Testing the Assumptions of the Network Paradigm for Studying Depression

Depression is a major public health problem. Decades of research have been conducted to create a classification system aligned with the complex phenomenological features of depression. The dominant classification system for depression is the latent paradigm, which conceptualizes observable symptoms of depression as effects of an underlying disorder. There is increasing evidence, however, that the latent model is inadequate to inform the prognosis and treatment of depression. Specifically, evidence is accumulating that symptoms of depression do not necessarily arise due to an underlying condition, but that symptoms occur as a network in which each one is causally related to a previous symptom. This dissertation critically evaluated the underlying assumptions of this “network paradigm,” one of the frameworks which had been proposed as an alternative to the traditional latent paradigm, as an appropriate model for studying depression. The first chapter systematically evaluated empirical depression network studies regarding whether the study design included an examination of the paradigm’s assumptions. In the second chapter, I investigated the relationships among depressive symptoms and determined whether causal relationships among depressive symptoms, a key assumption underlying this paradigm, could be a plausible explanation. The last chapter investigated a central controversy within the network literature regarding consistent findings and measurement error. The first chapter found that the majority of depression network studies published in the literature were not capable of providing empirical support of symptom causal relationships and often neglected to investigate the impact of measurement error. The second chapter estimated a significant relationship between two depressive symptoms - sadness and anhedonia, using an inverse probability treatment-weighted regression estimation approach in the context of longitudinal data. Causal relationships among symptoms, a key assumption underlying the network paradigm, may be a plausible explanation for the depressive symptom relationships. The third chapter found that statistical network models are not robust to measurement error through a series of simulation studies. Measurement error remained a general threat against the network paradigm, and existing network findings should be interpreted with caution. Overall, the network paradigm may be appropriate for study depression, but existing findings should be interpreted with caution. There is a need to explore the fundamental assumptions of paradigms prior to widespread application.