Simple constraint on logistic regression?

[AmberA]

Hi,

I wanted to run the logistic regression

Logit k x y z

Under the constraint c1+c2+c3=1 where c1,c2,c3 are the coefficients on x,y,z. This seems easy to do for least squares but I can’t see a way to do it for logit. Any help would be greatly appreciated.

Thanks!
Amber

[EViews Gareth]

Same as with Least Squares:

Code: Select all

k=c(1)*x+c(2)*y+(1-c(1)-c(2))*z

[AmberA]

That worked! Thanks!

Is there a way to figure out the Standard Error for the constrained variable (c(3) in this case)?

Amber

[startz]

sqrt(var(c(1))+var(c(2))+2cov(c(1),c(2))) I think.

[AmberA]

Yes, thanks, but where do you find the covariance matrix in the output?

It appears that another way to get the standard error is to run the regression several times while putting the constrained coefficient (1-c(1)-c(2)) on a different explanatory variable each time.

Thanks again!
Amber

[startz]

View/Covariance matrix

[AmberA]

Thanks!