Spatially Informed Estimation of Child Multidimensional Poverty at Local Government Area Level in Nigeria: A SAE-Inspired Regression Approach Using Mics 2021 And Worldpop 2020 Data

Abstract

This study produces Local Government Area (LGA) level estimates of child multidimensional poverty in Nigeria by integrating the 2021 Multiple Indicator Cluster Survey (MICS) with high-resolution WorldPop 2020 population density grids and conflict event data from ACLED. Using the Alkire-Foster framework, the national child Multidimensional Poverty Index (MPI) is estimated at 0.292, corresponding to a headcount ratio of 56% and an average intensity of 42.2%. This implies that approximately 55.7 million of Nigeria’s 99.6 million children experience deprivations in at least three indicators spanning education, health, and living standards. The MICS-derived estimate is conservative relative to official 2022 NBS/UNICEF/OPHI benchmarks (67.5% child headcount under the National MPI for ages 0–17; 83.5% for under-5s under the linked Child MPI), reflecting differences in survey design, indicator specifications, weighting procedures, and sampling adjustments. Because reliable direct MPI sampling variances are not available at the LGA level from public MICS outputs, the study adopts an SAE-inspired predictive framework. State-level MPI patterns are modelled as a function of logtransformed population density and conflict intensity, improving explanatory power to 75%. A pilot Fay-Herriot model is additionally fitted at the state level to illustrate hierarchical smoothing under approximate variance inputs. Geospatial processing aggregates auxiliary covariates to official LGA boundaries using GIS. Predicted LGA MPI values range from 0 to 0.569, with negative linear predictions constrained to 0 to respect the theoretical bounds of MPI. Model-based reliability assessment using coefficients of variation indicates that 651 of 774 LGAs (84%) have CV values below 15%. Results reveal strong north-south disparities, with the highest predicted burdens concentrated in northern zones. While the estimates remain exploratory and sensitive to omitted covariates, the study provides a transparent foundation for geographically targeted child poverty interventions and for future refinement through Bayesian SAE and expanded geospatial predictors