Bayesian network modelling of hierarchical composite indicators.

The water, sanitation and hygiene (WaSH) sector has witnessed the development of multiple tools for multidimensional monitoring. Hierarchical and composite indicator (CI)-based conceptual frameworks provide one illustrative example. However, this approach does not address the existing interrelationship of the indicators they comprise. Bayesian networks (BNs) are increasingly being exploited to assess WaSH issues and to support planning and decision-making processes. Here, we aim to evaluate the validity, reliability and feasibility of BNs in replicating an existing CI-based conceptual framework. We adopt a data-driven approach and propose a semi-automatic methodology. As a pilot study, we used the regional monitoring initiative Rural Water Supply and Sanitation Information System (SIASAR). Data from two different countries are processed and analysed to calibrate and validate the model and the method. The major findings show: i) the model inference capacity improves when structure is provided to the networks (according to the CI-based framework); ii) key components that explain a pre-defined objective variable are reduced and quantified (implying important advantages in data updating); and iii) interlinkages among these components can be identified (which might enhance multi- and trans-disciplinary actions). We conclude that BNs accurately replicate the CI-based conceptual framework, with great potential for a wider application.