viking1111 发表于 2010-2-26 12:08
各位,请问Logistics非线性模型中的异方差、序列相关分析用统计软件怎么实现?
理论上,异方差、序列相关是残差的问题,与估计方法无关,与模型是否为线性关系也不大。
但是Eviews我试了,不能对Logistics进行异方差的White检验。不知道STATA之类的软件有没有这个功能。有没有相关的软件可以给推荐一下的?
另外,由于Logistics用的是极大似然估计,不能用加权最小二乘估计,如果存在异方差,那应该怎么修正使用极大似然估计方法呢?
大家集思广益,互相学习,对热心人先说声谢谢啦
1) Is the dependent variable binary?
2) In that case, it is a probability model. The distribution is assumed as a standard because the variance is not identifiable. For probit is normal (0, 1) and logit is (0, pi/sqrt(3)).
3) To test if the data is from different distributions, you can use interaction to see if the interaction term is significant.
4) If you think the variance is a function of a covarince, you can write it into likelihood function to estimation that function.
HTH