PRECISION SPECIFICATIONS FOR THE ESTIMATION OF TREATMENT DIFFERENCES.
WISCONSIN UNIV MADISON DEPT OF STATISTICS
Pagination or Media Count:
Given that a completely randomized design is to be employed, it is assumed that the investigator wishes to estimate each of the treatment differences tau sub i - tau sub j with approximately equal precision. The objective in this paper is to suggest and compare different ways of specifying the precision which each estimated treatment difference is to have. It is assumed throughout that the residual variances corresponding to different treatments are unequal but known, and there are differential per unit costs of applying the treatments. For each type of precision specification, the number of replications of each treatment needed to attain the overall precision goals is determined so that the total cost of conducting the experiment is minimized. These specifications are then compared using a set of numerical examples. Author
- Medicine and Medical Research
- Statistics and Probability