Disaggregating Poverty Estimates to Sub-County Level using Small Area Model: Application of EBP

Purpose: The purpose of this study was to integrate Census data with sample survey data to disaggregate poverty estimates at the Sub County level. The aim was to produce estimates that provide more granular data, as current estimates are based on County level estimates. Methodology: Small Area Estimation (SAE) was utilized by applying the Empirical Best Prediction (EBP) Model. This model assumes a unit-level approach, making use of data sets collected at the household (individual/unit) level. To achieve this, household-level secondary data from the Kenya Continuous Household Survey (KCHS, 2019) were combined with auxiliary covariates from the 2019 Kenya Population and Housing Census (KPHC). Both descriptive and inferential analyses were performed. The data was analyzed using Stata and R-software, and the results have been presented using tables, diagrams, and charts. Findings: The results demonstrate similar poverty patterns as per county data findings on Kenya poverty report, 2022. The coefficients of variation (CV) for Small Area Estimation (SAE) estimates were consistently lower across most sub-counties. The bootstrap-based measures of uncertainty, including CV and mean squared error (MSE), confirmed that the Empirical Best Predictor (EBP) estimates were more precise. Unique Contribution to Theory, Policy and Practice: The study recommended that National Statistical Organizations (NSO) implement SAE to routinely produce disaggregated poverty and development indicators at the lowest administrative levels, including Enumeration Areas. By adopting this methodology, the high costs associated with large-scale surveys currently required to achieve sufficient sample sizes for such estimates can reduce significantly. Furthermore, the study emphasizes that strengthening capacity in SAE through targeted training is essential to ensure methodological rigor, maintain data quality, and uphold high statistical standards.