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

[Jianhong Chen <jianhongchen1985@gmail.com>]

Hi, Maarten

Thank you very much for your response. I read your article on Stata
Journal and I think you are completely right.  Just to make sure that
I understand right, are the exponentiated coefficients you mentioned
the same coefficients which I can get after runing negative binomial
model?

Also, since the reviewers require that we should compare marginal
effect, we have no chance to avoid that and we plan to use the
marginal effects as supplemental analysis, not main analysis. Do you
have any idea to do t-test of marginal effects of two subsample?

Thanks

Best,

Jianhong

On Fri, Oct 21, 2011 at 3:35 AM, Maarten Buis <maartenlbuis@gmail.com> wrote:
> On Thu, Oct 20, 2011 at 10:16 PM, Jianhong Chen wrote:
>> I am conducting two-way interaction with negative binomial model. The
>> reviewers asked us to do marginal effects because of non-linear model.
>> So, I splitted the sample according to the mean level of the
>> moderator.
>
> I would not use marginal effects in this case. The exponentiated
> coefficients are incidence rate ratios, and are as easy to interpret
> as marginal effects but without any of the disadvantages related to
> marginal effects.
>
> Consider the example below:
>
> The dependent variable (art) is the number of articles published in
> last three years of PhD (I believe these are biologists, but I am not
> certain).Women in an average status school (z_phd = 0) produce (1-
> .80)*100%= -20% less articles than men. This effect of being a women
> increases, i.e. becomes less negative, when the schools has higher
> status. For every standard deviation increase in status of the school
> the effect of women increases by a factor 1.14, i.e.
> (1-1.14)*100%=14%. In other words the effect of being a women in a
> school with 1 standard deviation more status than average is
> 1.1440*.7996= .91. Which means that women is such a school produce
> only 9% less articles than men.
>
> *--------------------- begin example ---------------------
> use http://www.stata-press.com/data/lf2/couart2, clear
> gen byte baseline = 1
> sum phd if !missing(art,fem,ment,kid5,mar)
> gen z_phd = (phd - r(mean))/r(sd)
> nbreg art i.fem##c.z_phd c.ment##c.ment kid5 mar baseline, irr nocons
> *---------------------- end example -----------------------
> (For more on examples I sent to the Statalist see:
> http://www.maartenbuis.nl/example_faq )
>
> For a more general discussion on how to interpret interactions in such
> non-linear models see:
> M.L. Buis (2010) "Stata tip 87: Interpretation of interactions in
> non-linear models", The Stata Journal, 10(2), pp. 305-308.
>
> 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/
>

*
*   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/