Alternative methods of causal inference in a multi-agent model of land use decisions under conditions of uncertainty

Most planning support tools for location decisions do not take uncertainty about the future distribution of land uses into account. To address this problem, the authors suggested a ‘Bayesian decision network’ approach, in which the uncertain future distribution of land use impacting the utility of any specific parcel of land for a particular land use is captured in terms of conditional probability distributions, representing the beliefs about the particular land use, conditional on a set of situational and contextual variables. The degree of uncertainty about the future distribution of land use is reflected by the uniformity of the probability distributions. We suggest a sampling method to address this problem and explore its performance against a simple expected value method in the context of a retail planning problem. The method involves calculating outcome variables for a solution considered under each of a number of random draws from the set of possible future spatial configurations of land uses. This paper specifies the sampling method, investigates the setting of parameters and explores the extent to which different ways of coping with uncertainty affects outcomes of decisions for a realistic case of retail planning. The results indicate that the sampling method can provide information about uncertainty. The degree of uncertainty indicates the range of the utility for a solution, thus the retail agent knows the maximum, average and minimum utility he can achieve for his decision. Furthermore, decisions to find the best option can be made based on different decision rules according to the retail agent’s risk attitude