Optimal Designs and Large Sample Tests for Linear Hypotheses.
VIRGINIA POLYTECHNIC INST AND STATE UNIV BLACKSBURG DEPT OF STATISTICS
Pagination or Media Count:
The study investigates the appropriateness of normal-theory inference for linear models having non-Gaussian errors. It is shown that bounds on the error of the Gaussian approximation depend on the design the optimal designs are characterized and shown to be orthogonal. Bounds on the actual probabilities associated with Scheffes projections, and with Dunnetts procedure for comparing several treatments with a control, are given in terms of their normal-theory approximations. Author
- Statistics and Probability