Software Alternatives for Variance Estimation in the Analysis of Complex Sample Surveys: A Comparison of SAS Survey Procedures, SUDAAN, and AM

The use of standard statistical software for the analysis of data collected from complex sample surveys yields biased point estimates of parameters and incorrect standard errors as a result of unequal probabilities of selection of observations, stratification and clustering, and unit nonresponse. The appropriate analysis of data from complex sample surveys requires the use of software that incorporates sample weights and accurate estimates of variance (e.g., Taylor series approximations or replication estimates). This paper provides a review of the special statistical requirements for the analysis of complex survey data; a comparison of a sample of available statistics and variance estimation strategies provided in SAS, SUDAAN SAS-Callable, and AM; and a contrast of results across these packages. The paper includes samples of code and illustrations of partial output from each software package.