st: Re: Stata tip 87: Interpretation of interactions in non-linear models

[Maarten Buis <maartenlbuis@gmail.com>]

On Wed, May 18, 2011 at 7:31 PM,  andre ebner wrote:
> Dear Maarten,
>
> thanks, I did not realize that I was fixing the individual effects at the
> average of the random effects of the individuals belonging to one of the
> four groups defined by the interaction term of two dummies (that is what
> you mean, not?), before computing marginal effects and then averaging
> them. Could you as a last thing be so kind to tell me, from what part of
> the STATA comand
>
> margins, over (inc_shock loan_const) expression(exp(xb)))
>
> I can see this? I was the opinion that RE would be not fixed at an
> average. This is of course a problem, as one should allow RE to vary
> across individuals within the four groups when computing marginal effects.

You can see that by noting that xb only contains the linear predictor
excluding the random effects. If you exclude something than that is
equivalent to fixing it at 0. Random effects are parameterised in such
a way in Stata that 0 equals the mean. Schematically, what you would
want is -expression( exp( xb + blup ) ), where blup is the best linear
unbiased prediction of the random effects. Unfortunately, you'd have
to derive and implement that for yourself...

> To be honest I have difficulties with your expression of „group constant“,
> as I intend RE as individual, time constant, unobserved effects that are
> not correlated with other covariates.

Ok, so you have observations nested in individuals. In that case the
"group" is the individual.

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/