Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: Testing interaction terms / mutually exclusive variables


From   Verena Dill <dill@uni-trier.de>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Testing interaction terms / mutually exclusive variables
Date   Thu, 20 Jun 2013 17:53:46 +0200

thank you so much again for your answer. After reading your reply and re-reading my emails I concluded that it may not have been the best way to explain my problem. I will try a better strategy:

I have two dummy variables that are mutually exclusive. For example, a firm belongs to the service sector, to the manufacturing sector or other sectors. I am running a regression and want to test whether or not the estimated coefficients are signficantly different from zero. One approach is to use the post-estimation command "test". Howover, "test" does not take into account that the two dummy variables are mutually exclusive. Is there another procedure that does this?


Thank you very much in advance
Verena

Am 19.06.2013 16:39, schrieb David Hoaglin:
Verena,

Not exactly.  I would not have included var00, which has no data to support it.

Your data do not allow you to test whether the effects of var1 and
var2 are additive.  You can estimate the effect of var1 when var2==1
and the effect of var2 when var1==1, but you cannot estimate the
"marginal" effect of either var1 or var2 (averaging over the levels of
the other variable).  The two comparisons that you can make are var11
- var10 and var11 - var01.

I assume that having 0 observations out of 700 with var1==0&  var2==0
reflects some structural aspect of your data, rather than a chance
occurrence.  That structure allows you to estimate the effects of
var01, var10, and var11 (perhaps with one category as the reference
category), but it does not allow you to estimate the interaction of
var1 and var2 (in the usual sense).  Perhaps the structure provides a
framework for restating the question and interpreting the three
effects that you can estimate.

David Hoaglin

On Wed, Jun 19, 2013 at 9:27 AM, Verena Dill<dill@uni-trier.de>  wrote:
What you wrote is exactly what I did in my regression (output below, just
for the matter of illustration I included the four categories;  var11:
var1==1&  var2==1, var10: var1==1&  var2==0, var01: var1==0&  var2==1,
var00: var1==0&  var2==0; because of the below mentioned structure of the
data two of the categories are omitted).
------------------------------------------------------------------------------
      partner |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+----------------------------------------------------------------
        var11 |   .7478253   .3528458     2.12   0.034     .0562602
1.43939
        var10 |   .9636673   .3315029     2.91   0.004     .3139335
1.613401
        var01 |          0  (omitted)
        var00 |          0  (omitted)
        _cons |    -.63364   .3095231    -2.05   0.041    -1.240294
-.0269858

But my question is: how can I test if the coefficients _b[var11] and
_b[var10] are equal taking the "interaction"-nature of the variables into
account? Using only "test _b[var11]= _b[var10 ]"  does not account for that.
Is there any other procedure I could use here (maybe similar to contrast)?
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index