SUDAAN

SUDAAN is a statistical software package for the analysis of correlated data, including but not limited to the analysis of correlated data encountered in complex sample surveys. Correlation between observations may exist, for example, in multi-stage studies where units such as geographic areas, schools, telephone area codes, or hospitals are sampled at the first stage of selection. A correlation may exist between observations due to the increased similarity of the observations located in the same cluster compared to observations between clusters. During any statistical analysis with survey data, the complex design features of the study, including clustering, unequal weighting, stratification, and selection with or without replacement, should be accounted for during the analysis. Standard software packages that do not account for the complex design features can produce biased results. In fact, in most instances the precision of statistics will likely be underestimated if one were to use a standard software package that does not account for the complex design features of a study. Software packages, such as SUDAAN, that are specifically designed for the analysis of data from survey studies will properly account for many of the complex design features of the study during an analysis.

SUDAAN originated in 1972 at Research Triangle Institute (RTI) and, over the years, has evolved into an internationally recognized, leading software package for analyzing correlated and complex survey data. SUDAAN 9.0, which was launched in August 2004, offers several additions and enhancements to prior releases. RTI statisticians are currently working on the next major release of SUDAAN and plan to have that available in 2008. SUDAAN can be run on any DOS or Windows platform and is also available for the SUN/Solaris and LINUX operating systems (32-and 64-bit operating systems). SUDAAN is available as a stand-alone software package or can be used in SAS-callable format. The SAS-callable format is of particular value to SAS users because it allows SUDAAN to be run directly within any existing SAS job. SUDAAN's syntax is very similar to SAS (e.g. a semicolon must appear at the end of every statement), [Page 854]and S UDAAN has the ability to read data files in various formats such as SAS, SAS XPORT, SPSS, and ASCII.

As noted earlier in this entry, SUDAAN provides estimates that correctly account for complex design features of a survey. These features include the following:

• Unequally weighted or unweighted data
• Stratification
• With- or without-replacement designs
• Multi-stage and cluster designs
• Repeated measures
• General cluster-correlation (e.g. correlation due to multiple measures taken from patients)
• Multiply imputed analysis variables

Currently, SUDAAN is a single program that offers nine analytic procedures. SUDAAN is one of the few software packages that offers three of the most commonly used robust variance estimation methods, including Taylor series linearization (generalized estimating equations for regression models), jackknife (with or without user-specified replicate weights), and balanced repeated replication.

All three variance estimation methods are available in all of SUDAAN's procedures. The nine procedures consist of three descriptive procedures, four regression procedures, and two survival procedures.