--- Chiara Mussida <[email protected]> wrote:
> I estimated a mlogit model by generating yearly dummy variables with
> the command xi. My model, indeed, refers to a ten years time span.
> And i would be pleased to know if the model could be defined as
> "homogeneous" over time. Is a simple testparm (with the equal option)
> between each couple of yearly dummies a proper tool for my purpose?
> i.e. if I reject the null of equality can I argue that my model is
> not homogeneous or are there alternative tests??
Tests always test a very specific hypothesis. Unless the meaning of a
term like "homogenous over time" is completely unambiguous for you and
your intended audience, I would avoid these terms and instead
explicitly state your hypothesis. In this case your hypothesis appears
to be that the probabilities of ending up in each category have not
changed over time. However, by entering time as a series of dummies you
have used a lot of decrease of freedom and thus reduced the power of
your test. More powerful, but also more restrictive, tests would test
the hypothesis of no linear effect of time, or no effect of time
entered as a (restricted cubic) spline. Also, you have not tested
whether the effects of other explanatory variables changed over time,
i.e. no interactions between these variables and time. The absense of
these interactions is something that could also be meant by "homogenous
over time". So in short, you can avoid all problems by just explicitly
stating your hypotheses and avoid ambiguous names for these hypotheses.
-- Maarten
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
*
* 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/
