# st: constraints

 From "Hannah Kewsberry" To statalist@hsphsun2.harvard.edu Subject st: constraints Date Mon, 16 Sep 2002 12:47:27 +0000

Thank you so much for reply. BTW, Sorry I didn't make myself clear in the previous question that I was actually looking for commands to constraint these estimates in multinomial situations.

So, again using the hypothetical example:
Dependant variable: PAR
Independent variables: e1, e2, e3, race, earn1, earn2, earn3.
If I want to constraint estimates of my earn variables only, ie. earn1=earn2=earn3.

What's the appropriate command I shall be using in the multinomial situation? Thanks again for all suggestions in advance.

Regards,

Hannah

--On Monday, September 16, 2002 2:33 -0400 Hannah wrote:

I have rephrased my question and hope to get more suggestions.Let say my dependant variable: PAR
Independent variables: e1, e2, e3, race, status, earn1, earn2,
earn3. If I want to constraint estimates of my earn variables only, ie. earn1=earn2=earn3.

What's the appropriate command I shall be using? Thanks for all

----
Presuming that you are using linear regression,

g totearn = earn1 + earn2 + earn3
reg par e1 e2 e3 race status totearn

does this by substituting the constraint into the model.
Alternatively,

cons def 1 earn1=earn2
cons def 2 earn1=earn3
cnsreg par e1 e2 e3 race status earn*, c(1 2)

Since you are applying two constraints to the coefficient vector
(i.e. if you ran the unconstrained regression and tested
earn1=earn2=earn3, you would be doing an F with 2 and ? d.f.), you
must define two constraints.

_________________________________________________________________