Path and Bayesian network analyses in the complex design of a well-being survey via New Zealand's Integrated Data Infrastructure.

Introduction: In mental health research involved with sensitive features and privacy issues, using integrated data offers an efficient alternative to traditional approaches such as interviews or in-person data collection. These conventional methods frequently face logistical barriers, including low response rates among study populations. Using integrated data from multiple existing administrative and survey sources provides protection for participants and economic savings for researchers, despite being constrained by the limitations of the original data sources. Our research questions were: 1) Can we conduct path and network analyses of school absenteeism, psychosocial factors, and mental health outcomes using the integrated survey data from New Zealand's Integrated Data Infrastructure (IDI)? and 2) How can we account for the complex design of the General Social Survey (GSS) to ensure representative inference? Materials and methods: The study population was New Zealand youth enrolled in school, aged 15 years and older in 2018, who participated in the 2018 GSS. The study analyzed the Ministry of Education data integrated with the 2018 GSS survey data from New Zealand's IDI. To explore the relationship between outcomes of including the WHO-5, "perceived life worthwhileness", "perceived general health", and predictors including school absenteeism, psychosocial factors, and family factors, quantile mixed-effects regression, path analysis, and Bayesian network (BN) analysis were used. The complex survey design was calibrated using replicated weights from the GSS, a design-based method for complex sampling, in path and regression analyses, and bootstrap resampling, in BN analysis. Results: In descending order from the weighted path model, overall life satisfaction (standardized path coefficient: 0.36), perceived health condition (0.34), ease in accepting cultural identity (0.12), and trust in the education system (0.07) were all significantly (positively) related to students' mental health well-being (WHO-5). Perceived life worthwhileness correlated with the WHO-5 but was only connected to overall life satisfaction. Similar factors were identified for perceived health, with additional attributing factors, such as fear of crime in the area and the family's overall well-being. Conclusion: The integration of Bayesian networks and path analysis offers rigorous methods for detecting complex interrelationships among integrated mental health outcomes from population data and variables from a complex survey design.