A new method for constructing networks from binary data

Abstract

Network analysis is entering fields where network structures are unknown, such as psychology and the educational sciences. A crucial step in the application of network models lies in the assessment of network structure. Current methods either have serious drawbacks or are only suitable for Gaussian data. In the present paper, we present a method for assessing network structures from binary data. Although models for binary data are infamous for their computational intractability, we present a computationally efficient model for estimating network structures. The approach, which is based on Ising models as used in physics, combines logistic regression with model selection based on a Goodness-of-Fit measure to identify relevant relationships between variables that define connections in a network. A validation study shows that this method succeeds in revealing the most relevant features of a network for realistic sample sizes. We apply our proposed method to estimate the network of depression and anxiety symptoms from symptom scores of 1108 subjects. Possible extensions of the model are discussed.

PTSD symptomics: network analyses in the field of psychotraumatology

“…We hope that this special issue encourages researchers to conceptualize and model PTSD data from a network perspective, which arguably has the potential to inform and improve the efficacy of therapeutic interventions (Borsboom, 2017; Cramer et al, 2010; Hayes & Strauss, 1998; McNally, 2016). The methodological field of psychological network psychometrics has moved remarkably quickly from visualizing correlation matrices in 2010 (Cramer et al, 2010) to using sophisticated statistical models in 2014 (Van Borkulo et al, 2014). The gap between clinical sciences and methodology is slowly closing, in part due to several tutorial papers (Costantini et al, 2017; Epskamp, Borsboom, & Fried, 2017; Epskamp & Fried, 2017) that have enabled clinical researchers to apply network models to a large number of disorders (for a review see Fried, van Borkulo et al, 2017).…”

False alarm? A comprehensive reanalysis of “Evidence that psychopathology symptom networks have limited replicability” by Forbes, Wright, Markon, and Krueger (2017).

“…In addition, network estimation techniques differ in sensitivity and specificity (Van Borkulo et al, 2014), meaning that some techniques more often err on the side of caution, and as such identify fewer edges, which should be accommodated in assessing replicability.…”

“…For instance, in relative importance networks all connections are estimated, while Ising models only estimate connections that improve the fit of the model (van Borkulo et al, 2014). Similarly, given the stronger causal interpretation of edges in a DAG opposed to Ising models, it is sensible that DAG estimation methods should be more conservative than Ising model estimation methods, leading DAGs to be sparser.…”

Estructura de la sintomatología depresiva en una muestra de mujeres con menos de un año postparto.

“…En dos estudios, utilizando análisis de redes para evaluar la estructura psicopatológica, la hipersomnia o el aumento de sueño tuvieron poca o ninguna relación con el resto de síntomas depresivos (6,7). En un trabajo de validación de una escala de depresión (la Escala de Psicopatología Depresiva), la hipersomnia no se correlacionó significativamente con el puntaje total de la Escala de Depresión de Zung ni con la severidad de la depresión, medida a través de la evaluación clínica realizada por el psiquiatra (16).…”

“…La estructura de la sintomatología de los trastornos mentales no ha sido adecuadamente investigada, debido a limitaciones metodológicas y a que los estudios generalmente asumían que los síndromes estaban formados por los síntomas descritos en las clasificaciones actuales; sin embargo, últimamente se están desarrollando métodos informáticos que permiten analizar gráficamente la vinculación entre los síntomas tal como sucede con el llamado análisis de redes (6,7).…”