Stochastic models of sociocentric networks were developed for testing hypotheses about micro-level dependencies (e.g., clustering, preferential attachment, or homophily) on the basis of empirical network data. Due to the complex nature of sociocentric networks, parameter estimates of these models are typically obtained by simulation-based inference. This opens up the possibility of using these models as simulation tools, and study emergent macro-level phenomena with them. The combination of fitting the models to empirical data sets and using them to explain macro-level outcomes renders these models powerful tools for sociological inquiry into interdependent social systems. In this presentation, the use of exponential random graph models and stochastic actor-oriented models as generative models for such networked social systems is discussed. As illustration, the case of achievement segregation in a highly competitive setting of tertiary education will be investigated, paying special attention to the relative contributions of peer influence processes and partner selection processes to the overall segregation level.