International Conference on Computational Statistics (COMPSTAT) edition:21 location:Geneva, Switzerland date:19-22 August 2014
Recently, a network perspective to psychopathology has been proposed, based on vector autoregressive (VAR) modeling of time series data. Contrary to the disease model that is operationalized in traditional psychometric analyses, the network approach focuses on the dynamic interplay between symptoms. To examine their vicious causal connections and formulate suggestions for treatment, one can inspect the VAR coefficients, which constitute the edges of the network. But what do these edges exactly mean? Which statements can or cannot be made? For instance, can the strength of the edges be directly compared? We will discuss several ways of scaling the VAR coefficients and list their pros and cons. It will be argued that different standardizations provide different angles on the same process. Moreover, we will discuss how the network representation can be enriched to facilitate a more nuanced interpretation of the edges.