This proposal seeks funding to improve estimation of variances and confidence intervals in analyses using the Integrated Public Use Microdata Series (IPUMS). The IPUMS makes decennial census microdata for the period 1850 through 2000 freely available to scholars in harmonized format through a user-friendly data access system with comprehensive documentation. Since its preliminary release in 1995, the IPUMS has become one of the most widely used demographic resources in the world, and has generated approximately a thousand books, articles, dissertations, and working papers. This supplement will support """"""""IPUMS Redesign"""""""" (HD43392), which aims to improve and modernize IPUMS data, documentation, and access tools. Most researchers using IPUMS apply inappropriate methods of variance estimation; they almost universally apply methods designed for simple random samples. Because of clustering, stratification, and weighting adjustments, however, such methods yield inaccurate p-values and confidence intervals. Bad variance estimates lead to erroneous research conclusions and policy recommendations. We plan three specific research activities to simplify the construction of reliable variance estimates in IPUMS-based research: (1) Construct and document new variables necessary to exploit variance estimation software for all U.S. census microdata. These include a stratification variable for the period from 1960 to 2000 and clustering and subsample-replicate variables for the earlier censuses. (2) Evaluate the reliability of variance estimates in the censuses using real-world applications, based on comparison of analyses that use Taylor series linearization, subsample-replicate estimates, and Census Bureau design factors. (3) Based on the results of this evaluation, develop and disseminate user-friendly documentation and recommendations for variance estimation using each statistical package, with examples of typical analyses and sample programs.
Showing the most recent 10 out of 25 publications
