Final Program - Society for Risk Analysis

this paper, we develop a novel model in which a government allocates defensive resourcesamong multiple potential targets, while reserving a portion of defensive resources(represented by the equity coefficient) for equal distribution (according togeographical areas, population, density, etc.). We consider that the defender is uncertainwhether the terrorist is strategic (adaptive) or non-strategic. The attack probabilitiesof a strategic terrorist are endogenously determined in the model, while theattack probabilities of a non-strategic terrorist are exogenously provided. By varyingthe equity coefficient, we compare the optimal defensive resource allocations amongmultiple targets, and the associated expected losses. We show that expected propertyloss increases in equity coefficient. We also conduct sensitivity analysis with regard tofour system parameters (i.e., equity type, total defense budget, cost-effectiveness ofdefense, and the probability that the terrorist is non-strategic). Extensive numericalexamples illustrate that the cost of equity (in terms of additional expected propertyloss) increases convexly in the equity coefficient. Furthermore, such cost would belower in: (a) type of equity (if the government would consider per-target equity); (b)the cost-effectiveness of defense; (c) the total defense budget; and (d) the probabilitythat the terrorist is non-strategic.172P.14 Shan X, Zhuang J; xshan@buffalo.eduUniversity at Buffalo, The State University of New YorkSUBSIDIZING TO DISRUPT A TERRORISM SUPPLY CHAIN - A FOUR-PLAYER GAMETerrorism with weapons of mass destruction (WMDs) is an urgent threat tohomeland security. The process of counter-WMD terrorism often involves multiplegovernment and terrorist group players, which is under-studied in the literature. Inthis paper, first we consider two subgames: a proliferation game between two terroristgroups or cells (where one handling the black market for profits proliferatesto the other one to attack, and this is modeled as a terrorism supply chain) and asubsidization game between two governments (where one potential WMD victimgovernment subsidizes the other host government, who can interfere with terroristactivities). Then we integrate these two subgames to study how the victim governmentcan use the strategy of subsidy to induce the host government to disrupt theterrorism supply chain. To our knowledge, this is the first game-theoretic study formodeling and optimally disrupting a terrorism supply chain in a complex 4-player scenario.We find that in the integrated game, when proliferation payment is high or low,the victim government will not subsidize the host government to destroy the blackmarket regardless of its cost. In contrast, in the subsidization subgame between thetwo governments, the decision of subsidization depends on its cost. When proliferationpayment is medium, the decision of subsidization depends on not only its costbut also the preparation cost and the attacking cost. We study three extensions: (a)a subsidization subgame of incomplete information, (b) a simultaneous-move integratedgame, and (c) an integrated game with a different sequence of moves. Findingsfrom our results would assist in government policy making.T2-C.1 Shao K; kshao@cmu.eduCarnegie Mellon UniversityBAYESIAN MODEL AVERAGING FOR BENCHMARK DOSE ESTIMA-TION FROM CONTINUOUS DATAThe use of Benchmark dose (BMD) with its lower limit, the BMDL, has beenaccepted by both government agencies and scientific communities since its introductionby Crump in 1984. Recently, Bayesian model averaging (BMA) has been proposedby a number of researchers as a method to take into account between-modeluncertainty for BMD estimation (Bailer et al. 2005, Morales et al. 2006, Shao andSmall 2011). However, the BMA method was mainly applied to estimate BMD fromdichotomous data (or quantal data) in the previous studies. In the present study, aBayesian framework is presented to calculate the BMD from continuous data basedon the concepts and methods introduced by Crump in 1995. The Bayesian methodsare used to fit alternative dose-response models for continuous data using MarkovChain Monte Carlo (MCMC) simulation for parameter estimation, and BMA (includingboth approximation and exact estimation) is used to compare and combine thealternative models. Additionally, BMA estimates for the BMD are developed, with theuncertainty in these estimates used to derive the lower bound BMDL. We believe thatthe BMA method may help risk assessors enhance the precision of BMD estimates.For the purpose of method demonstration and comparison, multiple dose-responsemodels for continuous data embedded in U.S. Environmental Protection Agency(EPA)’s Benchmark Dose Software (BMDS) are selected as examples to illustrate thismethodology.T2-J.4 Shaw H, Rocks SA, Denyer D; h.shaw@cranfield.ac.ukCranfield UniversityWHAT AFFECTS THE SHARING OF RISK KNOWLEDGE IN GOV-ERNMENT NETWORKS - A SOCIAL NETWORK ANALYSIS.Risk management is reliant on the availability and sharing of knowledge. Whilstthe exchange of codified knowledge informs practice and reduces organisational‘forgetting’, it is the sharing of tacit knowledge or advice with colleagues that offerssupport, aids decisions and fosters innovation. Interaction in social networks isparticularly important in knowledge intensive organisations or dynamic or uncertaincontexts. One such environment is a government policy group dealing with criticalrisk-based knowledge. We employ both quantitative and qualitative methods to assessthe effectiveness of knowledge sharing (KS) in a government policy group, and identifythe factors, conditions and mechanisms that affect KS in risk. A social networkapproach is employed to analyse a network of 23 officials (85.2% of the population)involved in risk-based decision making; including risk analysts, scientists, policy mak-