Estimating Standard Errors for International Census Data Samples: Cases from IPUMS-International

Lara Cleveland, University of Minnesota
Michael Davern, University of Minnesota

Social science research increasingly uses data from complex samples that incorporate clustering, stratification and weight adjustments. Standard error estimates from complex samples can often differ dramatically from those derived from simple random samples. Researchers using census microdata frequently employ data analysis methods designed for simple random samples. Estimates that fail to account for complex sample design can yield inaccurate p-values and confidence intervals. Recently, Davern et al. (forthcoming) suggested an improved method of accounting for implicit geographic stratification in historical IPUMS samples of U.S. census microdata. Using similar techniques, we evaluate four international data samples from IPUMS-International to suggest improved methods for estimating standard errors. Using full count census data, we evaluate the impact of sample design on standard error estimates, suggest a way to account for implicit geographic stratification using modern standard error estimation software and confirm the utility of a finite population correction for sample densities of 10%.