- Research Article
- 10.12681/healthresj.38338
How Social Care Services can be designed using cause-effect models and Bayesian analysis. A study in Scotland
- Jan 04, 2026
- Health & Research Journal
- Sotirios Raptis
Objective: The paper aims to model Health-and-Social-Care (H&Sc) services as Cause–Effect (CE) groups within a Bayesian framework, using cross- and self-causation dynamics. The contribution is that the data used are open and never studied before and are posted online by National Health Services Scotland (NHSS). Method and Material: The paper contributes to the ongoing discourse on how Machine Learning(ML), and Bayesian inference, can inform the purposes of policymaking. Cause–effect relationships and Bayesian methods can associate public services as causes-effects, through suitable likelihood functions and priors that binomial and normal distributions can imlplement. The study identified the optimal predictive distribution using the Maximum-a-Posteriori (MAP) estimation method. Moreover, the CE-matrix approach, enables the representation of multiple causes linked to a single effect-target in a tabular format, that facilitates interpretability and prediction. Results: The findings indicate that services related to 'Alcohol' can be predictors of other effect-services, while home-based services were identified as causes of subsequent hospital admissions. Moreover, low-demand services were observed in earlier years, particularly those with no records after 1997, whereas higher-demand services were newly introduced in later years. These findings may offer insights into latent inter-service relationships, and inform policy development. The cross- and self-causation in a Bayesian framework, determined that the posterior can be predicted by 5 to 10 previous observations and this is significantly affected by the level of zero-padding (percentage of past no-records). In later years, the CE models yield more probable demand patterns. Cause–effect relationships were identified between smoking-related services, mental-health support, and the epidemiological index of Primary-1-Education children's Body-Mass-Index (BMI). Conclusions: The conclusions drawn from this analysis may be particularly relevant for insurance providers and public policymakers, who can leverage Bayesian CE-linked service models for long-term care planning, especially for elderly and low-income populations. The validation of ser-vice interlinkages further enhances the potential for precise and efficient resource allocation.
Read more