Statalist The Stata Listserver

[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Meta regression

From   "Long Nguyen" <>
To   <>
Subject   st: Meta regression
Date   Mon, 14 May 2007 16:12:43 -0400


I'm a new user of STATA and trying to perform a meta regression.  I would
appreciate some detail instruction and examples or tutorials that can help
me do meta regression using this new tool.  I have the Getting Started..
Manual and installed the metareg module; however, the online help file
offered only limited information.

Thank you for your kind assistance,


-----Original Message-----
[] On Behalf Of Brian P. Poi
Sent: Monday, May 14, 2007 11:03 AM
Subject: Re: st: Different numbers of observations in sureg

On Mon, 14 May 2007 09:16:36 -0400 Alemu Mekonnen wrote:

> I would like to use different number of observations in the different 
> equations when using the seemingly unrelated regression command in stata 
> (i.e., sureg). For example, if there are two equations, I wanted to use 
> 600 observations for equation 1 and 900 observations for equation 2. My 
> question is how do I limit the number of observations to do this. I was 
> not able to use the IF command but this applies to all the equations in 
> the system.


Allen McDowell illustrates a clever way of fitting the SUR model with 
unequal numbers of observations by using the -xtgee- command in a "From 
the help desk" article in the Stata Journal (vol. 4, no. 4).

The trick is to recognize that the SUR model

    y1 = x1 * b1 + u1
    y2 = x2 * b2 + u2

can be "stacked" and written as

    [ y1 ] = [ x1  0 ][ b1 ] + [ u1 ]
    [ y2 ]   [  0 x2 ][ b2 ] + [ u2 ]

The stacked version looks just like a panel data model with two panels. 
One thing to note is in a SUR model, u1 and u2 are allowed to have 
different variances, while -xtgee- assumes they have a common variance. 
McDowell's article shows how to rescale the data using OLS regression and 
how to manipulate your dataset to use with -xtgee-.  Then to fit the model 
you simply use -xgee- with a Gaussian distribution, identity link 
function, and an unstructured correlation matrix.

    -- Brian Poi

McDowell, A.W. (2004). From the help desk: Seemingly unrelated regression
    with unbalanced equations.  Stata Journal, 4(4): 442-448.

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index