|
Activity Number:
|
176
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, July 30, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #307795 |
|
Title:
|
Internal Pilot Designs and Mixed Models
|
|
Author(s):
|
Matthew Gurka*+ and Christopher S. Coffey and Keith E. Muller
|
|
Companies:
|
University of Virginia and The University of Alabama at Birmingham and University of Florida
|
|
Address:
|
P.O. Box 800717, Charlottesville, VA, 22908-0717,
|
|
Keywords:
|
Internal pilot ; mixed model ; repeated measures ; sample-size re-estimation ; adaptive designs ; power
|
|
Abstract:
|
An internal pilot (IP) design uses interim sample size analysis to adjust the final number of observations. Although most research on IP designs has been conducted for univariate models, the common use of multivariate data naturally motivates interest in extending IP designs to more complex models. Mixed models are a popular tool for multivariate Gaussian data exhibiting missing values or mistimed observations. However, the development of IP methods for mixed models requires fixed sample tests that guarantee accuracy of inference in small samples and accurate power algorithms. Initially, we describe how to extend exact univariate IP methods to a restricted class of linear mixed models. We describe an example in medical imaging that directly benefits from this extension. We then discuss ongoing and future work for extending the use of such methods to more general mixed model settings.
|
