Enabling Mixed Method Modeling Through the Use of Causal Networks

Abstract

Mixed method modeling has been shown to be successful for understanding dynamic operating environments (e.g., geopolitics) by combining diverse computational models. These models tend to be fairly homogenous in that they focus on the same dependent variables, which both simplifies composition but also limits their applicability in broader domains. For complex problems, such as analyzing the factors that affect human populations in dense urban areas, there already exist many varied yet complementary models that could be leveraged in concert. However, challenges are encountered when combining independently developed models not designed with interoperability in mind. Such models can have a wide variety of data inputs, differing outputs at potentially differing timescales, implicit domain restrictions, and distinct underlying computational mechanisms. This paper investigates the use of causal networks of information to help apply these disparate models towards a common goal. We first discuss the challenges of model representation, provisioning, composition, and explanation in traditional mixed method modeling approaches. We present potential advantages and limitations that the addition of causal information provides. Finally, we offer an example of our approach using multiple models to understand security courses of action in an urban environment.