We study the over-time interactions of actors using a variety of quantitative and qualitative methods relevant to the research question at hand.
To study dynamics of market formation, this research integrates multiple methods, including: system dynamics approaches and the development and simulation of computational models to understand how over-time interactions across actors and market-specific factors may lead to failure or success; participatory approaches to engage stakeholders; and econometric methods, of spatio-temporal data sets, to estimate parameters important to understanding the strength of specific mechanisms and for making the models more robust. The integrative approach of these methods allows providing a broad picture and understanding of the underlying structure behind market formation dynamics.
System dynamics is both a field of study in business, engineering, and social and physical sciences and an approach to organisational problem solving at the professional level.
The field of system dynamics, created at MIT in the 1950s by Jay Forrester, is designed to help us learn about the structure and dynamics of the complex systems in which we are embedded, design high-leverage policies for sustained improvement, and catalyze successful implementation and change.
System dynamics uses informal causal maps and formal causal models with computer simulation to uncover and understand endogenous sources of system behavior. Drawing on engineering control theory and the modern theory of nonlinear dynamical systems, system dynamics often involves the development of formal models and management flight simulators to capture complex dynamics, and to create an environment for learning and policy design.
Unlike “pure engineering problems” human systems present unique challenges, including long time horizons, issues that cross disciplinary boundaries, the need to develop reliable models of human behavior, and the great difficulty of experimental testing.
Successful policy design in complex dynamic systems requires more than technical tools and mathematical models. For market formation and sustainability problems there are no purely technical solutions. To understand the problems and to be effective in identifying solutions one must consider the social, political, ecological and other impacts of proposed technical solutions. Failure to do so leads to unanticipated “side effects” that damage human welfare, and, more often than not, to the failure of the engineering solution on its own terms. To reduce the chance of falling into such traps, system dynamics is fundamentally interdisciplinary.
Computational models and simulation
As part of our research on how interactive processes of market formation unfold over time, we develop “behavioural dynamic computational models” and analyse those through simulation. The models are grounded in the system dynamics methodology described above. This method is particularly suited to developing endogenous explanations for dynamic phenomena. “Dynamics” are explained as arising primarily from the interactions among the elements and actors in the system as they unfold over time, rather than from exogenous inputs. “Behavioural” means that the models represent the decisions and actions by actors, considering the information available to those actors at each point in time and their information processing capabilities.
Simulation is particularly useful when the focus is longitudinal, nonlinear, or processual and involves dynamic interplay. Computational models, besides helping build confidence into dynamically complex empirical situations also allow counterfactual analysis. In such situations simulation allows developing an integrative and internally consistent dynamic theory, where induction from empirics alone is difficult. For example, in combining these forms analysis allows hypothesizing and testing the conditions under which stakeholders may more easily overcome barriers (thresholds) to market formation.
Participatory Modeling (Group Model-Building)
Group Model-Building is an approach where a group of stakeholders gathers in one or more sessions and is guided by a modeling team in the construction of the model. The goal is to increase insight into the problem, create alignment of mental models, and to develop a robust strategy to improve system performance. Group Model-building knows a few variations which primarily relate to qualitative and quantitative modeling, the use of preliminary models and whether to conduct sessions within a two day setting or to distribute these over a couple of months. (Source : VEnnix 1996, Encyclopedia of Life Support Systems (EOLSS))
Participative modeling workshop held in Lyon in 2018
To analyze empirically the strength of particular relationships hypothesized within the theories and models we develop, we use econometric analysis. To do this we use techniques specialized for dynamic models (such as derivatives of maximum likelihood and indirect inference). We often perform econometric analysis within a smaller, “reduced” version of the larger models, especially developed for this purpose. We estimate confidence bounds also using techniques that are appropriate for dynamics models (such as likelihood-ratio-based methods or bootstrapping).
Large empirical data sets
We compile large spatio-temporal datasets to study adoption of alternative products empirically. These sets may contain on the order of 10 million decision points. We seek research settings that provide a natural “shock,” for example related to the first product introductions within a novel product category. We then collect over-time product sales data at a fine grade level, such as ZIP or Postal code. Data must be collected for the focal innovation as well as for conventional choice options. One must also collect complementary data on relevant local demographic and environmental conditions – e.g. income, education, green preferences, fast food establishments – and product characteristics – product price, performance, and incentives – all potentially over-time evolving. A major task is to merge the data into a single set and prepare this for analysis. We complement the building of the datasets with collection of documents such as media reports providing additional background on the phenomenon we study. We use these data for populating the computer models as for econometric analysis.