Current sample size methods for multilevel and longitudinal data often get the wrong answer. Studies that are too small cannot achieve scientific goals. Studies that are too large expose human research participants to unnecessary harm. The ethical codes that bind NIH funded behavioral scientists doing research with human participants demand correct sample size selection. Flaws of current sample size methods are a critical barrier to progress. Behavioral scientists use community trials, longitudinal data, and multilevel models to compare diverse populations. Understanding addiction and gauging knowledge about head and neck cancer screening require measuring outcomes over time. The complex data yield complex variance patterns. Current software (such as Optimal Design) makes simplifying assumptions about the variance patterns, which may lead to the wrong sample size. No available software always gives the correct sample size analysis;the existing paradigm fails. There are three barriers to progress in designing multilevel and longitudinal studies: 1) inadequate tools to evaluate when current sample size methods fail;2) flawed methods and software to select sample size;and 3) insufficient training to find correct sample sizes. The successful completion of two aims will remove the three barriers.
The aims are 1) Create new sample size methods and software needed to accurately mirror multilevel and longitudinal models of normal, binary, and Poisson data common in studying health behavior in diverse populations and 2) Train behavioral scientists to use the new methods and software. Training will include template sample size analyses, tutorial publications, short courses, webinars and online tutorials. Better sample size calculations will produce better designed studies. Better designed studies will speed ethical, efficient, and effective behavioral research. Behavioral research lies at the heart of preventing addiction and cancer. Correct sample size selection is not merely a statistical nicety. The new methods will directly improve the health of millions of Americans.
The ethics and cost of human research demands accurate sample size. Current sample size selection methods can fail with multilevel and longitudinal data. We will develop new sample size tools for behavioral scientists conducting prevention research in communities with diverse populations.
|Andridge, Rebecca R; Shoben, Abigail B; Muller, Keith E et al. (2014) Analytic methods for individually randomized group treatment trials and group-randomized trials when subjects belong to multiple groups. Stat Med 33:2178-90|
|Simpson, Sean L; Edwards, Lloyd J; Styner, Martin A et al. (2014) Separability tests for high-dimensional, low sample size multivariate repeated measures data. J Appl Stat 41:2450-2461|
|Ringham, Brandy M; Alonzo, Todd A; Brinton, John T et al. (2014) Reducing decision errors in the paired comparison of the diagnostic accuracy of screening tests with Gaussian outcomes. BMC Med Res Methodol 14:37|
|Simpson, Sean L; Edwards, Lloyd J; Styner, Martin A et al. (2014) Kronecker product linear exponent AR(1) correlation structures for multivariate repeated measures. PLoS One 9:e88864|
|Chi, Yueh-Yun; Gribbin, Matthew J; Johnson, Jacqueline L et al. (2013) Power calculation for overall hypothesis testing with high-dimensional commensurate outcomes. Stat Med :|
|Guo, Yi; Logan, Henrietta L; Glueck, Deborah H et al. (2013) Selecting a sample size for studies with repeated measures. BMC Med Res Methodol 13:100|
|Chi, Yueh-Yun; Gribbin, Matthew; Lamers, Yvonne et al. (2012) Global hypothesis testing for high-dimensional repeated measures?outcomes. Stat Med 31:724-42|
|Gurka, Matthew J; Edwards, Lloyd J; Muller, Keith E (2011) Avoiding bias in mixed model inference for fixed effects. Stat Med 30:2696-707|