内生，参数与非参数的讨论

恩，相当给力的帖子！

pingguzy1 发表于 2013-5-14 02:26
请问一下大神，我cross sectional data, 两个dummy：y1(0,1) and y2(0,1). 分析y1=x1+x2+y2; y2=x1+x2+z1
...

   您帖子不要称～"大神"～ ，这违反Stata版规，会被删帖子。 总之，如果没啥意外，您那样的指令是ok的。 不过，我是想听听您的y1与y2到底是什么变量，您到底想进行什么研究…  另外，z1一般会被理解成工具变量，这个z1是须要进一步检测的。 通常biprobit的研究论文有上百上千篇了！ 除非您的研究议题相当有趣，否则不易发表，我建议您可以再找找另一有趣的dummy variable，改用mvprobit，不过这加深估计困难，但三条probit的paper比较少，也比较容易发表。 祝 顺心 自在

TO h3327156
Thank you very much!!! I could not reply in that poster because that is my first poster here and I am banned.

In my model, y2 is endogenous variable, x1 and x2 are exogenous variables, z1 is instrument.

In the biprobit estimation, the coefficient of inverse mills ratio is significant which means

there is endogenous problem.
I have run ivreg2 for Sargan test and weak instrument test and the result is good.
Just another question: can I use the result of  Sargan test and weak instrument test in ivreg2

to illustrate that there is no overidentification problem and weak instrument problem in my

biprobit estimation, please?


Sorry.

I have no right to send any message now to you because I am banned. I am sorry to have just posted the message here...

It is very interesting that you have read my poster and answered me before it was deleted...........

我的英文表达能力还没到可以轻松回您，所以还是以中文回您，请见谅！

In the biprobit estimation, the coefficient of inverse mills ratio is significant which means
there is endogenous problem.
我不太懂您这句话的意思，怎么又多出个inverse mills ratio？ 在您谈模型时，并没有指出您考量了sample selection的问题，如果您是忘了谈，那么您要参考的是Stata指令ssm

I have run ivreg2 for Sargan test and weak instrument test and the result is good.
为什么是使用ivreg2来检测？ y1与y2不都是dummy variable吗？ 我见过 Journal of Health Economics的paper，在biprobit的paper，检测instrumental variable都不是那样检测的，通常习惯上，喜欢在单条probit下进行【虽然我必须说，这不一定对】，以检测工具变量相关性来说，即在y2那条probit方程，摆不摆工具变量，进而得出Chi-square（这里有的列出Wald，有的列出LR），还有列出摆入工具变量后增加的Pesudo R-square参考。

Just another question: can I use the result of  Sargan test and weak instrument test in ivreg2
to illustrate that there is no overidentification problem and weak instrument problem in my
biprobit estimation, please?
您主要探索一个内生变量，使用一个工具变量，当然没有overidentification problem，至于weak instrument问题，当然，还是要注意的，只是，因为您只有一个工具变量，实务上很不保险，您确定您要把鸡蛋都丢到这一个篮子？

Sorry, it is because my computer does not work with Chinese inputting. I can read Chinese with no problem. Thank you very much for your reply!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Excellent for you!

Yes, for biprobit there is no inverse mills ratio, it is my fault, I mean the Likelihood-ratio test of rho=0, chi2, it is significant, which means there is endogeneity problem.


In fact, I have more than one instrument (2 or 3 :z1, z2, z3). Thus, I must check overid problem and weak instrument problem. Would you please point out the title of the paper you referred?

Thank you very much!

By the way, shall we talk about another two questions:

1. About treatreg command: in the result, it reports rho, sigma, lambda, lambda=rho * sigma, so this is the coefficient of inverse mills ratio, is it right that if I say in treatreg lambda is the coefficient of inverse mills ratio?

2. About movestay command: in the result, rho (0/1) stands for the coefficient of the correlation between the residuals of the two equation (first stage equation and second stage equation). Is it right, please?

pingguzy2 发表于 2013-5-14 05:40
Sorry, it is because my computer does not work with Chinese inputting. I can read Chinese with no pr ...

In fact, I have more than one instrument (2 or 3 :z1, z2, z3). Thus, I must check overid problem and weak instrument problem. Would you please point out the title of the paper you referred?
我上次点出的期刊是错的，正确是在Health Economics，见谅
一篇是Health Econ. 14: 537–544 (2005) The impact of diabetes on employment: genetic IVs
in a bivariate probit
另一篇是Health Econ. 18: 577–589 (2009) The impact of diabetes on employment in Canada
严格讲，这些文献并没有check overid problem，但有执行比较宽松的排除性限制，而按Wooldridge书中的说词，这些工具变数和主要方程误差项的联合不相关检定，也能提供一定的信心水平，如果要严格，理论上，check overid problem，您得变化您的工具变量组合才行【我的论文被要求过】。另外，提供给您的文献，前一篇是连这个检定都不做，直接采用卫生理论观点直接肯定他们的基因工具变量【没办法，第二作者Pagan很大牌，另外他们的基因工具变量很前沿，在2005年前就提出，后面的学者都跟著用，我也是】。

By the way, shall we talk about another two questions:
1. About treatreg command: in the result, it reports rho, sigma, lambda, lambda=rho * sigma, so this is the coefficient of inverse mills ratio, is it right that if I say in treatreg lambda is the coefficient of inverse mills ratio?
我觉得您跳地太快，不知道您扯到treatreg做啥，除非您的y1是连续变量。而您要那样说是可以啦！我没意见。另外，我建议您看看以下的两个帖子，有关treatreg，本版已讨论很多，看您是想手动计算，还是一步到位都ok，不过，除操作软件外，尽量看书理解计量经济理论，方为上策。
https://bbs.pinggu.org/thread-1387677-1-1.html
https://bbs.pinggu.org/thread-461658-1-1.html

2. About movestay command: in the result, rho (0/1) stands for the coefficient of the correlation between the residuals of the two equation (first stage equation and second stage equation). Is it right, please?
坦白说，我不会movestay，也不想探究这些，请见谅。

Thank you very much. Treatreg is for my another research and the dependent var of the 2nd stage is continuous variable.

Anyway, thank you very much.