Re: st: Chow test quadratic effect

[David Hoaglin <dchoaglin@gmail.com>]

Natalia,

Would you please explain why you do not include male==1 and male==0 in
a single model, perhaps with suitable interaction terms?

Does the nonlinearity actually have a quadratic form?  If the relation
between insure and age is not linear, addition of agesq should not be
automatic.

Age and agesq are both part of the relation between insure and age, so
you need to consider them together.

David Hoaglin

On Sat, Mar 30, 2013 at 5:18 PM, natalia malancu
<natalia.malancu@gmail.com> wrote:
> Hi all,
>
> I am estimating a mlogit model on two different groups. I then want to
> test if there is a structural change. My
> question is: how do I do Chow Test in stata if I have a quadratic term?
>
> example:
>
> webuse sysdsn4
> gen agesq=age*age
>
>  mlogit insure age agesq if male==1, base(1)
>      estimates store m1
> qui mlogit insure age agesq if male==0, base(1)
>      estimates store m2
>   suest m1 m2
>
> for the main effect I use:
>  test [m1_Prepaid]age =[m2_Prepaid]age
> test[m1_Uninsure]age=[m2_Uninsure]age
>
> If I am to consider age and agesq at the same time, then how do I do
> so? Or should I just test them separately?
>
> Thanks for helping,
> Natalia
*
*   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/