STATA mixlogit collinear 问题

大家好，请高手为我解答这个问题：我用这个命令“ mixlogit decision, group(count) id(id) rand(color size firmness sweetness  flavor  shelf  cost)”，运行后报错“Some variables are collinear - check your model specification”，可我的解释变量没有完全共线。数据及解释变量相关系数请看附件。提前谢谢帮忙的同学！ Doc11.docx (27.96 KB)

2014-2-23 16:44:18 上传

Book11.xlsx (94.42 KB)

2014-2-23 16:44:15 上传

原数据中忘了放入count这个变量，请重新下载这个数据 Book11.xlsx (106.37 KB)

2014-2-25 03:51:00 上传

谢谢了！

what I think is if you add all the terms, the model estimating will not converge based on how the MLE is estimated.  Have you looked the original paper and see if the model is estimable with collinearity (I read the paper some time ago, but never used this model)?

if you do

_rmcoll color size firmness sweetness  flavor  shelf  cost

does it show anything?

jjjj6666 发表于 2014-2-24 03:52
if you do

_rmcoll color size firmness sweetness  flavor  shelf  cost

Hi jjjj6666, 执行了你建议的口令后，返回的是“note: cost omitted because of collinearity”。可是我不想omit cost啊，如果不去掉任何变量，有没有解决的办法？谢谢！

继续等待高手的帮助！我用mixlogit命令分析了其他几个类似的数据，解释变量中cost是两个数值，其他是0和1虚拟变量，解释变量之间的相关系数也类似，STATA都运行了，没有出现任何报错。先谢了！

if force collinearity term, is the model unidentifiable?  At least, if force it, the model estimating may not converge.

jjjj6666 发表于 2014-2-25 01:11
if force collinearity term, is the model unidentifiable?  At least, if force it, the model estimatin ...

Thanks for the reply! I want to have a try "force the collinearity term", but how do I do it in STATA? Could you let me know the code to fulfill it? Thank you so much!

the ado file doesn't allow it. You need to modify the ado file (I think stata ml optimization can have the option to allow it), which may take some works.  But if looking the likelihood function, it seems the model itself has problem under this case, is this correct?

jjjj6666 发表于 2014-2-25 02:04
the ado file doesn't allow it. You need to modify the ado file (I think stata ml optimization can ha ...

Sorry but I do not see problems of the model. Could you give me a little more detail? Thanks!

真是抱歉，我上传的数据中没有count这个变量，请重新下载数据。