Comparison of the Five Test Statistics Currently Available in the SUDAAN® Modeling Procedures

• Activity Number: 342
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): G. Gordon Brown*+
• Companies: RTI International
• Keywords: SUDAAN ; Wald F test ; adjusted Wald F test ; Satterthwaite's adjusted Chi-sq test ; Satterthwaite's F test

Abstract

RTI's software package SUDAAN® is specifically designed to analyze cluster correlated data or complex data obtained from sample surveys. Currently, SUDAAN offers the choice of five different test statistics for hypothesis testing in the modeling procedures. These test statistics are the Wald Chi, Wald F, adjusted Wald F, Satterthwaite's adjusted Chi, and Satterthwaite's F. It is well known that the Wald Chi test tends to perform poorly when the data pertains to only a few clusters. The other four tests have been developed in an attempt to compensate for this problem and to better handle other intricacies of the data. While the existing research literature addresses merits of these tests, the question of "Which test should I use" remains at large. This work will attempt to answer this question under certain practical scenarios, such as when the design includes 50 or fewer clusters.