standard errors for sum of ols coefficient

[Mislav Sagovac]

I want o estimate the standard errors for sum of OLS coefficient. What would be the formula?

Let's say I have a model

reg y x1 x2 x1*x2 x3

and I want to estimate the standard errors of (b2 + b3), where b1 is coefficient on x2 and b2 on x1*x2.

How to proceed?

[Clyde Schechter]

Well, first let's get the model code right: you can't have x1*x2 as a term in the -reg- command.

reg y x1##x2 x3

(If x1 or x2 is continuous, then a c. prefix is needed in front of it.)

The sum of the coefficients of x2 and the x1#x2 interaction is then

lincom 1.x2 + 1.x1#1.x2

(I'm assuming x1 and x2 are 0/1 variables If they are coded differently, then the -lincom- statement needs corresponding modification.)

Moreover, if you are computing this sum because you are looking for the marginal effect of x2 conditional on x1 = 1 you can also get it from

margins, dydx(x2) at (x1 = 1)

[Richard Williams]

Also, I think using the coefl option can help so you get the coefficient names right, e.g.

sysuse auto, clear
reg price i.foreign weight i.foreign#c.weight, coefl
lincom _b[weight] + _b[1.foreign#c.weight]

[Mislav Sagovac]

Clyde, Richard

Thank you very much for replies!