<!-- s:cry: --><img src="{SMILIES_PATH}/icon_cry.gif" alt=":cry:" title="Crying or Very sad" /><!-- s:cry: -->
<!-- s:oops: --><img src="{SMILIES_PATH}/icon_redface.gif" alt=":oops:" title="Embarassed" /><!-- s:oops: -->作者: shiyiming 时间: 2003-10-9 14:03
I am not sure the meaning of "边际关联回归".
For my understanding, "边际关联" means "marginal association". Which means crude association between your dependent variable and one of you independent variables. So from this regression, you cannot tell whether your independent variables are correlated or not.
As you said too few postive results may cause your problem.
And I think you may need check other things, such as missing vales and highly correlated varibles.
For highly correlated variabs, I mean two or more variables just measure the same thing.作者: shiyiming 时间: 2003-10-9 21:57
谢谢 “没有” 对该问题的关注。
我说明一下,你的理解和我原表达的意思是有偏差的,可能怪我没有表达清楚。
我想再次说明一下,我现在分析的数据特点是因变量为类内相关的,即不满足独立要求。我采用的方法叫GEE2(广义估计方程2),在SAS中用ALR(替代logistic回归的方法)进行参数估计。GEE2可参阅文献"JRSS B 1992.54:3-40",ALR可参阅文献“Biometrika,1993. 80:517-26”。