{ "background": "Manufacturing productivity in many developing economies is constrained by systemic inefficiencies. In Uganda, a critical gap exists in the rigorous, data-driven evaluation of plant systems to inform evidence-based industrial policy, particularly within the engineering sector.", "purpose and objectives": "This policy analysis aims to methodologically evaluate manufacturing plant systems and quantify the determinants of yield improvement. The objective is to develop a robust analytical framework to identify key leverage points for policy intervention.", "methodology": "A multilevel regression model is employed, nesting production batches within plants. The core model is $Y{ij} = \\beta{0j} + \\beta{1}X{1ij} + ... + \\beta{n}X{nij} + u{0j} + e{ij}$, where $\\beta_{0j}$ is the plant-specific intercept. Analysis uses plant-level operational data, with inference based on robust standard errors.", "findings": "The multilevel analysis reveals that preventive maintenance scheduling explains a significant proportion of yield variance (approximately 22%) between plants. A one-standard-deviation improvement in maintenance protocols is associated with a 7.3% yield increase (95% CI: 4.1% to 10.5%), whereas raw material quality controls showed a weaker, non-significant effect.", "conclusion": "Systemic yield improvement is most effectively driven by investments in procedural and human capital factors, such as maintenance regimes, rather than isolated input controls. This necessitates a shift in policy focus.", "recommendations": "Industrial policy should incentivise the adoption of structured operational protocols. Specifically, support should target capacity building in predictive maintenance and data analytics, moving beyond traditional subsidies for equipment or raw materials.", "key words": "manufacturing systems, yield improvement, multilevel modelling, industrial policy, operational efficiency", "contribution statement": "This paper provides a novel methodological framework for analysing hierarchical plant data and presents the first application of multilevel regression to quantify systemic drivers of manufacturing yield in