| Since its inception in sociology, Social Network Analysis (SNA) has been transformed and transcended through the works in physics, computer science, and statistics. In this recent expansion, the literature has grown unevenly and in ways that are often far removed from practice. This dissertation studies four different complex systems which have been measured and reported as a network. There are three motivations for these applications. First, collaborators in other fields are often unfamiliar with the types of data analyses that are possible. Second, several of the previously proposed tools provide helpful ways to begin considering the problems. Third, we often need to refine these tools to suit the needs of the specific applications. These points are demonstrated in the efforts to (1) delineate healthcare community based on the physician network using an innovative implementation of agglomerative hierarchical clustering, (2) detect latent factors underlying the dyadic interaction network of students and teachers by improvising upon low-rank matrix completion algorithms, and (3) examine policy implications regarding contentious legislations of Affordable Care Act (ACA) and Wisconsin Act 10.;At the same time, advancements in computational and storage capacity allow researchers in all the disciplines to curate unprecedentedly large and diverse datasets. Making insights from them presents a challenge, and SNA comes in to partially fill this void. In this work, we present the applications of SNA in two different fields of healthcare research and education. It will proves the tremendous potential of SNA in bringing researchers a novel approach to even begin understanding the data. Each application is unique in its use of SNA. For example, to understand the structure of the labor market in Wisconsin, we improvise upon spectral clustering techniques along with stochastic block model. However, this approach is ill-suited in understanding the healthcare network in the US due to a large number of clusters needed. The four chapters in this dissertation demonstrate the importance of using SNA. The first chapter of this work proposes an automated approach in segmenting US healthcare market using the physician share patient network. Consequently, we examine the effect of Medicare shared saving payment plan, which is a part of the ACA, on the physician networks. In chapter three, we illustrates the use of low-rank matrix completion to discover the latent factors underlying the student-teacher bipartite interaction network. Lastly, using spectral clustering, we determine the partitions of Wisconsin labor network and the effect of Act 10 on this network. (Abstract shortened by ProQuest.). |
