Multilevel validation of agent-based information and innovation diffusion models
Spatially explicit agent-based models (ABM) of social diffusion processes have experienced growing attention over the last few years. These studies are grounded in rich empirical data sets for very specific application domains and aim to contribute to policy analyses and decision support. In such application contexts, the common strong argument of the method of ABM concerns the validity of the model structure for instance in terms of the representation of micro-level individual decision processes or inter-individual social dynamics. On the other hand, the models deliver very rich and detailed simulation results of future scenarios which can for instance be presented in the form of maps of certain model indicators. Once detailed empirical datasets and real-world geography are included in the simulations, the credibility of ABMs is significantly increased if the (degree of) fit of simulation results and empirical data can be stated, in particular when offering decision support in policy contexts. This retrodictive validation exercise comes to be increasingly challenging as more and more rich social science data sets become available. The purpose of the project is to propose a systematic validation procedure for the outlined class of ABMs and to illustrate its use in the context of finished, ongoing and future ABM projects at CESR.
A first set of validation measures was developed and applied to the SPREAD ABM of the demand side of the German green electricity market. A particular feature of the SPREAD project is that detailed customer data of a green electricity provider on spatial and temporal resolution are available for retrodictive validation. In principle, the format of this historical data matches the main modelled agent behavioural variable both in time and space. The rationale of the validation procedure is to start from most aggregated indicators derived from historical data and simulation results, and step-wise increase the temporal and spatial resolution. In the multilevel retrodictive validation, each level of aggregation is shown to offer different distinct insights. On a disaggregated level, the introduced measures objectify spatial properties of simulation results. The assessment procedure developed is generic in nature. While the method is illustrated for a specific case study ABM it may in principle be applied to other simulation contexts.
May 2013 − May 2016