For matrices that are used in several relations, another given abbreviation may be used for notational simplicity. Because of the number of matrices used, the abbreviation names will tend to indicate something about the relationship being modeled. All matrices will be indicated in bold.
Scalar values will be indicated in italics. Transpose matrices will be indicated by Matrix ' 3. Individual elements are coded between 0 and 1 such that the row total total of any specific actor's goals will sum to one. Individual elements are coded between -1 and 1, where a value of -1 indicates complete opposition, 0 indicates no relation, and 1 indicates complete correspondence. Individual elements are coded between -1 and 1, where a value of -1 indicates loss of all of a resource, 0 indicates no effect, and 1 indicates a gain of all of a resource n.
For instance, if money is a resource, a value of 1 would correspond to gaining an extremely large amount of money relative to a value of. Intuitively, actors will not consciously take actions that are contrary to their interests; however, they will not always take the action that best supports those actions.
Actors are also spurred to act by 'tactical preference,' which models both cultural preferences and institutional inertia for certain types of actions. This cultural preference is a product of dynamic social influences, however - as actors' feelings towards other actors change, so do the social influences effecting their behavior change. In the discussion of the calculation of actor action probabilities, we will discuss the mathematics by which the actor-action effectiveness relationship is calculated; how the actor-action preference relationship is calculated and socially influenced; and how an overall probability of action is derived from the two relations.
Goalij is artificially constrained between 0 and 1, but is not a probability i. The fundamental formula is the cornerstone formulation of Friedkin's structural influence theory see Friedkin , Influence values are constrained between 0 and 1, where 1 indicates that actor j is responsible for all of actor i's preferences, and 0 indicates that actor j's beliefs have no effect on actor i's.
The sum of all influences on actor i equals 1. Note that Influence1ii has special meaning - it is the effect of the actor's initial preferences on its preferences at any given time period; referred to, by the author, as the 'stubbornness' attribute. These are combined into the stubbornness weighting relation described below. Note that the influence is a dynamic relationship; that is, as discussed later in the model, actors change their influences based on the outcomes of other's actions.
As mentioned above, Sii equals Influence1ii. The tactical preference matrix is a dynamic matrix that indicates the willingness to adopt, or opinion about, certain actions; possessed by the various actors.
It is dynamic, because actors learn from each other over various time periods. For the purposes of this model, 'time period' represents both an arbitrary rate at which actors learn and the time it takes to make a decision.
The preferences for each actor do not have to sum to one; rather, they indicate a relative value compared to the actor's preferences for other actions. Equation 2: Tactical preference matrix. Or, Ptij equals the effectiveness of action j multiplied by actor i's preference for it at time period t, divided by the actor i's sum of the above for all actions.
Simulation and Experimentation 4. Both the influence and belief formulae represent inherently dynamic behavior; and the model itself describes complex, multi-agent, non-linear, dynamic behavior. In order to develop a simulation from the model described above, two primary modules must be implemented - a model to describe the decision that actors actually take, and a model to update the scenario state based upon the effects of the decisions made.
Figure 2: Basic simulation flow diagram 4. Given that the model outputs probabilities of action based on the actors' tactical preferences and goal effectiveness, the simplest decision algorithm would be to generate a random number for each actor and select the corresponding actions. The decision model actually used is slightly more complicated. After the probability matrix is generated in step 1, the simulation determines if each actor has enough resources to take the actions it would be inclined to take i.
At this point, a random action is selected. This decision model offers the advantages of modeling the resource dependence of actions - otherwise, actors have 'bottomless pockets' to act, regardless of other actors' actions against them.
These effects include both the direct effects on available resources, and indirect social effects. The updated resource available Avail matrix is then passed to the next run of the decision making model. Ideally, the more actors affect each other either positively, if both actions are positive; or negatively, if both actions are negative the more they will influence each other; while if one actor 'betrays' another one actor acts positive to the other while the other acts negative the less they will influence each other.
This relation is scaled against initial influences by an additional weight. Vii is constrained between 0 and 1. Actors become more or less hostile to other actors depending on the other actors' actions.
Once an actor's hostility towards another actor reaches a certain threshold, that actor will arbitrarily 'retaliate' against the other actor. Though not implemented, this feature demonstrates the flexibility of the decision model used in the simulation.
For each experiment, simulations were run for 20 time periods each. The dependent variable for each experiment were the 'tendency for violent actions' of actors in a scenario based on the second Intifada. The data is summarized below: Variance in violent tendencies 0.
Israel demonstrated the most variance between simulation runs, and the P-value leads us to confidently reject the null hypothesis - i.
Senstivity of volatility parameter 0. As additional search functionality is added to the model outputs, additional analysis will become possible. Conclusions and Areas for Future Research 5. This scenario model is well-oriented to the analysis of low-intensity conflict, for it can model multiple actors and take into account the opinions of parties related to, but not necessarily actively taking up arms in, the conflict of specific concern is the host population of a guerilla or terrorist movement.
Similarly, it is able to model evolution of cooperation or antagonism towards a primary actor in variegated or tribal populations e. This is especially relevant for situations in which actors have largely similar attitudes on a set of propositions, but varying goal weights. For example, in the case bargaining between industry and environmental regulators, while both parties would like to see a clean environment and abundant profits, the relative importance of the two can lead to divergent actions and subsequent breakdown in cooperation.
The meta-matrix approach is readily scalable, insofar as other relations between sets of actors can be modeled and reasoned upon in the same way as the relations presented and used in this model. Additional state metrics can be calculated for the various relations, and used in model calculations or actor reasoning.
As long as the simulation consistently evaluates the effects of actions, different decision making routines can be compared to each other across simulation runs or between actors in the same simulation. This coalition model, combined with a strategic influence model and the network-destabilization theories described in paragraph 2. Parties to the conflict can act cooperatively or antagonistically, and payoffs can be shared between actors. Sensitivity analysis was conducted to determine the effectiveness of the model.
Any opinions, findings, conclusions or recommendations expressed in this report are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the National Science Foundation, the Defense Advanced Research Projects Agency or the U. Zelikow New York, New York, Longman. Axelrod, R. Genetic Algorithms and Simulated Annealing. Los Altos, California, Morgan Kaufman: Andrews and David Knocke Eds. Stanford, CT, pp. In Dynamics of organizational societies : Models, theories and methods.
Carley, K. Forthcoming a. Arthur M. Forthcoming b. Intra-Organizational Computation and Complexity. In: Companion to Organizations, J. Baum, Ed. Forthcoming c. Smart Agents and Organizations of the Future. In: Handbook of New Media, L. Lieveroux and S. Livingstone, Eds. Friedkin, N.
A Structural Theory of Social Influence. Scribner, New York, New York. Hampton, J. Hobbes and the Social Contract Tradition. Krackhardt, D. Research in the Sociology of Organizations, Monterey, CA. Levinthal, D. Adaptation on rugged landscapes. Management Science, Laqueur, W. The Guerrilla Reader. A quantum advance… Expand.
View via Publisher. Save to Library Save. Create Alert Alert. Share This Paper. Background Citations. Methods Citations. Citation Type. Has PDF. Publication Type. More Filters. Psychology and social networks: a dynamic network theory perspective. The American psychologist. The Journal of Applied Behavioral Science.
View 3 excerpts, cites background. Dynamic network theory is a multidisciplinary framework that shows how social networks influence goal pursuits in social, organizational, and international systems Westaby, It integrates … Expand.
View 30 excerpts, cites background and methods. Extending dynamic network theory to group and social interaction analysis. This article proposes a markedly new conceptual approach to group and social interaction analysis, grounded in transformative advances in dynamic network theory. The framework first theoretically … Expand.
0コメント