"background": "Persistent yield inefficiencies in manufacturing systems represent a critical barrier to industrial productivity and economic development. Current diagnostic methods often lack the statistical rigour to disentangle plant-level effects from systemic process variations, particularly in contexts with heterogeneous operational data. ", "purpose and objectives": "This working paper develops and evaluates a novel Bayesian hierarchical model specifically designed for yield improvement diagnostics. The objective is to provide a robust methodological framework for quantifying and attributing yield gains within complex, multi-plant manufacturing environments. ", "methodology": "We propose a Bayesian hierarchical model where yield Y{ij \ (\, (1-) \), with () = \ + \ Xij + ui, and ui \ N (0, \²u). Here, uᵢ represents the random effect for plant i. Inference is performed via Hamiltonian Monte Carlo, with posterior credible intervals used for uncertainty quantification. ", "findings": "The model application to a case study demonstrates its diagnostic capability, isolating a dominant systemic factor accounting for approximately 60% of the explainable yield variance. Posterior distributions indicate a 95% credible interval of 0. 12, 0. 19 for the key process parameter \, confirming a positive but uncertain effect. ", "conclusion": "The Bayesian hierarchical framework offers a statistically principled approach for yield diagnostics, effectively partitioning variation and quantifying uncertainty in performance attribution. ", "recommendations": "Manufacturing engineers and plant managers should adopt hierarchical modelling for systematic yield analysis. Further research should integrate real-time data streams to transition from diagnostic to predictive yield management. ", "key words": "Bayesian inference, hierarchical modelling, manufacturing yield, process diagnostics, industrial engineering, probabilistic modelling", "contribution statement": "This paper introduces a novel application of Bayesian hierarchical modelling for manufacturing yield diagnostics, providing a new method to attribute
Asante et al. (Mon,) studied this question.