A primary focus of social network analysis (SNA) is to understand actor attributes from social structures in a network. It is an interdisciplinary research topic of statistics, sociology, graph theories, and computer sciences. Despite its popularity in other fields, SNA is under-utilized in psychological and educational research. This is largely due to the lack of easy-to-use models and user-friendly software. To fill the gap, this dissertation proposes three models for SNA under an extended structural equation modeling (SEM) framework. The first model is a latent space model with a factor structure. In this model, a social network is the outcome variable and the model intends to identify covariates predicting a network. As a generalization of the first model, the second model focuses on social networks with ordinal relations among actors. A Probit regression model is used to study the association of an ordinal social network and covariates. Both models are estimated using a two-stage maximum likelihood (ML) method. The performance of the two-stage ML method is assessed through Monte Carlo simulation studies. Simulation results show that the two-stage ML method can recover both model parameters and standard errors. The third model is a mediation model with a social network as a mediator. In this model, a latent space model is used to extract underlying factors of a social network, which directly participate in the causal process between two variables. To estimate the model, a Bayesian estimation method is used and its performance is evaluated through a simulation study. The usefulness of three models is demonstrated in analyzing a friendship network data set.