SAS Wald estimator LATE

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

在用2SLS解决内生性问题的时候，如果要考察的是一个treatment, (D), which is dummy varialbe that takes the value of 1 if receives the treatment. 它的instrument 是另一个或多个dummy variables, (Z1,Z2...).  用传统的2SLS来estimate很可能是biased的，最新的改进方案是用Wald estimator, 也叫LATE （Local Average Treatment Effect）. 这里我附件了一份参考文件，它后面有个例子，但是用STATA实现的，不是很懂。请问哪问高手知道Wald estimator,用SAS怎么实行的。跪拜，感谢.

Wald estimator.pdf (137.69 KB)

2012-6-24 21:11:35 上传

Wald estimator

那个例子用STATA是这样的：

clear
set seed 54687
set obs 20000
/* first, randomly assign the instrument - say half-half */
ge z = uniform()>.5
/* then, generate never-takers (d00), always-takers (d11) and compliers
(d01), independent of z */
ge d00=(_n<=5000)
ge d11=(_n>5000 & _n<=10000)
ge d01=(_n>10000)
/* observed outcomes: always zero for never-takers, always one for
always-takers, depends on the IV for compliers */
ge D=d11+z*d01
/* now give the three groups different LATE. Without loss of
generality, assume within group homogeneity. */
ge late=-1 if d00==1
replace late=0 if d11==1
replace late=1 if d01==1
/* next generate potential outcomes y0,y1 */
ge y0=0.25*invnorm(uniform())
ge y1=y0+late
/* actual outcome depends on treatment status */
ge y = D*y1+(1-D)*y0
/* the average treatment effect is simply the sample mean of late */
sum late
/* OLS doesn't give you ATE or LATE */
reg y D
/* IV gives you the LATE for the compliers */
ivreg y (D=z)
exit

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1 收藏1 回帖

关键词：estimator Wald Late Tima ATE instrument receives

The problem can also be solved as a region switch model. I posted an example before.

You may refer to Jeff Woodridge 's book for more details.

bobguy 发表于 2012-6-27 08:12
The problem can also be solved as a region switch model. I posted an example before.

You may refe ...

Thanks for your reply, Could you give me your previous posting link that addresses this issue.