From qualitative group modelling‐based exploration to rigorous simulation: Comment on the paper by de Gooyert, de Coninck and ter Haar

In commenting on the paper by de Gooyert et al. (2024), I have four aims: First, I emphasise that the paper represents a successful application of the qualitative group modelling method. Used with a group of experts from different disciplines, the method generated an integrative perspective on causal relationships in a very short time. Second, I acknowledge the post-workshop editing process that the authors undertook to arrive at the causal diagrams that are central to their paper. Third, I appreciate the authors' identification of five reinforcing loops that have the potential to achieve the transition to a climate-neutral energy system at an increasing speed. Fourth, I emphasise that the model described in the article provides an excellent basis for quantification and rigorous simulation-based analysis. Future research could thus contribute to the tradition of system dynamics work dealing with all aspects of climate change. Group model building has a long tradition in system dynamics. According to Torres (2019), it is one of the three topics that have attracted the most research. Early system dynamics studies were characterised by Forrester (2007) as being conducted in consultant mode. Computers were rare and difficult to use, so the programming part of modelling and running simulations was typically done by experts without direct client involvement. However, as Lane (2010, 2007) points out, the idea that modelling is best done with, or even by, the problem owners is already woven into the seminal publications that founded the field (Forrester, 1958, 1961). As system dynamics software improved and provided more intuitive user interfaces (e.g., Richmond, 1985), as stakeholder involvement came to be seen as a critical factor in implementation (e.g., Größler, 2007), and as the system dynamics methodology advanced (e.g., Morecroft, 2007; Richardson & Pugh, 1981; Sterman, 2000), group model building became more popular and more widely used. Milestones in the development of the group model building method include Morecroft (1984), Lane (1992), Vennix (1996), Richmond (1997), Andersen and Richardson (1997), Andersen et al. (1997), Vennix (1999) and, more recently, Wilkerson et al. (2020) and Brown et al. (2022). The methodological development of group model building was always accompanied by applications (e.g., Ansah et al., 2018; Rouwette et al., 2016; Vennix, 1996; Videira et al., 2010). The paper of de Gooyert et al. (2024) is a recent example. It reports how group model building is used to create a qualitative dynamic theory of the transition to a climate-neutral energy system from the knowledge of 11 experts with different disciplinary backgrounds who form the ‘Expert Team Energy System 2050’. Two workshops of 3 and 2 h were held to explore the topic from different perspectives and to create a causal loop model that shows interrelationships between variables from different domains at a high level of aggregation. The outcome of the workshops is included in the paper's appendix and underlines the messiness of the problem to be dealt with (even if one does not understand the Dutch words): many relevant variables (even at this high level of aggregation), many causal links with positive and negative polarities and quite a number of variables that stand outside and are not (yet) connected to other variables. It is normal for the raw output to be like a rough diamond: it needs to be cut and polished to make it shine. Similarly, the authors process the raw spaghetti diagram to include “shinier” illustrations in their paper. This brings me to the second point. After translating the variable names from Dutch to English, the authors went through a four-step polishing process: First, they simplified the diagram by clustering variable names and deleting isolated ones. Second, they changed some of the names to “avoid jargon and better follow system dynamics standards” (de Gooyert et al., 2024, Appendix B). Third, they modified the structure of the model to be more in line with the literature and best practice in system dynamics. It is worth emphasising that all causal links included in the diagram that do not describe purely formal relationships (e.g., “energy consumption economic activities” and “energy consumption citizens” add up to “total energy consumption”) are substantiated by references to published research—as suggested by Jalali and Beaulieu (2024). Fourth, the authors have added delays “based on their understanding of the energy system, supported by academic literature” (de Gooyert et al., 2024, loc. cit.). However, references to this literature are not included in the text, and explanations are rare, so the links that have been given a delay mark seem somewhat arbitrary. Although not described in great detail in the paper, the authors took a very inclusive approach, involving the members of the expert team in the refinement process at various stages to make the raw diamond shine. Immediately after the first group model-building workshop, the first author wrote a “workbook” containing both the original causal loop spaghetti diagram and simplified versions. This gave participants the opportunity to comment on the simplification and also to contribute any relevant thoughts that had emerged after the workshop. The simplified diagrams were then discussed intensively in the second workshop, which was held online. All participants in this meeting commented on the diagrams; that is, they took a few minutes to share their thoughts. This feedback was recorded and used in the further refinement process. During the writing of the paper, the causal loop diagrams were further polished, taking into account the results of the workshop, the results of the online discussion and the results of studies on similar topics. Several versions of the manuscript were circulated among the participants, with less and less use being made of the opportunity to provide feedback. In their search for policy recommendations to move away from the carbon-based energy system into which the Netherlands was locked by the discovery of the Groningen gas field in 1959 and its exploitation for more than 50 years, the authors rightly focus on reinforcing feedback loops. In doing so, they follow the advice that Lane and Husemann (2002, 2008, p. 45), with reference to the Odyssey, offer to every helmsman of a social system in an unattainable way: “attend to reinforcing loops.” With the deadly whirlpool of Charybdis on one side and the six-headed monster Scylla on the other side of his way, Odysseus had a difficult task to complete. The consequences of being caught by the currents of Charybdis are obviously worse: the boat gets faster and faster into the depths of the whirlpool, and the entire crew will perish. But there is another side to reinforcing loops. While Odysseus faces with the danger of a vicious circle, reinforcing loops can also act as virtuous circles that move a system faster and faster in a welcomed direction. In Odysseus' case, Charybdis as a virtuous circle would not pull the ship into the deep. On the contrary, it would spiral the ship safely through the strait. Virtuous circles are essential for moving systems out of lock-in states caused by strong balancing loops. De Gooyert et al. (2024) identify five reinforcing loops that have the potential to drive the decarbonisation of the Dutch energy system at an increasing speed—at least for some time before the inevitable balancing effects kick in. The reinforcing feedback loops R3 and R4, highlighted in Figure 1, consist of a few variables only, and links that all have the same polarity: either purely positive, as in R3, or purely negative and even in number, as in R4. This makes their reinforcing nature easy to understand. All five causal links are based both on the knowledge of the experts and on the literature referred to by the authors to substantiate the relationships. While R3 spans the public and private sectors (with business and consumers as important subsectors), R4 operates within the consumer sector. R5 extracted in Figure 2 includes four variables, of which the variable “effectiveness climate policies” belongs to the public sector and the other three variables belong to the consumer sector. Causal links with both negative and positive polarities are included. However, as the number of negative link polarities is even, the feedback loop is reinforcing. Following Moore et al. (2022), this loop could be named the “expressive force of law.” Loop R1, shown on the right-hand side of Figure 2, affects the business sector and creates the vicious circle with which the Netherlands is now struggling in the face of climate change: the more fossil assets and infrastructure there are, the less effective climate policy is, which increases the energy consumption of economic activities, leading to even more fossil assets and infrastructure. Finally, loops R2 and R6 (which was not identified by the authors) link all three sectors mentioned above and shown in Figure 3. The variable “sustainable innovation” is key to both loops (and to a group of balancing loops aggregated by the authors into B1), emphasising the importance of creating new products and processes that reduce the burden of effective climate policy on the private sector. The existence of these many reinforcing loops, involving different variables from all sectors, is promising because it allows for a wide range of possible interventions. The authors rightly focus on variables that could be changed by an exogenous influence in order to set the respective reinforcing loop in motion as a virtuous circle. For example, they suggest increasing social support for climate policy by strengthening the role of citizens and reducing the burden of climate policy by alleviating energy poverty (Figure 3). The exploratory qualitative analysis of the causal loop diagram generates promising ideas for policy levers. However, the limitations of a purely qualitative analysis of feedback loops are well known and widely discussed in system dynamics (Homer & Oliva, 2001; Lane & Husemann, 2002; Richardson, 1986; Schaffernicht, 2010; Sterman, 2018). Lane and Husemann (2008, p. 43) point out that it is “not enough merely to know that reinforcing loops might exist. It is necessary to assess whether they are active, how strong they are and whether they are truly dangerous.” I would change “dangerous” into “helpful” as R1 to R6 should operate as virtuous circles, but the conclusion to be drawn does not differ: Simulation must be used to rigorously analyse which policy or combination of policies achieves the desired transition of the Dutch energy system towards climate neutrality. However, the development of a comprehensive multisector model comprising many thousands of variables and equations appears to be the next but one step, if at all. Future research could and should start much simpler as demonstrated, for example, by Martínez-Jaramillo et al. (2022) or Homer (2020). Converting the highly aggregated causal loop diagram into a highly stylised quantitative model seems the more promising next step to go (Ghaffarzadegan et al., 2011). Complexity of such a model should be increased very carefully and only so far as the modelling of stocks and flows as well as the quantification of causal links requires. Such a stylised model would allow simulation experiments to be designed that can test the policy levers found by de Gooyert et al. (2024). In addition, sensitivity analysis and loop dominance analysis could be used to answer many additional questions, for example, how strong the reinforcing loops have to be to gain momentum and how dominance can be shifted from the balancing loops to the reinforcing ones (e.g., Hayward & Roach, 2024; Moxnes, 2005; Schoenberg et al., 2020). Simulation seems inevitable, but simulation of a small, stylised model seems the next step to take on a promising road. Open Access funding enabled and organized by Projekt DEAL.