Statistical Theory of Measurement Error in Self-Reported Survey Data and Correction Methods with Applications to Nigerian Health and Economic Surveys

Self-reported data in Nigerian household health and economic surveys are subject to systematic measurement error from recall bias, social desirability bias, and interviewer effects that bias statistical estimates in directions that remain analytically uncharacterised in the Nigerian survey research literature. This dissertation develops statistical theory of measurement error in self-reported survey data and derives correction methods with applications to Nigerian health and economic survey contexts. A unified measurement error model framework accommodating classical, systematic, and differential measurement error is developed. Theoretical bias formulae are derived for regression, logistic, and probit estimators under each error type. Correction methods examined include SIMEX, regression calibration, MCMC latent variable correction, and multiple imputation of true values using validation sub-samples. Asymptotic efficiency of correction estimators is compared theoretically. Empirical applications use data from the Nigeria Living Standards Survey (income self-report), the NDHS (breastfeeding duration recall), and the Nigeria Health Sector Expenditure Survey (healthcare cost recall). Validation sub-samples collected by comparison with administrative records and biomarker data confirmed the presence of systematic measurement error in all three surveys. Income self-report showed downward bias of 24 percent relative to administrative payroll validation. Regression calibration correction restored unbiased income effects in consumption determinant regressions. MCMC latent variable correction outperformed SIMEX for non-classical error. The dissertation provides theory and implementable correction tools for the Nigerian survey statistical community. Keywords: measurement error, survey data, regression calibration, SIMEX, Nigeria survey methodology