Robust Statistical Methods for Outlier Detection in Poverty Measurement Surveys in North West Nigeria

This study investigates the application of robust statistical methods to the problem of outlier detection in household income and expenditure data collected through poverty measurement surveys in North West Nigeria, addressing a source of measurement error that has received insufficient methodological attention in the subnational poverty analysis literature for this geopolitical zone. Household survey data on income and consumption, collected through nationally representative surveys including the Nigeria Living Standards Survey and the Harmonised Nigeria Living Standards Survey, frequently contain influential outliers attributable to data entry errors, extreme legitimate observations, and interview response errors, which can distort poverty headcount estimates and obscure genuine welfare patterns. Using microdata from the 2019 Harmonised Nigeria Living Standards Survey covering 12,000 households in Kano, Katsina, Sokoto, Zamfara, Kebbi, and Jigawa states, the study applies and compares Mahalanobis distance, minimum covariance determinant, least trimmed squares, and S-estimator approaches to outlier identification in multivariate household welfare data. The sensitivity of poverty headcount, poverty gap, and squared poverty gap indices to alternative outlier treatment strategies is assessed under the official Nigerian poverty line. Simulation studies are conducted to evaluate the performance of each detection method under scenarios with varying proportions and types of contamination. Results demonstrate that the minimum covariance determinant estimator provides the best overall balance of outlier detection power and false positive control. Keywords: robust statistics, outlier detection, poverty measurement, household surveys, North West Nigeria