Estimating a beta coefficient only for a subset of data

[Federico Stravo]

Hi all. I am trying to estimate a regression that looks like this:

use "https://www.stata-press.com/data/r17/nlswork.dta"
reg ln_wage tenure hours wks_work ttl_exp

Let's say that I want to estimate the beta coefficients for each variable on the right-hand side of the equation for the entire sample except for tenure, for which I want Stata to compute the coefficient only on the subset of people that are married, that is, married==1 in this dataset. In other words, I want the coefficients for hours, wks_work, and ttl_exp to be estimated on the entire sample, while the coefficient for tenure is estimated only on the subsample of people that are married.

How do I go about it? I do not know what to do. Simply doing c.tenure#i.married does not work of course.

Any help is appreciated. Thank you!

[Tiago Pereira]

Not clear what you want, but you can use
reg ln_wage tenure hours wks_work ttl_exp if msp==1

[Federico Stravo]

Hi Tiago. There are about 28,000 observations in the sample, 17,000 of which are married people. If I implement your line of code, the entire regression is estimated only for the 17,000 people who are married but that is not what I want. I want the beta coefficients for hours, wks_work, and ttl_exp to be estimated on all the 28,000 people in the sample, while at the same time I want the beta coefficient for tenure to be estimated only on the 17,000 people that are married.

Basically, I need my regression to look something like this:

ln_wage = beta₁ * tenure (if msp==1) + beta₂ * hours + beta₃ * wks_work + beta₄ * ttl_exp

I hope this is clear, sorry if not.

[Tiago Pereira]

I see. So this is not a Stata related issue.

[Carlo Lazzaro]

Federico:
you can't, as Stata internal reasoning is based on matrices, that cannot be shorthened for a set of variables only.
Therefore:
1) you can run:

use "https://www.stata-press.com/data/r17/nlswork.dta"
reg ln_wage tenure hours wks_work ttl_exp

considering all the observations included in the dataset (exception made for those with missing values in any variable, that Stata omit by default via casewise deletion);
2) the other approach is to limit your regression on a given subsample of the original dataset via an -if- clause:

use "https://www.stata-press.com/data/r17/nlswork.dta"
reg ln_wage tenure hours wks_work ttl_exp if msp==1

(the previous comment about casewise deletion still holds).