MODEL SELECTION · Application of Model Selection Principles to Distinguish Network and Latent Variable Models of Psychological Constructs

Psychological research aims to understand constructs that exist in the minds of individuals and that affect individuals and their societies, such as anxiety, racism, intelligence, and happiness. Researchers use statistical models to study these constructs, and the models they choose affect the conclusions that are made and future research questions.The current standard in psychology is to construe and model psychological constructs as latent causal variables that give rise to measurable variables. This standard is being challenged by a new movement to construe psychological phenomena as networks of interrelated variables in a causal system. The two approaches are radically different in what they imply about the nature and structure of psychological constructs and their causes and effects. Until now, the network and latent variable methods for representing and modeling psychological constructs have been developed and studied independently. The proposed research aims to integrate these fields of methodological research.The main research objectives are, first, to distinguish the network and latent variable modeling frameworks by explicating their divergent theoretical implications; second, to develop and test ways of comparing the statistical models implied by each framework in terms of parsimony and fit; and third, to develop and test statistical methods for comparing the validity of the two frameworks in terms of their ability to situate the construct within the larger theoretical space. The overarching project goal is to provide guidance to applied researchers in the social sciences as to how to choose a statistical model for their data based on both theoretical and empirical considerations.