Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

st: 3SLS with unequal number of observations in each equation

From   Anna Katharina <>
To   "" <>
Subject   st: 3SLS with unequal number of observations in each equation
Date   Sat, 19 Jan 2013 13:06:11 +0100

Dear Statalist,

I'm trying to estimate a system of equations by 3SLS, where the different equations represent different decades in a growth regression (Barro-approach) and look similar to

reg3 (eq70: gdppcgr70 $X70) (eq80: gdppcgr80 $X80) (eq90: gdppcgr90 $X90) (eq00: gdppcgr00 $X00) [...]

Most of my explanatory variables are endogenous and I wish to instrument them. Further, I need to use constraints in order to obtain (slope) coefficients that do not vary across decade (i.e. equation). 

As I do not have the same number of observations for each decade, I am loosing a lot of variables when using the command -reg3- (Stata Version 12.1), as Stata requires a 'balanced' data structe (equal no of obs in each eq).

I found an article written by Allen McDowell, published in The Stata Journal, regarding this issue, but based on the SUR approach. He suggests restructuring the data and using -xtgee- after that. This does not work out for me, as I have endogenous regressors that I wish to instrument (it would like to avoid doing the 2SLS-part 'by hand', as I would like to obtain valid standard errors in the last step.). Also, using this setup I cannot impute the constraints that guarantee equality of coefficients across equations.

Additionally, I tried to estimate the system by GMM, but also with this approach, existent observations were dropped in order to obtain the same number of observations in each equation.

I would be most grateful if anybody could suggest how I can estimate this system, with i) an unequal number of observations in each equation and with ii) the possibility to use instruments and with iii) the possibility to impute constraints on the coefficients.

Many thanks!

MdDowell, Allen: From the help desk: Seemingly unrelated regression with unbalanced equations. The Stata Journal, Vol 4(4), p. 442-448 		 	   		  
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index