A Study of Finding Optimum Variance Function in Generalized Mixed Models

• Main Authors: LI-TUNG-CHIANG, 李永卿
• Other Authors: YUN-MING-PONG
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/21062032621233310898

碩士 === 國立臺灣大學 === 農藝學研究所 === 90 === Because GLIM can be applied non-normal data in different situations in recent years, GLIMM with random factors make the applied area more widely. If the users decided the distribution in the program when they analyze by statistical software SAS MACRO GLIMMIX, variance function would be decided. But in residual plots, maybe we can't have reasonable explanation. So it is the purpose of this paper to choose the variance function properly. In the study, we offer three methods to choose the optimal variation function, and use one data group of the Phototaxis of Alates of Coptotermes formosanus Shiraki to explain the three methods. We find that it would have better explanation for data. Because residuals are better scattered and have no obvious pattern. So when the data for analyzing is mixed, we suggest that finding the optimal variation function first then analyze GLIMMIX.