dagga
如果协变量和因变量都不是正态分布,还能做协方差分析吗?我的数据经K-S正态检验几乎都是P<0.01,好恐怖啊,我该怎么认识和处理这个问题?
谢谢大家!请指教
micro@
did you use k-s test on the data or some kind of residuals? if on the data, then it's not quite informative to us.
tell us the design, the goal of you analysis, and some exploratory plots, so that people can help.
without knowing such and such, all the answers are not responsible for you.
dagga
我的数据(医学指标大约有十几个,都是计量资料)经K-S正态检验几乎都是P<0.01,我想先两两比较,如性别之间,城乡之间,后做相关影响因素的分析(尽量做多元回归,不作LODISTIC回归),但是由于正态分布的问题,让我感到很棘手,我也试过一些转换,如对数,平方根等等,都不行,我该怎么做?
另外,基本上我看到许多跟我测量的指标类似的文献,两两比较都是t检验,即便是在SD很大(与均数相比)的情况下也是如此,我很困惑。用非参数检验有怕说服力差,还有,两两比较的时候,想用协方差去控制一些影响的因素,不是协方差对状态性的要求有多高,我这种偏态的资料能做吗?请帮助我!
micro@
[quote]引用第2楼dagga于2006-11-19 14:01发表的“”:
我的数据(医学指标大约有十几个,都是计量资料)经K-S正态检验几乎都是P<0.01[/quote]
you didn't answer my first question. did you use k-s test on data, or residuals?
you also didn't answer my second question. what's the design?
you vaguely answered my third question. what's the goal of analysis?
ifeelyou
micro@说的对。
你先要搞清楚你的研究目的,设计类型。
robustreg
是个问题,文献中有看到过基于秩次的非参数协方差方法,我个人认为,还可以试试一些稳健回归方法,类似M估计或GM估计等等