Abstract
{ "background": "Longitudinal data from rural health clinics in low-resource settings are crucial for evaluating health system performance, yet methodological challenges in panel-data estimation for clinical outcomes persist. Existing approaches often fail to account for the complex, time-varying confounders and clinic-level heterogeneity characteristic of such systems.", "purpose and objectives": "This study provides a methodological evaluation of panel-data estimators for measuring clinical outcomes within a rural clinic network. Its objective is to determine the most robust estimation strategy for deriving causal inferences from observational, clinic-level longitudinal data.", "methodology": "We utilise a 26-year unbalanced panel dataset from a network of rural clinics. A suite of fixed-effects and dynamic panel models are estimated and compared. The core specification is a two-way fixed effects model: $Y{it} = \\alphai + \\lambdat + \\beta X{it} + \\epsilon{it}$, where $Y{it}$ is the clinical outcome for clinic $i$ in period $t$. Inference is based on cluster-robust standard errors at the clinic level.", "findings": "The methodological evaluation indicates that dynamic panel estimators (System GMM) outperform static models when accounting for persistence in outcomes. A key empirical finding from the preferred model is a statistically significant negative association between nurse turnover rates and antenatal care completion, with a coefficient of -0.15 (95% CI: -0.23, -0.07).", "conclusion": "The choice of panel estimator substantively influences the magnitude and significance of key determinants of clinical performance. Naïve pooled or static fixed-effects estimations can yield biased inferences in this context.", "recommendations": "Future research on health systems using panel data should employ and compare dynamic estimation techniques. Programme evaluations should prioritise the collection of consistent clinic-level data over time to facilitate such robust analyses.", "key words": "health systems research, panel data, fixed effects, dynamic models, causal inference, rural health, Ethiopia", "contribution statement": "This paper