old (19%), and 

more than 65 years old (1%). The occupation of respondent is dominated by businessman (47%) and 

followed by entrepreneur (18%), civil servant (16%), student (10%), house wife and etcetera have same 

proportion, that are 4%, part timer has very small proportion (1%). Number of commute in a week 

conducted by respondents are 7 times a week (58%), 5 times (20%), more than 7 times (15%), and 3 

times (7%). By taking notice to respondent who majority have job as businessman, probably there is 

correlation between occupation and number of commute in a week, that is 7 times a week.  

Respondents’ trip characteristics and level of service of transport facilities are shown as follow. The 

reason respondent not using public transportation dominated by long travel time (37%) followed by long 

waiting time (21%), uncomfortable (20%), having far access from home (11%), travel cost is expensive 

(11%), and etcetera (5%). 

 

Table 1. The result of respondents’ characteristics 
Item Range % Item Range % 

Sex Male 61 Vehicle Ownership (%) Motorbike 70  
Female 39 

 
Car 30 

Age < 25 19 Income (USD) < 88 29  
25 - 45 58 

 
88 - 177 27  

46 - 65 22 
 

>177 44  
65 > 1 People in Group 1 67 

Occupation Civil Servant 16 
 

3-Feb 23  
Entrepreneur 18 

 
>3 10  

Businessman 47 Reason not using public 

transport 

Long travel time 37 

 
Student 10 

 
Expensive 11  

Part Timer 1 
 

Uncomfortable 20  
House Wife 4 

 
Far access 11  

Others 4 
 

Long waiting 

time 

21 

Last Education Elementary school 7 Frequency per week ≤3 7  
Junior High School 14 

 
5 20  

Senior High School 45 
 

7 58  
University 27 

 
>7 15  

Post Graduate 5 
   

 
Others 3 

   

 

 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

38 

 

Table 2. Respondents current and alternative mode and their income 

 

From Table 2 we can relate that respondents whom using car as their current mode prefers to use 

microbus and motorbike as their alternative mode are 19% and 80%, respectively. For microbus user, 

their alternative modes are car (2%) and motorbike (69%), while motorbike user prefers car (10%) and 

microbus (52%) as their alternative modes. Diagonally, we can see that there are respondents answered 

the alternative mode exactly the same as current mode, i.e. car respondent using car as their alternative, 

microbus respondent using microbus as their alternative and motorbike user using motorbike as their 

alternative. They are categorized as captive respondent, that is respondent who has no any alternative 

mode if current mode cannot be used. 

Similar to the current mode of respondent over the alternative mode, income of respondents affecting 

modal choice preferences. From survey result, 80% of car respondent have high-income. Middle and 

high-income for motorbike respondent have almost identical numbers, that are 51 and 55 respondents, 

respectively. In contrary, microbus respondents have the opposite result of income. Majority of 

microbus user is a low-income respondent, which is 64 respondents (55%). 

Afterwards, we could say that car user who uses motorbike as their alternative mode of transport is 

high-income respondent. Microbus user who categorized as captive respondent are middle-income. Only 

few microbus users who uses car categorized as high-income respondent. Motorbike respondent who 

uses microbus as their alternative mode mostly are middle-income respondent. High-income motorbike 

users mostly are captive respondent. 

3.3. RP/SP Model Estimation 
Three models were estimated: RP model, SP model and RP/SP model using Multinomial Logit 

(MNL) model, each of which was estimated by maximizing the log-likelihood function. In order to ease 

the explanation, artificial tree structure has been made. Given the way the data was collected, the 

structure in Figure 1 should represent these condition appropriately. 

 

 

   Figure 1. The artificial Tree Structure 

 

 

Current 

mode 

Alternative mode Income 

Car Microbus Motorbike Total 
Low 

income 

Middle 

income 

High 

income 
Total 

Car 39 19 80 138 12 16 110 138 

Microbus 2 46 69 117 64 38 15 117 

Motorbike 10 52 83 145 39 51 55 145 

Total 51 117 232 400 115 105 180 400 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

39 

 

Table 3. The result from estimation data 
 RP model SP model RPSP model 

ConstantCar 2.155 (5.97) 
 1.201 (12.85) 

ConstantBus 1.089 (1.59) 
 0.953 (3.99) 

ConstantTrain 
 3.214 (7.75) 0.267 (10.59) 

Cost 0.008 (2.04) -0.061 (-10.81) -0.005 (-11.27) 

Time -2.927 (-2.68) -3.843 (-7.05) -0.302 (-6.93) 

Age 0.025 (1.33) -0.013 (-2.72) 0.0168 (2.65) 

Sex -1.381 (-3.46) 0.345 (3.22) -0.904 (-6.23) 

FrequencyTrain 
 0.123 (10.88) 0.0098 (9.59) 

Income -0.144 (-1.71) 0.013 (0.96) -0.121 (-4.45) 

μ   11.682 (10.38) 

ρ2 0.744 0.164 0.408 

N 400 2000 2000 

 

From RP data, a binary choice data set was created by regarding the current mode and the alternative 

mode as the first and the second options of chosen mode. The data set consist of 400 respondents, after 

eliminating incomplete and data error. In this data set, data collected only available in two alternatives, 

namely the current mode and the alternative mode. The current mode is the mode that respondent used 

every day for commuter trip, while the alternative mode is his/her alternative mode if the current mode 

is cannot be used. The current and alternative mode considered as the best choice or have the highest 

utility for respondent from the choice set.  

The estimation of RP, SP and RP/SP model presented in Table 3. Not all models having correct signs 

due to some errors. Cost variable has sign that should not be positive. Logically, cost variable should 

have negative sign which means the higher the travel cost, the most likely respondent not using the 

current mode as their transport mode. The explanation of positive sign in cost variable could be as 

follow: 

1. There is a correlation between distance and travel fare. The longer the distance of the travel, the 
higher travel fare should be paid by respondent 

2. The current mode has higher cost than the alternative mode. The expensive current mode is the 
product of the number of respondent who have travel expenses using the current mode higher than 

the alternative mode divided by number of non-captive respondents and multiplied by hundred to 

obtain the results in percentage. Captive respondent excluded in the calculation because there is no 

significant difference in travel cost.  

3. There were mistakes in translating questionnaire form from English version into Indonesian version 
before distribute it to surveyor. 

 

As for SP model, stated preference or stated intention of using the new commuter train were 

employed to create binary choice data. If the respondents have willingness or intention to switch mode 

using commuter train, he or she is considered to have chosen the commuter train over the current mode 

he/she currently used. Otherwise, the respondent is considered to choosing the currently mode they used 

over the commuter train. Thus, a binary choice can be created. In the data, each respondent facing five 

different level of services of new commuter train. With number of observation of 2000, respondent will 

accept and have intention to use new commuter train if the level of service match his/her condition. 

Commuter train constant is introduced in the model so that it may capture the attributes of new 

commuter train that are not include in the model and response bias toward the new commuter train. The 

commuter train constant has a significant positive coefficient, probably reflect the overstated use of the 

commuter train. 



 

 

 

 
Civil and Environmental Science Journal 

Vol. I, No. 01, pp. 034-040, 2018 

 

 

 

40 

 

The last model is combined RP/SP model. The framework is shown in equation (3) to (9). Due to the 

difference in variance of RP and SP data, variable µ is employed. The value of scale parameter µ is 

expected to be less than one. If scale parameter µ has value less than one, it scales down the explanatory 

variables in the stated preference model because SP model has more random noise than RP model. 

However, if the µ value greater than one, the RP model has more random noise.  

The estimation results of RP/SP model are shown in the third column of table 3. It shows that train 

frequency has positive coefficient. The results also show that µ has estimated value greater than 1. This 

probably because of the errors obtained from RP data explained previously. The value more than 1 scales 

up the SP model. The estimation of RP/SP model is almost similar to RP model which indicate that the 

joint estimation successfully replicates the RP model, except in variable cost. 

4. Conclusions 
The combined estimation of discrete choice models from RP and SP data was presented. The strategy 

combining both types of data can benefit with explicit consideration of their merit and demerit. The case 

study is modal split model under hypothetical scenario, namely introducing commuter train as new 

alternative. In estimating RP and SP data simultaneously to estimate the mode choice model, alternative 

specific constant was estimated separately. In modelling using MNL model, the most significant model 

from RP model, SP model and combined RP/SP model is RP/SP model than other models alone. 

Combining RP data with SP data increase the accuracy of parameter estimates in the model. In our result 

shows that RP model contain more random noise than SP model. To perform this model in real 

condition, more data need to be collected, such as number of passenger for each mode and OD table, 

thus the probability for each mode can be calculated. From academic point of view, this research need 

more advanced model such as Nested Logit (NL), Cross Nested Logit (CNL), Generalized Nested Logit 

(GNL) to resulting higher quality in estimation and research due to the estimated modes are able to 

categorized in private vehicle and public transport. 

References 

[1] M. Ben-Akiva and S. R. Lerman, Discrete Choice Analysis: Theory and Application to Travel 

Demand. MIT Press, 1985. 

[2] T. Morikawa, M. Ben-Akiva, and K. Yamada, “Forecasting intercity rail ridership using 

revealed preference and stated preference data,” Transp. Res. Rec., vol. 1328, pp. 30–35, 

1991. 

[3] J. J. Louviere et al., “Laboratory-Simulation Versus Revealed-Preference,” Transp. Res. Rec. 

794, pp. 42–51, 1981. 

[4] D. a. Hensher, P. O. Barnhard, and T. P. Truong, “The role of stated preference methods in 

studies of travel choice,” J. Transp. Econ. policy, vol. 22, no. 1, pp. 45–58, 1988. 

[5] P. Cattin and D. R. Wittink, “Commercial Use of Conjoint Analysis: A Survey,” Source J. 

Mark., vol. 46, no. 3, pp. 44–53, 1982. 

[6] M. Ben-Akiva et al., “Combining revealed and stated preferences data,” Mark. Lett., vol. 5, no. 

4, pp. 335–349, 1994. 

[7] T. A. Domencich and D. McFadden, Urban Travel Demand: A behavioral Analysis. Amsterdam 

and Oxford, 1975. 

[8] J. De Vos, P. L. Mokhtarian, T. Schwanen, V. Van Acker, and F. Witlox, “Travel mode choice 

and travel satisfaction: bridging the gap between decision utility and experienced utility,” 

Transportation (Amst)., vol. 43, no. 5, pp. 771–796, 2016.