Casual Structural Modeling of Survey Questionnaires via a Bootstrapped Ordinal Bayesian Network Approach (Updated June 20, 2024).

This article discusses the use of survey questionnaires in psychology and social science research to measure latent traits in study subjects. While structural equation models have traditionally been used for causal analysis, their reliance on knowledge of true causal structure limits their reliability. The article proposes the use of a Bayesian network, which can learn causal structure objectively from data without prior knowledge. The authors develop a learning algorithm based on hill-climbing and bootstrapping that can identify a unique causal structure and quantify associated uncertainty. The algorithm is demonstrated using a dataset from a psychological study on the relationship between symptoms of obsessive-compulsive disorder and depression. The article also mentions the availability of a user-friendly open-source R package for implementing the proposed algorithm. [Extracted from the article]