Observation or intervention? Inference with Bayesian belief networks in tourism research

Bayesian belief networks (BBNs) are increasingly applied in tourism research to model complex systems with uncertainty and interdependence. BBNs support two distinct modes of inference: observational inference, which updates beliefs based on observed evidence, and interventional inference, which estimates the consequences of deliberate actions. Conflating these inferential modes risks misinterpretation of model outputs and undermines the policy relevance of research findings. By situating BBNs within Bayesian epistemology and causal inference and drawing on Pearl’s causal framework, this paper clarifies the philosophical, conceptual, and methodological foundations of these two forms of inference and emphasises that valid interpretation of BBN outputs requires explicit assumptions about the data-generating process, not probabilistic associations alone. Using an illustrative example, we demonstrate that observational and interventional probabilities can diverge because of spurious associations due to confounding and collider bias and outline principles that enable causal reasoning to avoid these biases. To support responsible and replicable application, we propose a structured conceptual workflow for BBN-based inference that integrates problem formulation, causal structure specification, parameterisation with model checking, causal intervention, and transparent reporting. A review of a published tourism study illustrates how departures from this workflow can lead to observational results being misinterpreted as causal effects.