[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: heckman two-step selection estimation for survey data

From   Shehzad Ali <>
Subject   Re: st: heckman two-step selection estimation for survey data
Date   16 Nov 2008 11:21:12 +0000

Dear listers,

Following on from the discussion on the use of weights with - heckman twostep - can I ask would the same logic apply to treatment effects models like -treatreg twostep- ? To be precise, can -treatreg twostep- be used for survey data although the twostep function does not allow use of weights or cluster?

Thank you,


On Nov 1 2008, Joao Ricardo F. Lima wrote:

Dear Mandy,

you can read the svy manual (Stata 10), page 75: "When used with
independently distributed, nonweighted data, the likelihood to be
maximized reflects the joint probability distribution of the data
given chosen model. With complex survey data, however, this
interpretation of the likelihood is no longer valid, as survey data
are weighted, not independently distributed, or both. Yet for survey
data, (valid) parameters estimates for a given model can be obtained
using the associated likelihood function with the appropriate
weignting. Since the probabilistic interpretation no longer holds, the
likelihood is instead called a pseudolikelihood....".

With survey data, you can´t use OLS (second-stage of the Heckman
two-stage) to get consistent estimators. With svy data you will max
the log pseudolikelihood.

Other experts can explain it better, but keep in mind that with svy
date, you can´t do heckman two-step.


Joao Lima

2008/11/1 Mandy fu <>:
Hi all,

I notice -svy:heckman- is not allowed to use -twostep-option,while
-heckman- does not allow to use pweight. So, I got the silly question:
how to do the heckman two-step estimation for survey data?

[I checked the archives and find log time ago someone else has asked
the question why -heckman-does not allow to use pw. But nobody
answered yet. So, I ask again and feel free to let me know if the
question does not make sense at all. So, I'll not waste time on this.]

Thanks for your time!

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2023 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index