Environment and Planning A 2004, volume 36, pages 1673 ^ 1686

DOI:10.1068/a36138

The choice of park and ride facilities: an analysis using

a context-dependent hierarchical choice experiment
Ilona D M Bos

Faculty of Technology, Policy and Management, Department of Transport Policy and Logistics
Organization, Delft University of Technology, The Netherlands; e-mail: [email protected]

Rob E C M Van der Heijden

Nijmegen School of Management, Human Geography and Spatial Planning, University of

Nijmegen, The Netherlands; e-mail: [email protected]

Eric J E Molin

Faculty of Technology, Policy and Management, Department of Transport Policy and Logistics
Organization, Delft University of Technology, The Netherlands; e-mail: [email protected]

Harry J P Timmermans

Faculty of Architecture, Building and Planning, Urban Planning, Eindhoven University of Technology, The Netherlands; e-mail: [email protected]
Received 7 July 2003; in revised form 18 December 2003

Abstract. Park and ride (P&R) facilities have been proposed in several countries to alleviate the
accessibility problems in cities. Despite growing accessibility problems, these facilities do not seem
to attract the expected number of car drivers and are underused. In an attempt to measure consumer
evaluations of the attributes of P&R facilities, a stated choice experiment, based on the method of
hierarchical information integration, was conducted in the city of Nijmegen, The Netherlands. This
paper documents the major results of this study, which differs from previous research in that a large
range of attributes is examined, including accessibility of the facility, the quality of the facility, and the
features of connecting transport. In addition, context variables affect the decisionmaking process of
car drivers such as weather, having heavy luggage and travel purpose were incorporated in the study
design. The results indicate that social safety, quality of the connecting public transport and relative
travel times by transport modes are key attributes to the success of P&R facilities. Contextual
variables seem to have only a minor impact.

1 Introduction
Growing car use in and around cities in industrialized countries has led to increased
accessibility problems, reflected in traffic jams and parking problems. To alleviate this
problem, park and ride (P&R) facilities on the outskirts of cities, where car drivers can
switch to connecting public transport, have been proposed in the Netherlands and in
several other countries. P&Rs are supposed to combine the advantages of a highquality public transport system in densely populated cities with those of the car in
more thinly populated areas. However, despite the growing accessibility problems,
P&Rs do not seem to attract the expected number of car drivers.
In an attempt to identify the reasons why P&Rs are underused, several evaluation
studies have been conducted in the Netherlands (for example, Ministry of Transport,
2000; Muconsult, 2000; Van der Heijden and Molin, 2002; Van der Heijden et al, 2000).
However, such evaluations face the challenge of how to collect the data and how to
infer the decisionmaking process of car drivers. The decision whether or not to choose
the facility is quite complex, and involves at least the accessibility of the facility, the
quality of the facility, the features of the connecting transport, and the availability of
information. Furthermore, certain context variables affect the decisionmaking process
of car drivers, for example, weather, having heavy luggage, and of the purpose of travel.

1674

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

Given this large number of attributes and context variables, it is not easy to disentangle
the effects of all these attributes on the choice decision.
Past research has typically relied on (a combination of ) revealed preference (RP)
and stated preference (SP) approaches. Ghali et al (1997), Guan and Nishii (2000),
Lo and Lam (2001), and Van der Heijden and Molin (2002) used an SP experiment, but
their results are of limited use as they varied only time and costs attributes. Bradley et al
(1993) estimated joint RP/SP models, but also considered only time and costs attributes.
Some studies have been carried out which have included more than time and costs
attributes in their experiment. Polydoropoulou and Ben-Akiva (2001) estimated joint
RP/SP models wherein not only the cost and time attributes, but also comfort attributes were included. Besides the time and cost aspects, they found that the number
of transfers is most important, followed by mode of transport and the probability of
having a seat. The probability of delay still has a significant influence but is least
important. Van der Waerden et al (1997) applied an SP approach, using a larger but
still limited set of attributes. They found that time and cost attributes are of great
importance whereas additional provisions do not significantly contribute to the choice
to use or not to use the P&R facility. However, social safety aspects and the quality of
the connecting public transport (with the exception of the waiting time at the P&R)
were not included in their model. Moreover, the quality of the door-to-door car
alternative is not included in this study, which means that this model is not able to
measure the influence of, for example, parking costs on P&R choice behaviour. Finally,
Van der Heijden et al (2000) incorporated in their SP study the probability and the
costs of parking at the destination, but social safety aspects were not taken into
account. They also found that the time and costs aspects are the most important
attributes and the additional provisions least important.
Because of the inclusion of a limited number of attributes in the models, the scope
of previous studies seems too limited to understand fully the influence of the large
set of potentially relevant attributes. As a consequence, developing and planning P &R
facilities cannot be fully based on the results of these studies. To that end, the effect of
a more complete set of attributes describing P&R facilities, the quality of connecting
public transport, and the characteristics of the destination should be measured. In
addition, modal choices may be dependent on temporal conditions such as weather
or having heavy luggage. Hence the effects of these temporal conditions, also referred
to as context variables, should be studied as well. As only a relatively small number of
P&R facilities have been built, the RP approach will not provide the required insight
because of the limited variety in P&Rs. Therefore, the SP modelling approach seems the
only possible one for the current state of affairs. SP approaches rely on the preferences
or choice expressed by respondents for hypothetical choice alternatives. Because the
researcher has complete control over the covariance between attributes of the choice
alternatives, a preference or choice model can be estimated efficiently. However, the
application of an SP modelling approach can pose problems as a large number of
attributes influences P&R choice, whereas traditional SP experiments can handle only
a limited number of attributes.
The aim of this paper is to present the results of a stated choice experiment to model
P&R choice in the city of Nijmegen in the Netherlands. Because of the large number
of potentially influencing attributes, the hierarchical information integration (HII)
approach, originally proposed by Louviere (1984), was applied. The HII approach
is an extension of the traditional information integration approach (better known
as stated preference in transportation research) and allows one to handle a large
number of attributes. This approach has been successfully used in other applications,
including the choice of retail facility (Louviere and Gaeth, 1987), recreation destination

The choice of park and ride facilities

1675

choice (Louviere and Timmermans, 1990), residential choice (Molin, 1999; Molin et al,
2003), and freight mode choice (Norojono and Young, 2003). To the best of our knowledge, this is the first application of HII to model passenger mode choice behaviour.
This paper is organized as follows. First, the method of HII is briefly introduced to
those readers not familiar with this methodology. Next, the research design and data
collection are discussed, followed by a discussion of the analyses and results. Finally,
some conclusions are drawn and implications for P&R policies and the planning and
development of such facilities are discussed.
2 Research design
2.1 Methodology: the hierarchical information integration approach

Louviere (1984) proposed the HII approach as a way to deal with complex decision
problems that involve many attributes. He assumed that, when individuals are confronted with decisions that involve many attributes, they process information in a
hierarchical manner. Individuals are assumed first to group the attributes into higher
order decision constructs. Then they evaluate each construct separately, and finally
integrate these evaluations to arrive at an overall preference or choice. In line with
these assumptions, a separate experiment for each decision construct is devised to
estimate the contribution of each attribute to the evaluation of the corresponding
higher order decision construct (see figure 1). In addition, a bridging experiment is
devised to estimate how the evaluations of the higher order decision constructs
are integrated to arrive at an overall preference (Louviere, 1984; Louviere and
Timmermans, 1990). Based on the responses observed in each experiment, a utility
function can be estimated for each decision construct. Furthermore, an overall integrative model can be estimated which links the separate construct evaluations with the
overall evaluation or choice.

.
.
.

Attribute K

Attribute grouping G
Attribute 1
Attribute 2
..
.

Individuals'
judgment or score
for an attribute
combination
(profile) in
grouping K

Predicted
scores

Score on judgment
dimension used in
grouping 2

.
.
.

Attribute grouping 2
Attribute 1
Attribute 2
..
.

Score on judgment
dimension used in
grouping 1

Individuals'
overall
judgment
response on
dimension(s)
of interest

.
.
.

Predicted
scores

Attribute J

Individuals'
judgment or score
for an attribute
combination
(profile) in
grouping 2

Attribute I

Predicted
scores

Overall response
dimension of
interest

Individuals'
judgment or score
for an attribute
combination
(profile) in
grouping 1

Experiment(s) 2

Attribute grouping 1
Attribute 1
Attribute 2
..
.

Model(s) of
experiment(s) 1

Response(s) to
experiment(s) 1

Experiment(s) 1

Score on judgment
dimension used in
grouping K

Figure 1. Flowchart of proposed hierarchical judgment process (source: Louviere, 1984).

1676

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

2.2 Construction of the experiments

The first stage in applying the HII approach is concerned with the grouping of the total
set of potentially relevant attributes into logical or at least useful decision constructs.
In previous studies, the grouping of attributes has typically been a subjective decision of
the researcher, based on a literature review or logic. Because such subjective decisions
may not reflect the grouping process of travellers, we decided to conduct a separate
project that aimed to identify the higher order constructs used by travellers when
choosing or evaluating P&R facilities and the relationships between attributes and
higher order decision constructs. The results of this project are reported in detail
elsewhere (Bos et al, 2002; 2003). Using Likert scales, and multidimensional scaling
(MDS), we identified five decision constructs.
The first construct is `parking', and included attributes about information, the
chance of finding a parking place, the possibility of reserving one, and the walking
distance from car to public transport. The second decision construct was labelled
`P&R facilities', and was defined by social safety attributes, such as supervision at
the P&R, a lighted pedestrian route, and liveliness at the P&R facility, and attributes
about additional provisions, such as a heated waiting room or a supermarket. The
third construct is `connecting public transport', and was composed of attributes
about the reliability of public transport and attributes about comfort. Finally, the
constructs `time' and `costs' were identified. First `time' was related to attributes such
as time needed to look for a parking place at the destination, the amount of traffic
in the city, and the extra travel time from the principal road to the P&R. `Costs' was
defined by attributes such as total costs of transferring, costs of road pricing, and
parking costs at the destination.
A straightforward application of the HII approach would thus result in six experiments, one for each of the five decision constructs, and one integrative, bridging,
experiment. However, as the total task load for respondents may become too high,
the number of experiments and the number of attributes was further reduced by:
(1) deleting the attributes whose importance scores were very low in the preliminary
research, (2) assuming that modern P&R designs would always satisfy certain attribute
levels, making several attributes superfluous, and (3) a closer inspection of the likely
covariance of attributes, allowing us to combine some attributes which are considered
to be similar according to the preliminary research. These considerations led to the
following simplifications. First, as only a single attribute of the original `parking'
construct attributes was left, this attribute was merged into the `P&R facilities' construct. In some sense, this operational decision was also supported by the MDS
solution showing that the `parking' and the `P&R facilities' constructs were close in
the MDS solution. Moreover, the MDS solution shows that time and costs aspects
may also be regarded as constructs. From RP studies it is well known that different
time aspects, such as the time needed to reach a public transport stop, waiting time for
the vehicle to arrive, travel time, and time required to reach the final destination, may
be weighed differently (for example, Wardman, 2001). For example, it is often found
that waiting time influences utility more negatively than travel time. Hence, to examine
the trade-off between time aspects, an experiment that varies different time attributes
should be constructed. However, from our own experience in a pilot SP experiment
regarding P&R choice, in which different time aspects were varied, it was found that
these aspects were not weighed significantly differently (Van der Heijden and Molin,
2002). It appeared that respondents simply added up all the different time aspects,
and reacted to the total time. Respondents even asked the interviewer why they had
to do that themselves and indicated that they would have preferred a total time
attribute. The same applied to different cost aspects. Assuming this finding can

The choice of park and ride facilities

1677

be generalized, separate time and cost experiments would not reveal any relevant
insight and only extend the respondents' task, with response fatigue as a likely
consequence. Instead, the time and costs decision constructs were included as total
time and total cost attributes in the choice experiment. One might argue then that
this is not a bridging experiment in the sense that the time and costs attributes are
single attributes and not construct evaluations.
These simplifications meant that three experiments were constructed. One experiment was intended to estimate the contribution of the underlying attributes to the `P&R
facilities' decision construct, another was intended to estimate the influence of attributes
on the `connecting public transport' construct, and there was a bridging experiment.
The construction of these experiments involved combining the underlying attribute
levels into profiles. In order to limit the number of profiles, the `smallest orthogonal
fraction' of the full factorial design was chosen for each experiment. This operational
decision implied that none of the interaction effects could be estimated. Thus, it was
assumed that the part-worth utilities of each attribute level defining a particular
decision construct are added to obtain the overall preference for that decision construct. This decision resulted in eighteen profiles for the `P&R facilities' experiment and
nine profiles for the `connecting public transport' experiment. Respondents were asked
to evaluate each profile on a ten-point rating scale, ranging from very unattractive (1) to
very attractive (10).
In addition to the two construct experiments, a bridging experiment was devised.
Originally, Louviere (1984) framed this experiment as a preference task, but Timmermans
(1989) (see also Louviere and Timmermans, 1992) generalized this approach into a choice
task. The latter approach was followed in this study. It was assumed that travellers have
three options to travel to the city centre. They can drive from door to door, they can
use P&R (that is, drive to the outskirts of the city and then switch to public transport),
or they can use public transport for the whole trip. Consequently, choice sets consisted
of a P&R alternative, a car alternative, and a door-to-door public transport alternative.
The P&R alternative was described by the respondent evaluations of `quality of
connecting public transport', `quality of P&R facilities', `extra total time needed
when using P&R', and `extra total costs when using P&R'. The car alternative was
described by `extra total time when using the car' and `extra total costs when using
the car'. Extra time and extra costs are defined as additional time and costs as
compared with a car trip without delays or parking fees. The door-to-door public
transport alternative was treated as the base alternative, which does not vary
between the choice sets. As for the public transport alternative, travellers had to
assume that it is free of delays.
The choice sets and choice alternatives were simultaneously constructed from a
fractional factorial design, resulting in eighteen choice sets. The four P &R attributes
and the two attributes for the car alternative were assigned to different columns of that
design, ensuring that the attributes are not only orthogonal within but also between the
P&R and car alternatives. The door-to-door public transport alternative was added to
each choice set as a base alternative. Respondents were asked to choose the single
alternative in each choice set that they liked best.
As the choice may be different for certain conditions, the following context variables are also varied in the experiment; (1) weather (dry or rainy), (2) travel purpose
(working or recreational), (3) heavy luggage or no luggage, (4) car-pooling or not, and
(5) the time of the day (daytime or evening). A fractional factorial design of these five
attributes was constructed, resulting in eight context situations. The choice design was
nested under each of the eight situations. As a consequence, eight different versions
of the choice experiment were constructed, to be evaluated by different respondents.

1678

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

2.3 Sampling method and sample characteristics

Data to estimate the model were collected in the city of Nijmegen, a medium-sized city in
the east of the Netherlands. It was selected because it is well known for its accessibility
problems, especially from the north side because of its location directly along the river
Waal. From the north the city is reachable only by the `Waalbridge', which leads to a dense
stream of traffic. Parking problems in the historic centre of Nijmegen further complicate
access to the city. Car drivers who visit Nijmegen for work or recreation were approached
in this area. Hence, the target group was formed by car drivers who live outside the city of
Nijmegen and work or spend their free time in Nijmegen on a regular basis.
The target group was approached in two different ways. First, in the historical
centre of Nijmegen and in another major, suburban, shopping centre in Nijmegen
car drivers who had just parked their cars were approached by interviewers and asked
whether they were willing to fill out a questionnaire. If they said they would, interviewers checked whether they belonged to the target group and, if so, they were asked
to provide their home address. The questionnaire was mailed to this address, together
with a return envelope. Second, car drivers working in Nijmegen and living outside the
city were approached through a selected number of larger companies. These companies
could choose between sending an e-mail address with a link to an Internet questionnaire or sending a paper version by mail to the home address. The data collection took
place in the second half of June 2002.
Table 1. Response group characteristics.

1 Sex
Male
Female
Missing values
2 Education level
Bachelor's or master's degree
Lower or intermediate education
Missing values
3 Age (years)
18 30
31 50
5 51
Missing values
4 Class of car
City car or compact class
(Compact) middle class
Highly middle class/top class/others
Missing values
5 Ownership of car
Own car
Lease car
Missing values
6 Experience with park and ride facilities
Experience
No experience (rarely or never used)
Missing values
7 Experience with public transport
Experience
No experience (rarely or never used)
Missing values

Absolute
(N 805)

Relative
(%)

439
356
10

54.5
44.2
1.2

445
351
9

55.3
43.6
1.1

154
448
166
37

19.1
55.7
20.6
4.6

253
439
101
12

31.4
54.5
12.5
1.5

730
66
9

90.7
8.2
1.1

285
503
17

35.4
62.5
2.1

567
232
6

70.4
28.8
0.7

The choice of park and ride facilities

1679

In total, 805 people filled out the questionnaire; 500 completed the paper version, and
305 the questionnaire on the Internet. The characteristics of the response group are listed
in table 1, showing that (1) as many men as women filled out the questionnaire; (2) as
many highly educated people filled out the questionnaire as middle or lower educated
people; (3) most respondents were between 30 and 50 years old, but the younger and older
groups were substantially represented as well; (4) most respondents have a (compact)
middle-class car; (5) almost all respondents own their cars; (6) most respondents have
no experience with P&Rs; and (7) seven out of ten respondents have experience with
public transport in general. From these results, there are no reasons to believe it was a
rather unusual group of respondents.
3 Analysis and results
A preference function was estimated for each construct design separately. In addition,
a choice model was estimated using the data of the choice experiment. The results are
discussed in the following subsections.
3.1 Quality of P&R facilities

Regression analysis was used to estimate a main-effects-only preference model. To that

end, the attribute levels were effect coded. This implies that the intercept of the regression
equation is equal to the average rating, whereas the regression coefficients represent the
part-worth utilities of the attribute levels as deviations from this average rating.
The model, presented in table 2 (see over), has been estimated on the basis of
aggregated data. Moreover, the importance of the attributes is expressed as the absolute
difference between the utility of the highest valued attribute level and the utility of the
lowest valued attribute level. The R 2, being a measure of the goodness of fit of the model,
is equal to 0.996, which means that the model has an excellent fit. However, when
interpreting these outcomes, one should realize that the degree of freedom is small.
Table 2 clearly shows that the three attributes related to the general notion of social
safety are more important than the facility attributes. `Supervision at P &R' is obviously
the most important attribute. When cameras are placed at the P&R facility, attractiveness increases by almost one point, which is a substantial increase considering the fact
that a ten-point scale has been used. The additional presence of supervisors also
increases the attractiveness, albeit to a much smaller extent. Clean, well-maintained
P&Rs also contribute positively to the evaluation of P&Rs, but graffiti affects utility less
negatively than might be expected. As expected, surveyable and lively pedestrian routes
also increase P&R attractiveness.
Of the facilities attributes, the attributes `additional provision', `walking time', and
`waiting room' are almost equally important, while `paying facilities' are clearly least
important with none of the part-worth utilities significant at the 5% level. However,
`paying facilities' are not really unimportant, but additionally estimated individual models
(not presented in this paper) reveal that preferences differ widely between the respondents.
Some respondents reacted positively towards paying electronically with a chip card,
whereas others prefer a paying machine or manned ticket service, with the result that
effects averaged out at the aggregate level. The data indicate that the availability of a
supermarket influences P&R evaluation positively. Furthermore, it is remarkable that
`walking distance from car to public transport' appears to be of relatively minor importance to the traveller. This result may be plausible when considering the fact that the range
of times was only up to 5 minutes. A greater range for that attribute might have
demonstrated a stronger effect on P&R choice. However, in practice, walking distances
longer than 5 minutes are not plausible. Apparently, travellers are not really unwilling to
walk a few minutes. Finally, as expected, a heated waiting room adds to the overall utility.

1680

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

Table 2. Part-worth utilities quality of park and ride facilities (N 805).

Attribute
Overall utility (intercept)

Part-worth
utility
4.99

1 Supervision
No supervision
0.75
Cameras
0.21
Cameras and supervisors
0.54
2 Maintenance
Clean, good state of repair
0.66
Holes in asphalt
0.26
Graffiti and holes in asphalt
0.40
3 Pedestrian route from car to public transport
Obscure and abandoned
0.48
Surveyable but abandoned
0.03
Surveyable and lively
0.46
4 Additional provisions
No additional provisions
0.32
Kiosk
0.04
Supermarket
0.35
5 Walking time from car to public transport (minutes)
1
0.28
3
0.10
5
0.38
6 Waiting room
No waiting room
0.34
Covered but unheated
0.03
Covered and heated
0.31
7 Paying facilities
Pay machine
0.11
Manned ticket service
0.00
Electronic with chip card
0.11

Significance
level

Importance
(rank order)

0.00
0.00
0.02
0.00
0.01
0.00
0.61
0.01
0.50
0.01
0.15
0.01
0.63
0.12
0.96

1.29 (1)

1.06 (2)

0.94 (3)

0.67 (4)

0.66 (5)

0.65 (6)

0.22 (7)

R2 0:996.

In general, it can be concluded that all part-worth utilities were in the anticipated direction,
giving face validity to the estimated model.
3.2 Quality of connecting public transport

Linear regression analysis was also applied to estimate the influence of the relevant
attributes on the evaluation of the public parking attribute. The results of the model,
also based on aggregated data, are presented in table 3. Because a saturated model was
estimated, the model fits the data perfectly and therefore significance levels are not
provided.
Table 3 clearly shows that `certainty of a seat' is the most important attribute. An
explanation might be that the car drivers who completed the questionnaire are
used to having comfort in their cars. The second most important attribute is `number
of transfers'. The improvement in going from one unguaranteed to one guaranteed
transfer increases utility by the same amount as going from a guaranteed transfer to
no transfers. Next in importance is `frequency of connecting public transport'. This
result suggests that, if the frequency decreases below once every 10 minutes, utility
decreases rapidly. By far the least important attribute is `mode of public transport'.
However, further individual-level analysis of the data reveals that the preference
differences average out between the respondents at the aggregate level.

The choice of park and ride facilities

1681

Table 3. Part-worth utilities quality of connecting public transport (N 758).

Attribute
Overall utility (intercept)
1 Certainty of a seat (percentage chance)
5
50
95
2 Number of transfers
0
1 guaranteed
1 unguaranteed
3 Frequency of connecting public transport
Once in 5 minutes
Once in 10 minutes
Once in 15 minutes
4 Public transport mode
Metro or train
Tram or light rail
Bus

Part-worth
utility

Importance
(rank order)

5.36
1.04
0.11
0.93
0.77
0.00
0.77
0.44
0.11
0.55
0.04
0.08
0.13

3.3 Choice between P&R, car, and door-to-door public transport

1.97 (1)

1.54 (2)

0.99 (3)

0.17 (4)

Having estimated the part-worth utility functions, we next estimated a multinominal

logit model of P&R choice. To that end, the relative choice frequencies were aggregated across respondents for each choice set. In the choice model, not only were
the attributes describing the P&R and car alternatives considered, but the influence
of context variables was also taken into consideration. In this way, not only were
the part-worth utilities for the attributes derived but also those for the context
variables.
The r 2 of this model is equal to 0.793, which suggests that the estimated model performs well. Comparing the model with context variables [log-likelihood (B) 822:797]
with the model without context variables [log-likelihood(B) 1010:69] suggests that
context variables significantly affect the choice of P&R facilities, as a w 2-value of 375.8
can be calculated with a p-value of 0.000. The part-worth utilities for the attributes,
together with significance levels and importance scores are presented in table 4 (see
over), and those for the context variables in table 5 (over).
Because door-to-door public transport constituted the base alternative, travellers'
utility of public transport is zero by definition, and other alternatives are compared
with public transport. The positive value of the intercepts both of the P&R alternative
and of the car alternative suggests that, on average, these two alternatives are preferred
to the public transport alternative. The intercept of the car alternative is higher than
the intercept of the P&R alternative, indicating that, on average, the car is preferred
to the P&R alternative.
When the part-worth utilities and the importance scores of the attributes are
examined, it seems that the time attributes have the highest impact on choice behaviour. Taking into account the range differences, extra time when using P&R (the
utility of each minute of extra travel time per P&R decreases by 0.053) influences
choice behaviour slightly more than extra time using a car (the utility of each minute
of extra travel time per car decreases by 0.049). Additional analyses indicated that
this difference is statistically significant.

1682

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

Table 4. Part-worth utilities of attributes in the choice model (N 708).

Attribute

Part-worth
utility

Intercept park and ride

1 Extra time using park and ride (minutes)
0
10
20
2 Extra costs using park and ride ( )
0
2
4
3 Quality of park and ride
Value 4
Value 6
Value 8
4 Quality of public transport
Value 4
Value 6
Value 8
Intercept car
5 Extra time using car (minutes)
0
20
40
6 extra costs using car ( )
0.50
3.50
6.50

Significance
level

0.52

0.00

0.46
0.15
0.61

0.00
0.00

0.49
0.05
0.54

0.00
0.14

0.48
0.02
0.47

0.00
0.61

0.39
0.02
0.40

0.00
0.64

1.31

0.00

1.00
0.04
0.96

0.00
0.15

0.48
0.02
0.50

0.00
0.00

Importance
(rank order)
1.07 (2)

1.03 (3)

0.95 (5)

0.79 (6)

1.96 (1)

0.98 (4)

Table 5. Part-worth utilities of the context variables in the choice model (N 708).
Context variables

Luggage
No luggage
Heavy luggage
Travel purpose
Work
Recreation
Passengers
None
One
Weather
Good
Rainy
Time of day
Daytime
Evening

On P&R

On car

Between P&R and car

partworth
utility

significance
level

partworth
utility

significance
level

partworth
utility

significance
level

0.25
0.25

0.00

0.39
0.39

0.00

0.06
0.06

0.00

0.32
0.32

0.00

0.30
0.30

0.00

0.00
0.00

0.95

0.07
0.07

0.02

0.18
0.18

0.00

0.04
0.04

0.00

0.08
0.08

0.00

0.02
0.02

0.54

0.04
0.04

0.00

0.05
0.05

0.13

0.00
0.00

0.94

0.02
0.02

0.11

The choice of park and ride facilities

1683

When we look at the part-worth utilities of the time attributes, we see that an
increase in extra time per P&R of more than 10 minutes affects P&R choice more
negatively than an increase of less than 10 minutes, whereas the increasing extra time
of the car alternative influences utility almost linearly. Thus, if using P&R takes more
than 10 minutes extra time, time loss due to traffic jams will cause a lower percentage
of car drivers to switch to P&R than is the case when the extra time of using P&R is
quite limited.
For the costs attributes, each additional euro for using P&R influences choice
behaviour more negatively than a comparable increase in car costs (the utility of each
euro of extra travel costs is 0.249 for P&R and for the car 0.173). Furthermore, utility
decreases linearly with increasing P&R costs. This is not true for the case of increasing
car costs: the increasing car costs above 3.50 affect utility more than an increase
below 3.50.
Finally, the quality of P&R facilities has a slightly higher impact on P&R choice
than quality of public transport (the utility of each value of extra quality is 0.240 for
P&R and 0.174 for connecting public transport). For each construct, utility increases
linearly with increasing quality of P&R and quality of connecting public transport.
Hence, every increase in quality of both constructs in any part of the evaluation scale
affects choice in the same way.
3.4 Effect of context variables on choice

Table 5 shows the effects of context variables on mode choice. In the first two columns
of the table the effects of the context variables on the P&R alternative are presented,
with their significance levels. In the next two columns the effects of the context
variables on the car alternative are shown and in the last two columns the difference
of the effects of context variables on the P&R and car alternatives are presented.
The table shows, first, that car drivers with heavy luggage are more likely to use
both P&R and the car than are car drivers without luggage. When comparing P&R with
car, it is observed that the utility of the car alternative increases in a significantly
greater degree in case of heavy luggage than the P&R alternative. These results could
be explained by the fact that car drivers prefer to use the car for (a part of ) the trip
when they have to carry heavy luggage. Moreover, car drivers travelling to work are
less likely to use P&R and the car than are car drivers travelling for recreational
purposes. This might be explained by the fact that people are less likely to use a
P&R facility every day than to use the P&R occasionally. No significant differences
appear when the P&R alternative is compared with the car for travel purposes. Furthermore, car drivers without a passenger are less likely to use P&R instead of the car.
However, the P&R is preferred to door-to-door public transport when one is travelling
with passengers. This reflects the notion that one feels more comfortable using public
transport if one is not alone. In addition, the P&R is preferred to both the car and
door-to-door public transport in the case of bad weather. This may reflect that, under
such conditions, congestion in and around cities may be worse and door-to-door public
transport uncomfortable. Finally, considering the influence of the time of day, neither
the influence on the P&R alternative nor on the car alternative is significant.
Overall, it can be concluded that the importance scores do not diverge that much
between attributes. All attributes, including those being considered in the choice model,
contribute substantially to the choice behaviour of the traveller. When the influence of
context variables is compared with the influence of the attributes describing costs,
times, and quality of P&R and connecting public transport, the context variables
seem to have a smaller influence. However, the influence of travel purpose on the
P&R and car alternatives and the influence of heavy luggage on the car are larger.

1684

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

4 Conclusions
In this paper the car drivers' preferences for P&R facilities were traced by applying the
HII approach. A stated preference model was measured for both `quality of P&R
facilities' and `quality of connecting public transport'. These models were then integrated
into the choice model, with the part-worth utilities required to estimate the percentages
of use of P&R, car, and door-to-door public transport. In addition, the effects of the
context variables for weather, heavy luggage, passengers, travel purpose, and time of
the day on decisionmaking were estimated. All estimated part-worth utilities were in the
expected directions and the goodness-of-fit indices indicated a high model fit. This gives
confidence in the estimated model and also in the application of the HII approach to
model mode choice.
The results of applying the HII method, first of all, show that much attention
should be paid to social safety aspects, such as supervision by cameras and a safe
pedestrian route. Additional provisions (for example, a heated waiting room or a
supermarket) are less important in respondents' evaluation of the quality of a P&R
facility. The success of a P&R facility is also influenced by the quality of public
transport connecting the P&R with the destination of the traveller. The results show
that the certainty of a seat was the most important attribute in the decision construct
`quality of connecting public transport'. Because this is the most important public
transport quality attribute, seat availability should be taken into full account when
the public transport from P&R facilities is planned. For example, one should be very
careful about planning an additional stop at the P&R facility or diverting existing
regional bus or train services, because those vehicles can already be rather crowded
when they arrive at the outskirts of the city. As an alternative, separate shuttles from
the P&R facility to the city centre should be considered.
The part-worth utilities of the attributes in the choice model show that the car
driver is more willing to use P&R if the extra travel time or extra costs related to car use
are high. Matched with actual developments, this indicates that in the future an
increasing number of drivers will use P&R facilities more regularly because inbound
congestion is still getting worse in many cities. Moreover, inner cities are increasingly
introducing restrictive parking policies, either by putting the building of new parking
facilities on hold or by increasing parking fees. Such policies mean that more time is
needed to find cheap parking lots and more time is required to walk to the final
destination. These measures will make it increasingly unattractive to enter inner
cities by car and, hence, will stimulate drivers to use alternative transport modes,
including P&R.
The willingness of car drivers to use P&R also increases if the extra travel time
when using P&R is low. Offering high-speed connecting public transport, realized, for
example, by a dedicated (bus) lane to the city and enabling efficient transfer at the P&R
facility could produce this low extra travel time. The results of the analyses indicate
that the type of transport is not relevant in this context. Moreover, the extra costs of
P&R (such as parking costs at the P&R facility and costs of the connecting public
transport) should be kept relatively low.
Most of the conclusions regarding the influence of the attributes considered on
P&R choice are consistent with earlier findings, although a larger set of attributes were
included in the present study. Attributes about time and costs appear to be most
important whereas additional provisions do not have a large influence on P&R choice.
However, one remarkable result was obtained. In this study, the chance of having a
seat is the most important attribute describing the quality of the connecting public
transport, whereas Polydoropoulou and Ben-Akiva (2001), who estimated joint RP/SP
models, found that the number of transfers and the mode of transport had a larger

The choice of park and ride facilities

1685

influence on P&R choice. The difference in influence of the mode of transport is

explainable by the fact that in their study the frequency of the connecting public
transport was not included and this is assumed to be larger in the case of a metro
than of a tram or a bus. Moreover, the range of attribute levels which is chosen
influences the ranking results of the attributes. Polydoropoulou and Ben-Akiva's study
gives limited insight into the exact attribute levels which were varied in the experiment.
When the influence of context variables is compared with the influence of the
attributes describing costs, times, quality of P&R, and quality of connecting public
transport, the context variables seem to have a smaller influence. However, some
relevant recommendations can be derived from the effect of context variables on
decisionmaking. For instance, historic cities often attract many visitors, who visit
the city infrequently for recreational purposes. According to our findings, these visitors
are relatively willing to use P&R. The same holds true for visitors to special events or
certain attractions. The design and `selling' of P&R facilities could therefore focus on
these target groups.
Although there seem to exist possibilities to improve the success of P&R facilities
for such specific target groups, it remains important to alleviate day-to-day traffic
management problems to stimulate commuters to use these facilities. We have argued
that the success of such facilities depends on the extent to which policymakers and
marketing have been successful in meeting user preferences. The results of this study
have identified the most critical attributes in this regard. It shows that the success
of P&R facilities depends ultimately on integral policies that not only improve the
quality of the P&R facility as such but also improve the quality of the full multimodal
chain and discourage the use of the car in the city centre. It would be interesting to
analyze whether P&R facilities designed in terms of such an integral policy perspective indeed perform better than the first generation of P&R facilities, which were
typically developed from a `we build it and they will come' perspective.
References
Bos D M, Molin E J E, Van der Heijden R E C M, 2002, ``User-based definition of higher order
decision constructs to model preferences for P&R transfer points'', in Urban Transport VIII.
Proceedings of Urban Transport and the Environment in the 21st Century Eds L J Sucharov,
C A Brebbia, F Benitez (WIT Press, Southampton) pp 417 ^ 426
Bos D M, Molin E J E, Timmermans H J P, Van der Heijden R E C M, 2003, ``Cognitions and
relative importances underlying consumer valuation of park and ride facilities'' Transportation
Research Record number 1835, 121 ^ 127
Bradley M A, Kroes E, Hinloopen E, 1993, ``A joint model of mode/parking type choice with
supply-constrained application: transportation planning methods'', in Proceedings of Seminar D
held at the PTRC European Transport Highways and Planning 21st Summer Annual Meeting,
13 ^ 17 September, University of Manchester Institute of Science and Technology (PTRC,
London) volume P363, pp 61 ^ 73
Ghali M O, Pursula M, Milne D, Keranen M, Daleno M, Vougiokas M, 1997, ``Assessing the
impact of integrated trans modal urban transport pricing on modal split: transportation
planning methods, volume 1'', in Proceedings of Seminar E held at PTRC European
Transport Forum, Brunel University, England, 1 ^ 5 September (PTRC, London) volume
P414, 1997/09, pp 341 ^ 352
Guan H Z, Nishii K, 2000, ``A modelling method for estimating the P&BR demand'', in Proceedings
of the 2nd International Conference on Traffic and Transportation Studies (ICTTS 2000),
No jiaotong University, Beijing, Peoples Republic of China, 31 July ^ 2 August (American Society
of Civil Engineers, Preston, VA)
Lo H P, Lam W, 2001, ``A latent class model applied to stated preference data'', in Travel Behaviour
Research: The Leading Edge Ed. D A Hensher (Pergamon Press, Oxford) pp 425 ^ 438
Louviere J J, 1984, ``Hierarchical information integration: a new method for the design and analysis
of complex multiattribute judgement problems'' Advances in Consumer Research 11 148 ^ 155

1686

I D M Bos, R E C M Van der Heijden, E J E Molin, H J P Timmermans

Louviere, J J, Gaeth G J, 1987,``Decomposing the determinants of supermarket choice: a hierarchical

information integration approach'' Journal of Retailing 63(1) 25 ^ 48
Louviere J J, Timmermans H J P, 1990, ``Using hierarchical information integration to model
consumer responses to possible planning actions: recreation destination choice illustration''
Environment and Planning A 22 291 ^ 308
Louviere J J, Timmermans H J P, 1992, ``Testing the external validity of hierarchical conjoint
analysis models of recreational destination choice'' Leisure Science 14 179 ^ 194
Ministry of Transport, 2000 Transferia: een handreiking bij voorbereiding en realisatie [P&R facilities:
assistance in preparation and realization] Ministry of Transport, Public Works and Water
Management, The Hague
Molin E J E, 1999 Conjoint Modeling Approaches for Residential Group Preferences dissertation,
Faculty of Architecture, Building and Planning, Eindhoven University of Technology,
Eindhoven
Molin E J E, Oppewal H, Timmermans H J P, 2003, ``Conjoint modeling of residential group
preferences: a comparison of hierarchical information integration approaches'' Journal of
Geographical Systems 14 343 ^ 358
Muconsult, 2000 Evaluatie transferia (Module I) [Evaluation of P&R facilities (module 1)]
Muconsult, Stadhuisplein 4, 3811 LM, Amersfoort
Norojono O, Young W, 2003, ``A stated preference freight mode choice model'' Transportation
Planning and Technology 26 195 ^ 212
Polydoropoulou A, Ben-Akiva M, 2001, ``Combined revealed and stated preference nested logit
access and mode choice model for multiple mass transit technologies'' Transportation Research
Record number 1771, 38 ^ 45
Timmermans H J P, 1989, ``Een decompositioneel hierarchisch model voor woningkeuze: theorie
en illustratie'' [A decompositional hierarchical model for residential choices: theory and
illustration], in Methoden voor Woning en Woonmilieubehoefte Onderzoek Ed. J Musterd
(SISWO, Amsterdam) pp 46 ^ 72
Van der Heijden R E C M, Molin E J E, 2002, ``Locating P&R facilities based on travel behaviour:
a Dutch case study'', in Urban Transport VIII. Proceedings of Urban Transport and the
Environment in the 21st century Eds L J Sucharov, C A Brebbia, F Beritez (WIT Press,
Southampton) pp 733 ^ 742
Van der Heijden R E C M, Molin E J E, Bos D M, 2000, ``Parking at a distance: option for reducing
traffic and parking pressure in urban areas?'', in Urban Transport VI. Proceedings of Urban
Transport and the Environment for the 21st Century (WIT Press, Southampton) pp 145 ^ 177
Van der Waerden P, Borgers A, van Schaijk E, 1997, Parkeergelegenheden met overstapfaciliteiten
in het stedelijk verkeer'' [Parking lots with transfer facilities in urban transportation], in
Colloquium Vervoersplanologisch Speurwerk 1997. Sprong in het duister? Lange termijn
ontwikkelingen in het vervoersplanologisch onderzoek, Amsterdam, 27 ^ 28 November 1997
Part III (Transportation Planning Research Colloquium, Amsterdam) pp 1523 ^ 1540
Wardman M, 2001, ``A review of British evidence on time and service quality valuations''
Transportation Research, Part E 37 107 ^ 128

2004 a Pion publication printed in Great Britain