| Summary: | 碩士 === 國立政治大學 === 統計研究所 === 102 === Optimal design under the survival analysis models has rarely been considered in the literature. In this article, exponential failure time is assumed and the expected failure time which is inversely related to two explanatory variables, one is controlled variable and the other is uncontrolled variable, through a linear function is considered. Since the time each experimental unit entering into the study is not known, and the corresponding uncontrolled variable is also unknown, assumptions on the distributions of the entering time and the uncontrolled variable are made in order to find optimal designs. Upon entering into the study, an “optimal” decision is made on the experimental unit, and whether the unit will fail or be censored is also considered. To improve efficiency of parameter estimation, D-optimal criterion is employed, and conditional D-optimal designs are found. Under different setting of values of the parameters and with the help of computer programming, conditional D-optimal designs are found and are listed for a clinical medicine problem. Design robustness on unknown parameters is also investigated.
|
|---|
