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]

Re: st: Different results from interaction term and Suest Command

From   Maarten buis <>
Subject   Re: st: Different results from interaction term and Suest Command
Date   Wed, 16 Jun 2010 06:32:40 -0700 (PDT)

--- On Wed, 16/6/10, natasha agarwal wrote:
> First Method with the interaction term:
> xtreg lnrval lnk lnw lnvfdi fdev1 y14-y18, fe vce(robust)
> where fdev1=1 if region higher developed and zero
> otherwise. In this method fdev1 (which is the difference
> between the two groups is not statistically significant)

This is not an interaction effect, but just a dummy. It
tells you wether the average lnrval differs differs between
developed and devoloping countries
> Second Method with sample splitting
> bysort number: center lnrval lnk lnw lnvfdi y14-y18, casewise
> regress c_lnrval c_lnk c_lnw c_lnvfdi c_y14-c_y18 if fdev1==1
> est sto fe2
> reg c_lnrval c_lnk c_lnw c_lnvfdi c_y14-c_y18 if fdev1==0
> est sto fe3
> suest fe2 fe3, vce(cluster number)
> est sto fe4
> test [fe2_mean]c_lnvfdi==[fe3_mean]c_lnvfdi (the slopes
> are significantly different

This is a completely different hypothesis: now you test wether
the effect FDI differs between developed and developing countries.

Given that you are testing different hypotheses, it is no surprise
that you get different answers.

If you want to test for interactions in using your first method
you will have to create a variable: lnfdiXfdevl = lnvfdi * fdev1
and add that variable to your -xtreg- model. see for instance:

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

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