Statistical Methods

Internal Pilots for Repeated Measures ANOVA

Repeated measures of continuous data often provides the best combination of cost and statistical sensitivity for a wide range of health research in many disciplines. As driving examples, consider comparing methods of displaying digital mammograms, or radiotherapy planning films. The limited number and cost of qualified readers encourage choosing a minimum sample size, while scientific goals demand great sensitivity to differences. An accurate sample size analysis allows the scientist to resolve the conflict.

Unfortunately, accurate sample size choice requires an accurate value for error variance of Gaussian data. Uncertainty surrounding the variance makes an Internal Pilot very appealing. Such designs use the first fraction of the data to estimate the variance and then adjust the sample size up or down, as needed to achieve the target power. However, appropriate analysis methods are not available for repeated measures with internal pilots. Our research will meet the need for such new methods.

  1. We will develop better statistical power approximations for the "univariate approach" to repeated measures (UNIREP ANOVA), including exact properties and more accurate approximations.
     
  2. We will derive exact and approximate properties of the distribution of the final sample size of Internal Pilot designs used with UNIREP ANOVA.
     
  3. We will describe the analytic properties of UNIREP ANOVA in Internal Pilot designs, including some exact and large sample distributions, as well as practical algorithms.
     

Website: http://www.soph.uab.edu/faculty/bst/christopher_coffey

Student supported by project: David Sawrie

Staff: Christopher Coffey - Principal Investigator