Lana Carneiro Almeida wrote:
> I want to test for interaction in linear regression, between two
> categorical variables, and I used the following comand in STATA:
>
> xi: regress y i.var1 i.var2 var3 i.var1*i.var2
>
> so the model gives me more than one beta/p-values of interaction, but
> ommit one category, and I would like to know the p-value of all beta.
> How can I get it?
Any single first-order variable can only be interpreted when the other
first-order variables are zero, which may or may not be meaningful to
you, and in any case are only included so that you obtain the
best-possible estimates on the coefficients of your interaction terms
(otherwise, you are assuming that these terms equal zero: almost
always an incorrect assumption). If you are looking to fit two
interaction terms (e.g., -d1*d2- and -d1*d3-), then -d2*d3-, much like
-d1-, is redundant, as this will be defined by the other two
interaction terms.
> Another question is about interpretation of interaction: if only one
> beta is statisticaly significant, can I interpretate that there are
> interaction between the two variables?
As said before, the interaction is the _main_ action. If the
coefficient on this term is insignificant, then you have insufficient
evidence that, say, the effect of -d1- on -y- varies by values of
-d2-. Unless you have other interactions to test, you would then
revert to your previous model _without_ the interaction term.
--
Clive Nicholas
[Please DO NOT mail me personally here, but at
<[email protected]>. Please respond to contributions I make in
a list thread here. Thanks!]
"My colleagues in the social sciences talk a great deal about
methodology. I prefer to call it style." -- Freeman J. Dyson.
*
* 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/
