OLS regression with factorial variable coefficients summing to zero

[Christian Proebsting]

Hello,

I'm estimating a simple ols regression on several sets of factor variables:

reg y i.dest i.orig i.time

I'd like to constrain the sum of coefficients for each factor variable to zero. How can I achieve this? I'd also like to know the standard errors in that case.

Thank you,
Christian

[Maarten Buis]

A constraint like that typically happens when you want to parametrize the effect of a categorical variable as deviations from the grand mean. If that is the case, take a look at help contrasts

[Christian Proebsting]

Thanks, Maarten. I had seen that you had posted that advice before and I had looked at this example
http://www.stata.com/features/overview/contrasts/
But I'm not sure how to apply it to my examples. I don't want to do adjacent contrasts, but a contrast relative to the 'average' so that all coefficients add up to 0.

[Maarten Buis]

That website is not meant to help you do an analysis. It is meant to help you decide whether or not Stata is worth buying for you.

The help files and manuals are for helping you do research using Stata. So that is the information you should consult.

Here is a brief example, but read to documentation to decide if you really want the g. or the gw. prefix. The g. prefix literally answers your question, but I suspect that it is gw. that you actually want.

// open example data
sysuse auto, clear

// estimate the model
reg price i.rep78

// get the comparison with the mean
// assuming a balanced design
contrast g.rep78

// check if it adds up to 0:
mata : sum(st_matrix("r(b)"))
// 0.000000000000062 is close enough to 0