Re: st: Nonlinear least squares restrictions

["G. Anderson" <ga274@cam.ac.uk>]

Hi Maarten,

Thanks a lot for your suggestions- they are proving very useful. I am 
currently using the propcnsreg function using the "lcons" identification 
option.

My variables included in lambda are annual dummies. Since I am looking at 
2000-2005, I therefore have included a dummy for 2001,2002,2003,2004,2005- 
with no dummy for 2000 to avoid the collinearity. Does this sound sensible?

Am I right in thinking that with the "lcons" identification option, I can 
interpret the constrained coefficients the indirect effects of my 
constrained variables on the dependent variable, through the latent 
variable in the year 2000 (since as noted above 2000 is omitted as an 
annual dummy)?

Ideally I'd like to be able to interpret the constrained coefficients as 
the average effect over the years 2000-2003, rather than just in the year 
2000- although at the moment I'm struggling to do this in this program.

Thanks for all your help,

Gareth




On Jun 5 2013, Maarten Buis wrote:

> On Wed, Jun 5, 2013 at 4:34 PM, G. Anderson wrote:
> > Is the right setup using  propcnsreg:
> >
> > yield dum00 dum01 dum02 dum03 dum04 dum05 ,constrained(gdp unemployment
> > inflation) lambda(dum00 dum01 dum02 dum03 dum04 dum05)
> >
> > Where dum are annual dummies. In particular- do I include the annual 
> > dummies as both independent variables and variables in lambda?
>
> Yes, I did it this way to allow for a more flexible specification of
> time as the main effect than in the interaction effect (-lambda()-).
> So, you could have time as dummies for the main effect, but constrain
> time to be linear in the interaction effect. The other way around is
> technically possible, but I would not recommend it.
>
> > Is it also possible to impose a normalization so
> > that on average the coefficients on the dummies is equal to 1?
>
> Just create your variables using an effect or sigma or deviation
> coding scheme (different names for the same thing).
>
> Hope this helps,
> Maarten
>
> ---------------------------------
> Maarten L. Buis
> WZB
> Reichpietschufer 50
> 10785 Berlin
> Germany
>
> http://www.maartenbuis.nl
> ---------------------------------
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/