Department of Statistical Science
Duke University
presents:
M.J. Susie Bayarri
susie@stat.duke.edu
Universitat de Valencia, Valencia, Spain
"Optimal Sample Size When Designing "Successful" Replications"
Abstract: Replications of experiments is a common practice in most areas of applied research. Nevertheless, and in spite of its widespread use, a systematic study and/or discussion of the meanings and goals when replicating seems to be virtually absent in the literature. As a consequence, there does not seem to exist a clear understanding of what a successful replication means. In this talk, we consider several possible goals when replicating, which lead to different (but precise) definitions of successful replications. Bayesian hierarchical models allow for a flexible quantification of the assumed relationship among the different experiments. Bayesian predictive distributions are the natural tool to compute the probability that the replication would be successful, thus allowing for designs for which that probability would be high enough. The results are exemplified with data coming from a non-central t.
October 3, 1997
4:00 pm - 5:00 pm
116 Old Chem Building Any questions concerning the seminar may be addressed to Cheryl McGhee @ [919] 684-8029 or e-mail cheryl@stat.duke.edu. Please contact the author(s) directly for reprints etc.