Abstract for: Disaggregation of a Stock Variable Based on Attribute Distribution
A co-flow structure, with built in table functions based on cumulative distribution properties, is used to approximately disaggregate a perfectly mixed stock into two sub-stocks. An implementation of this structure requires a priori knowledge of the distribution of allied co-flow attribute that is presumed to be a random variable. A user can specify the fractile threshold (Z value) for this attribute around which the stock can be split. This structure is tested for a variety of conditions. The goal of these tests is to examine whether the co-flow based partitioning is robust to variations in (i) different structural parameters (e.g. time needed to adjust the stock) and (ii) distribution properties of the co-flow attribute. Results show that the specified table function based on cumulative distribution functions can correctly separate a perfectly mixed stock, on average, for normal or exponentially distributed attributes. Limitations of the approximation for estimating attribute averages are documented. Implications of these findings for comparing system dynamics models against agent based models, and for inferring the results of Monte Carlo simulations that involve smoothing of flows, are discussed.