||A major concern in asymmetric warfare is the threat coalition operations face from insurgent group activity , . In many cases, such groups are loosely and chaotically organised, but their ideals are sociologically and psychologically embedded across members such that the group has expected behaviours that can represent a threat. Therefore understanding how groups change, interact and conflict in different situations is of significant interest. However it is frequently the case that limited pre-existing data is available from which techniques such as machine learning can be employed to make predictions of the future based on past events. As such, alternative approaches to gain insights into possible future behaviour are needed. In this context generative modelling is a promising approach which has gained considerable interest for studying social phenomena. Developing models of this nature is complex as it involves emulating human behaviour. To make progress, models need to be grounded in relevant psychological and sociological theory that can be translated to a quantitative representation. Furthermore, for modelling to support decision making, there needs to be accurate representation of a physical scenario, and confidence in the model's ability to generate “real world” behaviours needs to be examined. In this paper we consider both of these issues by examining the use of existing datasets towards parameterising and testing generative models.