Design of Experiment: Determination of Sample Size.
STANFORD UNIV CALIF DEPT OF STATISTICS
Pagination or Media Count:
The paper considers the two procedures used most often for the determination of sample sizes in the design of experiments viz. the test procedure and the estimation procedure. The statistically most sound of the existing methods are described. Attention is drawn to the fact that, in most cases, these methods are only satisfactory if an experimenter has a rather precise prior knowledge about chi square, the variance of the experimental error. Next, some new methods are derived which do not have this drawback and which include the earlier methods as special cases. The new methods are based on the unconditional with respect to chi suare distributions of the usual test statistics and estimators, derived by weighting their usual distributions with a prior distribution on chi square chosen fom the natural conjugate family to the s square-distriution family. An approximation to the unconditional noncentral F-distribution is derived, and some tables are given to facilitate the computation. The determination of the number of replications is illustrated by some examples. Author
- Statistics and Probability