Add non-positive constraints to regression

[Megan Sheahan]

I am trying to constrain several regression coefficients to be non-positive (generally negative). How can I do this with either the nl or constraint commands? I find lots of advice for how to specify non-negative constraints, but not the opposite.

For reference, here is what I have tried so far, which returns an error:

constraint 1 inj1 < 0
cnsreg hrql inj1, constraint(1)

[Andrew Musau]

Are you running a linear regression? If so, check out

help cnsreg

although zero would be okay

But isn't constraining the coefficient to zero the same as omitting the variable from the regression?

[Andrew Musau]

I did not notice your edit. You should always continue the thread if you are replying to someone.

constraint 1 inj1 < 0

This is not valid. Being that you say that you know how to constrain the coefficient of a variable to non-negative values, to constrain the coefficient of the same variable to be non-positive values, constrain the coefficient of (-1*variable) to non-negative values.

[Nick Cox]

Constraints can make perfect sense, but not to me on this evidence. You have a regression as simple as it could be, one response and one predictor and you want to force the coefficient to be negative, regardless? Something needs explaining here: is it a problem with outliers, marginal distributions, functional form, or something else? Why not show us

scatter hrql inj1

and ideally explain what is going on and show example data too?

[Megan Sheahan]

Thank you for your helpful replies! Apologies for not providing more detail up front. The sample regression above is just a simplified version of what I am trying to estimate.

I believe I solved my issue using the nl command. For reference if others have the same challenge, here is what I did (again, simplified):

nl (hrql = -exp({lna1})*inj1 + {c}), nolog
nlcom (a1: exp(_b[lna1:_cons]))

I believe that simply adding a negative sign in front of the output of the second line of code yields the constrained coefficients I'm looking for. It's possible, however, that I'm misinterpreting this output.