two step Heckman 的一个问题 - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Economics - 经济

two step Heckman 的一个问题

two step Heckman 的一个问题# Economics - 经济

r*u2008-10-29 07:10

1 楼

I didn't get any $$ last week?
发信人: caty (caty), 信区: Avatar
标题: MITBBS 10周年主题服装秀，红包丰厚，请勿错过！！！
发信站: BBS 未名空间站 (Mon Jun 9 22:16:11 2008)
为庆祝 MITBBS 10周年，6月服装周周秀活动以各类庆祝，节日，搞怪为主题。胡搭不
限。
入选范围扩大，凡入选者可奖励新伪币100元。（此活动为6月周周秀主题秀）参选者请
在本帖跟贴。凡参选报名id，奖励20元着装补贴。祝大家周周娱乐无极限！

C*12008-10-29 07:10

2 楼

问题是这样的：
我的数据是基于一个survey，不能完全反应population的分布，分析的时候用了
weight
在stata中，用sampling weight 纠正estimator.
我现在的问题是：用这个数据，应用two step Heckman 的procedure 估计 wage
equation with potential selection bias and predict the wages at the sample
mean. The second is my ultimate goal.
在stata中， any weight is not allowed in Heckman's two step estimation.
－－我还没找到reference来证明这个， my guess is due to predicted inverse
mills ratio whose distribution is inconsistent with the distribution of X,
which alters the variance-cova

f*r2008-10-29 07:10

3 楼

Check out "svy: heckman". Frankly, I don't see your algorithm generates
consistent estimator. svy: heckman is build-in Stata command, should be
consistent.

C*12008-10-29 07:10

4 楼

Thanks for responding.
I checked svy: heckman, but it can not be applied to two-step procedure.

f*r2008-10-29 07:10

5 楼

I think there is a misconception. You don't need the Heckman two-step
estimator per se. Heckman model usually refers to the model which adjust
for selection bias or selectivity. There is so called Heckman two-step
estimator which is one way to estimate the model, but not the only one. In
fact, maximum-likelihood estimator is more efficient. I have not used it;
but I think svy: heckman is a ML estimator which allows for sampling weight.
It does not do Heckman's two-step estimator for a rea

C*12008-10-29 07:10

6 楼

Thank you~