An Evaluation of the Use of Personal Computers for Variance Estimation with Complex Survey Data
Barbara Lepidus Carlson, Ayah E. Johnson, and Steven B. Cohen
Over the past decade, software with the capability of handling data resulting from surveys with complex designs has become more readily available, primarily for use on mainframe computers. More recently, a few of these have been released in versions designed for the personal computer (PC). As the speed and capacity of PCs increase, there is the potential for more widespread use of this specialized software.
In this paper, the authors present the results of comparisons between two of these PC software packages, SUDAAN and PC CARP, and their mainframe counterparts with respect to computer run-time, programming difficulty, special features and flaws, and data preparation issues, as well as space and memory requirements for the PCs and cost considerations for the mainframe. Analyses are limited to computations of variance estimates, using Taylor linearization, for weighted proportions and means, and are carried out on files from the 1987 National Medical Expenditure Survey, which has a stratified multi-stage area probability design.
Statistical software; SUDAAN; PC CARP.