Learning Network Structure for Ordinal Outcomes: An Application of Inductive Causation to Food Safety in Cambodia

Abstract The theoretical frameworks of the Capability, Opportunity, Motivation-Behavior (COM-B) model and the Theoretical Domains Framework (TDF) are often used by social scientists to study interconnections between determinants of human behavior. In this study, we seek to empirically validate COM-B by learning network structures from data on behavioral determinants to food safety practices in Cambodian markets. To this end, we implemented a carefully tailored combination of statistical methods in a novel context that is uniquely aligned with the motivating problem. Specifically, we leveraged a search algorithm for network structures with methodological adaptations for ordinal survey data. Data consisted of responses from 169 participants to 18 survey items formulated according to TDF domains within COM-B constructs for food safety and measured on an ordinal 1-to-7 Likert scale. We adapted the Inductive Causation algorithm to learn network structure from ordinal data. Marginal and partial Spearman rank correlations for all pairs of survey items yielded an undirected dependency graph, followed by directed-separation to identify unshielded colliders. Results showed a dense network structure of survey items depicting closely interconnected behavioral determinants of food safety practices in Cambodian markets. Findings were consistent with theoretical expectations and provided empirical support for COM-B. Yet, only a limited number of network edges were oriented based on these data. Additional empirical and methodological work is warranted to further refine insight into a data-informed COM-B network, with implications for development of targeted interventions that promote behavioral change. Supplementary materials accompanying this paper appear on-line.