c**********w
发帖数: 1746 | 1 包子求解啊，我有四个predictors，一个outcome，散点图都附上了。四个因素好像都
有点影响，比如只有在x1或者x3比较低的时候，outcome绝对值才会比较大，散点图上
都出现三角形的形态。但是我用四个因素做anovan，总共才解释20%的variance，而且
x1,x3基本没有解释variance（明明x1,x3看起来形态更明显啊）
是我还没有找到更合适的predictor吗？还是这种限制性的因素不适合用anova分析？谢
谢分析！包子致谢。 |
c********h
发帖数: 330 | 2 你的y基本都是非正的，而且大部分都是0，变量与y也没有什么线性关系，做anova肯定
不好啊 |
k****n
发帖数: 165 | 3 Anova is for the case where your explanatory variables X are categorical. In
your case, try a regression.
After fitting your current simple linear model with interaction, do a
residual check. As catforfish mentioned, you might need to adjust some
higher order terms of X to improve the goodness of fit. |
q******n
发帖数: 272 | 4 try regression first.
【在 c**********w 的大作中提到】
: 包子求解啊，我有四个predictors，一个outcome，散点图都附上了。四个因素好像都
: 有点影响，比如只有在x1或者x3比较低的时候，outcome绝对值才会比较大，散点图上
: 都出现三角形的形态。但是我用四个因素做anovan，总共才解释20%的variance，而且
: x1,x3基本没有解释variance（明明x1,x3看起来形态更明显啊）
: 是我还没有找到更合适的predictor吗？还是这种限制性的因素不适合用anova分析？谢
: 谢分析！包子致谢。
|
s*r
发帖数: 2757 | |
c**********w
发帖数: 1746 | 6 matlab has a "continuous" anova? I can specify a variable as continuous
In
【在 k****n 的大作中提到】
: Anova is for the case where your explanatory variables X are categorical. In
: your case, try a regression.
: After fitting your current simple linear model with interaction, do a
: residual check. As catforfish mentioned, you might need to adjust some
: higher order terms of X to improve the goodness of fit.
|
c**********w
发帖数: 1746 | 7 包子已转，还有两个follow up问题：
1，为什么anova可以把一定的variance attribute到某个因素上呢？这个好像比较
handy，可以说，某个因素有多么重要云云。。。regression就不可以，因为有
correlation，这个是个大缺陷
2，如果都是continuous data，使用限定一阶interaction的multiple regression和使
用anova，区别在哪里啊？
寄信人: deliver (自动发信系统)
标 题: 本站转帐通知单
发信站: BBS 未名空间站 (Tue May 27 21:25:01 2014)
来 源: mitbbs.com
chairmanmeow，您好：
您转给 kalvin,现金(伪币):40,收取手续费:0.4
同时附加了如下留言给 kalvin.
thank you for your reply!
站务 |
k****n
发帖数: 165 | 8 Got it. Thanks! For your question,
1. For one predictor variable anova, variance = sum(y_ij-y_bar)^2
if predictor variable X is categorical, it makes sense to rewrite it as sum(
y_ij-y_i.+y_i.-y_bar )^2, where y_i. is the average for group level i; y_
bar is the grand average for the total observations; y_ij-y_i. is the within
group variability
while y_i.-y_bar is the between group variability (group is defined by X).
If X is continuous, it defines infinite many of such groups. Such technique
to break down the variance will no longer work.
2. Usually anova for continuous predicator variable involves discretizing
the predictor variables (don't quote me on this). You better check how the
continuous anova is fitted first. |
