An Evaluation of Statistical Software Procedures Appropriate for the Regression Analysis of Complex Survey Data
Steven B. Cohen, Judy A. Xanthopoulos, and Gretchen K. Jones
Data from complex survey designs require special consideration with regard to variance estimation and analysis, because of design components that include unequal selection probabilities, stratification, and clustering. Statistical software package programs are currently available which accommodate a complex survey design, and allow for the generation of centrality parameters and variance estimates for statistics expressed in terms of means, totals, ratios, and multivariate regression coefficients. The methods of variance estimation include the Taylor series linearization method and balanced repeated replication. Using data from the National Medical Care Expenditure Survey, which is characterized by a highly complex survey design, the following four statistical programs appropriate for multivariate analysis of complex survey data are compared: SURREGR, SUPERCARP, REPERR, and NASSREG. The comparison focuses on cost-efficiency, user facility, and program capabilities for a series of regression analyses that are representative of the analytical requirements of the National Medical Care Expenditure Survey.
SURREGR; SUPERCARP; REPERR; NASSREG.