Hybrid simulation involves the use of multiple simulation paradigms, and is becoming an increasingly common approach to modelling modern, complex systems. Despite growing interest in its use, little guidance exists for modellers regarding the nature and variety of hybrid simulation models. Here, we concentrate on one particular hybrid -that involving agent-based and system dynamics models. Based on an up-to-date review of the literature, we propose three basic types of hybrid agent-based system dynamics simulations, referred to here as interfaced, integrated and sequential hybrid designs. We speculate that the classification presented may also be useful for other classes of hybrid simulations. on the part of the modeller. In such cases, it may be that an alternative simulation approach, either using another modelling paradigm or a hybrid approach, could provide a simpler, more natural or more efficient solution.This paper is not intended to provide a general review of hybrid simulation. Nor does it consider all of the hybrid combinations in equal measure. Instead, it focuses primarily on the class of hybrid simulation involving agentbased simulation and system dynamics, referred to here as AB-SD simulation. Our reasons for focusing on this particular combination are firstly that it is less well researched and understood than the SD-DES combination and secondly that we believe it offers a potentially useful approach to the modelling of complex adaptive systems (CAS). Such systems are defined by the US Argonne National Laboratory [5], a leading exponent of agent-based simulation, as 'fluidly changing collections of distributed interacting components that react to both their environments and to one another'. For example, McCarthy et al. view the development of new products as a complex adaptive system [32]. They state that 'Complex adaptive systems consist of a nested and scaleable system of agents; that is, the level of system abstraction could be an individual, a group, or an organization' [32, p 442]. They go on to say that '…nonlinearity and feedback can occur at multiple levels between individual agents and between groups of agents' [32, p443].Since nonlinearity and feedback are essential parts of the SD worldview and are explicitly represented within SD models, it is little wonder that SD has often been applied to model CAS. A fine example is provided by Sterman and Wittenberg [51] who developed an SD model to illustrate Kuhn's arguments concerning the rise and fall of academic paradigms. An implicit notion within this model was that each paradigm could be regarded as an agent.This notion of agency can also be observed in several other SD models of CAS. In fact, it is not always clear whether a particular model should be regarded as purely an SD model or as a hybrid AB-SD model. Perhaps what Duggan [13] refers to as agent-oriented SD models is a more accurate description of such models. It is our belief, however, that some CAS are best represented by truly hybrid AB-SD models. In this paper, we review several ...

In this paper we propose and evaluate a method for studying technology adoption at the national level using hybrid simulation. A hybrid simulation model is developed which combines elements of system dynamics and agent-based modelling, and treats nations as adopting agents. International diffusion is modelled as a social system where the adoption of an innovation, or even just growing pressure to adopt an innovation, in one nation can then influence its adoption in others. The model is used to investigate nine different technological innovations for which sufficient international data are available. Using the available empirical data, the method of differential evolution is used to configure the model which allows the parameter space to be explored in an efficient manner, without bias or subjective disagreement. Good agreement is found between the parameters derived in this way and those reported to configure analytic models. For each of the nine innovations, we report the rank order correlation between the actual order of adoption of the innovations by nations and the order predicted by the simulation model. We also report the rank order correlations between the actual order and the order predicted by a much simpler statistical model. Improvements in the rank order correlation are shown when some form of social influence between nations is included, although there is no significant difference in results between the four different types of social influence considered by the simulation. The nine technologies investigated also appear to fall into two groups with significantly different uptake speeds. Advantages and limitations of the approach are discussed along with suggested implications for practice.

The potential of hybrid models to enhance simulations of the real world is explored. Whilst the scope for design of such models is large, the focus here brings together agent-based and system dynamics modelling within a defined architectural framework. Comprising a number of modules, each of which is implemented in a single modelling paradigm, the design of hybrid models looks to exploit the potential from a range of approaches and tools. Coded within a single programming environment, the international diffusion of technological innovation is used as a case study to highlight hybrid simulation model design and implementation. An integrated hybrid simulation design which incorporates feedback between modules in a continuous, fluid, process is employed to develop a model comprising two system dynamics modules and one agent-based module. The predictions from the hybrid model are compared to known outcomes regarding the national adoption of mobile telephony, fixed internet and fixed broadband. We conclude with some thoughts on the design of hybrid simulation models.