Complex Contagion of Desirable Behavior in Adolescent Social Networks - a Simulation Model

This thesis proposes a network spreading model for the simulation of complex contagion processes in social networks. Current models for political decision support often fail in reproducing macro phenomena that emerge from micro behavior. The approach aims at overcoming those shortcomings related to restrictions of current Dynamic Stochastic General Equilibrium (DSGE)- models to rational homogeneous individuals on the ground of the connection of Network Science and Agent-based Simulation. Hereby special attention is drawn to applications in the field of Conditional Cash Transfer Programs (CCT). Using a case study that concerns the educational commitment of adolescents in northeastern Brazil, a step by step description of model setup is given. The contribution to the current state of the research is hereby fourfold. A novel approach to model the diffusion of educational commitment among adolescents (the effort they put into learning) as a Coordination-Game is proposed and it is demonstrated that it adequately represents reality. Moreover, the problem of missing data is addressed in this thesis from the perspective of a modeler that aims at creating meaningful large-scale network simulations. Adaptions of existing link-prediction and network generation approaches as well as a combination of both are proposed as a new, well performing method to impute missing links in social networks, stemming from surveys and online sources. It is shown that both, the "Boundary Specification Problem" and the "Fixed Choice Effect" can be tackled successfully with this techniques. Moreover, the thesis proposes an implementation of a Learning Classifier System (LCS)-based decision module for the agents within the simulation model. This novel adoption of the well known approach provides the agents with bounded rationality and hence enables more realistic simulations. For the first time, it is demonstrated that this decision module mimics human reasoning about educational commitment well. Eventually, an adaption of the standard Genetic Algorithm is proposed and developed for the task of parameter estimation and fitting the simulation model to real data. It is demonstrated that the Genetic Algorithm is well suited for this task.