Re: st: Fwd: Comparing marginal effects of two subsamples

[Maarten Buis <maartenlbuis@gmail.com>]

On Mon, Oct 24, 2011 at 5:49 PM, Jianhong Chen wrote:
> Thank you very much. Just to confirm,  there is no need to compare
> marginal effect by splitting sample. As long as I provide the
> incidence-rate ratio for interaction term and it is sigificant, I can
> say that the hypothesized interaction is supported. Right?

I cannot confirm that statement, in the sense that I will not give a
"recipe" for how to get to a conclusion(*). The point of both
incidence rate ratios and marginal effects to get interpretable
numbers. At the beginning of this thread I gave an example of how to
do so <http://www.stata.com/statalist/archive/2011-10/msg00938.html>.
So the first step is to interpret your coefficients. If you want to do
testing, you need to figure out what the null hypothesis is that you
want to test, i.e. you need to translate a statement like "the
interaction term is significant" to either incidence rate ratios or
marginal effects. After that, performing the test is trivially easy.

-- Maarten

(*) I know I am being mean by not giving you an easy answer (which
exists), but when dealing with interactions there are just so many
things that can go horribly wrong that following some "recipe" will
only lead to more errors. The only option to do such things right is
to really understand the numbers you are producing. It is not hard (I
gave you an example), and that bit of effort will really pay off.

--------------------------
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/