{ "background": "District hospital systems in South Africa face persistent challenges in service delivery and health outcomes. Robust, longitudinal methods are required to isolate the effect of systemic interventions from secular trends and confounding factors.", "purpose and objectives": "This study aimed to quantify the causal impact of a national district hospital support programme on intervention yield, defined as the composite rate of key service outputs per facility-month, using a quasi-experimental design.", "methodology": "We employed a difference-in-differences model, $Y{it} = \\beta0 + \\beta1 (\\text{Treated}i \\times \\text{Post}t) + \\gammai + \\deltat + \\epsilon{it}$, where $Y_{it}$ is the yield for hospital $i$ in period $t$. The analysis used panel data from a national administrative dataset, with facility and time fixed effects. Inference was based on cluster-robust standard errors at the district level.", "findings": "The intervention was associated with a statistically significant increase in mean intervention yield. The estimated average treatment effect on the treated was 8.7 percentage points (95% CI: 5.2, 12.1). The parallel trends assumption was validated using lead terms, which were statistically insignificant.", "conclusion": "The support programme had a positive, significant effect on district hospital performance as measured by intervention yield. The difference-in-differences approach provided a rigorous counterfactual for policy evaluation in a complex health system.", "recommendations": "Policy makers should consider the scaled implementation of the support programme, with integrated monitoring using the yield metric. Future evaluations should incorporate cost-effectiveness analysis.", "key words": "health systems evaluation, quasi-experimental design, fixed effects, health policy, service delivery", "contribution statement": "This paper provides a novel application of the difference-in-differences framework to evaluate a large-scale health system intervention in a sub-Saharan African context, demonstrating its utility for isolating causal effects from