[求助]若数据同时存在多重共线性，异方差，自相关问题

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

若数据同时存在多重共线性，异方差，自相关问题：

1，应以何顺序削除他们的影响，

     先消除多重共线性，再WLS消除异方差，最后解决时间序列自相关？

     or

     先消除多重共线性，再解决时间序列自相关，最后WLS消除异方差？

2，以某种顺序为优的依据又是什么？

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏3 回帖

关键词：多重共线性 自相关问题 多重共线 自相关 共线性 数据 方差 线性 自相关

顶起

在spss里可以采取step method，然后同时做异方差，自相关的诊断，不知道，对不

谢谢回复

LS可以说得详细些吗？

我只知道stepwise 方法做OLS，楼上说的step 是指stepwise 吗？stepwise方法不是用来剔除变量方法得到变量最优的回归方程的吗（我猜是对付共线性问题的一种方法）？

自相关和异方差的解决的先后顺序是怎样的呢？先用WLS解决异方差，还是先用ARMA模型解决自相关？

[此贴子已经被作者于2008-7-14 7:30:24编辑过]

顶啊顶

顶

这个问题怎么没人回答呢

纸上谈兵一下：

顺序应该是自相关最后处理，多重共线性与异方差的顺序随意。

原因也不复杂：

处理多重共线性与处理异方差之间并不会产生什么的影响，所以二者的顺序无所谓；

但没有多重共线性和异方差的序列自相关显然要较之前容易得多。

l       It all depends on your research model, i.e. if you have time series autocorrelation and multicollinearity occurs due to same reason, having time series day as independent variables. So, if you have time series apply autocorrelation first, multicollinearity will be solved as well.

l       In general if not time series multicollinearity is the result of the research model, review your variables, maybe some of them are not necessary or maybe you can build indices or scales to aggregate some of the independent variables so that you get rid off the multicollinearity. However, in social sciences it almost impossible to avoid it, anyhow some of the independent variable will have correlation, 0.4 -0.5 is acceptable in social sciences.

l       Heteroscedasticity is the case where residuals are not independent and correlated with dependent variable; this is one of the assumptions to use OLS. You can transform the dependent variable, apply non-linear regression or use MLE instead of OLS, or you can have mixed models where you assign probability to error distributions. This problem is more technical rather than research model driven,

 

I would be more concerned with the first two issues. Again, it depends on your domain (what is acceptable) and your research model. Your theory should lead the research model.

谢谢楼上各位大侠的回复，看到一丝光明

It all depends on your research model, i.e. if you have time series autocorrelation and multicollinearity occurs due to same reason, having time series day as independent variables. So, if you have time series apply autocorrelation first, multicollinearity will be solved as well.

[此贴子已经被作者于2008-7-18 5:25:04编辑过]