RE: st: restricting margins to significant variables only

[Nick Cox <n.j.cox@durham.ac.uk>]

Either way, one over-arching point is that modifications to a model make nonsense of the classical idea of hypothesis testing and of what P-values mean. 

Lots of us, and I include myself, view modelling as a highly iterative process guided by data, and indeed what else should it be? 

Some of us, ditto, sit loose to P-values and don't take them very seriously. But those who want to select their models in the light of the data  and then write them up keeping exactly the same view of P-values as if the model published is exactly the model first thought of are playing a rather strange game. 

Nick 
n.j.cox@durham.ac.uk 

Richard Williams

At 10:34 AM 3/18/2011, Maarten buis wrote:
>--- On Fri, 18/3/11, Richard Williams wrote:
> > Having said that, you can always have margins fix some
> > variable values at 0. That might make good substantive sense
> > in some cases (e.g. you want to make predictions for a group
> > coded 0) but I don't see why you would want to do so in this
> > case. Indeed, it would be silly to fix a variable at 0 when
> > 0 is not a possible or likely value for the variable.
>
>The original question was "Is it advisable to get margins to
>base its calculations on only those variables which are
>significant?" In that case you do not want to set the variable
>to zero but its effect, i.e. a form of stepwise selection,
>which I told him was a bad idea.

If i understand the original query, what was being proposed is even 
worse than stepwise. It sounded like he wanted to leave the 
insignificant variables in the model but treat their effects as 
though they were zero. Either way, it sounds like a choice between 
bad and worse.

*
*   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/