Out of Bounds? The Boundary Specification Problem for Centrality in Psychological Networks

摘要

The analysis of psychological networks has become common in multiple subfields including clinical, social, and personality psychology, where the focus is often on identifying highly central nodes that represent symptoms, beliefs, or traits. However, the boundaries of these networks are often ambiguous and relevant nodes are often missing from the network. In this article, we use a series of simulations to show that even under typical conditions of missingness, the centrality of nodes in an empirical psychological network are poorly correlated or uncorrelated with their centrality in a hypothetical "true" psychological network, and thus are invalid. We illustrate the implications of this lack of validity using an empirical example drawn from a recent study of political belief system networks, demonstrating that the original study would have drawn incorrect conclusions about American's most central political beliefs. We conclude by recommending that centrality measures should be computed and interpreted only in psychological networks that include (nearly) all the nodes inside a theoretically meaningful boundary. Translational Abstract Psychologists have recently begun to study psychological networks, which capture the connections between psychological constructs such as beliefs or personality traits. In many cases, these networks are used to identify the most central or important beliefs or traits. However, conclusions about which beliefs or traits are most important can be incorrect when they are based on incomplete networks. In this article, we use simulations and an empirical example to illustrate why incomplete data can lead to conclusions that are wrong. Based on these findings, we recommend that psychologists should only compute and interpret centrality in complete psychological networks.