Previous studies on water treatment systems in the region have often employed single-level regression models, which may inadequately account for the hierarchical structure of facility data and lead to biased inferences regarding system performance. This replication study aims to critically evaluate the methodological robustness of prior yield optimisation analyses and to apply a multilevel modelling framework to provide more accurate estimates of key performance determinants. We replicated the data collection protocol from a foundational study, gathering operational data from a stratified sample of treatment facilities. A three-level linear mixed model was fitted: $Y_{ijk} = \beta_0 + \beta_1 X_{1ijk} + u_{k} + v_{jk} + e_{ijk}$, where $u_{k}$ and $v_{jk}$ are random intercepts for region and facility, respectively. Inference was based on robust standard errors. The multilevel analysis revealed that the effect of coagulant dosage on yield was substantially overestimated in prior single-level models by approximately 18%. The facility-level random effects were significant (p < 0.01), confirming the necessity of the hierarchical approach. The application of a multilevel regression framework corrects for previous methodological shortcomings, offering more reliable parameter estimates for engineering decision-making in yield optimisation. Future performance analyses of distributed infrastructure systems should adopt hierarchical modelling techniques to account for clustered data structures. Practitioners should recalibrate dosage models using the corrected coefficients. multilevel modelling, replication study, water treatment, yield optimisation, Kenya, hierarchical data This study provides a novel methodological diagnostic and a corrected statistical model for infrastructure performance analysis, demonstrating that accounting for regional and facility-level clustering alters key engineering parameters.