Content uploaded by Amin Dehdarian
Author content
All content in this area was uploaded by Amin Dehdarian on Sep 13, 2022
Content may be subject to copyright.
A multi-method approach to develop dynamic hypotheses for systemic
challenges in the Swiss freight transport system
Amin Dehdarian, the Learning Lab: Future Transport Systems, ETH Zürich, Switzerland
Abstract
Freight transportation sector in Switzerland is undergoing rapid changes. Developments such as decarbonization,
digitalization of the consumer market, advances in powertrain technologies and innovative solutions in the
supply chain have influenced the freight market. However, uncertainties about the potential impacts of these
developments have shaped a variety of ideas among freight market actors (Liimatainen et al., 2019). This paper
tries to explore these uncertainties, and analyze potential challenges and solutions, by mapping the mental
models of the primary stakeholders and transforming them into dynamic hypothesis. A multi-method approach
is used for this transformation and defining strategic decision areas for the future of freight transport system in
Switzerland. The method starts with semi-structured expert interviews, and uses ideas from the Grounded
Theory approach to map the mental models and find the primary strategic areas. Then, Vester’s sensitivity model
is used for impact analysis and identifying the main factors and dynamics within each strategic area, which results
in developing a dynamic hypothesis. Then, Social Network Analysis is used to aggregate the impacts and quantify
the properties of the main factors, which provides input for Causal Loop Diagramming (CLD).
The analysis leads to the emergence of eight strategic decision areas, where road freight electrification and rail
digitalization are the primary technological trends, and the need for cooperation in the forms of combined
transport and collaborative logistics are highlighted. For each strategic area, the key factors (as leverage points,
stabilizer and indicators) are identified and a CLD is developed to visualize the dynamic hypothesis. The results
can enable strategists, policy makers and freight operators to make more informed decisions, by looking at
primary market actors’ expectations of the future trends, technological developments, challenges and solutions
of the sector.
Method and Sample
In order to gather expert opinions and transform them into dynamic hypothesis, a multi-method approach is
used, which consists of five steps. These steps are different and complementary, where the results of each step
provides the input for the next step; thus, together they provide objective and quantitative insights from
subjective opinions. Then, the main factors constituting these mental models are specified and classified into the
key themes. For each theme, analyzing how these factors are interdependent leads to the identification of the
critical factors. Combining these critical factors and their dependencies is a prerequisite for defining the dynamic
hypotheses and developing CLDs.
Figure 1. A systemic approach for deriving dynamic hypothesis from expert opinions
Step 1: Expert interviews and the coding process
In the first step, a total number of 31 expert interviews were conducted. The experts were chosen from both
road and rail freight transport, as the dominant modes in Switzerland. They were mostly top-level managers from
different types of companies and organizations, as freight operators, policymakers, technology providers,
research institutions and shipping companies. Therefore, they could represent the primary actor groups of the
sector. The interviews were transcribed and compiled by a modified version of “the Grounded theory” approach.
This approach starts with extracting all quotes from interviews and summarizing them into open codes. Each
open code paraphrases the main message of the quote. The main modification is that instead of classifying open
codes into higher-level code (i.e. axial and selective coding), the classification was done through analyzing the
link between the main concepts in each code, and how higher-level patterns or clusters emerge in a network of
concepts.
2
Step 2: Impact analysis and network construction
In the second step, in order to understand how the quotes are related, each open code is transformed into at
least one “impact”, based on the context of discussion, and subjective interpretation of the main message. Each
impact highlights the existence of two factors in a quote, where a factor influences another factor. In network
terminology, the influencing factor is called the “source”, while the other is called the “target”. A graphical
representation of the impacts can be seen in figure 2. Each factor is represented as a node, while the influence
of one factor over another is depicted as a directed edge from the source to the target. The edge thickness
(weight) is a measure of the number of times each link has been mentioned. By aggregating all impacts a network
can be constructed, which enables us to do cluster analysis and calculate node properties in the next step.
Figure 2. Impact visualization of factors
Figure 3. interpretation of node positions in the network
Step 3: Cluster analysis and factor properties
After constructing the network of factors, cluster analysis identifies the main categories and themes of the data.
Cluster detection is done at two levels. In the first level, it is implemented on the whole network of impacts for
rail and road freight transport systems (all interview transcripts), which results in the emergence of big clusters
or ‘themes’ of the study. Then, at the second level, it is used in each of these clusters to identify smaller clusters
or ‘categories’. Each category in a theme is a group of factors, which their relationships is stronger than
relationship with the nodes outside the category. Then, for each node in the network three measures are used
to calculate node properties. Out-degree is the number of edges going out of each node, meaning the number
of influences it has on other nodes. Similarly, in-degree is the number of edges going into each node and
represent the number of influences it has received from other nodes. Finally, eigenvector centrality is a centrality
measure that calculates the level of influence of a node over a network based on its connection to other nodes
(Dehdarian & Tucci, 2021).
Step 4: The Sensitivity model and factor analysis
The node properties in the network are used as inputs for an analytical tool called Vester’s Sensitivity Model
(SM). It is a semi-quantitative and systematic approach, which focuses on pattern recognition in system variables,
and analyzes how different factor groups have different levels of criticality in the system, and how we can
interpret the importance of these groups of factors (Vester and Hessler, 1982). First, an “impact matrix” is created
based on the impact analysis explained above. Then, these calculated values for each factor based on network
analysis are used for classifying them into different groups. Figure 3 shows three groups of factors based on a
combination of the activity (out-degree to in-degree) and connectivity (centrality) measures.
Step 5: Dynamic hypothesis and Causal Loop Diagramming
In the last step, the main factors and their dependencies are used for communicating the main story
or the underlying structure behind the dynamics of each strategic area and define the dynamic
hypothesis. Then, the CLD combines the results of network analysis and the Sensitivity model to
summarize the positive and negative feedback loops, responsible for the main behaviors in the system
(Sterman, J.D., 2001; Dehdarian, A., 2017).
3
References
Dehdarian, A. (2017). Three Essays on Methodologies for Dynamic Modeling of Emerging Socio-
technical Systems. EPFL.
Dehdarian, A., & Tucci, C. L. (2021). A complex network approach for analyzing early evolution of smart
grid innovations in Europe. Applied Energy, 298, 117143.
Liimatainen, H., van Vliet, O., & Aplyn, D. (2019). The potential of electric trucks–An international
commodity-level analysis. Applied energy, 236, 804-814.
Sterman, J. D. (2001). System dynamics modeling: tools for learning in a complex world. California
management review, 43(4), 8-25.
Vester, F., & Hesler, A. (1982). Sensitivity model. Frankfurt/Main: Umland-verband
