The past two decades have given rise to significant advances in the development and application of methods for analyzing networks, particularly within the behavioral and health sciences. Network analysis considers how a set of units (or nodes) are connected to one another through directional or non-directional links (or edges). Analyzing the structure of a network involves the use of both graphical and statistical procedures, for instance, to determine the centrality of particular nodes and to detect community subgroup structures within the network. Classic examples include social networks among peers, connectivity networks within fMRI data, community structures in internet usage, and management networks within the workplace. A particularly exciting recent development is the study of networks of psychopathology symptom data. Whereas classic approaches assume that sets of symptoms reflect shared underlying pathology (e.g., latent depression causes sleep disturbances, feelings of hopelessness, etc.), a network approach conceptualizes symptoms as multiple causal elements within a complex network of behaviors. This is a powerful new development and an excellent exploration of this topic is presented in a recent paper by Borsboom and Cramer in the Annual Review of Clinical Psychology that we recommend highly. Doug Steinley will also address this application of network analysis, as well as many others, in his upcoming workshop in Chapel Hill on May 31st to June 2nd.
Selecting the number of classes (or components) is one of the most challenging decisions to make when fitting a finite mixture model (including latent class analysis and latent profile analysis). In this post, we talk through the conventional wisdom on class enumeration, as well as when this breaks down.