Analysis of Route Choice Behaviour Based on
Questionnaire and 3D Traffic Flow Simulator
Toshihiro Hiraoka*, Kouhei Okabe*, Hiromitsu Kumamoto*, Osamu Nishihara*
and Kenji Tenmoku**
*Dept. of Systems Science, Graduate School of Informatics, Kyoto University,
Yoshida Honmachi, Sakyo-ku, Kyoto, Japan
**Sumitomo Electric Industries, Shimaya 1-1-3, Konohana-ku, Osaka, Japan
Abstract. To model driver's route choice behaviour, we executed a questionnaire and
an experiment with a simplified 3D traffic flow simulator. Questionnaire result shows
many drivers prefer the criteria such as cognitive resource demands minimization to
travel time minimization when a destination is unfamiliar. Simulation result also shows
the existence of the criteria, and that VICS information reduces the travel time.
Annual traffic congestion losses in Japan amount to about 24 hours per 1 person; in money
conversion, about 12 thousand billion yen nationwide. Some traffic management systems in
order to ease traffic congestion have been introduced, for example, VICS (Vehicle
Information and Communication Systems), signal control systems and so on. VICS provide
car navigation systems with the information about congestion, route guidance, predicted
travel time and parking area almost real time.
Generally, traffic flow simulator is used in verification of the traffic management systems,
but a route choice model embedded in the simulator is very simple one such as choosing the
route whose predicted travel time is minimum. That model lacks the reproduciblity of the
actual driver’s behaviour, so it is difficult to verify what kind of influences or what kind of
phenomena happen by the introduction of navigation systems and VICS.
In this paper, we analyse the actual route choice behaviour. At first, the questionnaire
result about route choice behaviour is analysed, such as route choice criteria and information
collection methods. Experiments are performed using the simplified 3D traffic flow simulator
having driver’s front view and a navigation screen. We verify the consistency between
questionnaire and experiment.
2. Questionnaire about route choice behaviour
2.1 Questionnaire contents
A questionnaire is composed by a question about 1) attributes of examinees (sexuality, age,
driving experience, driving skill level, etc.), 2) attitude to navigation system and VICS, 3)
Table 1 Attributes of Examinees
Age Under 19 20 to 24 25 to 29 30 to 39 40 to 49 50 to 59 Over 60 No
Male 4 75 41 32 22 21 8 1 204
Female 2 13 14 7 6 6 1 0 49
Total 6 88 55 39 28 27 9 1 253
no answ er have
others needless others
needless 0.8% 5.5%
4.3% 5.9% 0.4%
several times 4.7% w ant
per year 8.7%
18.2% everyday 18.2%
30.0% w ant
several times 9.9%
per month don't have
several times don't know
per w eek 80.2%
Fig 1 Driving frequency Fig 2 Attitude to navigation Fig 3 Attitude to VICS
route choice criteria in the case where a destination is unfamiliar or familiar, 4) information
collection methods for route choice.
2.2 Examinee attributes and attitude to navigation system and VICS
A questionnaire was executed on paper and World Wide Web. Table 1 shows attributes of
examinees. This proportion is similar to actual data in Japan, except for the smaller middle
age and elder drivers in questionnaire. Driving frequency, attitudes to navigation system and
VICS are shown in Fig 1, Fig 2 and 3 respectively.
According to recent researches, the number of passenger cars in Japan amounts to a little
more than 50 million, and 7.2 million navigation systems and 2.8 million VICS units are
delivered. The ownership rate of navigation system and VICS unit in this questionnaire are
similar to actual rate. Fig 3 indicates that the popularity and interest to VICS are very low.
2.3 Cluster analysis
Cluster analysis is applied to the result of following Question 1 and 2. It is an algorithm that
can classify the population into finite groups with similar members. There are many methods
that calculate similarity. Ward method based on Euclid distance is used in this paper.
2.4 Route choice criteria
Questions about route choice criteria are asked in the both case where the destination is
unfamiliar and familiar. Only unfamiliar case is shown in this paper.
Most general route choice criterion is “travel time”, in other words, driver chooses the
route whose travel time is expected to be minimum. Some researches suggest criteria such as
mileage, runnability and comfortableness. To survey what criteria affect driver’s route choice
behaviour, Question 1 in Fig 4 was given to examinees.
Four clusters, CLA1 to CLA4, are classified based on a cluster analysis for Question 1
(Table 2). It is shown that runnability and comfortableness are taken seriously in the largest
cluster CLA1. Runnability is a driver’s desire to arrive the destination certainly, and
comfortableness is a desire to drive easily. Therefore, we define CLA1 as cognitive resource
demands minimization type. And Cluster CLA2, CLA3 and CLA4 are also defined as travel
Q1. You are going to drive a car toward unfamiliar destination. What are your route choice criteria in order to
decide the route? Please answer the priority percentage of the following route choice criteria. (Sum of the
percentages must be 100%)
a) Travel time priority: % b) Mileage priority: %
c) Runnability / comfortableness priority: % d) Others priority: %
Fig 4 Question 1 (About Route Choice Criteria)
Table 2 Cluster Analysis Result for Route Choice Criteria (Question 1)
Criteria Travel Runnablity &
Mileage Others Numbers(%) Type
Cluster time Comfortableness
CLA1 25.4 16.1 57.1 1.3 150 (59.3%) Cognitive resource demands minimization
CLA2 81.9 8.4 8.4 1.0 54 (21.3%) Travel time minimization
CLA3 42.5 42.4 14.2 1.3 29 (11.5%) Synthetic evaluation
CLA4 16.9 12.7 15.3 55.3 20 (7.9%) Other
time minimization type, synthetic evaluation type and other type respectively.
The cluster size of CLA2, which is regarded as conventional route choice criteria, is
21.3 % and this is about one-third of that of CLA1. This result shows that CLA1 dominates
CLA2 because the former is most major route choice criteria when the destination is
By investigating the correlation between the result of route choice criteria and the
attributes of examinees, it is cleared that a driver who is young, beginner and poor at driving
tends to belong to CLA1, and that a driver who is elder, veteran and good at driving tends to
belong to CLA2 or CLA3.
We have to pay much attention to the followings when route choice criteria are modelled.
It’s better to use multiple criteria types such as a combination of CLA1 and CLA2, because
this questionnaire result reveals that actual route choice criteria are not so simple and uniform.
The occurrence ratio of criteria types must be considered.
There is a difficult problem on modelling the route choice criteria CLA1. It is how to
formulate the criteria of runnablility and comfortableness, e.g., right / left turn, type of road.
As one solution, there is a method to convert the cost of right turn into travel time, but another
problem how to set up the cost takes place because it is mental value of each drivers.
2.5 Information collection methods to evaluate route choice criteria
Question 2 in Fig 5 was asked to investigate what information drivers refer to in order to
decide the route by evaluating their route choice criteria. It is assumed that all drivers have
Table 3 shows the result of cluster analysis. Four clusters, CLB1 to CLB4, are classified
based on the cluster analysis. This result shows that the largest cluster is CLB1, which is
44.7 % of all, where drivers attach weight to map information. We define the information
collection method CLB1 as map oriented type. CLB2, CLB3 and CLB4 are also defined as
navigation oriented type, navigation dependent type and navigation independent type
If a driver has a navigation system and a VICS unit, he/she knows congested areas and
estimates the travel time more precisely. In this way the navigation system brings many merit
Q2. You are going to drive a car toward unfamiliar destination. What Information do you use in order to decide
the route? Please answer the priority percentage of the following route choice criteria. (Sum of the
percentages must be 100%)
a) Navigation system guidance priority: %
b) Road Map priority: %
c) Guidance sign on the road priority: %
Fig 5 Question 2 (About Information Collection Methods)
Table 3 Cluster Analysis about Information Collection Methods (Question 2)
Navigation Map Guide sign Numbers(%) Type
CLB1 16.8 62.3 20.8 113 (44.7%) Map oriented
CLB2 53.5 26.9 19.6 66 (26.1%) Navigation oriented
CLB3 89.7 2.3 8.0 42 (16.6%) Navigation dependent
CLB4 7.0 36.6 56.4 32 (12.6%) Navigation independent
to drivers, but the cluster CLB1 is a slightly larger than the sum of CLB2 and CLB3. This will
be derived from the following reason. 1) Drivers don’t notice the merit because of a low
ownership rate of navigation system and a low popularity of VICS. 2) The present problem of
the navigation system may lower user's trust even though he has it.
The correlation of the result of Question 1 and Question 2 proves that the navigation
information is taken more seriously (a rate to belong to CLB2 or CLB3 is higher) from the
drivers who belong to CLA2 than those who belong to CLA1, CLA3 and CLA4. This result is
thought to reflect that the navigation information is useful for the estimation of travel time.
The questionnaire shows that the navigation information is, at present, almost equivalent to
the map information. It is expected that drivers recognize the utility of the navigation
information and the rate of drivers who belong to CLB2 or CLB3 increases, when navigation
system becomes more popular from now on.
3. Simplified 3D traffic flow simulation
3. 1 Outline of simulation
Fig 6 shows a simulator screen. Experimental subjects choose a route and drive from origin to
destination in the simulation area shown in Fig 7 (about 5 5km).
Three types of experiments were executed. In Experiment 1, information about the area
map and current position is provided on the navigation screen. In Experiment 2, shortest route
information calculated by Dijkstra method is provided in addition to the condition of
Experiment 1. In Experiment 3, VICS information (congestion information) is provided in
addition to the condition of Experiment 2.
Main congestion sections are shown by dark lines in Fig 7. The congestion situation is
renewed every 5 minutes, and experimental subjects can check the situation with 5 minutes
maximum delay on the navigation screen.
Acute congestion section:
Driver’s front view
Medium congestion section:
Fig 6 Simulator Screen Fig 7 Simulation Area Map
(a) Traveling Routes of Experiment 1 (b) Traveling Routes of Experiment 2 (c) Traveling Routes of Experiment 3
(Provided wiith Map Information) (Provided with Map / Shortest (Provided with Map / Shortest
Route Information) Route / VICS Information)
Fig 8 Simulation Area and Simulation Results
Table 4 Simulation Results
Subjects Exp.1 Exp.2 Exp.3
#1 1266 1219 726
#2 486 1766 737
#3 2132 2111 989
#4 1279 1233 739
#5 1626 1233 746
#6 1547 1267 725
#7 1627 1627 750
Ave. 1423 1494 773
Experiments were executed to seven men and women in twenties and thirties who have
driver’s licence. Route data as well as intentions of behaviours such as congestion avoidance
or turns were recorded by direct interviews.
3. 2 Simulation results
Table 4 shows that VICS information reduces travel time almost 50%. In Experiment 3, each
experimental subject succeeded to avoid congestion and travelling routes concentrated into
Routes A and Routes B shown in Fig 8(c). Routes A is short mileage one with avoiding
congestion, and Routes B is simple one with smaller number of turns as well as congestion
avoidance. Routes A results from the route choice criteria CLA1 and Routes B from CLA2 or
The average number of turns at Experiment 1 is 3.3, and that of Experiment 3 is 6.1. The
minimization of the number of turns is one of the criteria satisfying cognitive resource
demands minimization. But providing the congestion information changes the order relations
between the cognitive resource demands minimization criteria and the travel time
minimization criteria, and causes drivers to turn actively. In other words, it represents that
drivers change their route choice criteria corresponding to the difference of the information
collection methods or the information itself which drivers can get at that time.
Fig 8(b) shows that only one subject chooses the shortest mileage route provided on the
navigation screen, because this route intersected many congestion sections.
As mentioned above, the congestion information provision may cause route concentration.
This suggests that congestion sections may vary as VICS becomes more popular, and we have
to cope with the new congestions.
The result of a questionnaire about route choice behaviour shows the criteria such as
cognitive resource demands minimization is major one when a destination is unfamiliar. The
simulation result shows 1) the existence of the criteria, 2) the effectiveness of VICS
information for reduction of travel time and 3) the possibility of route concentration caused
by VICS information.
We have to analyse other case, such as where a destination is familiar, and study the
method how to formulate the criteria in order to embed in a traffic flow simulator.
 H. Shimoura et al., Evaluation of the Effect of DRGS using Traffic Flow Simulation System, the Second
World Congress on Intelligent Transport Systems, 1995.
 T. Saito et al., Analysis about route choice characteristics of the driver by questionnaire investigation, Proc.
of IEEJ Meeting on Road Transportation (RTA-98-27), 1998, pp.25-29 [in Japanese].
 T. Hiraoka et al., A Study of Dynamic Route Selection Based on Simplified 3D Traffic Flow Simulator,
Proc. of 40th SICE Annual Conference, CD-ROM, 2001 [in Japanese].