This paper describes some of the central, non-procedural aspects of sensitivity analysis in system dynamics.First section focuses on the objectives of sensitivity analysis in this particular field of modeling.The second section concentrates on the types of model change involved, with emphasis on changes in model structure and parameters.The third section discusses the interpretation of model response to changes. The central questions are how the sensitivity is judged and by whom.The final section discusses the parts in the modeling process entailing sensitivity testing.Overall the paper asserts a more comprehensive role for sensitivity analysis than seems to be commonly accepted among model builders and model users. The subjectivity and individuality of sensitivity analysis is also emphasized.
The basic assumption of this paper is that system dynamics in its original form was developed to suit policy-making in small organizations and that application of system dynamics in the field of public policy must be accompanied by change in research methodology and organization. To support this view, the paper describes experiences from a study of the Scandinavian forestry and forest industry.The model building process, interaction with decision-makers, and the organization of empirical research are analyzed separately. Based on the analysis a procedure for using system dynamics in public policy analysis is recommended. In the recommended procedure a reference group representing various client groups serves a source of qualitative information and as a channel for implementation. The need to keep model building well focused is stressed. Parallel studies of historical development on the micro- and the macro-level are suggested as a means to speed up modeling. It is finally recommended that the major results from the analysis are presented in a non-technical report.
Model building standards within the field of system dynamics are still evolving. This paper offers some general guidelines for development and presentation of refined models. Model refinement, the core of the modeling process, encompasses incremental structural and/or parametric changes to existing models. Development and presentation of refined models are enhanced through comparison of original and refined model behaviour and through comparison of policy response. Model comparison aids the modeler in identifying misspecification of new structure. In addition, presentation of comparison results assists the reader in evaluating the merits of the refined as compared to the original model, and helps to insure that the builder and user of the refined model is familiar with original model assumptions.
System Dynamics (SD) may be viewed as a process of designing ROBUST systems. The concept of ROBUSTNESS leads to a need for analyzing the effects on SD models of both parameter changes and stochastic inputs. It is demonstrated that the effects of large parameter changes can be measured by the use of hill climbing techniques given efficient computation. The paper describes the traditional ways of assessing sensitivities in SD models, together with methods based on perturbation techniques which unify the parameter and stochastic sensitivity problems. The computational characteristics of the various methods are analysed and the factors that affect their computational efficiency are discussed.The paper discusses the results of experiments to determine the accuracy and speed of the various methods on a 7 state variable, 16 parameter model and on a 70 state variable, 160 parameter model derived from it. The perturbation methods yield acceptable accuracy and for the models described reduce computer time by a factor of between 9 and 25. Compiler changes discussed in the paper would make sensitivity analysis easier and quicker and would improve techniques elsewhere in System Dynamic.
This paper establishes the importance and usefulness of a well-defined reference mode as a guide to developing transparent causal structures for system dynamics models. The importance of a transparent causal structure is two-fold: it enhances understanding the model dynamics, and it facilitates communicating to others the model and the insights derived from model simulations. The paper offers a fundamental guideline for selecting transparent causal structures the following: strive for as highly-aggregated and as simple a structure that will generate the dynamics of interest. Ability to follow the guideline depends on a well-defined reference mode, which in turn requires a clear model purpose. To illustrate how a well-defined reference mode can guide the selection of a transparent causal structure, the paper traces the development of a model of the labor market. First, the model purpose is described. Next, the evolution of the basic causal structure is discussed, utilizing the reference mode embodied in the model purpose to select a transparent structure. Finally, the causal influences on model rates of flow are highlighted. To establish the suitability of the selected structure, the paper then summarizes the results of model tests. As the paper shows, the relatively transparent causal structure chosen for the model appears capable of providing insight into the real-world labor market, and of enhancing labor-market policy analysis.
This paper presents a system dynamics model of worker mobility and wage determination in a multi-sector economy. The paper reviews the background and structure of the model, illustrates the model validation process, and sheds light on the dynamics of the labor market.
Observations of modeling efforts suggest that many models fail for managerial reasons. This paper is based on the hypothesis that 1) managerial failures occur because various facets of the modeling process are inherently hard to manage, and 2) that deliberate management can reduce or eliminate many common problems. The hypothesis is pursued by breaking the modeling procedure into a series of steps, sketching what typically does but should not happen at each of them, and putting forth some thoughts about what can be done to avoid the normal pitfalls. Particular attention is paid to mundane variables such as time allocations and finances and attitudes and emotional considerations. In general, when modeling study is not deliberately managed, the construction phase preempts the bulk of time and resources to the detriment of planning, conceptualization, testing, documentation, and client-modeler interaction. This phenomenon appears to be caused, in part, by an over-emphasis on the “harder”, more technical work of construction; by difficulty justifying work that produces no direct, tangible product; and by mental resistance to testing.
According to an implicit “start simple” principle widely accepted by system dynamics practioners, model’s complexity must be progressively increased during the modeling process. How this increase in complexity should come about has yet to be explained. In this paper, two strategies are discussed and evaluated. Since a top-down strategy starts with a high level of aggregation but includes in the model all the main variables since the first formulation, it is to be preferred to a bottom-up scheme. Moreover, the top-down strategy ensures the global coherence of the model at any stage of its conception and appears to be much more consistent with the system dynamics philosophy. This paper emphasizes the need for an adequate computer modeling language and briefly describes a first attempt. The main property of such a language is to allow a hierarchical description of models, where any composing unit can be altered without the need for a complete recompiling of the whole.
The process of attaining a useful model embraces the conceptualization, formulation, and testing stages. This paper argues that effective conceptualization can be achieved through a dynamic hypothesis (that is, a chosen time development of interest and hypotheses about the underlying mechanisms). The resulting rough, conceptual model should then be improved gradually through a recursive procedure where the model is tested, redesigned and tested again, in as many ways as possible and as long as is feasible. The paper attempts to structure the hazy topic of model construction by defining a number of terms and presents lists of dysfunctional tendencies in and guidelines for model construction.
The Scandinavian countries are approaching full utilization of the regrowth in domestic forests, and the forest industry is facing a period of much slower expansion in volume than in the past. Slower growth implies problems for the industry, forestry, and society at large. The “transition” from ample to scarce wood resources could take several forms, depending on actions taken both inside and outside the forest sector. A system dynamics simulation model has been constructed to describe different possible transition paths, and to highlight potential problems. The model purpose is not to predict what will actually happen in the future, but to describe possible futures in an internally consistent way. Such insights about the consequences of various management strategies are useful to interest groups as a basis for discussing how to reach their goals. Within the industry, there is a tendency toward temporary overexpansion of capacity. The forest sector's ability to survive under slow growth conditions could be enhanced by technological and organizational remedies. The necessary remedies will be less traumatic the earlier one accepts and acts upon the problems of finite wood supply.