ABSTRACT The global PET (polyethylene terephthalate) sector is transitioning towards potentially sustainable options, such as rPET (recycled PET), bPET (bio‐based PET), and PLA (Polylactic Acid). This work develops a system dynamics model to examine adoption pathways and evaluate policy interventions for the PET sector. The Multinomial LOGIT model allocates market share from a vendor's perspective based on price and GHG emissions. The model is applied to India, considering a horizon of 2020 to 2045. Five scenarios are simulated, including business‐as‐usual (BAU) and circular economy intervention. The results showed that PLA, bPET, and rPET significantly decreased reliance on virgin PET (vPET) in the future. Compared with the BAU scenario, aggressive CE implementation reduced the vPET share to 30% by 2045, down from 60%, reducing sectoral GHG emissions by 5.8%. Overall, the results recommend that policies should incentivize the adoption of rPET by enhancing recycling infrastructure to foster a sustainable plastic industry.
ABSTRACT Model conceptualization in system dynamics (SD) is a critical yet relatively underexplored phase of the modeling process, often described as both an art and a science. Practitioners employ a wide range of approaches, frequently combining multiple methods to address complex or unfamiliar problems. The literature on conceptualization remains fragmented, making the process particularly challenging for newcomers and practitioners from other fields. This study addresses this fragmentation by providing a structured overview of conceptualization approaches in SD. It identifies 10 distinct approaches, illustrates each with representative examples, and offers a systematic comparison of their perspectives, strengths, and limitations. It further identifies key dimensions along which these approaches differ, providing a basis for more structured selection and combination. The findings provide structured guidance for modelers, particularly novices, by highlighting how complementary approaches can be combined. The study also identifies gaps in the literature and suggests directions for future research.
ABSTRACT İstanbul plays a significant role in the economy of Türkiye due to its leading position in almost all economic activities. However, the city is prone to a major earthquake risk which may have devastating impacts on human lives as well as the regional and national economy. The possible economic losses associated with an earthquake affecting İstanbul have been a concern for several decades, but scientific studies estimating such losses remain limited, except for some expert opinion‐based assessments. Thus, this study introduces a hybrid methodology to estimate the economic impacts of a potential earthquake that would affect İstanbul. The method integrates an input‐output model with a system dynamics model and makes it possible to examine the effectiveness of several policies to alleviate the negative impacts of the earthquake on the economy under various earthquake scenarios. Here, a bi‐regional demand‐side input‐output table for the İstanbul Region and the rest of Türkiye is generated for the first time using a non‐survey method and the dynamic relationship between the fundamental components of an economic system is modeled using the system dynamics approach. Then, the bi‐regional input‐output table and system dynamics models are integrated through a coupling mechanism, and the economic impacts of possible earthquakes that may affect İstanbul are evaluated for four different scenarios and three policy families with six policy variants. The findings demonstrate that the hybrid framework can reveal the sector‐specific and dynamic economic consequences of an earthquake and help identify policies that may enhance post‐disaster recovery.
ABSTRACT Port terminal operators have evolved from merely handling cargo to serving as strategic nodes that connect hinterland and foreland transportation while coordinating the broader supply chain. This transformation integrates multiple operational activities involving various stakeholders, making it challenging to oversee all port terminal functions. This study adopts a system‐based perspective, viewing port terminals as interconnected processes that generate value for operators. Using a system dynamics approach, it quantifies the value generated by different port operations at the Port of Tanjung Priok in Jakarta, Indonesia. The findings reveal that seaside operations—such as berthing and container loading/unloading—contribute the most value to terminal operators, followed by transit and storage and landside operations. Quantifying the contribution of different operational activities for value creation, this study demonstrates how value creation can be operationalized and systematically evaluated, thereby moving beyond predominantly qualitative approaches in existing research. These insights are valuable not only for port managers seeking to optimize resource allocation but also for policymakers developing national frameworks to enhance port performance and regional development.
Many global institutions and countries now recognize the need for adaptation to climate-induced events and are embracing renewable energy to mitigate climate change. These efforts will be undermined, however, because of the two neglected feedback loop processes addressed in this paper. Analyses incorporating the feedback loops are explained below in detail. The analyses indicate that expected efforts to mitigate climate change will not prevent unacceptable consequences. Countries will therefore need to accommodate polycrisis management dynamics. This is a task that would be well-suited to the System Dynamics community.
The rise of generative artificial intelligence (AI) has introduced new possibilities for automating qualitative system dynamics modeling, lowering barriers to entry while raising concerns about methodological rigor and ethical integrity. This paper presents the qualitative engine, an AI-assisted modeling tool developed as part of the open-source sd-ai platform, designed to help users generate causal loop diagrams (CLDs) using large language models (LLMs). Unlike prior systems such as SDBot, the qualitative engine employs a zero-shot, one-pass prompting strategy with structured JSON outputs to ensure consistent, parseable results. It supports incremental model building and contextual integration of problem statements and background information, enabling iterative human-AI collaboration. The paper also describes the results of assessing the engine on evaluation frameworks: causal translation (accuracy in extracting trivially stated causal relationships) meant to assess only the most basic form of competency and conformance (adherence to user instructions on scope and level of complexity) meant to assess basic capabilities related to abstraction. Using the gemini-2.5-flash-preview-09-2025 LLM, the engine achieved 100% accuracy in a causal translation evaluation and 78% success in a conformance evaluation. These findings demonstrate the potential of open, modular AI-powered tools to enhance users' abilities to carry out SD modeling.
System dynamics models offer powerful tools for understanding complex systems and informing policy. Yet, their impact often falls short due to limited stakeholder engagement and poor implementation. Here, we explore how system dynamics models can be designed for real-world impact by integrating analytical rigor, transparent modeling, and interactive stakeholder engagement. We build on the literature on engaging stakeholders through management flight simulators, gamification and storytelling, and utilize the C-ROADS and En-ROADS climate simulators and insights from the MIT Climate Pathways Project. To date, more than 492,000 people in 185 countries, including more than 23,000 leaders in government, business, investing, and civil society, have participated in interactive sessions with these simulators. We present three key design principles: (1) rigorous modeling and model transparency, (2) intuitive model interfaces, and (3) facilitated, interactive simulation-based experiences. These principles enable users-from students to senior policymakers-to challenge mental models, explore trade-offs in a safe, engaging environment, and learn for themselves. We discuss limitations and directions for refinement of these principles. The paper contributes to the system dynamics and broader practice literature by offering actionable insights for designing models and protocols for their use that catalyze learning and informed decision-making in complex policy environments.
Compartment models are widely used in system dynamics to analyze complex systems. However, determining an appropriate level of aggregation remains a methodological challenge. This paper explores how parallel and serial (dis-)aggregation affect a model's ability to reproduce dynamics of interest, using the beer distribution game (BDG) as a case study. We review existing aggregation guidelines that emphasize similarity of decision rules, flow purpose, and residence times, and assess their completeness through comparative simulations of (dis-)aggregated BDG models. Our results suggest that while these guidelines are generally useful, they are incomplete: symmetry of flows in parallel structures emerges as an additional, previously neglected criterion, and including fast dynamics in models focused on long-term behavior may add excessive detail without improving insight. These findings provide an initial step toward guiding modelers in balancing model simplicity and usefulness and motivate further research on whether, how, and why they extend beyond the BDG.
Speculative bubble formation has been a long observed feature of markets, and continues to be a major focus of behavioral finance, economics, marketing, and operations management research. While these bubbles arise from specific physical or informational features of specific markets, their formation is also driven by the interplay between realized prices and expected rewards, balanced against the increasing risk of loss from speculation. As in macroeconomic stability literature of the prior century, this work applies classical proportional-integral-derivative (PID) controller design to a sociotechnical context. However, unlike this prior approach, this work extends this classic model for speculative bubbles by explicitly incorporating feedback between the those making choices and the systems in which they are embedded. Three case studies are presented to illustrate how this parsimonious and context-agnostic approach can classify and describe both value-creating and value-destroying speculation.
This paper explores the integration of Artificial Intelligence (AI) via large language models (LLMs) into the practice of System Dynamics to lower barriers to entry and accelerate model development, understanding, and application. The core of this work consists of two new engines within the sd-ai platform, an open-source framework that links natural language interfaces to simulation-ready models: the quantitative engine, which generates fully specified stock-and-flow models complete with equations, units, and documentation; and Seldon, an LLM-based modeling companion that uses natural language dialogue to explain model behavior, feedback structures, and key insights. The paper details the sd-json schema for encoding model structure, the prompting strategies that minimize hallucinations and enable iterative, human-in-the-loop model development, and the use of Loops that Matter (LTM) analysis to ground Seldon's explanations in verifiable model behavior. Using a fermentation case study, we demonstrate how these tools can be used to collaboratively build, critique, and refine a model, as well as to identify structural limitations, and generate policy recommendations. The results show that LLM-assisted modeling can replicate core phases of the System Dynamics methodology: problem definition, model formulation, behavior explanation, and policy design, while reducing the time traditionally required. We argue that these tools signal a new "third age" of System Dynamics modeling, in which human modelers focus on framing problems and interpreting results, while AI systems help the user to handle technical implementation and explanation.
A high level of entrepreneurship is a positive indicator of a country's potential for economic growth. However, some countries are more entrepreneurial than others, and this trend is often difficult to change. This paper addresses why the entrepreneur-to-population ratio remains so consistent within a country over time. The analysis is based on the Spanish case, which is one of the countries with the lowest entrepreneurship rates worldwide, despite the regulative efforts of governments over the past 20 years. Drawing on Bass's adoption-diffusion theory of innovations, while also considering the regulative, normative, and cultural-cognitive institutional dimensions affecting the entrepreneurs' behavior, we have designed a model that reproduces the flow of entrepreneurs through an entrepreneurial process consisting of three stages: potential entrepreneurs, early-stage entrepreneurs, and established entrepreneurs. Our main findings suggest that the number of entrepreneurs goes after a dynamic of growth and stagnation throughout the process. The Adoption and Diffusion of Entrepreneurs' Behavior (ADEB) model proposed herein explains both the time-limited effect of regulative incentives on the diffusion of entrepreneurs through the entrepreneurial process, as well as the long-term effect of normative and cultural-cognitive institutional dimensions. The theoretical discussion derived from this dynamic analysis provides relevant implications for academics, policy makers, and entrepreneurs.
Management research has been dominated by variance theories that explain outcomes through statistical associations. While process theory complements variance theory by focusing on how and why events, activities, and choices interact over time, it remains challenging to acquire rich, longitudinal process data and translate them into novel, abstract process theories. We propose integrating qualitative meta-analysis (QMA) with system dynamics (SD) modeling to address these challenges. QMA systematically synthesizes temporally rich evidence across multiple qualitative case studies, while SD provides a formal language for articulating causal mechanisms, accommodating temporal complexity, and testing dynamic hypotheses. Using a recent application to servitization, we illustrate how QMA-SD can be implemented in a step-by-step manner. Beyond its substantive contribution to the servitization literature, we demonstrate the broader potential of QMA-SD for generating empirically grounded, generalizable, and testable process theories.
Loops that matter (LTM) has gained traction in system dynamics for its ability to algorithmically identify feedback loops and shifts in loop dominance. Yet, its unstructured use may encourage superficial analyses and hinder systems understanding among new learners. To maximize the pedagogical value of LTM, we developed a structured curriculum that integrates LTM into existing feedback analysis pedagogy, emphasizing feedback narratives and dynamic insights. Delivered to two cohorts of new system dynamics students, the curriculum comprises two lesson plans: (1) feedback loop analysis and narrative building and (2) identifying leverage points for systems change. Our evaluation of the curriculum's impact on learner performance, across three cohorts of master's students (2022-2024), show significant and meaningful improvements in feedback analysis scores among participants. We further reflect on the insights and lessons we learned and provide pedagogical recommendations for other educators.
This paper applies system dynamics modeling to compare the effectiveness of cash-based assistance (CBA) and in-kind aid in conflict-affected Sudan. Using primary data collected from local stakeholders, the model simulates aid distribution dynamics under three conflict intensity scenarios: Constant violence, intensified violence, and declining violence. The study evaluates both reachability and cost efficiency of each aid modality across these scenarios. Findings suggest that Policy A—expanding the number of capable local actors—improves aid delivery in all contexts but is, particularly, effective under intensified violence, where in-kind aid supported by widespread local engagement is more resilient. In contrast, Policy B, which shifts aid allocation toward CBA, proves more efficient in the constant and declining violence scenarios, where reduced transaction costs and greater flexibility enhance delivery. The research highlights the need for adaptive aid strategies tailored to evolving security conditions and emphasizes the critical role of local actors in optimizing humanitarian response.
This paper introduces the Qualitative Systems Exploration Model (QSEM), a new semi-quantitative framework for systematically interpreting and analyzing Causal Loop Diagrams within participatory system dynamics. QSEM is applied at late-stage model conceptualization and offers researchers and modeling practitioners a set of tools and techniques to improve transparency and reproducibility in model assessment, ensuring that component and feedback structure selections are traceable and well justified. Throughout its three core phases: (i) System Factor Classification; (ii) Loops of Interest; and (iii) Archetype Identification and Analysis, QSEM integrates with established Group Model Building scripts to facilitate structured participant engagement and collaborative sensemaking. Real-world application is demonstrated in a commissioned government project aimed at understanding factors influencing dietary choices and food systems, where the framework helped identify potential policy-relevant system drivers. Future directions involve applying QSEM in other projects, evaluating its robustness, consistency, and scalability, refining archetype detection, enhancing data visualization, and exploring dedicated software solutions to expand its utility in qualitative SD.
Deep learning provides a set of techniques for detecting complex patterns in data and is a critical component of the burgeoning artificial intelligence revolution, enabling transformative advancement in a variety of fields. However, when the causal structure of the data-generating process is underspecified, deep learning models can be brittle, lacking robustness to shifts in data-generating distributions. In this paper, we demonstrate that methods and concepts familiar to system dynamics modelers can be used to address this problem of brittleness, thereby improving the efficacy of deep learning systems. Specifically, we turn to loop polarity analysis as a tool for specifying the causal structure of a data-generating process, in order to encode a more robust understanding of the relationship between system structure and system behavior within the deep learning pipeline. We use simulated epidemic data based on an SIR model to demonstrate how measuring the polarity of the accumulations of the different feedback loops that compose a system can lead to more robust inferences on the part of neural networks, improving out-of-distribution performance and infusing a system-dynamics-inspired approach into the deep learning pipeline. This case study provides one example of how to leverage an understanding of the causal structure of a data-generating process to extract low-dimensional summary statistics that in turn allow us to build more robust deep learning pipelines. Code for this paper is available at https://github.com/davidbkinney/loop_polarity_underspecification .
Some of the greatest developments in our field contain Mixed Blessings: positives and negatives, good stuff and bad stuff, brilliant aspects and actual flaws. In this paper I will call them 'Blessings' and 'Bugs'. The Blessings help us do our strongest work, increase our skills, develop our insights, and expand our reach. But sometimes Blessings can turn into Bugs. They can get in our way by appearing to be indispensable, by insisting on being used, by substituting seductive simplicity for necessary deep thinking, by distracting us away from what really is going on, by cluttering our reports when they do not apply, by confusing our audiences when something else would work better. And when they are actually, subtly, wrong. This note offers reflections on the Blessings and Bugs in system dynamics practice as I have seen them, with the goal of strengthening and sharpening our field.
A high-quality simulation model should help its users to easily and appropriately calibrate their trust in the model. Traditional evaluation metrics such as validation and robustness are necessary but insufficient for this task. Trust calibration depends on factors like the model's transparency, applicability to intended use, usability, reputation, and consideration of potential bias. This article proposes a framework for designing and evaluating system dynamics models by considering factors that contribute to the proper calibration of user trust. This framework takes inspiration from trusted artificial intelligence, broadening our traditional concept of model quality and explicitly focusing on what users need to consider a model trustworthy and to understand the model's relevance to its intended purpose. The trusted simulation framework can improve our integration of model quality activities throughout the modeling process, leading to more impactful and better-targeted model design, development, and evaluation.
Although maternal morbidity and mortality are persistent public health burdens in the U.S., few studies have addressed these outcomes from a complex systems perspective. To bridge this knowledge gap, two SD modeling projects that apply community-based system dynamics group model building (SD GMB) are ongoing. With core SD GMB meetings with community participants having taken place within the same calendar month, the close timing of these processes provided a unique opportunity for rapid learning by the research team and the adaptation of the specific approaches employed across projects. Applying tenets of reflexivity, this paper describes how SD GMB activities were conceptualized, planned, and implemented, offering reflections on the effectiveness of our approach by drawing upon our own insights about successes, failures, and apparent enigmas in each project. These reflections are described as five 'lessons' that may offer valuable insights for others engaging in the formative phase of SD GMB projects.
