B0658
Title: Evaluating latent structures in the graphical network model: visual exploration and hypothesis testing
Authors: Jinyuan Liu - Vanderbilt University (United States) [presenting]
Abstract: While the classical factor models assume the covariance between observed items arises from some latent factors, a recent graphical network model has been formalized to conceptualize such covariance as a result of their pairwise interactions. These two perspectives complement each other to advance the standardized measurements of many complex traits, such as neuro-degenerated or psychiatric diseases. However, of growing interest is to demonstrate the validity of the latent factor structure in the network model by harmonizing these two perspectives. Current visual inspection as an indirect validation is unsatisfactory, hence it is proposed to establish a formal hypothesis testing utilizing a permutation-based multivariate ANOVA framework to overcome the challenges including non-independence in the network structure. This timely solution for evaluating the psychometric factors in a graphical network model is illustrated with the positive and negative syndrome scale (PANSS) for measuring the symptom severity of schizophrenia.
