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

Re: st: testing the equivalance of the coefficient of the same variable in two (nested) models

 From Maarten buis To statalist@hsphsun2.harvard.edu Subject Re: st: testing the equivalance of the coefficient of the same variable in two (nested) models Date Sun, 15 Aug 2010 09:35:00 +0000 (GMT)

--- On Sat, 14/8/10, Evelyn Ersanilli wrote:
> I want to test whether the estimated coefficient of a
> variable (a dummy variable, let's call it X) is the same in
> two models.
> In the second model I add three additional variables, and I
> want to know whether the observed decrease in size of the
> estimated coefficient of X is significant.
> X has a significant effect in both models, but in the
> second model the estimated coefficient is smaller than in
> the first model.
>
> Based on a search through the statalist archives I wrote
> the following command lines;
>
> **************************
> local controls "(varlist)"
> reg hcident `controls' if genb>1 & nomiss2==1
> est sto hc1
> reg hcident `controls' seet seehc dscrfrq if genb>1
> & nomiss2==1
> est sto hc0
> suest hc0 hc1
> test [hc0_mean]X=[hc1_mean]X
> ****************************
> (In case you are wondering: X is significant in both models
> at p<.001. I used markout to run both models on the same
> sample)
>
> I have two questions
> 1. What does the '_mean' mean? Is it the mean of the size
> of the estimation of X or the mean of something else? If I
> leave it out or replace it by '_b' or 'coef' I get an error
> message.

A regresion model results in two types of parameters: the
effects of the explanatory variables on the average explained
variable and the variance of the residuals. We usualy ignore
the latter, but it is part of the model, and Stata needs to
store it somewhere when you call -suest-. The numbers in the
panel labeled hc1_mean and hc0_mean are the regular regression
coeficients: the effects on the mean dependent variable, the
numbers in the panels hc1_lnvar and hc0_lnvar are the log of
the variances of the residuals.

> 2. The test is significant (Prob > chi2 =
> 0.0001), so the H0 [hc0_mean]X=[hc1_mean]X is refuted. I
> interpret this as meaning that the three variables I add in
> the second model (seet seehc dscrfrq) in part account for
> the effect of X. Is this a correct interpretation?

Yes, you can do even more with that, as you can read up in
<http://www.ats.ucla.edu/stat/stata/faq/pathreg.htm>.

Hope this helps,
Maarten

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

http://www.maartenbuis.nl
--------------------------

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/