Re: st: MIME-Version: 1.0

[maarten buis <maartenlbuis@googlemail.com>]

On Thu, May 5, 2011 at 1:05 PM, Fabio Zona wrote:
> do you agree that - for an interaction effect to be significant - not only
> the interaction term needs to be significant, but also the wald test of
> the difference between the interaction term versus each of the two
> main effects need to be significant?

No I do not agree with that statement. The interaction term alone is
enough. Moreover, what would the null hypothesis in such a test
mean? The interaction effect tells you how much the effect of x1
changes when x2 changes, the main effect of x1 tells you how
much the expected value of y changes when x1 changes when
x2 == 0 (and vice versa for the main effect of x2). What would be
the substantive interpretation of a null hypothesis that sets an
interaction term and an main effect equal to one another? Moreover,
the size and significance of the main effects trivially depend on how
you centered (or whether you centered) the other variable without
changing the fit of your model to the data, so the test you propose
is pretty meaningless.

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/