A Comparison of Median Testing Methods for Complex Survey Data Using Various Statistical Packages
Warren, L. K., & Davies, C. (2012, July). A Comparison of Median Testing Methods for Complex Survey Data Using Various Statistical Packages. Presented at JSM 2012, San Diego, CA.
When dealing with highly skewed data, medians are often more informative for comparing the centers of the distributions of two populations than the means. The SUDAAN® software package is capable of accounting for complex survey designs when estimating a single median and its standard error, but does not have an option for comparing population medians from dependent samples. This paper explores several methods for comparing population medians while accounting for possible correlation within a complex sample. Capabilities and limitations of statistical software packages, including SAS, SUDAAN, R, and STATA, are outlined. A test inversion method introduced by Binder and Patak (1994), and modified for use in SUDAAN, is compared with Jackknife, Bootstrap, and a naïve method of comparing the medians. We conduct a simulation study which suggests that the modified test inversion method performs as well or better than the other methods in larger sample sizes. A real data example using the NSDUH Public Use Files is also discussed.