Abstract
{ "background": "Evaluating the impact of technological and managerial interventions on manufacturing systems efficiency requires robust quasi-experimental methods. The difference-in-differences (DiD) model is widely applied in econometrics but its methodological rigour and assumptions are less frequently scrutinised within industrial engineering contexts, particularly in developing economies.", "purpose and objectives": "This case study aims to methodologically evaluate the application of the DiD model for measuring efficiency gains within manufacturing plants. It assesses the model's suitability, key assumptions, and practical implementation challenges in this specific industrial setting.", "methodology": "The study employs a longitudinal panel dataset from a sample of plants, distinguishing between treatment and control groups. The core statistical model is specified as $Y{it} = \\beta0 + \\beta1 \\text{Treat}i + \\beta2 \\text{Post}t + \\delta (\\text{Treat}i \\cdot \\text{Post}t) + \\epsilon_{it}$, where $\\delta$ is the average treatment effect. Inference is based on cluster-robust standard errors at the plant level to account for serial correlation.", "findings": "The methodological evaluation reveals that the parallel trends assumption, critical for DiD validity, held for core productivity metrics but was violated for energy intensity. The estimated average treatment effect on overall equipment effectiveness (OEE) was a 7.5 percentage point increase, with a 95% confidence interval of [5.2, 9.8]. Practical challenges included defining a valid control group and managing intermittent treatment adoption.", "conclusion": "The DiD model provides a structured framework for causal inference in manufacturing efficiency studies, but its application demands rigorous pre-testing of assumptions and careful design to ensure the control group is appropriate. Its strength lies in accounting for time-invariant unobserved confounders.", "recommendations": "Practitioners should formally test the parallel trends assumption using pre-intervention data and consider staggered adoption designs. Future research should explore synthetic control