Supporting policy analysis in the Dutch
rail sector using System Dynamics
A. Mannaerts’, C.E. van Daalen’, J.van Luipen’, S.A. Meijer’
1Delft University of Technology, Faculty of Technology, Policy and Management,
P.O. Box 5015, 2600 GA Delft, The Netherlands
2 ProRail, P.O. Box 2038, 3500 GA, Utrecht, The Netherlands
email: c.vandaalen@tudelft.nl
With a sizeable expected growth of demand for rail transport in the Netherlands in the
coming decades, and limited resources for expansion of the rail network, intensified
utilization of the infrastructure is to be expected. To adequately manage this growth,
appropriate tools for policy analysis are needed. The possibilities and pitfalls of using
System Dynamics for policy analysis in the Dutch rail system have been explored by
performing a modelling study into the interrelations of modal split, mobility and
operations using System Dynamics. Additional scrutiny is placed on the method,
because of the unstructuredness of many problems in the rail sector, and decision-
making in a network type environment. Results show that the reliability of
infrastructure is a major component in the extent of delays. Furthermore, the effect of
unreliability in a train trip and the characteristics of a car trip are important for the
choice between train and car. Although classical policy analysis has proven to be
possible, modelling the operational part of the system has proven challenging due to
the spatial and discrete characteristics of parts of the system. Recommendations are
given to improve the model and model use to better suit the unstructuredness of the
problems.
Keywords: Rail System, Netherlands, Policy Analysis, System Dynamics
1. Introduction
The Netherlands has one of the heaviest utilized railway networks in the EU (CBS,
2009). In 2006, trains travelled over 135 million kilometres on the network. This traffic
is mainly generated by passenger trains, which account for 80% of all reserved train
paths. Combined passenger and freight train paths total around 2.5 million each year
(ProRail, 2011). All this is done on a network which in 2004, was only 2,796 km long,
and which consisted of 6,517 km of track. The Dutch railway system is very complex,
due to its heavy utilization and network design (ECMT, 2005), organizational and
institutional arrangements (Tijdelijke Commissie Onderhoud en Innovatie Spoor,
2012), and number of stakeholders (ProRail, 2011).
In the Netherlands a lot of train movements take place on a relatively small network.
Additionally the structure of the rail network in the Netherlands adds to the overall
complexity. The network can best be described as having a polynuclear structure, with
several cores. This creates a criss-cross of traffic between and inside agglomerations
(Nijman, 2012a, 2012b). To complicate matters more, both local and intercity trains
operate on the same network. They must share the same infrastructure, complicating
operations further. First of all, local problems can spread through the network because
of local and intercity trains influencing each other. Secondly local trains have a lower
average speed than intercity trains. Speed differences on a railway track severely
influence the capacity of the track.
The coming decade a further growth of traffic is expected, and the rail infrastructure
manager of the Netherlands, ProRail, has set itself the goal to increase the capacity of
the network by 50% in 2020 (ProRail, 2012). With only limited financial resources and
an already complex network the goal is to achieve this increase in capacity by more
efficient planning and scheduling of railway traffic (MinlenM, 2011). Measures to
increase capacity through heavier utilization of the network can harm the robustness
of the network. Both may be achieved, but at a very high cost. The real challenge
therefore is striking a new balance between capacity, costs and robustness.
Policy Analysis and the resulting decision making process takes place in an
environment which can be described as a network. The rail sector in the Netherlands
has a separation of infrastructure manager and train operators. These organizations
are independent of the Ministry of Transport, although the ministry has the tools and
obligation to steer the sector.
The Dutch railway system is complex in many ways: whether you look at the technical
infrastructure, organizational layout, operational planning, (number of) actors (i.e.
stakeholders) involved, goals to be reached or decisions to be made; all of these are
complex in themselves. Due to the high interdependence of all these parts the overall
picture is even more complicated, and in this environment sound decisions have to be
made.
Further muddying the waters is the fact that when looking at the policy problems
facing the railway system, these problems can only be described as unstructured.
Unstructured problems are defined as problems where there is no consensus on values
and neither a consensus on knowledge (Hisschemdller, 1993; Hisschemdller & Hoppe,
1995). Although the main actors are all invested in delivering the best train services
possible, the definition of this value ‘best’ may vary. Any policy will be a trade-off
between these values, and all of these values will be weighted differently by the
actors.
For analysing and designing policies in this complex system the System Dynamics (SD)
methodology can be used. It supports not only the design of policies themselves
(Forrester, 1961), but can also help understand the complexity of a system.
Additionally it can also be used in a multi actor environment to communicate about
findings and for collaborative analysis and design. Enhancing learning about complex
dynamics systems is one of SDs major purposes (Sterman, 2001). This can be done by
qualitative analysis of models, but also by using simulation to show users the effects of
their decisions. Feedback is not only used in the models themselves, but is central to
the methodology.
2. Approach
The goal of this research is to explore the possibilities of System Dynamics to better
understand the complexity of the Dutch railway system, by modelling the relationships
in and between sub-systems. This understanding will have to be used and
communicated in a complex multi-actor setting when designing policies.
The use of SD modelling in the rail sector has mostly been limited to the modelling of
vehicles and vehicle interactions. In the last three decades four studies have been
performed into the dynamic effects of the overall railway system. These focussed on:
the effect of maintenance on performance (Gottschalk, 1983); strategic management
with a focus on competitiveness with regard to maintenance and investment strategies
(Schmidt, 1989); a strategic planning model (Homer, Keane, Lukiantseva, & Bell, 1999);
and a study of the performance of the Indonesian railways (Lubis, Pamungkas, & Tasrif,
2005).
In light of the limited literature on SD for analysis of the rail system, an SD simulation
study was undertaken to experience first-hand the pros and cons of using SD for
analysis of the rail system. This was done by modelling the relations between traveller
choice of transport modes and the effect this has on the operations on the network.
The SD approach to this problem facilitated a structured approach to system analysis,
identification of the feedback structure of the modelled system, evaluation of
uncertainty and identification of directions for further policy analysis.
The model itself, the results of qualitative and quantitative analysis, the modelling
process and the results of validation and verification have been used to evaluate the
usability of SD for policy analysis in this specific case. Recommendations will be given
on how the SD methodology can be used for policy analysis in the Dutch rail sector.
The article is structured in sections as follows. Section 3 describes the conceptual
model of the railway system and the most important concepts that have been
included. In section 4 the implementation of the model is discussed as well as
verification and validation. Section 5 discusses the results of simulation and further
quantitative analyses. Section 6 discusses the validity of the model in the context of
policy analysis in a network environment. In section 7 conclusions will be drawn and
recommendations for use of SD in the Dutch rail sector will be given.
3. System Conceptualization
In the model that describes the relations between the choice of travel mode and the
operations on the rail network, three distinct subsystems can be found: one that
describes the modal split, one that describes demand for mobility and one that
describes operations. These subsystems influence each other as depicted in Figure 1.
Each of them will be described briefly. After that a distinction between trip types will
be made. Finally the overall feedback structure will be presented.
Figure 1: General depiction of interrelations between the three subsystems.
Modal split
Given the distance of trips performed by train in the Netherlands, the car is often the
only viable alternative. In the model the modal split therefore represents the ratio
between train and car usage for a certain trip type.
When a traveller wants to take a trip, the modes which are available can be seen as
products that satisfy this need to a certain degree. The characteristics of a product
provides benefits and satisfies needs to varying degrees (Kotler & Armstrong, 2001).
Rating of the train service in the Netherlands has revealed ten unique dimensions on
which passengers rate a trip (Brons & Rietveld, 2009). The three most prominent
characteristics on which trips are rated are: the price-quality ratio, travel comfort and
travel time reliability.
The characteristics are operationalized by: determining the monetary costs of a trip;
the valuation of travel time; and the costs of unreliability. The monetary costs are
determined for a whole trip, including parking costs or costs for access and egress to
stations, if applicable. The valuation of travel time is modelled using the disutility
travellers experience during a trip, which relates the time spent traveling and the
comfort of different part of the trip (Vaessens, Van Hagen, & Exel, 2008; Wardman,
2004). This concepts is graphically depicted in Figure 2. The costs of unreliability in a
trip are modelled by determining the rescheduling costs, which are the costs of early
and late arrival due to unreliability, and takes into account the tendency of travellers
to leave early in order to prevent arriving late at their destination (Brons, 2005). The
higher the unreliability, the higher the rescheduling costs will be.
High
A Amenity
low Time
Figure 2: The disutility of time as experienced by a railway passenger. Adapted from: (Peek & van
Hagen, 2004)
Mobility
Of the total amount of kilometres travelled in the Netherlands, only a small amount is
done by train. Based on the feasibility of making a trip by train three groups can be
distinguished: the train is unfeasible (car captives); the train is an option; the train is
the only possibility (train captives) (Van Hagen, 2011). Of the trips in which the train is
an option, about 9.5% is actually done by train. The distribution of mobility by
feasibility is shown in Figure 3.
No train market
Objective car captives
Hard barriers
Subjective car-captives
Choice traveler: car
Choice traveler train
Train capti
Figure 3: Distribution of mobility in km per year. The trips can either not be done by train (red),
possibly be done by train (yellow) or only be done by train (green). Adapted from: (Van Hagen, 2011)
Operations
The demand for transport by rail leads to a capacity requirement which must be
fulfilled by train services on the rail network. Additional equipment will lead to an
increase of incidents related to equipment. Incidents related to infrastructure are
influenced by the quality of the infrastructure. Besides equipment or infrastructure
‘other’ type of incidents are distinguished, that are often caused by passengers,
personnel or third parties. The time needed to recover from an incident and the
frequency of the train service determine how many trains are affected by an incident.
Besides these delays that are directly caused by incidents, disruptions will also lead to
a spread of delays further through the network, caused by interactions between
equipment, personnel or
infrastructure. The amount of secondary delays will increase when the utilization
(complexity) of the network increases.
Distinction of Trip Types
The rail network and road network in the Netherlands both have a dual function. They
are used to transport people within agglomerations, as well as between
agglomerations. For the rail network this means that different types of services have to
be offered: local and intercity.
Different trips will have different characteristics. For a long train trip a transfer is, for
instance, more likely than for a short trip. The effects of access and egress costs and
time will relatively be higher for a short trip than for a long trip by train. The same is
true for the parking costs of a car. Furthermore a trip can be made with different
purposes such as leisure or business.
In the model a distinction is made between trips performed during peak-hours and
between short and long distance trips. This results in four trip types as displayed in
Figure 4, each with their own set of characteristics, such as value of time.
Peak Off-peak
Short Distance | Short Distance
Peak Off-peak
Long Distance | Long Distance
Figure 4: The four types of trips in the model.
Feedback Structure
Analysis of the feedback structure leads to the identification of seven unique feedback
loops, as shown in Fout! Verwijzingsbron niet g d Each of the seven feedback
loops will be discussed briefly. Feedback loop:
1. Describes the relation between demand for train transport and frequency. A higher
demand can lead to an intensified train service. This leads to a higher frequency of
trains, limiting the time lost when a connection is missed. In turn this reduces
uncertainty about the arrival time, increasing the quality of the service, leading toa
higher demand.
2. A higher train frequency will also lead to less waiting for a train to arrive. This will
reduce the waiting time at a station between transfers. This will reduce the travel
time of the trip by train, leading to an increased quality, leading to a higher
demand.
3. Is almost the same as the previous feedback loop. An increased frequency will also
reduce the waiting time when arriving at the first station of a trip. This time can
have a different time value than time spent waiting in between trains.
4. A higher train frequency will also lead to higher utilisation of the infrastructure.
This means that incidents with other trains will have a higher indirect effect,
leading to larger delays and waiting time. This will reduce the quality of the service
and lead to a lower demand.
5. Another result of a higher frequency, and thus more trains on the network is that
the number of trains affected by an incident increases. When part of the network is
out of operation for an amount of time a high frequency will mean more trains are
impacted by this. This will increase delays and, thus reduce the quality of the
service.
6. An increase in required capacity will lead to an increase of equipment in use. As a
results of this also the number of equipment related incidents will increase. This
will then lead to an increase of total incidents which will lead to an increase of
delays and reduced quality of the service.
7. Finally and increase of incidents will also increase the number if trains indirectly
affected, further increasing delays and reducing service.
The main conclusion that can be drawn from this causal model is that an increase of
capacity through a higher frequency of train services will lead to a decrease of travel
time and uncertainty about arrival, but will also complicate operations leading to an
increase of delays and thus of travel time.
hobility choice
elas en
re oy
modal spi in attractiveness tun to i
x tocar tract tae ae.
early departure mobility by train
train trip
eee i) 2. On
time
anatertiig time train trip \
required
equipment inde
O1)
equipment
infrastructure
utilisation
4) 4. ) t=) 5.
incidents
trans inditecly
affected by incidents. *
ea P
B total number of
zat incidents
i) é -
trains diteetly afe
ted by incidents
a ee
Figure 5: structure of the model.
4. Model Specification
The model was implemented in Vensim Professional 6.0. Besides the basic
functionality required for the implementation and simulation of SD models, it supports
the use of subscripts, which means part of the model can be reused if similar concepts
(such as different types of passengers) have to be implemented. A full overview of the
model equations and values of constants can be found in Appendix A.
Data on the valuation of travel time and reliability was mainly found in scientific
articles. Extensive research on this subject has been performed in the United Kingdom
(Wardman, 2001, 2004) and the Netherlands (Tseng, Verhoef, & Rietveld, 2012).
Values for variables regarding operations were often found and derived from reports
by Dutch Rail (NS) and the network manager (ProRail) that contained information on
the network and operations.
Most of the data on mobility was derived from ‘Onderzoek Verplaatsingen in
Nederland 2011’, performed by Statistics Netherlands (CBS, 2011). This dataset
provides information on the daily mobility of the Dutch population and contains
responses of 37,754 persons. The total dataset contains 127,410 cases which relate to
parts of a trip. For these cases 150 variables are defined, which relate to characteristics
about household, trip purpose, mode of transportation, departure, arrival, etc. For the
purpose of this research this dataset was reduced to trips of interest: namely where
car or train were the main mode of transportation. These trips were then categorized
into four groups, reflecting the four trip types.
Modal split
For the modal split the cost component was implemented using a simple summation of
costs such as ticket price, parking and fuel costs. The value of time (VOT )was
determined by the VOT of the parts of a trip. A car trip consists of a single part (the
drive), but a train trip consists of time for access and egress, waiting, transferring and
in vehicle time. The costs of unreliability were determined by estimating the average
early and late arrival of trips, based on a standardised log-normal distribution which is
scaled based on the percentage of trip arriving on time and the time at which 95% of
the travellers have arrived.
Monetary costs, time value of the trip and the costs of unreliability were traded-off
based on a per characteristic basis train vs. car. A non-linear function was used in
which large differences between car and train per component have a larger impact
than small differences. This equation is presented in Equation 1, with: w; being the
weight for quality aspect i; and q; being the value of that quality aspect, for train or
car; and c determining the effect of the difference of a quality aspect between train
and car. This results in quality aspects with a difference between car and train being
weighted heavier than quality aspects which are almost equal.
/
a 4i,train®
modal split = wi =
d (2 / +1/ )
tes Witrain dicar
With S = {costs,time value, reliability}
Equation 1
Mobility
The total demand for train transport was determined by the effect of the modal split
on the number of choice passengers and the mobility of train captives. To reflect the
inertia in travel choice (Chorus & Dellaert, 2009) and the assumption that a change in
travel choice is caused by changes in the environment (Van Dalen, 2012), a delay in
change from choice car to choice train traveller and vice versa was implemented.
Operations
For the operations the effect of incidents on the operations was estimated based on
the causal model describing the links between incidents, primary and secondary
delays.
Validation and verification
Validation and verification cannot prove that a model is correct and possible for all
possible scenarios, but it can provide evidence (and build trust) that the model is
sufficiently accurate for its intended use (Thacker et al., 2004). The model has been
evaluated using a wide array of tests as suggested by (Sterman, 2000) and
(Wolstenholme, 1989).
The structure of the model and the adequacy of the model were evaluated during
separate discussions about it with two system experts. Dimensional consistency of the
model and equations was verified, partial model testing was used to test and correct
model parts. The presence of integration errors in the numerical results was disproved.
Finally an extensive sensitivity analysis was performed on variables and model parts, to
evaluate the sensitivity of model results to these parameters and determine the effect
of uncertainty in the model. The sensitivity analysis was the main quantitative result of
the model. The outcome of this analysis is discussed in the next section.
5. Simulation Results
Because of high uncertainty in the model, variables and structure, the model is not
suitable for predicting and forecasting. Therefore the model was used for a structured
analysis of the effects of uncertainty and sensitivity on the model results.
The univariate sensitivity analysis performed allows for a structured comparison of the
model outcomes. When the model is sensitive to a variable or component of the
system this can lead to two conclusions, or a combination thereof: (1) That variable or
component of the system can be used to design a high leverage policy; (2) Because of
the impact of this variable or component, uncertainty surrounding it must be reduced
in order to improve the validity of the model. Whether conclusion one or two applies
will depend on whether this component or variable can be influenced by stakeholders
in the system and how much is known about this component, qualitatively and
quantitatively.
Base Run
The results of the base run show a stable system, where the increase of train travel can
be explained by the overall increase in demand for mobility. The modal split increases
only slightly in favour of train travel, caused by an improvement of the quality of a
train trip because of higher frequency services to deal with the increase in demand.
The results of four of the key performance indicators of the system are shown in
through.
10
modal split total mobility train
om 48 «72 «96 «(120 144 168 192 216 240 2 48 72~«6~S«DS«ASCBSC9~—«21G 240
Time (Month) Time (Month)
BaseRun total mobility BaseRun
+ BaseRun total mobility BaseRun
+ BaseRun total mobility BaseRun
BaseRun total mobility BaseRun
Figure 6: Development of Modal Split over time, per Figure 7: Total passenger kilometres travelled per year
trip type. by train.
Selected Variables avg train delay
100,000
75,000
i 50,000
4
A
25,000
0
0 24 48 72-96 120 44 168 192 216 240 0.02
Time (Month) 2 48 «72 +~«96~«120=«MS«CBS«C92—«G 240
ajott peak cp masta Time (Month)
req peak cap : BaseRun) avg train delay : BaseRun
Figure 8: Required capacity in passengers per hour Figure 9: Development of average delay per train.
during and outside peak hours.
Sensitivity Analysis
For the sensitivity analysis 43 constants or variables were selected, that were of
interest because of uncertainty about their values, because they can be influenced by
actors in the system or because they represent a component of the system about
which uncertainty regarding the structure exists. 21 variables were selected as criteria
that indicate the performance of the main components of the model: modal split,
mobility and operations. This resulted in 86 model runs, with their respective results
combined in a single spreadsheet.
The model results were compared to the base run. When a 10% change of a variable
resulted in a change of more than 10% for one of the criteria, this criterion was
considered sensitive to that variable. A histogram of the results of the sensitivity
analysis can be found in Figure 10. The results of this analysis have been grouped into
four categories: external factors; reliability; effects of demand; and the trade-off
function. The numerical results of the sensitivity analysis can be found in Appendix B.
11
200 1g
s
: reall tt Po Pe |
Sob oe eee OR SS
BFP PS PF FH FF SH oF oF oS
Range
Figure 10: Distribution of results of the sensitivity analysis.
RELIABILITY IMPORTANT FACTOR IN MODAL CHOICE COMMUTE
Rescheduling costs during peak hours have the most impact on the total kilometres
travelled by train in the sensitivity analysis. Additionally the modal split and passenger
kilometres travelled by train is also sensitive to the predictability of train arrival times.
The reliability ratio of the car compared to the train is too high outside peak hours to
have effect, but during peak hours the train is a better match. Improvements of
reliability will therefore mostly lead to increased usage of the train service during peak
hours.
EFFECTS OF INCREASED DEMAND: MORE TRAINS LEAD TO HIGHER AVERAGE DELAY
Analysis of the feedback structure of the conceptual model in Section 3 already
suggested that an increase in demand will have an impact on the performance of the
rail network due to increased complexity of operations if the frequency of train
services was increased. This was confirmed by the sensitivity analysis.
Growth of mobility leads to more train usage at the cost of more delays
The growth of mobility leads to an increased usage of the train for transport, but this is
reflected in an intensified utilization of the rail infrastructure. The increase of the
number of trains will lead to more incidents, and increased spread of delays. This will
lead to an increase of the average delay of trains, curbing demand. This increase of
delay and its effects can be explained by the causal model, namely loop 4 to 7
describing the relations between increased frequency and equipment usage and
(in)direct incidents, infrastructure usage and delays.
Infra reliability and repair time major influence on average delay
The reliability of the infrastructure and the time needed to restore it in case of
incidents is a major factor in determining the average delay. This is becomes more and
more important when the frequency of trains is increased because it affects more
trains and spreads more through the system.
TRADE-OFF FUNCTION VERY SENSITIVE TO VALUATION OF DIFFERENCE BETWEEN
QUALITY ASPECTS
The overall model performance is very sensitive to the parameters of the trade-off
function. This function is also one of the softest parts in the model. It represents a
generalization of human behaviour. The way the trade-off of quality aspects are
12
modelled can be seen as the most important part of the system in terms of influence it
has on model outcomes and because uncertainty about the real-world decisions allow
for different trade-off functions.
EXTERNAL FACTORS: QUALITY CAR TRIP IMPORTANT FOR ATTRACTIVENESS TRAIN
The external factors are variables that are determined outside the railway system and
on which the stakeholders in the system have little to no direct influence. For most
trips the choice is between taking the train of the car. It is therefore no surprise that
characteristics of the car trip are important for the usage of the train service.
Raising speed limits leads to increased competition for the train outside rush hour for
long distance trips
Increase of the average car speed will lead to increased competition for the train,
especially on long distance trips performed outside rush hours. Since there will be little
to no traffic jams outside rush hours, the main cause for this would be a raise of the
speed limit.
Improvements of predictability of car arrival times will lead to reduced train usage
The other major car related factor that affects the modal split is the reliability of arrival
time. If this reliability increases further this will negatively affect the portion of train
users for all trip types. Improvements of the road networks, local, regional or national,
that lead to an improvement of the predictability of a car trip will negatively influence
train usage.
Time value of access and egress important for short distance trips
A change in the value of time of access and egress costs will lead only lead to a
significant improvement of short distance train trips. This can be explained by the fact
that in a short distance trip the ratio of access and egress time to in train time is much
higher than for longer trips.
6. Value and Validity of the Model Analysis in a Complex Dynamic Network
Because of the separation of operations, management and oversight in the Dutch
railway system, decision making will require the cooperation of stakeholders. The
policy analysis and decision making process become even more complex when taking
the institutional arrangements into account. None of the stakeholders is able to
impose their own will upon the others. Any collective decision will therefore be the
result of a process of consultation and negotiation, which allow actors to use all sorts
of strategies to maximize their influence on the final decision (de Bruijn & ten
Heuvelhof, 2002).
The decision-making also takes place in an environment that corresponds to the
definition of a network: the stakeholders are interdependent, unable to impose their
own problem definition, aims and information on others and not able to make a
unilateral decision (de Bruijn & ten Heuvelhof, 2002). This poses a threat to a decision
making process when the problems involved are contested and unstructured. De
13
Bruijn and ten Heuvelhof also list four main reasons why a policy analysis may not be
authoritative in a network environment:
1. The quality of the analysis;
2. Stakeholders do not understand the analysis;
3. Stakeholders do not commit themselves to the way the analysis is carried out
and therefore do not commit to the results;
4. The analysis does not match the game playing during the decision-making
process.
The main remedy for the first reasons is improvement of the analysis. For the other
reasons the main remedy involves improving communication about the analysis and
improving interaction between the analysts and the stakeholders. In fact, inadequate
communication between policy analysts and policy actors is one of the reasons for the
limited impact that policy analysis has on policy making (Geurts & Joldersma, 2001).
In the following paragraphs methods will be discussed that can improve the validity of
the analysis and the value of it to the decision making process. This will be done by
discussing the ways the model can be improved, what knowledge gaps should be
addressed, and how policy actors can be involved.
MODEL IMPROVEMENTS
Improving the model can be achieved by expanding the model boundaries and adding
additional components to the model structure. Adding these components can help by
improving the quality of the analysis because of the inclusion of additional feedback
loops. Inclusion of concepts and models that are not yet in the model, that are deemed
important by stakeholders, can also help convince them of the validity of the model.
Also during development of the model some concepts were implemented using the SD
methodology that would be easier to represent in a different type of model. This
resulted in a very complex structure of that part of the model. A hybrid combination of
multiple modelling methods could help improve the validity of the model by providing
more accurate results, but also reducing the complexity of the SD model.
An example of this is the calculation of unreliability of arrival times in a chain of
transport modes: the effect of the unreliability of the arrival time of a train was used to
determine the unreliability of a trip. Due to limitations of the SD approach and the
simulation package, this was modelled using a single arrival distribution which values
would be determined based on the probability of making a connection. This resulted in
a distribution that would have the same properties of the distribution of arrival times
for a trip, but would not take into account specific characteristics of such distributions
such as the impact of service frequencies on delays when a connection is missed.
During development of the SD model, a very simple model of arrival distribution was
developed in an Excel spreadsheet. This model was used to calibrate a generalized
version of the arrival distribution in the SD model. The spreadsheet model however did
represent the actual distribution for a trip under specific conditions more naturally
than the one in the SD model. It could however not be directly used in the SD
14
simulation because the conditions would change over time. Implementing the
simulation model in a package that would allow the import and export of values during
simulation and the execution of other programs would allow the coupling of the model
for policy analysis to specific and detailed models that could better represent the
operational effects of policies.
RESEARCH KNOWLEDGE GAPS
During the modelling process knowledge gaps were encountered that limited the
validity of parts of the model. Additional research into the these specific areas is
required before the model can be improved to better reflect the real world system and
thus improve the validity and authority of the model.
Trade-off Function
During sensitivity testing of the model it was found that the model results were very
sensitive to the trade-off function itself, as well as the aspect of how heavily large
difference are weighted. To improve the validity of the model it is suggested that more
research is performed in determining which kind of trade-off function is most
appropriate for the model. This trade-off function would have to take into account the
modularity of the model, which supports adding any finite number of trade-off aspects
by trading off the train to car values per quality aspect, to allow a weighted averaging
regardless of the unit the quality aspect is measured in.
Effect of Utilization on Reliability
In the model increased utilization of the infrastructure results in an increase of delays
because incidents affect more trains and because of smaller buffer times they spread
more easily through the system. The effect of increased utilization of the network was
not linked to an increase in unreliability of the arrival times of trains. The sensitivity
analysis of the simulation model revealed that the model results were significantly
influenced by the reliability of arrival times. Although the effects of unreliable train
services on customer satisfaction has been the focus of many studies, quantification of
the effect of operational aspects on the reliability of arrival times has not. A statistical
study of the operational results of rail networks or a simulation study of such a system
could improve the quantitative insight in this relation.
Trip Data
The parameters that were used for description of different trip types were extracted
from the OViN database (CBS, 2011). Most of the trips of the database concerned car
travel, and although the results were weighted for the frequency of trip types this
posed some problems during implementation. For example the number of long
distance trips was very limited, which may result in unreliable averages for the trip
types. Furthermore some data such as the average speed had to be calculated from
the data based on departure and arrival times and the distance travelled. The results
of the model could be improved if more specific and reliable data was gathered
tailored to the data needs of the current model.
15
INVOLVE POLICY ACTORS
To improve the authority of a policy analysis, resulting in trust in and acceptance of the
results, interaction and communication between the analyst and the stakeholders is
very important. Furthermore, most of the insight in a complex system is generated in
the modelling process itself. Involving stakeholders can thus not only result in
increased acceptance of the system, but also in enhancing the understanding of the
actual decision-makers in the system.
In participatory policy analysis the focus is on the network perspective in policy
making. It focusses on improving the process of communication between the policy
analyst and the stakeholders in the network. The emphasis in this process is not on
providing an analysis of policies options, but on increasing the problem solving
capacities of the stakeholders. It is directed at improving as well as integrating the
mental models of different actors in a policy network (Geurts & Joldersma, 2001).
Two ways of conducting participatory policy analyses using System Dynamics are group
model building and gaming. Group model building focusses on integrating divided or
subjective knowledge, different views and values, mediation and the generation of a
shared system view (Vennix, 1996, 1999). Gaming focusses on improving the
understanding of participants of the relation between the structure and the behaviour
of the system by means of role-playing and interaction of stakeholders in a simulated
environment (Lane, 1995; Geurts & Joldersma 2001). It is often supported by or based
ona simulation model.
Both participatory modelling and gaming allow the transfer of knowledge acquired
during the analysis to be transferred to stakeholders while avoiding some of the
validation problems encountered in a ‘classical’ policy analysis setting. With
participatory modelling validity is less important, as long as there is agreement
between participants regarding the relations in the model it satisfies its purpose. With
gaming key learning concepts identified during the modelling process can still be
transferred, in an environment where the results of a formal modelling process will
and can be endlessly scrutinized. Participating in a game can also be considered less of
an obstacle by participants than committing themselves to the results of a policy
analysis. This does not prevent the game from being able to influence the perception
of the system, problems and solutions.
7. Conclusi and Rec d
Modelling the rail system using SD facilitated a structured approach to system analysis,
identification of the feedback structure of the system, evaluation of uncertainty and
identification of directions for further policy analysis.
Analysis of the feedback structure of the system has shown that a further growth of
passenger transport can both lead to shorter travel times and higher reliability of the
rail network, but also to an increase of delays due to the added complexity of the
Operations.
16
This was confirmed by the quantitative analysis which has shown further that the
reliability of infrastructure and the recovery time is a major component in the extent
of this delay. Furthermore the effect of unreliability in a trip was quantified and was
found to be of significant importance in determining the choice of travellers between
the car or train. Finally it was found that characteristics of a car trip such as average
speed and improvement of reliability of car travel was of significant effect on this
choice. Improvements to the road network could therefore be a threat to the
competitiveness of the train.
Because of network type of decision making surrounding policy design for the Dutch
rail network, the validity, trust and authority of a policy analysis is very important.
Because of the complexity of the system, unstructuredness of the problems and
different stakeholders, performing an authoritative and acceptable policy analysis is
difficult. The modelling process undertaken for this research has shown that in general
System Dynamics can be valuable and is up to this task, but that for modelling part of
the operational aspect of the system it is not the most suitable method.
This problem can be handled in three different ways: first as was done in this research,
relations can be simplified and represented on a higher level of aggregation. Second
the relations can be represented and estimated by using additional methods such
intensive modelling and validation supported by experts, performing additional
research to uncover empirical evidence to support these relations or perform
additional simulation studies to support them. Thirdly more appropriate models or
simulation could be coupled to the SD model to better represent these relations.
The high requirements for validity and acceptance of the model, due to the
unstructuredness of the problems and the network type decision making, means that
the first option is not viable. Simplification of the model would reduce the authority of
the analysis and would give ample opportunity to criticize it. Performing additional
research or developing additional models to support the SD model would be both
costly and labour-intensive. The relative newness of System Dynamics for policy
analysis in the rail sector in general, and in the Netherlands in specific might pose a
problem to the willingness of making this investment.
Besides the classical usage of System Dynamics for policy analysis it can also be used in
different ways, that would better fit with the problem, the environment wherein the
policy analysis takes place, and be less costly while still staying true to the main
purpose of System Dynamics: enhancing learning about complex dynamic systems.
This leads to the following three recommendations for use of System Dynamics for
policy analysis in the Dutch Rail Sector:
In the context of a single organization or department System Dynamics can be used as
problem structuring method. Modelling of the system has supported a guided search
into concepts and interactions, leading to a formalization of the interactions and
assumptions about the system. Qualitative analysis revealed important trade-offs and
feedback in the system. Implementation of the model revealed knowledge gaps and
the need for data essential for any analysis of the system. System Dynamics can be
17
used to research other problems as well and lead to a comprehensive overview and
better understanding of the workings of the system.
System Dynamics can be used as a tool in a group model building process. Participatory
modelling can be used for the creation of a shared problem perception. The causal
diagrams are easy to understand and use, but also allow for representation of a
complex system structure. They can be used to structure debate and better
understand the effects of feedback. If such a process would result in a shared system
view, the conceptual model can then be converted and simulated to allow quantitative
analysis.
Because the needs for an authoritative analysis requires substantial research,
development and validation of a model for classical policy analysis, this does not mean
simpler, less substantiated models developed within one organization cannot be used
in a multi actor environment. Many of the findings about the effects of feedback and
the need for effective policies can also be represented in a game. This game can be
developed based on a causal model, or be supported by a quantified simulation. Due
to the nature of gaming the requirements for validity of the model will be less high.
Important insights gained from an analysis, such as the importance of reliability in a
train trip, can in this way still be conveyed to policy makers.
References
Brons, M. (2005). De beleving en waardering van onbetrouwbaarheid in het
personenvervoer: een conceptueel kader (concept). Amsterdam: Vrije Universiteit
van Amsterdam.
Brons, M., & Rietveld, P. (2009). Improving the Quality of the Door-to-Door Rail
Journey: A Customer-Oriented Approach. Built Environment, 35(1).
de Bruijn, H., & ten Heuvelhof, E. (2002). Policy analysis and decision making in a
network: How to improve the quality of analysis and the impact on decision making.
Impact Assessment and Project Appraisal, 20(4), 232-242.
CBS. (2009). Hoe druk is het nu werklijk op het Nederlandse spoor? Den Haag: Centraal
Bureau voor de Statistiek.
CBS. (2011). Onderzoek Verplaatsingen in Nederland 2011. Den Haag.
Chorus, C., & Dellaert, B. (2009). Inertia in travel choice: The role of risk aversion and
learning. Paper presented at the 12th International Conference on Travel Behaviour
Research, Jaipur, Rajasthan, India.
Van Dalen, L. J. (2012, August 15th). [Choice of modality].
ECMT, European Conference of Ministers of Transport. (2005). Railway Reform &
Charges for the Use of Infrastructure - Conclusions and Recommendations. Paris,
France.
Forrester, J. W. (1961). Industrial dynamics. Cambridge, Mass.: MIT Press.
Geurts, J. L. A., & Joldersma, C. (2001). Methodology for participatory policy analysis.
European Journal of Operational Research, 128(2), 300-310. doi: 10.1016/s0377-
2217(00)00073-4
18
Gottschalk, P. (1983). A system dynamics model for long range planning in railroad.
European Journal of Operational Research, 14(2), 156-162. doi: 10.1016/0377-
2217(83)90309-0
Van Hagen, M. (2011). De klant centraal. February 17th 2011: Delft Top Tech: Master
of business in Rail Systems.
HisschemOller, M. (1993). De Democratie van Problemen. De Relatie tussen de Inhoud
van Beleidsproblemen en Methoden van Politieke Belsuitvorming. Amsterdam: Free
University Press.
Hisschemdller, M., & Hoppe, R. (1995). Coping with Intractable Controversies: The
Case for Problem Structuring in Policy Design and Analysis. Knowledge and Policy:
The International Journal of Knowledge Transfer and Utilization, 8(4), 40-60.
Homer, J. B., Keane, T. E., Lukiantseva, N. O., & Bell, D. W. (1999, 1999). Evaluating
strategies to improve railroad performance-a system dynamics approach. Paper
presented at the Simulation Conference Proceedings, 1999 Winter.
Kotler, P., & Armstrong, G. (2001). Principles of Marketing. Upper Saddle River, New
Jersey: Prentice-Hall Inc.
Lane, D. C. (1995). On a resurgence of management simulations and games (Vol. 46).
Basingstoke, United Kingdom: Palgrave Macmillan.
Lubis, S., Pamungkas, & Tasrif, M. (2005). Rail Sector Policy Analysis Using System
Dynamic Approach. Paper presented at the Eastern Asia Society for Transportation
Studies, Bangkok.
MinlenM, Ministerie van Infrastructuur en Milieu. (2011). Brochure Programma
Hoogfrequent Spoorvervoer.
Nijman, G. (2012a, May 15th). [Discussion concerning dynamic behaviour of the
railway system].
Nijman, G. (2012b). Inleidende syllabus Openbaar Personenvervoer.
Peek, G., & van Hagen, M. (2004). Eén verbinding is géén verbinding - Van harde én
zachte bereikbaarheid. Paper presented at the Vervoersplanologisch Speurwerk
2004, Zeist, the Netherlands.
ProRail. (2011). Beheerplan 2012.
ProRail. (2012). Ruimte op de rails Retrieved february 16, 2012, from
http://www.prorail.nl/Bedrijfsinformatie/Nieuws/Pages/Ruimteopderails.aspx
Schmidt, D. (1989). Computerbased decision support of strategic planning and strategic
management with system dynamics models illustrated by the example of the german
federal railway. Paper presented at the The 7th International Conference of the
System Dynamics Society, Stuttgart, Germany.
Sterman, J. D. (2000). Business dynamics systems thinking and modeling for a complex
world. London: Irwin Professional.
Sterman, J. D. (2001). System dynamics modeling: Tools for learning in a complex
world. California Management Review, 43(4), 8-+.
Thacker, B. H., Doebling, S. W., Hemez, F. M., Anderson, M. C., Pepin, J. E., &
Rodriguez, E. A. (2004). Concepts of Model Verification and Validation Other
Information: PBD: 30 Oct 2004 (pp. Medium: ED; Size: 41 pages).
Tijdelijke Commissie Onderhoud en Innovatie Spoor. (2012). Parlementair onderzoek
onderhoud en innovatie spoor. 's-Gravenhage: Tweede Kamer der Staten-Generaal.
19
Tseng, Y., Verhoef, E., & Rietveld, P. (2012). Valuation of travel time reliability for
railway passengers. Department of Spatial Economics. Vrije Universiteit Amsterdam.
Amsterdam.
Vaessens, B., Van Hagen, M., & Exel, M. (2008). Auto vs trein - De context bepaalt de
keuze. Utrecht: NS Commercie.
Vennix, J. A. M. (1996). Group model building : facilitating team learning using system
dynamics. Chichester [etc.]: Wiley.
Vennix, J. A. M. (1999). Group model-building: tackling messy problems.
[10.1002/(SICI)1099-1727(199924)15:4<379::AID-SDR179>3.0.CO;2-E]. System
Dynamics Review, 15(4), 379-401.
Wardman, M. (2001). A review of British evidence on time and service quality
valuations. Transportation Research Part E: Logistics and Transportation Review,
37(2-3), 107-128. doi: 10.1016/s1366-5545(00)00012-0
Wardman, M. (2004). Public transport values of time. Transport Policy, 11(4), 363-377.
doi: 10.1016/j.tranpol.2004.05.001
Wolstenholme, E. F. (1989). A current overview of system dynamics. Transactions of
the Institute of Measurement and Control, 11(4), 171-179.
20
