Abstract
{ "background": "Process-control systems are critical for infrastructure efficiency, yet robust longitudinal evidence of their performance gains in developing contexts is limited. A seminal study on such systems in a Central African nation provided initial evidence but used a conventional linear model, potentially mis-specifying the hierarchical structure of site-level data.", "purpose and objectives": "This study replicates and extends the original analysis by implementing a multilevel modelling approach. The objectives are to verify the previously reported efficiency gains and to evaluate whether accounting for data clustering yields different, more reliable inferences about system performance.", "methodology": "We conducted a replication study using the original project dataset. A three-level random intercepts model was fitted: $y{ijk} = \\beta0 + \\beta1 X{ijk} + uk + v{jk} + e{ijk}$, where $uk$ and $v_{jk}$ are random effects for region and site, respectively. Inference was based on robust standard errors and 95% confidence intervals.", "findings": "The multilevel analysis corroborated the direction of positive efficiency gains but attenuated the estimated effect size. The original model overestimated the aggregate gain by approximately 18 percentage points. The intra-class correlation coefficient indicated that 32% of the variance in efficiency was attributable to regional-level clustering.", "conclusion": "While the core finding of positive gains is robust, the magnitude is sensitive to model specification. Ignoring hierarchical data structures can lead to overstated precision and inflated effect estimates in engineering performance studies.", "recommendations": "Future evaluations of distributed engineering systems should adopt multilevel modelling as a default when data is nested. Practitioners should prioritise control-system calibration in regions showing the highest random effects to maximise aggregate efficiency.", "key words": "replication study, multilevel modelling, process control, efficiency gains, infrastructure, random effects", "contribution statement": "This study provides a novel methodological reassessment, demonstrating that a hierarchical model specification significantly alters the quantified benefits of automated control systems,