Simulation Studies on Rank Estimating Equations for Survival Data

• Main Authors: Wang, An-Shih, 王安仕
• Other Authors: Chen Chu-Chih
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/26580489255924767185

碩士 === 淡江大學 === 數學學系 === 85 === The purpose of this thesis is computer simulation,under the acceleratedfailure time model(AFTM),to compare the power performance of the rankestimating equation(REE,Chen and Hsieh,1997) with that of the linear rankstatistic(LRS) of Prentice(1978) in the survival time of two-group treatmentwith right-censored data. The REE applies the Gehan''s(1965) generalized Wilcoxon score,while theLRS is applicable with both Wilcoxon and log-rank score.Without covariates,it is known that the log-rank score will have bigger power than the Wilcoxonscore when the hazard rates of two groups are parallel,and the situationis reversed when this assumption is violated.Here we consider the presenceof a concomitant covariable as well. The main result simulation results are as follows:(1)All the powers increases with the sample size,and decreases with the censored percentage.(2)The Prentice''s LRS can not detect the presence of a covariable,thus is affected in power performences.(3)When the censored percentage is none or small,the REE is much better than the LRS for alll cases studies,but the power of the REE deteriorates as the percentage increases,which needs further investigation.