Abstract
{ "background": "Transport maintenance depot systems in developing nations often operate with suboptimal yield, leading to resource wastage and service delays. In Rwanda, systematic evaluation of depot performance has been hindered by data sparsity and the hierarchical nature of operational data across regions and time.", "purpose and objectives": "This case study presents and evaluates a novel Bayesian hierarchical modelling framework designed to quantify and improve yield within the nation's transport maintenance depot system. The objective is to provide a robust methodological tool for performance assessment under uncertainty.", "methodology": "A Bayesian hierarchical model was developed, explicitly modelling depot-level yields as functions of regional effects and time-varying covariates. The core structure is $y{i,t} \\sim \\text{Normal}(\\alpha{j[i]} + \\beta X{i,t}, \\sigma^2)$, with $\\alphaj \\sim \\text{Normal}(\\mu{\\alpha}, \\tau^2)$, where $y{i,t}$ is the yield for depot $i$ at time $t$, and $\\alpha_j$ are regional intercepts. Inference was performed using Hamiltonian Monte Carlo.", "findings": "The model successfully identified significant regional heterogeneity in baseline performance, with posterior probability exceeding 0.95 that the central region's intercept parameter was highest. A key finding was a modelled average yield improvement of approximately 17% attributable to the systematic intervention period, with the 90% credible interval for this effect being [12.3%, 21.8%].", "conclusion": "The Bayesian hierarchical model provides a statistically robust framework for analysing depot system yield, effectively handling data limitations and quantifying uncertainty. It moves beyond descriptive metrics to offer a probabilistic assessment of improvement drivers.", "recommendations": "Adopt the presented modelling framework for ongoing performance monitoring. Depot managers should prioritise resource allocation to regions with lower modelled baseline intercepts. Future system expansions should incorporate the collection of covariate data specified by the model.", "key words": "Bayesian inference, hierarchical modelling,