Professional Documents
Culture Documents
net/publication/262689818
CITATIONS READS
7 625
3 authors, including:
SEE PROFILE
All content following this page was uploaded by Arun Baby Zacharia on 30 May 2014.
Abstract: Travel behavior plays an important role in transportation planning. Mode choice analysis is the process of arriving at a
decision about which mode to use under a set of circumstances. Choice of a particular mode by a commuter affects the general efficiency
of travel within the city. Various techniques are available in the literature for mode choice modeling. Among these, multinomial logit
models are found to be efficient in estimating the different mode shares in a scenario where more than two choices of modes of travel
are available for a commuter. The present study used multinomial logistic regression to analyze the mode choice behavior of commuters
in Thiruvananthapuram, a typical Indian city. The findings from the study revealed that as age increases preference to car increases and
preference to two-wheelers decreases in comparison with public transport. Increase in time per distance and increase in cost per distance
cause the commuters to switch to car and two-wheelers from public transport. Identification of such factors and its variations will help city
planners to formulate effective transport policies to improve the transport within the city. DOI: 10.1061/(ASCE)TE.1943-5436.0000533.
© 2013 American Society of Civil Engineers.
CE Database subject headings: Logistics; Urban areas; Commute; Travel patterns; Transportation management; India.
Author keywords: Mode choice; Revealed preference technique; Multinomial logistic regression; Odds ratio; Significance level.
A pilot questionnaire survey was first conducted. The questions Gender Male 1 340 46.0
pertained to the socioeconomic condition and travel characteristics Female 2 399 54.0
of the trip maker. The socioeconomic characteristics were assessed Monthly Less than 10,000 1 230 31.1
by considering factors such as age, gender, nature of job, monthly income 10,000–15,000 2 192 26.0
income, and vehicle ownership. Travel characteristics included dis- 15,000–20,000 3 110 14.9
20,000–25,000 4 72 9.7
tance from home to office, mode of conveyance, reasons for choos-
Above 25,000 5 135 18.3
ing the mode, waiting time, in-vehicle travel time, cost of travel, Vehicle Does not own a vehicle 1 222 30.0
etc. Commuters’ satisfactory levels with respect to present mode, ownership Owns car only 2 194 26.3
their willingness to switch over to another mode, and reasons for Owns two-wheeler 3 181 24.5
it were also arrived through the questionnaire survey. It also Owns both car and 4 142 19.2
contained questions that enabled ranking of mode attributes in two-wheeler
general and ranking of performance of the existing KSRTC service Distance (km) 0–5 1 292 39.5
in the city. Efficiency level of existing KSRTC service were ranked 6–10 2 185 25.0
based on factors such as comfort, time of travel, cost of travel, 11–20 3 136 18.4
safety, ease of boarding, frequency/availability, travel in peak 21–50 4 126 17.1
Time/distance 0–2 1 400 54.1
hours, reliability, travel in difficult weather. A pilot survey was
(min=km) 3–5 2 232 31.4
conducted and 90 samples were obtained. Based on the response, 6–10 3 107 14.5
preferences, and the suggestions of the commuters, the variables Cost/distance 0–2 1 323 43.7
were shortlisted and the questionnaire was further refined. A large- (Rs:=km) 3–5 2 247 33.4
scale survey was conducted among the commuters using this 6–12 3 169 22.9
refined questionnaire. Valid 739 100.0
Total 739 —
Socioeconomic Variables
Sampling Procedure
Users’ socioeconomic characteristics as reflected through their
age, gender, designation, monthly income, and vehicle ownership Sufficient care was taken to ensure that the sample is of optimum
are believed to be important factors in mode choice. The categories size. The sample size depends on the population of the city under
of different socioeconomic variables used for this study are shown study. In the city of Thiruvananthapuram, the total employed
in Table 1. In the vehicle ownership category, the captive riders people in public and private sector is 190,000 (Census of India
were also considered, because there are chances of them hiring a 2001). Assuming the population to be normally distributed, empir-
taxi or opting for a car pooling. ical formulas given by Levy and Lemeshow (2008) [shown in
Eqs. (2) and (3)] were used to determine the sample size.
Transport System Variables Z2 pq
n0 ¼ ð2Þ
The travel cost and travel time data per kilometer were extracted e2
from the data collected. The time of travel (min) for bus users where n0 = sample size for infinite population; Z = statistical
was taken as the summation of time taken for accessing the nearest parameter corresponding to confidence level (Z is 1.96 for 95%
bus stop, waiting time at bus stop, and in-vehicle travel time. The confidence interval); e = desired margin of error (adopted as 5%);
categories of different transport system variables used for this study p = hypothesized true proportion for population (adopted as 0.5 to
are shown in Table 1. account for the worst case) and
n0
Attitudinal Variables q¼1−p n¼ ð3Þ
1 þ ðn0N−1Þ
Attitudinal variables are those attributes for which no objective
measurement is possible. The questionnaire included variables where n = sample size for finite population; N = population size.
about travel mode such as comfort, safety, ease of boarding, travel Based on Eqs. (2) and (3) the minimum sample size was ob-
in peak hour, reliability, and travel in difficult weather. Attitudinal tained as 385. For the present study, a total of 739 samples were
variables were given least preference by the commuters during collected. It was ensured that the data represented all age groups
A preliminary analysis of the coded data was performed to yield preference to bus mode and highest preference to car mode.
an overview of the mode choice behavior of commuters in the city
of Thiruvananthapuram. Various cross-classification charts were
prepared with respect to age, gender, vehicle ownership, income, Analysis Using MNL Regression
etc. The various analysis conducted are explained in the following
sections. When using MNL regression, one category of the dependent var-
The effect of age of commuters on the choice of mode is shown iable is chosen as the reference category. All parameters in the
in Fig. 1. It can be observed that higher age groups preferred model are interpreted with reference to it. It is the standard/
bus mode. Among the personalized vehicles, car was preferred by reference category to which others (response category) would
higher age group, whereas two-wheelers were more preferred naturally be compared. The coefficients are estimated through an
by lower age groups. iterative maximum likelihood method. Odds ratios are determined
Gender was also found to have an influence on the choice for all independent variables for each category of the dependent
of mode as can be observed from Fig. 2. Car as mode of travel variable except the reference category. Odds ratio represents the
is mostly preferred by males than females. Similar trend was change in the odds of being in the dependent variable category ver-
observed in the case of two-wheelers as well. Of the various modes sus the reference category associated with a one unit change in the
of travel within the city, females preferred public transport system independent variable.
(buses) more than any other mode, whereas males preferred MNL regression was done to identify the influence of various
two-wheelers. factors on the switching behavior of commuters from bus to car and
two-wheeler. Such an analysis will help policymakers to improve
the government-owned buses and attract more commuters from per-
sonalized modes such as cars and two-wheelers. Therefore, for the
analysis, the mode of conveyance is chosen as dependent variable,
as the problem is concerned with choice of mode. In order to get
the preference of bus relative to other modes, bus is chosen as the
reference category. Bus was compared with car and two-wheeler
separately to understand the factors and their degree of influence
in shifting commuters from bus (public transport) to personalized
modes (car and two-wheeler). The independent variables chosen
are age group, gender, monthly income, vehicle ownership, dis-
tance from home to office, time per distance, and cost per distance.
The independent variables are classified into various categories.
The output of analysis is shown in Table 1, in which N repre-
sents the number of observations under each case. The marginal
percentage lists the proportion of valid observations found in each
of the outcome variable’s groups. Valid indicates the number of
observations in the data in which the output variable and all pre-
Fig. 1. Age group–based classification of mode preference dictor variables are nonmissing.
The likelihood ratio tests that show the contribution of each var-
iable to the model are shown in Table 2. From Table 2, it can be
seen that all variables have significance less than 0.05. Therefore,
it can be concluded that all the variables used in the model have
significant contribution toward predicting the mode choice behav-
ior of commuters. Of these, age, vehicle ownership, distance to
travel, and cost of travel are found to be the most significant ones.
Table 3 shows a statistical comparison of intercept-only model
and final model. Intercept only describes a model that does not in-
clude any predictor variables and simply fits an intercept to predict
the output variable. Final describes a model that includes the speci-
fied predictor variables and has been arrived at through an iterative
process that maximizes the log likelihood of the outcomes seen in
the output variable (Table 3). By including the predictor variables
and maximizing the log likelihood of the outcomes seen in the data,
the final model is an improvement on the intercept-only model.
Fig. 2. Gender-based classification of mode preference
The likelihood ratio test shows the contribution of each variable
70 −2 Log Degree of
60 Model likelihood Chi-square freedom Significance
50 Intercept only 1,430.833 — — —
40 Final 556.465 874.369 36 0.000
30
20
10 that larger percentage of variation can be explained by the model.
0 Thus, based on the pseudo R2 values it can be concluded that
does not own owns car owns two- owns two- the model developed in the present study explains approximately
vehicle wheeler wheeler & car
56–79% variation. Therefore, the model can be considered statisti-
Fig. 4. Vehicle ownership–based classification of mode preference cally significant.
The overall goodness of fit of the MNL model developed is
shown in Table 4. It can be seen that for each case, the predicted
response category is chosen by selecting the category with the
to the model. This fact is revealed by the statistical tests on the two highest model-predicted probability. Cells on the diagonal are
models as given in Table 3. The chi-square statistic is the difference correct predictions. Cells off the diagonal are incorrect predictions.
between the –2 log-likelihoods of the null/intercept-only and final It can be observed that out of the 338 commuters who are found to
models. Because the significance level of the test is less than 0.05, use bus, the MNL model predicted total number of bus commuters
it can be concluded that the final model is outperforming the null as 263 and wrongly predicted that 46 bus commuters used two-
model. The pseudo R2 value of the final model according to Cox wheelers and 29 used cars. Therefore, the accuracy of prediction
and Shell, Nagelkerke, and McFadden tests were 0.694, 0.789, for bus is 77.8%. Similar analysis on two-wheeler commuters gave
0.560, respectively. Pseudo R2 value indicates the proportion of a prediction accuracy of 92.4%, and car as 75.6%. The model has
variance of the response variable explained by the predictors an overall accuracy of 82%.
and its maximum value is 1. Larger pseudo R2 statistics indicate Tables 5 and 6 give the parameter estimates that summarize the
effect of each predictor for commuters’ choice of car relative to bus
and car relative to two-wheeler, respectively. B values are the esti-
Table 2. Model-Fitting Information mated MNL regression coefficients for the models. Parameters with
Model-fitting negative coefficients decrease the likelihood of that response cat-
criteria Likelihood ratio tests egory with respect to the reference category. ExpðBÞ values are
−2 Log Degree of
Parameter likelihood Chi-square freedom Significance
Intercept only 1,430.833 0.000 0 — Table 4. Prediction Accuracies
Age group 593.120 36.656 6 0.000 Predicted
Gender 568.837 12.372 2 0.002
Percent
Income 579.880 23.415 8 0.003
Observed Bus Two-wheeler Car correct
Vehicle 1,133.842 577.377 6 0.000
ownership Bus 263 46 29 77.8
Distance 588.936 32.472 6 0.000 Two-wheeler 10 219 8 92.4
Time/distance 567.748 11.283 4 0.002 Car 23 17 124 75.6
Cost/distance 580.186 23.721 4 0.000 Overall percentage (%) 40.1 38.2 21.8 82.0
the other variables in the model are held constant, are discussed
in the following sections (as given in Table 6). age groups, whereas preference to cars is higher among higher age
Based on the parameters estimated through MNL regression groups. Two-wheeler is a preferred mode of conveyance among
given in Table 6, the following observations are made. lower age group. Therefore, it can be concluded that with increase
• Age group. MNL regression analysis of those who choose in age, preference to car increases and preference to two-wheelers
two-wheelers relative to bus observed that compared with com- decreases in comparison with bus.
muters of age group (51–60), commuters of age group (21–30) Gender-wise comparison revealed that preference to car as well
have 1.26 times as much chance of choosing two-wheeler rather as two-wheeler is higher among male commuters compared with
than bus. Therefore, it can be concluded that the preference of females. Female commuters preferred bus given a choice between
two-wheeler relative to bus is higher for lower age groups and car and bus as well as between two-wheeler and bus.
lower for higher age groups. Income-wise comparison revealed that those with monthly in-
• Gender. Regarding the influence of gender over choice between come less than Rs. 10,000 and up to Rs. 15,000 showed more pref-
two-wheeler and bus, compared with females, males have al- erence to bus. With increase in monthly income above Rs. 15,000
most 3.02 times more chance of switching to two-wheeler from commuters preferring bus as a mode of transport decreased. As ex-
bus. The reason may be because females are not comfortable in pected, among the higher income group, car is the preferred mode
using two-wheeler as it is an unsafe mode, and therefore they of travel. This trend is expected in a city like Thiruvananthapuram
prefer bus to two-wheeler. where there is no mass transport system and the existing public
• Monthly income. The preference of two-wheeler relative to transport system is very inefficient and unreliable. Therefore, it
bus is lower for low-income groups (monthly income less can be observed that there is a trend of shifting to personalized ve-
than Rs. 10,000–15,000) and higher for high-income groups hicles with increase in financial position of commuters. Moreover
(monthly income Rs. 15,000 to greater than 25,000). The trend with increase in travel distance (intercity traveling), commuters
were found to give less preference to car and two-wheeler relative
is similar to the observation regarding choice between car and
to bus. This also points to the nonavailability and nonreliability of
bus given in the previous section. Therefore, the same argument
intracity public transport system.
holds good for this case as well.
With increase in travel time per distance, commuters were found
• Vehicle ownership. The preference of two-wheeler relative to
to give more preference to car and two-wheelers relative to bus.
bus is lower for those who own only car and higher for those
Similar trend was also observed with increase in travel cost per
who own both two-wheeler and car. The reason for this might be
distance. Among the personalized modes, preliminary analysis
because those who own only car, preference of two-wheeler to
showed that two-wheeler is the most preferred mode. With increase
bus does not make any sense. Only teenagers might prefer
in travel time per distance as well as cost per distance, two-wheeler
such a shift if they get an opportunity to own a two-wheeler. commuters switch from bus. Therefore, transport planners should
However, those who own both two-wheeler and car have shown ensure that time per distance as well as cost per distance is kept
a greater trend to switch to two-wheeler from bus compared with to the minimum to attract more commuters to the public transport
those who own only two-wheeler. The better manoeuvrability of system.
two-wheelers and their ability to travel faster than car under con- MNL analysis on the influence of vehicle ownership over choice
gested conditions could be the reason for this trend. Moreover, of bus revealed that those who own both car and two-wheeler prefer
considering the total cost, two-wheelers are more economical. two-wheeler for shorter trips and car for longer trips. Commuters
The MNL model has correctly predicted captive riders as their who own only car showed lower preference of two-wheelers to bus,
odds ratio is near to zero. whereas those who own both two-wheeler and car showed higher
• Distance (km). Multinomial analysis reveals that as travel dis- preference of two-wheeler especially for shorter trips.
tance increases, the chances of switching to two-wheeler from A comparison between two-wheeler users and car users with
bus decreases. The chance of switching to two-wheeler from bus reference to their preference to bus revealed that two-wheeler own-
is almost 9.42 times for short-distance commuters (0–5 km), ers showed a greater preference to bus compared with car users.
whereas it is only 7.24 for commuters having travel distance This may be because users of two-wheelers generally fall under
in the 6–10 km range. The preference of two-wheeler relative low/middle income category, and therefore their next natural choice
to bus is higher for short-distance commuters. It decreases with will be bus. The commuters, who own both two-wheeler and car,
increase in travel distance. gave least preference to bus and higher preference to car. This is
• Travel time per distance (min=km). The analysis revealed that because in a developing country like India, given the poor state of
the preference of two-wheeler relative to bus is higher for higher public transport, those owning car as well as two-wheelers prefer
values of time per distance. As travel time per distance in- car owing to its safety and flexibility. The study also revealed that
creases, commuters prefer two-wheeler over bus. Infact increase the bus commuters who own both car and two-wheeler preferred
in travel time per distance is an indication of congestion, and two-wheeler when faced with a situation that demands switching
therefore it can be concluded that as congestion increases, com- from bus. This shifting from bus is mainly due to the inefficient and
muters switch from bus to two-wheelers. nonreliability of bus transport. This also points to the fact that in the
residential location choices in an urban area with surface streets and rail
choked urban corridors, and will reduce vehicular emission, delay, transit lines.” J. Urban Plann. Dev., 132(4), 235–246.
and fuel consumption. Koppelman, F. S., and Bhat, C. (2006). A self instructing course in mode
choice modeling: Multinomial and nested logit models, U.S. Department
of Transportation, Federal Transit Administration, Washington, DC.
Acknowledgments Levy, S. P., and Lemeshow, S. (2008). Sampling of populations: Methods
and applications, 4th Ed., Wiley, Hoboken, NJ.
The authors would like to thank Ms. Anu P. Alex, Assistant Liu, G. (2007). “A behavioral model of work-trip mode choice in
Professor, Dept. of Civil Engineering, College of Engineering, Shanghai.” China Econ. Rev., 18(4), 456–476.
Trivandrum for the meticulous proofreading and her help in inter- Mahlawat, M., Rayan, S., Kuchangi, S., and Patil, S. (2007). “Exami-
preting the results. They would also like to thank the authorities nation of student travel mode choice.” Rep. 07-2616, Transportation
and employees of St. Mary’s School, State Bank of Travancore, Research Board 86th Annual Meeting, Transportation Research
Indian Bank, Bharat Sanchar Nigam Limited, Mar Ivanios College, Board, Washington DC.
and Medical College Trivandrum for co-operating with the ques- Nurdeen, A., Rahmat, R. A. O. K., and Ismail, A. (2007). “Modeling of
tionnaire survey. transportation behavior for coercive measures for car driving in Kuala
Lumpur.” ARPN J. Eng. Appl. Sci., 2(2), 18–24.
Palma, A., and Rochat, D. (2000). “Mode choices for trips to work in
References Geneva: Anempirical analysis.” J. Transp. Geogr., 8(1), 43–51.
Rastogi, R., and Rao, K. V. (2003). “Travel characteristics of commuters
Arasan, V. T., Rengaraju, V. R., and Krishna Rao, K. V. (1994). “Character- accessing transit: Case study.” J. Transp. Eng., 129(6), 684–694.
sitics of trips by foot and bicycle modes in Indian city.” J. Transp. Eng., Seneviratne, P. N. (1985). “Acceptable walking distances in central areas.”
120(2), 283–294. J. Transp. Eng., 111(4), 365–376.
Arasan, V. T., Rengaraju, V. R., and Krishna Rao, K. V. (1996). “Trip SPSS. (2006). “SPSS for Windows.” Statistical analysis software, Release
characteristics of travellers without vehicles.” J. Transp. Eng., 122(1), 15, SPSS, Chicago.
76–81. Spurr, T., and Chapleau, R. (2007). “The contribution of the confusion
Badoe, D. A., and Wadhawan, B. (2002). “Jointly estimated cross- matrix to the analysis of mode choice in Montreal.” Rep. 07-2925,
sectional mode choice models: Specification and forecast performance.” Transportation Research Board 86th Annual Meeting, Transportation
J. Transp. Eng., 128(3), 259–269. Research Board, Washington DC.