A Comparative Empirical Analysis of Spatial Econometric Models for Poverty Mapping in Nigeria

Small area poverty estimation using spatial econometric models has advanced significantly in methodological sophistication, yet comparative empirical evidence on the relative performance of different spatial specifications for poverty mapping at the LGA level in Nigeria is absent from the published literature. This study conducts a comparative empirical analysis of spatial econometric model specifications for poverty rate mapping across Nigeria's 774 LGAs. Poverty headcount data from the 2022 NBS Multidimensional Poverty Index and LGA-level covariates from administrative and survey sources were used. Six spatial model specifications were compared: OLS, spatial lag, spatial error, spatially lagged X, spatial Durbin model, and geographically weighted regression (GWR). Comparison criteria included Akaike Information Criterion, log-likelihood, residual spatial autocorrelation, and leave-one-out cross-validated prediction error. Moran's I of OLS residuals confirmed significant spatial dependence (I = 0.48, p < 0.001). Spatial Durbin model achieved the best AIC and lowest cross-validated prediction error (RMSE = 5.8 percent), outperforming both spatial lag (RMSE = 7.1 percent) and spatial error (RMSE = 6.9 percent) models. GWR revealed substantial spatial non-stationarity in covariate effects on poverty, with education effects three times stronger in South West than North East LGAs. The study contributes a systematic model comparison framework for Nigerian poverty mapping and recommends Spatial Durbin Model as the default specification for national poverty map production by the National Bureau of Statistics. Keywords: spatial econometrics, poverty mapping, spatial Durbin model, geographically weighted regression, Nigeria