All Times ET
Virtual
Stepdown Simulated Multiple Comparisons Tests (305328)
*Lonnie Grantham, US Army Institute of Surgical Research
Laura L.F. Scott, US Army Institute of Surgical Research
Keywords: Multiple comparisons tests, multiple comparisons procedures, closed testing methods, simulated multiple comparisons, stepdown methods
Background: Control of type I error when using well known multiple comparisons procedures (MCP), such as Dunnett’s and Tukey’s tests, are common in practice. Closed testing methods (CTM) provide an alternative, flexible approach that can be used effectively for any collection of linear combinations. For complicated study designs requiring fixed and random effects, stepdown simulated (SDS) tests use CTM to correctly incorporate both logical dependencies among hypotheses and correlations. Methods: Data collected from 2 porcine burn studies with nested or crossed factors were used. The LSMESTIMATE statement, with ADJUST=SIMULATE and STEPDOWN options, was specified in the SAS® GLIMMIX procedure. The ADJDFE=ROW option was specified to assign the correct denominator degrees of freedom for the multiplicity adjustments. One-sided tests were conducted using the UPPER and LOWER options. Results: SDS MCPs outperformed the more commonly used procedures, providing better statistical power with sufficient control of familywise type I error. Conclusions: SDS MCPs provide a targeted, powerful method for adjusting p-values to control type I error rates when complex linear mixed models are required.